JP5212204B2 - Response generating apparatus and program - Google Patents

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, the user utterance is analyzed to extract the predicate and case element, the extracted predicate or the response sentence candidate to confirm the case element, the response sentence candidate to query the omitted case element, the reason why the predicate was performed, There has been proposed a response generation device that generates a response sentence candidate that asks a place and a response sentence candidate that confirms the relationship between predicates, and outputs a response sentence candidate that is randomly selected from the generated response sentence candidates. For example, see Patent Document 1).

  Also, a word that is an adjective, adverb, or noun is used as a concept word, and a concept attribute database that defines the attribute information of the concept word based on the modification relation and thesaurus is constructed, and the user utterance is analyzed to extract the concept word Then, an interactive device has been proposed in which attribute information registered corresponding to the extracted concept word is searched from the concept attribute database, and a response sentence corresponding to the attribute information is generated and output (for example, a patent) Reference 2).

JP 2007-206888 A JP 2009-3811 A

  However, in the response generation device of Patent Literature 1 and the dialogue device of Patent Literature 2, it corresponds to the part of speech and attribute information of the extracted word regardless of whether it is appropriate or inappropriate to ask the user about the content of the response sentence. A response sentence determined as described above is generated. Therefore, for example, there is a problem that an unnatural response such as “Why is it big?” May be generated in response to a user utterance “That stone is very big”.

  The present invention has been made to solve the above-described problem, and an object of the present invention is to provide a response generation apparatus and program capable of preventing natural generation of an inappropriate response sentence and performing natural conversation. To do.

  In order to achieve the above object, a response generation device according to a first aspect of the present invention includes a situation extraction unit that extracts a word representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence, and the situation A word that represents the situation extracted by the extraction means is determined whether it is registered in the emotion questionable judgment dictionary as an appropriate word or an inappropriate word for which an emotion is asked, and an appropriate word to ask an emotion If it is, a response sentence candidate prepared in advance for asking feelings for the word representing the extracted situation, or a response sentence in which the word representing the extracted situation is added to a response sentence format prepared in advance And an emotional question response generating means for generating candidates.

  According to a second aspect of the present invention, there is provided a response generation program comprising: a situation extraction unit that extracts a word representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence; and the situation extraction unit. When the word representing the extracted situation is a word that is appropriate for asking for emotion after judging whether it is registered in the emotion questionable judgment dictionary as an appropriate word or an inappropriate word for asking for emotion Generates a response sentence candidate prepared in advance for asking feelings about the word representing the extracted situation, or a response sentence candidate in which the word representing the extracted situation is added to a prepared response sentence format It is a program for functioning as an emotional question response generating means.

  According to the response generation device according to the first aspect of the invention and the response generation program according to the second aspect of the invention, the situation extraction unit extracts a word representing the situation from the input sentence based on the analysis result obtained by analyzing the structure of the input sentence. Extract. A situation refers to a person's behavior or an event or event that a person experiences. Then, the emotion question response generating means determines whether or not the word representing the situation extracted by the situation extracting means is registered in the emotion questionable judgment dictionary as an appropriate word or an inappropriate word for asking an emotion. If it is an appropriate word to ask for emotion, a response sentence candidate prepared in advance for inquiring feelings for the word representing the extracted situation, or a situation extracted in a response sentence format prepared in advance A response sentence candidate to which a word representing is added is generated.

  In this way, since it is determined whether or not it is appropriate to ask the emotion based on the extracted word, the response sentence candidate is generated. Dialogue can be conducted.

  Further, in the response generation device according to the first invention and the response generation program according to the second invention, the emotion questionable judgment dictionary is asked for a word without a classification name related to emotion in the thesaurus. Can be registered as an appropriate word, or a word with a classification name related to emotion in the thesaurus can be registered as an inappropriate word for asking for emotion. By using such an emotion questionable determination dictionary, it is possible to easily determine whether it is appropriate or inappropriate to ask an emotion.

  Further, the response generation device according to the third invention is extracted by the situation extracting means for extracting a clause representing the situation from the input sentence based on the analysis result obtained by analyzing the structure of the input sentence, and the situation extracting means. If the section that describes the situation is registered in the frequency questionable decision dictionary as an appropriate section or an inappropriate section to ask about the frequency, and if it is an appropriate section to ask about the frequency, A frequency question for generating a response sentence candidate prepared in advance for asking the frequency of the clause representing the extracted situation or a response sentence candidate in which a clause representing the extracted situation is added to a prepared response sentence format Response generation means.

  According to a fourth aspect of the present invention, there is provided a response generation program comprising: a situation extraction unit that extracts a clause representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence; and the situation extraction unit. When the section that indicates the extracted situation is a section that determines whether or not the frequency is asked in the frequency questionable dictionary as an appropriate section or an inappropriate section that asks the frequency, and asks the frequency Generates a response sentence candidate prepared in advance for inquiring the frequency of the clause representing the extracted situation, or a response sentence candidate obtained by adding the clause representing the extracted situation to a prepared response sentence format It is a program for functioning as a frequency question response generation means.

  According to the response generation device according to the third aspect of the invention and the response generation program according to the fourth aspect of the invention, the situation extraction means adds the clause representing the situation to the input sentence based on the analysis result obtained by analyzing the structure of the input sentence. Whether or not a clause representing the situation extracted by the situation question extracting means is registered in the frequency questionable judgment dictionary as an appropriate clause or an inappropriate clause to ask for the frequency. If it is an appropriate clause to determine the frequency and ask the frequency, it is extracted to a response sentence candidate prepared in advance or a response sentence format prepared in advance to ask the frequency for the section representing the extracted situation. A response sentence candidate with a clause indicating the situation is generated.

  In this way, since it is determined whether or not it is appropriate to ask the frequency based on the section representing the extracted situation, a response sentence candidate is generated. Dialogue can be conducted.

  Further, in the response generation device according to the third invention and the response generation program according to the fourth invention, the frequency questionable determination dictionary has a high frequency of experience, a low frequency of experience, and a specific time. A situation that can only happen, that always accompanies another situation, that cannot occur alone, that always occurs for a given purpose and that cannot be completed alone, that means start or end, and a situation that means existence Either register a section that represents a situation that falls into any of the sections where it is inappropriate to ask about the frequency, or a section that represents a situation other than the situation that falls into any of the above sections that is appropriate to ask about the frequency. Can be registered as By using such a frequency question impossible determination dictionary, it is possible to easily determine whether it is appropriate or inappropriate to ask the frequency.

  Further, the response generation device according to the fifth invention is extracted by the situation extracting means for extracting a clause representing the situation from the input sentence based on the analysis result obtained by analyzing the structure of the input sentence, and the situation extracting means. It is appropriate to ask whether the clause that represents the situation is registered in the parallel object questionable decision dictionary as an appropriate clause or an inappropriate clause to ask for the parallel object, and ask the parallel object. In the case of a clause, the response sentence candidate prepared in advance for asking a parallel object for the clause representing the extracted situation, or the clause representing the situation extracted in a response sentence format prepared in advance. And a parallel object question answer generating means for generating a response sentence candidate to which is added.

  According to a sixth aspect of the present invention, there is provided a response generation program comprising: a situation extraction unit that extracts a clause representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence; and the situation extraction unit. Judgment is made as to whether or not the clause representing the extracted situation is registered in the parallel object questionable decision dictionary as an appropriate or inappropriate clause to ask for the parallel object, and the parallel object is asked Is an appropriate clause, a response sentence candidate prepared in advance for asking a parallel object for the clause representing the extracted situation, or the situation extracted in a response sentence format prepared in advance Is a program for functioning as a parallel object question answer generation means for generating a response sentence candidate to which a clause representing is added.

  According to the response generation device according to the fifth aspect of the invention and the response generation program according to the sixth aspect of the invention, the situation extraction means extracts a clause representing the situation from the input sentence based on the analysis result obtained by analyzing the structure of the input sentence. The parallel object question answer generation means extracts the clause representing the situation extracted by the situation extraction means to the parallel object questionable dictionary as an appropriate clause or an inappropriate clause to ask for the parallel object. If it is an appropriate clause to determine whether or not it is registered and ask for parallel objects, prepared response sentence candidates for asking parallel objects for the clause representing the extracted situation Alternatively, a response sentence candidate is generated by adding a clause indicating the extracted situation to a response sentence format prepared in advance.

  In this way, since it is determined whether or not it is appropriate to ask for parallel objects based on the clauses representing the extracted situations, the generation of response sentence candidates is prevented, thereby preventing the generation of inappropriate response sentences. Can have a natural dialogue.

  Further, in the response generation device according to the fifth invention and the response generation program according to the sixth invention, the parallel object questionable determination dictionary uses a plurality of parallel nouns from a plurality of input sentences as objects. Extracting a clause representing a situation, dividing each of the plurality of parallel nouns into clauses having the target word, and paralleling the divided clauses each having a frequency of appearance of a predetermined value or more Asking a word can be registered as an appropriate clause, or the divided clauses whose appearance frequency is smaller than a predetermined value can be registered as an inappropriate clause for asking parallel objects. By using such a parallel object questionable determination dictionary, it is possible to easily determine whether it is appropriate or inappropriate to ask for parallel objects.

  Moreover, it is also possible to configure a combination of at least two response generation devices selected from the response generation device according to the first invention, the response generation device according to the third invention, and the response generation device according to the fifth invention. Thereby, the variation of a response sentence candidate increases and it can communicate flexibly.

  In addition, the response generation device according to the first invention, the response generation device according to the third invention, and the response generation device according to the fifth invention, the part of speech is a noun based on the analysis result obtained by analyzing the structure of the input sentence. A noun extraction means for extracting the word from the input sentence, and a noun detail questionable judgment dictionary as a word that is appropriate for asking for details or an inappropriate word for a word whose part of speech is extracted by the noun extraction means. If it is appropriate to determine whether it is registered and ask for details, the extracted part of speech is a response sentence candidate prepared in advance for asking details about a noun word, or prepared in advance. And a noun detail question answer generation unit for generating a response sentence candidate in which a word representing the extracted situation is added to the response sentence format.

  Further, based on the analysis result obtained by analyzing the structure of the input sentence, an adjective expression extracting means for extracting a word or clause of the adjective expression from the input sentence, and a word or clause of the adjective expression extracted by the adjective expression extracting means , It is appropriate to ask the reason, or whether it is registered in the adjective reasoning questionable judgment dictionary as an inappropriate word or clause, and it is appropriate to ask the reason In the case, the previously prepared response sentence candidate for inquiring the reason for the extracted adjective expression word or clause, or the extracted adjective expression word or clause is added to the prepared response sentence format An adjective expression question answer generation means for generating a response sentence candidate can also be included.

  In addition, in the noun detail questionable determination dictionary, a word that is suitable for asking for details is a noun word that has a part of speech that is a noun and a questionable word and whose frequency of appearance on the web is a predetermined value or more. Or the part of speech is a noun word that has a frequency of appearance on the web of a combination of a noun word and a question word that is less than a predetermined value. can do.

  In addition, in the adjective expression reason question possible judgment dictionary, a word with a classification name related to a person's emotion and sense in the thesaurus is registered as an appropriate word to ask the reason, or in the thesaurus A word without a classification name relating to a person's emotion and sense can be registered as an inappropriate word for asking a reason.

  As described above, according to the response generation device and the program of the present invention, it is possible to prevent the generation of an inappropriate response sentence and to perform a natural conversation.

It is a block diagram which shows schematic structure of the response generation apparatus which concerns on this Embodiment. It is an image figure which shows an example of a situation noun dictionary. It is an image figure which shows an example of the example sentence of an emotional question. It is an image figure which shows an example of the response sentence example of a frequency question, and a response sentence example format. It is an image figure which shows an example of the response sentence example of a parallel object question, and a response sentence example format. It is an image figure which shows an example of the response sentence example of an adjective expression reason question, and a response sentence example format. It is a flowchart which shows the production | generation process of emotion questionable judgment DB. It is an image figure which shows an example of the result of a morphological analysis. It is an image figure which shows the content of a thesaurus. It is a figure which shows an example of the rule for determining whether an emotion can be asked. It is another image figure which shows the content of a thesaurus. It is an image figure which shows an example of emotion questionable judgment DB. It is a flowchart which shows the production | generation process of frequency question impossible judgment DB. It is an image figure which shows the other example of the result of a morphological analysis. It is a table showing the types of situations where it is inappropriate to ask the frequency. It is an image figure which shows an example of frequency question impossible judgment DB. It is a flowchart which shows the production | generation process of parallel object questionable judgment DB. It is an image figure which shows the other example of the result of a morphological analysis. It is a figure which shows an example of the language pattern which judges a parallel object. It is an example of the count result of the appearance frequency of the clause showing the situation of a parallel object. It is an image figure which shows an example of parallel object questionable judgment DB. It is an image figure which shows the other example of parallel object questionable judgment DB. It is a flowchart which shows the production | generation process of noun detailed question possibility judgment DB. It is an image figure which shows the other example of the result of a morphological analysis. It is a figure which shows an example of the co-occurrence pattern of a question word and a noun. It is an example of the count result of the appearance frequency of the co-occurrence pattern of an interrogative word and a noun. It is another example of the count result of the appearance frequency of the co-occurrence pattern of an interrogative word and a noun. It is an image figure which shows an example of noun detailed question possibility judgment DB. It is a figure which shows an example of the question word for asking the detail with respect to the word whose part of speech is a noun. It is a flowchart which shows the production | generation process of a description expression reason question possibility judgment DB. It is an image figure which shows the other example of the result of a morphological analysis. It is another image figure which shows the content of a thesaurus. It is a figure which shows an example of the rule for judging whether the reason of an adjective expression can be asked. It is another image figure which shows the content of a thesaurus. It is an image figure which shows an example of a description expression reason question possible judgment DB. It is a flowchart which shows the response production | generation process routine in the response production | generation apparatus of this Embodiment. It is an image figure which shows the other example of the result of a morphological analysis.

  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 shown in FIG. 1, the response generation apparatus 10 according to the present embodiment is connected to a microphone 12 that collects user utterances to generate an audio signal, a speaker 14 that outputs audio, and the microphone 12 and the speaker 14. And a computer 16 for executing 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. A situation determination unit 22 that determines whether or not a user utterance includes a situation (word) extraction unit 24 that extracts a word representing the situation from the user utterance, and a situation (word) extraction unit 24 The emotion question response generation unit 26 generates a question asking for an emotion as a response sentence candidate when the word extracted in (2) is registered in the emotion question possibility determination database (DB) 28. When a situation is included in the user utterance, the situation (section) extraction unit 30 that extracts a section representing the situation from the user utterance, and the section extracted by the situation (section) extraction unit 30 are stored in the frequency question impossible determination DB 34. When the question extracted by the frequency question response generation unit 32 and the situation (section) extraction unit 30 is registered in the parallel object question possibility determination DB 38 when the question that is not registered is generated as a response sentence candidate. Whether or not a word of noun is included in the user utterance based on the analysis result by the parallel object question answer generation unit 36 and the language analysis unit 20 that generate a question asking for a question that asks for the parallel object The noun discriminating unit 40 for discriminating the words, the noun extracting unit 42 for extracting the noun word from the user utterance when the user utterance includes the noun word, and the noun extracting unit 42 Based on the analysis result by the noun detail question response generation unit 44 and the language analysis unit 20 that generate a question asking for details when the word whose noun is a noun is registered in the noun detail question possibility determination DB 46. An adjective expression discriminating section 48 for discriminating whether or not an utterance includes a word or a clause of an adjective expression; when a user utterance includes a word or a clause of an adjective expression, the word or clause of the adjective expression from the user utterance When a word or clause of an adjective expression extracted by the adjective expression extracting unit 50 is registered in the adjective expression reason questionable possibility judgment DB 54, a question asking a reason is generated as a response sentence candidate. Adjective expression reason / question response generation unit 52, a response response generation unit 56 that generates a response message when a response message candidate is not generated, and a generated response message candidate Can be expressed by a configuration including an output unit 58 that converts one response sentence candidate selected from the above, or a response sentence of a conflict into an audio signal and outputs it from the speaker 14.

  The situation determination unit 22 determines whether or not a word representing the situation is included in the analysis result of the language analysis unit 20 for the user utterance. A situation is a person's behavior or an event or event that a person experiences, and it is assumed that some emotion is associated with the situation. Here, if the analysis result includes (i) "verb" and (ii) "situation noun + (" da "or" is ")" includes "situation" It is determined that “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 situation (noun) extraction unit 24 extracts a word representing the situation from the analysis result of the user utterance determined that the situation is included by the situation determination unit 22. In the case of (i) above, a word that is a “verb” is extracted, and in the case of (ii) above, a word that is a “situation noun” is extracted.

  The emotion question response generation unit 26 determines whether or not the word extracted by the situation (noun) extraction unit 24 is registered in the emotion question possibility determination DB 28 to be described later. Is generated as a response sentence candidate. Response sentence candidates are selected as basic question sentences, for example, “What do you think?” And “How do you feel?” In advance, and selected from these basic question sentences based on the analysis results of user utterances. The response sentence candidate 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 response sentence candidate is temporarily stored in a predetermined storage area.

  The situation (section) extraction unit 30 extracts a node representing the situation from the analysis result of the user utterance determined that the situation is included by the situation determination unit 22. In the case of (i) above, a clause including the corresponding “verb” is extracted, and in the case of (ii), a clause including the corresponding “situation noun” is extracted.

  The frequency question response generation unit 32 determines whether or not the clause extracted by the situation (section) extraction unit 30 is registered in the frequency question impossibility determination DB 34 to be described later. A question sentence that asks the frequency is generated as a response sentence candidate. For example, a response sentence example and a response sentence example format as shown in FIG. 4 are prepared in advance, and the response sentence candidates are randomly selected from the response sentence examples or generated using the response sentence example format. The response sentence example format can be, for example, a format such as "Do you do well?", And insert a clause indicating the situation extracted from the analysis result of the user utterance in the "~" part. Sentence candidates can be generated. The generated response sentence candidate is temporarily stored in a predetermined storage area.

  The parallel object question answer generation unit 36 determines whether or not the clause extracted by the situation (section) extraction unit 30 is registered in a parallel object questionable determination DB 38 to be described later. Then, a question sentence that asks the user for a parallel object is generated as a response sentence candidate. 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 candidate is selected at random from the response sentence example or generated using the response sentence example format. The response sentence example format can be, for example, a format such as “What else?”, And a “verb” of a clause representing a situation extracted from the analysis result of the user utterance in the “˜” part. A response sentence candidate can be generated by inserting a certain word. The generated response sentence candidate is temporarily stored in a predetermined storage area.

  The noun discriminating unit 40 discriminates whether or not the analysis result of the language analyzing unit 20 for the user utterance includes a noun word.

  The noun extraction unit 42 extracts a word whose part of speech is a noun from the analysis result of the user utterance which is determined that the noun determination unit 40 includes a word whose part of speech is a noun.

  The noun detail question response generation unit 44 determines whether or not the word extracted by the noun extraction unit 42 is registered in a noun detail question possibility determination DB 46 to be described later. A question sentence that asks for details about a word is generated as a response sentence candidate. Although details will be described later, the noun detail question possible judgment DB 46 has a configuration in which a word is registered for each question word (for example, “what” or “who”) used for asking details of a noun. ing. The response sentence candidate is a combination of the interrogative word corresponding to the word registered in the noun detail questionable judgment DB 46 and the extracted part of speech combined with the noun word, and "What is the extracted part of speech?" Can be generated as follows. The generated response sentence candidate is temporarily stored in a predetermined storage area.

  The adjective expression discriminating unit 48 discriminates whether or not the word or clause of the adjective expression is included in the analysis result of the language analysis unit 20 for the user utterance. “Adjective expression” means (i) a word whose part of speech is “adjective” and (ii) “a part of speech is a noun word + (” is ”or“ is ”)”. Is determined to contain the word or clause of “adjective expression”.

  The adjective expression extraction unit 50 extracts words or clauses of the adjective expression from the analysis result of the user utterance determined that the adjective expression discriminating section 48 includes the adjective expression. A word corresponding to the above (i) or a clause corresponding to the above (ii) is extracted.

  The adjective expression reason question response generation unit 52 determines whether the word or clause extracted by the adjective expression extraction unit 50 is registered in the adjective expression reason question possibility determination DB 54 described later. Then, a question sentence that asks the user for the reason for the adjective expression is generated as a response sentence candidate. For example, a response sentence example and a response sentence example format as shown in FIG. 6 are prepared in advance, and the response sentence candidates are randomly selected from the response sentence examples or generated using the response sentence example format. The response sentence example format can be, for example, a format such as “Why?”, And a word or clause of an adjective expression extracted from the analysis result of the user utterance is inserted in the “˜” part. Response sentence candidates can be generated. The generated response sentence candidate is temporarily stored in a predetermined storage area.

  Next, various questionable (impossible) determination DB generation processing will be described.

  First, the generation process of the emotion questionable determination DB 28 will be described with reference to FIG.

  In step 100, a sentence is extracted from a large corpus such as text on the web. Here, it is assumed that the sentence “I went to the zoo” has been extracted. Next, in step 102, morphological analysis is performed on one sentence extracted in step 100. Here, as shown in FIG. 8, the analysis results of “zoo (noun)” “ni (case particle)” “go (verb)” “ta (auxiliary verb)” are obtained. Next, in step 104, it is determined whether or not a situation is included in the analysis result. If the situation is included, the process proceeds to step 106, and if the situation is not included, the process proceeds to step 114. Since “go (verb)” is included here, it is determined that the situation is included, and the process proceeds to step 106.

  In step 106, words representing the situation are extracted. Here, “go (verb)” is extracted. Next, in step 108, it is determined whether or not the extracted word has been determined to be registered in the emotion questionable determination DB 28. If not determined, the process proceeds to step 110, and if determined, the process proceeds to step 114. Here, since it is still the first sentence, it is determined that it has not been discriminated, and the routine proceeds to step 110.

  In step 110, as shown in FIG. 9, the words are classified on the hierarchy according to the semantic similarity between the words in the plurality of hierarchies (here, five hierarchies), and the classification according to the semantic similarity is performed. It is determined whether or not the word extracted in step 106 is a registration target in the emotion questionable determination DB 28 based on a rule using a classification name on the thesaurus with names assigned to each classification of each hierarchy. If it is a registration target, the process proceeds to step 112. If it is not a registration target, the process proceeds to step 114. For example, as shown in FIG. 10, a rule using a classification name on the thesaurus can be defined so as to register a word without a classification name related to emotion in the fourth layer on the thesaurus. If the classification name on the thesaurus of the word extracted from the analysis result of the user utterance has a classification name related to emotion, it is assumed that the content of the user utterance already contains emotion. It is inappropriate to output a response sentence that asks for emotion in response to the user's utterance, and the use of the emotion questionable determination DB 28 generated according to the above rules prevents the generation of such an inappropriate response sentence. Can do. Here, the class name of the fourth hierarchy of the “go (verb)” thesaurus is “round trip”, and since it is determined to be a registration target based on the rule, the process proceeds to step 112.

  In step 112, the words determined to be registered in step 110 are registered in the emotion questionable determination DB 28 as words suitable for asking for emotions.

  Next, in step 114, it is determined whether or not the above processing has been performed for all sentences on the corpus. If there is an unprocessed sentence, the process returns to step 100 to process the next one sentence. Next, assuming that the sentence “I went to the amusement park and enjoyed it for a long time” is extracted, “go (verb)” and “enjoy (verb)” are extracted in step 106, and “go (verb)” in step 108. ) ”Is determined, and“ Enjoy (Verb) ”is determined as not determined. In step 110, as shown in FIG. 11, when the classification name “Pleasure / Pleasure” in the fourth hierarchy on the “Enjoy (Verb)” thesaurus is determined based on the rule, the operation “Enjoy (Verb)” is performed. Since it can be determined that it is not appropriate to ask for emotion, “enjoy (verb)” is not registered in the emotion questionable determination DB 28. If it is determined in step 114 that the above processing has been completed for all sentences, the processing is terminated. FIG. 12 shows the emotion questionable determination DB 28 generated by this processing.

  Next, the generation process of the frequency question impossibility determination DB 34 will be described with reference to FIG.

  In step 120, a sentence is extracted from a large corpus such as text on the web. Here, it is assumed that the sentence “I brushed my teeth” has been extracted. Next, in step 122, morphological analysis is performed on the one sentence extracted in step 120. Here, as shown in FIG. 14, analysis results of “tooth (noun)”, “wo (case particle)”, “brush (verb)”, and “ta (auxiliary verb)” are obtained. Next, in step 124, it is determined whether or not a situation is included in the analysis result. If the situation is included, the process proceeds to step 126. If the situation is not included, the process proceeds to step 132. In this case, since “brush (verb)” is included, it is determined that the situation is included, and the process proceeds to step 126.

  In step 126, a clause representing the situation is extracted. If it is determined in step 124 that (i) “verb” is included, a clause having the corresponding verb as a predicate is extracted. (Ii) If it is determined that “situation noun + (“ da ”or“ is ”)” is included, the predicate is “situation noun + (“ da ”or“ is ”)” Extract a clause. Here, “brush teeth” is extracted. Next, in step 128, it is determined whether or not the extracted clause has been registered in the situation DB. The situation DB is a database for temporarily registering clauses representing a large number of situations extracted from a large-scale corpus for processing to be described later. If not registered, the process proceeds to step 130, where the extracted clause is registered in the situation DB and the process proceeds to step 132. On the other hand, if registered, the process proceeds to step 132 as it is.

  In step 132, it is determined whether or not the above processing has been performed for all sentences on the corpus. If there is an unprocessed sentence, the process returns to step 120 to process the next sentence. Next, if the sentence “I went to the amusement park” is extracted, “Go to the amusement park” is extracted in step 126 and registered in the situation DB in step 130. If it is determined in step 132 that the above processing has been completed for all sentences, the process proceeds to step 134.

  In step 134, when each of the situations registered in the situation DB corresponds to a situation where it is inappropriate to ask the frequency as shown in FIG. 15, it is registered in the frequency question impossible judgment DB 34. The determination of whether or not it is inappropriate to ask the frequency is done manually. “Brush your teeth” corresponds to “1. Everybody usually experiences as usual”, which is listed as an unsuitable situation to ask about the frequency. . “Going to the amusement park” does not correspond to any of the situations listed as inappropriate situations for asking the frequency, and is not registered in the frequency question impossible judgment DB 34. That is, the frequency question impossibility determination DB 34 indicates that it is inappropriate to ask the frequency for the situation registered in the frequency question impossibility determination DB 34. When the above process is completed for all the situations registered in the situation DB, this process is terminated. FIG. 16 shows the frequency question impossibility determination DB 34 generated by this processing.

  Next, the generation process of the parallel object questionable determination DB 38 will be described with reference to FIG.

  In step 140, a sentence is extracted from a large corpus such as text on the web. Here, it is assumed that the sentence “I ate bread and salad at noon” was extracted. Next, in step 142, morphological analysis is performed on the one sentence extracted in step 140. Here, as shown in FIG. 18, “noon (noun)” “ni (case particle)” “pan (noun)” “to (parallel particle)” “salad (noun)” “wo (case particle)” “ An analysis result of “eating (verb)” and “ta (auxiliary verb)” is obtained.

  Next, in step 146, it is determined whether or not a verb whose object is parallel noun is included based on whether or not the language pattern shown in FIG. If it is included, the process proceeds to step 148. If it is not included, the process proceeds to step 152. Here, “Pan (noun)” “To (parallel particle)” “Salad (noun)” “O (case particle)” “Eat (verb)” is the pattern of “1. Noun and noun case particle verb” Since they match, it is determined that a verb whose object is parallel nouns is included, and the process proceeds to step 148.

  In step 148, a sentence containing a verb whose object is parallel nouns is divided into clauses each having a plurality of parallel nouns as objects, and each clause is extracted as a clause representing the situation. To do. Here, “eat bread” and “eat salad” are extracted.

  Next, in step 150, the node representing the extracted situation is registered in the situation DB in which the node representing the situation and the count value obtained by counting the appearance frequency of the clause are determined. In the case of the same clause as that already registered, the count value is incremented by one.

  Next, in step 152, it is determined whether or not the above processing has been performed for all the sentences on the corpus. If there is an unprocessed sentence, the process returns to step 140 and the next one sentence is processed. Next, if the sentence “I ate bread or sandwiches” is extracted, then in step 146, since it matches the pattern of “2. Noun or noun case particle verb”, a verb whose object is a parallel noun is included. In step 148, “eat bread” and “eat sandwich” are extracted. In step 150, in the situation DB, “eating bread” is incremented by one, and “ “Eating sandwiches” is newly registered. If it is determined in step 152 that the above processing has been completed for all sentences, the process proceeds to step 154.

  In step 154, with reference to the situation DB, a clause representing a situation whose appearance frequency is equal to or higher than the threshold is registered in the parallel object questionable judgment DB. An example of the situation DB after processing all the sentences in the corpus is shown in FIG. Here, if the threshold is “10”, in FIG. 20, “eating bread” to “buying a mouse” are registered in the parallel object questionable determination DB 38. A high count value (occurrence frequency) in the situation DB often appears as a sentence whose object is parallel nouns originally, and it is assumed that there are other omitted objects. The question asking for the object to do is appropriate. When all the clauses whose count value is greater than or equal to the threshold value are registered in the parallel object questionable determination DB 38, the present process is terminated. FIG. 21 shows the parallel object questionable determination DB 38 generated by this processing.

  In addition, paying attention to the “verb” of the clause registered in the situation DB, the count value of all the clauses including the “verb” for any “verb” in any clause May exceed the threshold, only the “verb” may be registered in the parallel object questionable determination DB 38 instead of the clause representing the situation. For example, based on the situation DB of FIG. 20, the clause containing “eating (verb)” is counted regardless of the object of “bread”, “salad”, and “sandwich”. Exceeds the threshold. In addition, for “go (verb)”, the count value exceeds the threshold value when any of the objects “to the movie theater”, “to the coffee shop”, and “to the bookstore” is taken. Therefore, as shown in FIG. 22, only the verbs “eat” and “go” are registered in the parallel object questionable determination DB 38, not each of the clauses representing the situation. On the other hand, for “buy (verb)”, when “keyboard” and “mouse” are the objects, the count value exceeds the threshold, but “house” and “land” are the objects. In this case, since the count value does not exceed the threshold value, “buy the keyboard” and “buy the mouse” whose count value exceeds the threshold value are respectively registered in the state of the clause indicating the situation.

  Next, the generation process of the noun detail questionable determination DB 46 will be described with reference to FIG. Here, a case of generating a database for determining whether or not a question for asking details of a noun using the question word “what” is appropriate will be described.

  In step 160, a sentence is extracted from a large corpus such as text on the web. Here, it is assumed that a sentence “I bought a car” has been extracted. Next, in step 162, morphological analysis is performed on one sentence extracted in step 160. Here, as shown in FIG. 24, the analysis results of “car (noun)”, “wo (case particle)”, “buy (verb)”, and “ta (auxiliary verb)” are obtained. Next, in step 164, it is determined whether or not a noun is included in the analysis result. When the noun is included, the process proceeds to step 166, and when it is not included, the process proceeds to step 176. Since “car (noun)” is included here, it is determined that a noun is included, and the process proceeds to step 166.

  In step 166, a word whose part of speech is a noun is extracted. Here, “car (noun)” is extracted. Next, in step 168, it is determined whether or not the extracted word has been determined to be registered in the noun detail questionable determination DB 46. If not determined, the process proceeds to step 170. If determined, the process proceeds to step 176. Here, since it is still the first sentence, it is determined that it has not been discriminated, and the routine proceeds to step 170.

  In step 170, the frequency of appearance on the web of combinations of extracted words and interrogatives corresponding to co-occurrence patterns of interrogatives and nouns as shown in FIG. 25 is counted. Here, for the question word “what” and the extracted word “car (noun)”, for example, when “what car” appears on the web, the co-occurrence pattern “1: question word noun case particle” (Besides “more”) ”, so count.

  Next, in step 172, it is determined whether or not the appearance frequency on the web of the combination of the extracted word and questionable word is equal to or higher than a threshold value. If it is equal to or greater than the threshold, the process proceeds to step 174, and if it is smaller than the threshold, the process proceeds to step 176. FIG. 26 shows an example of the count result. Here, assuming that the threshold value is “100”, the sum of the appearance frequencies of the combination of the extracted word “car (noun)” and the question word “what” is equal to or greater than the threshold value. In FIG. 26, the appearance frequency is counted for each co-occurrence pattern, but it is only necessary to be able to count the total appearance frequency.

  In step 174, the noun detail questionable determination DB 46 determines that the word that is the combination of the word and the question word whose appearance frequency is determined to be greater than or equal to the threshold value in step 172 is an appropriate word to ask for details using the question word. sign up.

  Next, in step 176, it is determined whether or not the above processing has been performed for all the sentences on the corpus. If there is an unprocessed sentence, the process returns to step 160 to process the next one sentence. Next, if the sentence “I went to eat yakiniku by car” is extracted, “car (noun)” and “yaki (noun)” are extracted at step 166, and “car (noun)” is extracted at step 168. "" Has been determined, and "Yakiniku (noun)" is determined to have not been determined. In step 170, the combinations of the question word “what” and the extracted word “yaki (noun)” corresponding to the co-occurrence pattern are counted. FIG. 27 shows an example of the count result. Since the total frequency of appearance of the combination of the extracted word “Yakiniku (noun)” and the interrogative word “what” is smaller than the threshold, it is not registered in the noun detail questionable judgment DB 46. That is, it is unsuitable to ask the details of “Yakiniku (noun)” using the question word “what”.

  If it is determined in step 176 that the above processing has been completed for all sentences, the processing is terminated. FIG. 28 shows words registered for the question word “what” in the noun detail questionable judgment DB 46 generated by this processing. Note that, as shown in FIG. 29, a plurality of interrogatives for inquiring details on a word whose part of speech is a noun are determined in advance, and each of these interrogatives corresponds to the interrogative by the same processing as described above. Appropriate words are registered to generate a noun detail questionable judgment DB 46.

  Next, with reference to FIG. 30, a description will be given of the generation process of the adjective expression reason questionable determination DB 54.

  At step 180, a sentence is extracted from a large corpus such as text on the web. Here, it is assumed that the sentence “It was boring” has been extracted. Next, in step 182, morphological analysis is performed on the one sentence extracted in step 180. Here, as shown in FIG. 31, the analysis results of “boring (noun)” “da (auxiliary verb)” “ta (auxiliary verb)” “yo (final verb)” are obtained. Next, in step 184, it is determined whether or not an adjective expression is included in the analysis result. If an adjective expression is included, the process proceeds to step 186. If an adjective expression is not included, the process proceeds to step 194. Since “boring (noun +“ it ”)” is included here, it is determined that an adjective expression is included, and the process proceeds to step 186.

  At step 186, words or clauses of the adjective expression are extracted. Here, “boring (noun +“ it ””) is extracted. Next, in step 188, it is determined whether or not the extracted word or clause has been determined to be registered in the adjective expression reason question possibility determination DB 54. If not determined, the process proceeds to step 190. If determined, the process proceeds to step 194. Here, since it is still the first sentence, it is determined that it has not been discriminated, and the routine proceeds to step 190.

  In step 190, based on the rule using the classification name on the thesaurus as shown in FIG. 32, it is determined whether or not the word or clause extracted in step 186 is a registration target in the adjective expression reason questionability determination DB 54. To do. If it is a registration target, the process proceeds to step 192. If it is not a registration target, the process proceeds to step 194. For example, as shown in FIG. 33, a rule using a classification name on the thesaurus is defined to register a word or a clause with a classification name related to human emotions and senses in the fourth layer on the thesaurus. Can do. The classification names related to human emotions and sensations refer to things related to human subjectivity, such as “sense”, “fatigue / sleep, etc.”, “pleasure / joy”. When the classification name on the thesaurus of words or clauses extracted from the analysis result of the user utterance is given a classification name related to human emotions and senses, the contents of the user utterance include the user's subjectivity. It is appropriate to ask why. Here, the category name of the fourth level of the “boring” thesaurus is “fatigue / sleep, etc.”, and based on the rule, it is determined to be a registration target, and the process proceeds to step 192.

  In step 192, the words or clauses determined to be registered in step 190 are registered in the adjective expression reason questionable determination DB 54 as words or clauses that are appropriate to ask the reason.

  Next, in step 194, it is determined whether or not the above processing has been performed for all the sentences on the corpus. If there is an unprocessed sentence, the process returns to step 180 to process the next one sentence. Next, if the sentence “I bought a red car” is extracted, “red (adjective)” is extracted in step 186, and on the thesaurus of “red (adjective)” in step 190, as shown in FIG. When the classification name “color” in the fourth hierarchy of is judged based on the rule, it can be judged that it is not appropriate to ask the reason for the adjective expression “red (adjective)”, so “red (adjective)” It is not registered in the adjective expression reason questionable judgment DB 54.

  If it is determined in step 194 that the above processing has been completed for all sentences, the processing is terminated. FIG. 35 shows the adjective expression reason questionable judgment DB 54 generated by this processing.

  Next, with reference to FIG. 36, a response generation processing routine in the response generation device 10 of the present exemplary embodiment will be described. This routine is performed by the CPU executing a response generation program stored in the ROM.

  In step 200, it is determined whether or not a user utterance has been input from the microphone 12. If a user utterance has been input, the process proceeds to step 202. If not, the determination in this step is repeated until it is input. . Here, it is assumed that the user utterance “I watched a movie” has been input.

  In step 202, the input voice signal indicating the user utterance is recognized as voice string information, and morphological analysis is performed on the character string information. Here, as a result of the analysis, as shown in FIG. 37, “movie (noun)” “is (case particle)” “watch (verb)” “ta (auxiliary verb)” “yo (final particle)” is obtained.

  Next, in step 204, based on the analysis result of the morphological analysis, it is determined whether or not a situation is included in the user utterance. If the situation is included, the process proceeds to step 206. If not, the process proceeds to step 222. Here, since “view (verb)” is included, it is determined that the situation is included, and the process proceeds to step 206.

  In step 206, a word representing the situation is extracted from the analysis result of the user utterance. Here, “view (verb)” is extracted. Next, in step 208, it is determined whether or not it is appropriate to generate a question asking for emotion by determining whether or not the extracted word is registered in the emotion questionable determination DB 28. If it is registered, the process proceeds to step 210. If it is not registered, the process proceeds to step 212. Here, referring to the emotion question possible determination DB 28 shown in FIG. 12, since “watch” is registered, it is determined that it is appropriate to generate a question asking for emotion, and the process proceeds to step 210.

  In step 210, response sentence candidates for asking for emotions are generated and temporarily stored in a predetermined storage area. Here, “How was it?” Is selected from the question sentence examples shown in FIG. 3 and generated as a response sentence candidate.

  Next, in step 212, a clause representing the situation is extracted from the analysis result of the user utterance. Here, “watch movie” is extracted. Next, in step 214, it is determined whether or not it is appropriate to generate a question asking for the frequency by determining whether or not the extracted clause is registered in the frequency question impossible determination DB 34. If not registered, the process proceeds to step 216. If registered, the process proceeds to step 218. Here, referring to the frequency question impossibility determination DB 34 shown in FIG. 16, since “watch movie” is not registered, it is determined that it is appropriate to generate a question asking for frequency, and the process goes to step 216. move on.

  In step 216, a response sentence candidate for inquiring the frequency is generated and temporarily stored in a predetermined storage area. Here, the response sentence example format “well? Do you do?” Is selected from the response sentence example and the response sentence example format shown in FIG. Response sentence candidates such as “Do you want to watch?” Are generated.

  Next, in step 218, it is appropriate to generate a question asking for parallel objects by determining whether or not the clause extracted in step 212 is registered in the parallel object question possibility determination DB 38. Judge whether or not. If registered, the process proceeds to step 220, and for example, a response sentence candidate is generated using the response sentence example and the response sentence example format shown in FIG. On the other hand, if the extracted clause is not registered in the parallel object questionable determination DB 38, the process proceeds to step 230. Here, referring to the parallel object question possibility determination DB 38 shown in FIG. 21, since “watch movie” is not registered, it is determined that it is inappropriate to generate a question asking for parallel objects. Then, the process proceeds to step 238.

  If it is determined in step 204 that no situation is included in the user utterance and the process proceeds to step 222, whether or not a noun is included in the user utterance is determined based on the analysis result of the morphological analysis. Determine. If the noun is included, the process proceeds to step 224. If the noun is not included, the process proceeds to step 230. Here, since “movie (noun)” is included, it is determined that a noun is included, and the process proceeds to step 224.

  In step 224, a word whose part of speech is a noun is extracted from the analysis result of the user utterance. Here, “movie (noun)” is extracted. Next, in step 226, it is determined whether or not it is appropriate to generate a question asking for details about the extracted word by determining whether or not the extracted word is registered in the noun detail question possible determination DB 46. to decide. If registered, the process proceeds to step 228. If not registered, the process proceeds to step 238. Here, referring to the noun detail question possible judgment DB 46 shown in FIG. 28, since “movie” is registered corresponding to the question word “what”, it is appropriate to generate a question asking for details. After the determination, the process proceeds to step 228.

  In step 228, a response sentence candidate for asking details about the extracted word is generated and temporarily stored in a predetermined storage area. Here, the question sentence “what” and “movie” stored in correspondence with “movie” in the noun detail questionable determination DB 46 are combined to generate a response sentence candidate such as “what movie?”.

  If it is determined in step 222 that nouns are included in the user utterance and the process proceeds to step 230, whether or not an adjective expression is included in the user utterance based on the analysis result of the morphological analysis. Is determined. If an adjective expression is included, the process proceeds to step 232; otherwise, the process proceeds to step 242. Here, (i) the part of speech is “adjective” and (ii) “part of speech is a noun word + (“ da ”” or “is”) ”is not included, so an adjective expression is included If it is not determined, the process proceeds to step 242.

  If an adjective expression is included, in step 232, words or clauses of the adjective expression are extracted from the analysis result of the user utterance, and then in step 234, the extracted word or clause is the reason for the adjective expression. By determining whether or not it is registered in the questionable determination DB 54, it is determined whether or not it is appropriate to generate a question that asks the reason for the description expression. If registered, the process proceeds to step 236, for example, a response sentence candidate is generated using the response sentence example and response sentence example format shown in FIG. 6, and if not registered, the process proceeds to step 238.

  Next, in step 238, it is determined whether or not there are at least one response sentence candidate generated in steps 210, 216, 220, 228, and 236. If a response sentence candidate is stored in a predetermined storage area, the process proceeds to step 240, and one response sentence candidate is randomly selected from the generated response sentence candidates. Here, response sentence candidates such as “How was it?”, “Do you often watch movies?”, And “What kind of movie?” Are generated, and for example, “How was it?” Is selected. . On the other hand, if no response sentence candidate has been generated, the process proceeds to step 242 to generate a compatible response sentence such as “Hey ~” or “So?”. The response sentence of the companion can be generated by selecting from a standard sentence prepared in advance or adding a word extracted from the analysis result of the user utterance to a standard format prepared in advance.

  Next, in step 244, the response sentence candidate selected in step 240 or the response sentence generated in step 242 is converted into an audio signal and output. Here, “How was it?” Is output as a voice and the process is terminated.

  In addition, you may perform the process of the said steps 204-220, steps 222-228, and steps 230-236 in parallel. Moreover, you may perform the process of said step 206-210, step 212-216, and step 212-218-220 in parallel.

  As described above, according to the response generation device of the present embodiment, a word or clause representing a situation extracted from the analysis result of the user utterance, a part of speech is a noun word, and an adjective expression word or clause. Because it is possible to determine whether or not the content of the question is appropriate by determining whether or not these words and clauses are registered in the database generated to determine whether or not it is appropriate to ask in advance It is possible to prevent the generation of inappropriate response sentences and perform natural dialogue.

  In the present embodiment, an appropriate word or clause is registered in the emotion questionable determination DB, parallel object questionable determination DB, noun detail questionable determination DB, and adjective expression reason questionable determination DB. In the question impossibility determination DB, the case where an inappropriate clause is registered has been described. However, since it can be determined that it is inappropriate if it is not appropriate based on similar rules, the emotion question possible determination DB, In the parallel object questionable judgment DB, the noun detail questionable judgment DB, and the adjective expression reason questionable judgment DB, an inappropriate word or clause may be registered, or in the frequency question impossible judgment DB An appropriate clause may be registered.

  Further, in the present embodiment, the case where sound output by a speaker is performed has been described as an example, but 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.

DESCRIPTION OF SYMBOLS 10 Response production | generation apparatus 12 Microphone 14 Speaker 16 Computer 20 Language analysis part 22 Situation determination part 24 Situation (word) extraction part 26 Emotion question response generation part 28 Emotion question possibility judgment DB (database)
30 Situation (section) extraction unit 32 Frequency question response generation unit 34 Frequency question possible determination DB
36 Parallel Object Question Answer Generation Unit 38 Parallel Object Questionable Judgment DB
40 Noun discrimination unit 42 Noun extraction unit 44 Noun detail question response generation unit 46 Noun detail question possible judgment DB
48 Adjective expression discriminating unit 50 Adjective expression extracting unit 52 Adjective expression reason question response generating unit 54 Adjective expression reason questionable judgment DB
56 Affinity Response Generation Unit 58 Output Unit

Claims (15)

A situation extracting means for extracting a word representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence;
It is appropriate to determine whether or not the word representing the situation extracted by the situation extracting means is registered in the emotion questionable determination dictionary as an appropriate word or an inappropriate word for which an emotion is asked. If it is a simple word, a response sentence candidate prepared in advance for inquiring feelings for the word indicating the extracted situation, or a word indicating the extracted situation is added to a response sentence format prepared in advance. An emotional question answer generation means for generating a response sentence candidate;
A response generation device including:   A word that is not assigned a classification name related to emotions in the thesaurus is registered in the emotion questionable judgment dictionary as an appropriate word to ask for emotions, or a classification name related to emotions in the thesaurus is added. The response generation device according to claim 1, wherein the word is registered as a word that is inappropriate for asking an emotion. A situation extracting means for extracting a clause representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence;
It is appropriate to ask whether or not the clause representing the situation extracted by the situation extraction means is registered in the frequency questionable judgment dictionary as an appropriate clause or an inappropriate clause to ask the frequency. If it is a long section, a response sentence candidate prepared in advance for inquiring the frequency of the extracted situation is added, or a section representing the extracted situation is added to a prepared response sentence format. A frequency question answer generating means for generating a response sentence candidate;
A response generation device including:   In the frequency questionable judgment dictionary, a situation that is frequently experienced, a situation that is experienced less frequently, a situation that can occur only at a specific time, a situation that always accompanies another situation, and that cannot occur independently, always has a predetermined purpose Whether a section that represents a situation that occurs for the purpose of and is not complete by itself, that means start or end, and that that means existence is registered as a section where it is inappropriate to ask the frequency 4. The response generation apparatus according to claim 3, wherein a clause representing a situation other than the situation corresponding to any one of the above is registered as a clause appropriate to ask the frequency. A situation extracting means for extracting a clause representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence;
It is determined whether or not the clause representing the situation extracted by the situation extracting means is registered in the parallel object questionable determination dictionary as an appropriate clause or an inappropriate clause for asking for a parallel object. If it is an appropriate clause to ask for an object to be executed, a response sentence candidate prepared in advance for asking a parallel object for the section representing the extracted situation, or a response sentence format prepared in advance is used. Parallel object question answer generating means for generating a response sentence candidate to which a clause representing the extracted situation is added;
A response generation device including:   In the parallel object questionable judgment dictionary, a clause representing a situation in which a plurality of parallel nouns are set as objects is extracted from a plurality of input sentences, and each of the plurality of parallel nouns is divided into clauses having the object as a target. Then, register the divided clauses with the appearance frequency of each of the divided clauses equal to or higher than a predetermined value as appropriate clauses for asking parallel objects, or the division with the appearance frequency smaller than the predetermined value. The response generation apparatus according to claim 5, wherein the registered clauses are registered as clauses for which it is inappropriate to ask for parallel objects.   A combination of at least two response generation devices selected from the response generation device according to claim 1 or claim 2, the response generation device according to claim 3 or claim 4, and the response generation device according to claim 5 or claim 6. Response generator. Noun extraction means for extracting a word whose part of speech is a noun from the input sentence based on an analysis result obtained by analyzing the structure of the input sentence;
Determine whether a word whose part of speech has been extracted by the noun extraction means is registered in the noun detail questionable dictionary as an appropriate word or an inappropriate word to ask for details, and ask for details Is an appropriate word, the extracted part of speech is a response sentence candidate prepared in advance for asking details of the noun word, or a word indicating the situation extracted in a response sentence format prepared in advance. A noun detail question answer generation means for generating a response sentence candidate to which is added,
The response generation device according to claim 1, comprising:   In the noun detail questionable judgment dictionary, the part-of-speech noun word that has a part-of-speech noun-word combination and a question-phrase on the web that has a frequency of appearance of a predetermined value or more is registered as an appropriate word to ask for details. Or a request that the part-of-speech noun is a noun word whose frequency of appearance on the web of a combination of a noun word and a question word is less than a predetermined value is registered as an inappropriate word to ask for details Item 9. The response generation device according to Item 8. Based on the analysis result obtained by analyzing the structure of the input sentence, an adjective expression extracting means for extracting words or clauses of the adjective expression from the input sentence;
Whether or not the word or clause of the adjective expression extracted by the adjective expression extracting means is registered in the adjective expression reasonable questionable dictionary as an appropriate word or clause for which the reason is asked, or as an inappropriate word or clause If it is an appropriate word or clause to ask for the reason, a response sentence candidate prepared in advance for asking a reason for the word or clause of the extracted adjective expression, or a response prepared in advance An adjective expression question answer generating means for generating a response sentence candidate by adding the extracted word or clause of the adjective expression to a sentence format;
The response generation device according to claim 1, comprising:   In the adjective expression reason question possible judgment dictionary, a word with a classification name related to a person's emotion and sensation in the thesaurus is registered as an appropriate word to ask the reason, or a word of the person in the thesaurus The response generation device according to claim 10, wherein a word without a classification name related to emotion and sense is registered as an inappropriate word for asking a reason. Computer
A situation extracting means for extracting a word representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence;
It is appropriate to determine whether or not the word representing the situation extracted by the situation extracting means is registered in the emotion questionable determination dictionary as an appropriate word or an inappropriate word for which an emotion is asked. If it is a simple word, a response sentence candidate prepared in advance for inquiring feelings for the word indicating the extracted situation, or a word indicating the extracted situation is added to a response sentence format prepared in advance. An emotional question answer generation means for generating a response sentence candidate;
A response generator to make it function. Computer
A situation extracting means for extracting a clause representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence;
It is appropriate to ask whether or not the clause representing the situation extracted by the situation extraction means is registered in the frequency questionable judgment dictionary as an appropriate clause or an inappropriate clause to ask the frequency. If it is a long section, a response sentence candidate prepared in advance for inquiring the frequency of the extracted situation is added, or a section representing the extracted situation is added to a prepared response sentence format. A frequency question answer generating means for generating a response sentence candidate;
A response generator to make it function. Computer
A situation extracting means for extracting a clause representing a situation from the input sentence based on an analysis result obtained by analyzing a structure of the input sentence;
It is determined whether or not the clause representing the situation extracted by the situation extracting means is registered in the parallel object questionable determination dictionary as an appropriate clause or an inappropriate clause for asking for a parallel object. If it is an appropriate clause to ask for an object to be executed, a response sentence candidate prepared in advance for asking a parallel object for the section representing the extracted situation, or a response sentence format prepared in advance is used. Parallel object question answer generating means for generating a response sentence candidate to which a clause representing the extracted situation is added;
A response generator to make it function.   The response production | generation program for functioning a computer as each means which comprises the response production | generation apparatus of any one of Claims 1-11.

Images