JP4316839B2 - Conversation control device and conversation control method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a conversation control apparatus and a conversation control method for outputting answer contents corresponding to input information in a natural format close to a dialog format based on input information input from a user.
[0002]
[Prior art]
In the conventional conversation control system, since the answer contents stored in advance corresponding to the utterance contents are output based on the utterance contents from the user, the user can use the answer contents answered from the conversation control system. I was able to enjoy the feeling of talking with other users in a pseudo manner.
[0003]
[Problems to be solved by the invention]
However, in the above conversation control system, the utterance content (input information) from the user is collated with the utterance content (phrase) group of the user stored in advance, and matches the input information from the phrase group. Since it searches for a phrase and outputs the response content associated with the searched phrase, if it does not store a large number of phrases that are as close as possible to the above input information, it will output the response content that is more appropriate for the input information. I couldn't.
[0004]
On the other hand, if the developer who constructs the conversation control system stores in the conversation control system in advance a huge number of phrases that are as close as possible to the input information and the answer contents associated with the large number of phrases, the conversation control system Can search for a phrase that matches the input information from a huge group of phrases stored in advance, and output the answer contents associated with the searched phrase, so that the answer is more suitable for the input information. The contents can be output.
[0005]
However, a developer who builds a conversation control system can search for a phrase that matches the input information using the conversation control system, and the user will have a huge phrase group (a phrase group near infinity). Must be stored in the conversation control system in advance, and it is difficult to construct a conversation control system that outputs answer contents more suitable for input information.
[0006]
In addition, since each answer content is stored in a predetermined format, it is difficult for the conversation control system to output in a format that answers the utterance content from the user as it is. Could not be output in a more natural format. For example, when a user utters “Are you a new singer?”, It is difficult for the conversation control system to answer in the form “A new singer's score is XX”. .
[0007]
Therefore, the present application has been made in view of the above points, and each morpheme constituting input information input from the user is specified, and the semantic content grasped from each specified morpheme is analyzed and analyzed. By obtaining pre-prepared answer contents associated with semantic contents, the optimum answer contents corresponding to the input information from the user are obtained, and the input information can be stored even if each answer contents stored in advance is reduced. It is an object of the present invention to provide a conversation control apparatus and a conversation control method capable of acquiring the optimum answer content corresponding to the above and further outputting the obtained answer content in a natural format closer to the dialog format.
[0008]
[Means for Solving the Problems]
The invention according to the present application stores a plurality of second morpheme information indicating a morpheme composed of one character, a plurality of character strings, or a combination thereof, and a response content to the user corresponding to the input information in advance. The character string indicating the input information is specified based on the input information input from the user, and at least one morpheme constituting the minimum unit of the character string is extracted as the first morpheme information based on the specified character string And collating the extracted first morpheme information with each second morpheme information, and searching from each second morpheme information for second morpheme information including at least one morpheme constituting the first morpheme information, Based on the retrieved second morpheme information, the response content associated with the second morpheme information is acquired, and based on the acquired response content, the extracted first morpheme information or the retrieved second morpheme It is characterized in adding to the answer content to at least one of morphemes constituting a broadcast.
[0009]
According to the invention according to the present application, the conversation control device acquires the response content associated with the second morpheme information based on the searched second morpheme information, and extracts based on the acquired response content Since at least one morpheme constituting the first morpheme information or the searched second morpheme information can be added to the response content, the conversation control device can display the response content corresponding to the input information in a natural format close to the interactive format. Can be output.
[0010]
For example, when the user inputs input information “Is there a new singer A”, the conversation control device acquires the response content “Song Name 1” corresponding to the input information, and this response content “Song Name”. The morpheme “A singer's new score” constituting the input information can be added to “1”, so that ““ A singer's new score ”is“ Song name 1 ”” can be output as an answer to be output.
[0011]
As a result, an answer that directly answers the input information input by the user is output from the conversation control device, so that the user can feel as if the user is having a conversation with another user. Can do.
[0012]
Further, the conversation control device searches the second morpheme information including the morpheme constituting the first morpheme information (the element constituting the input information from the user) from each second morpheme information, and the second searched Since the response content associated with the second morpheme information can be acquired based on the morpheme information, the conversation control device is constructed by each morpheme based on each morpheme constituting the input information from the user. Answer contents created in advance can be acquired based on the semantic space (semantic content grasped from the character string consisting of each morpheme), and the entire input information is simply set as a keyword and associated with the keyword. It is possible to obtain answer contents more suitable for input information than to obtain answer contents.
[0013]
Furthermore, since the conversation control device can search for the optimum answer content for the input information of the user based on each morpheme constituting the input information from the user, the developer who develops the conversation control device It is only necessary to create answer contents based on the semantic space constructed from each morpheme and a search function for searching the answer contents. Artificial intelligence for analyzing input information from users, neural networks, etc. There is no need to build a program consisting of
[0014]
Furthermore, since the conversation control device retrieves the second morpheme information including the first morpheme information, it is not necessary to retrieve the second morpheme information that completely matches the input information from the user, and the conversation control device is developed. The developer who does not need to store a large amount of response contents corresponding to the input information that will be input from the user in advance, and can reduce the capacity of the storage unit.
In the invention according to the present application, each morpheme of the extracted first morpheme information is classified into each attribute such as a subject case consisting of a main case and a target case consisting of a target, and a morpheme belonging to at least one of the classified attributes is acquired. It is a conversation control system characterized by being added to the answer contents.
[0015]
The conversation control device classifies each morpheme constituting the second morpheme information for each attribute such as a subject case and a target case, stores the morpheme in advance, and stores the second morpheme information based on the searched second morpheme information. A morpheme belonging to at least one attribute constituting may be added to the acquired answer content.
[0016]
According to the invention according to the present application, the conversation control device classifies each morpheme constituting the extracted first morpheme information into each attribute such as a subject case composed of a main case, a target case composed of a target, and classified. Since the morpheme belonging to at least one attribute can be added to the answer content, the conversation control device can output the answer content corresponding to the input information in a more natural form, that is, in a form close to an interactive form. .
[0017]
For example, when the user inputs input information “Is there a new singer A”, the conversation control device acquires the response content “Song Name 1” corresponding to the input information, and this response content “Song Name”. Since only the morpheme “new score” (target morpheme) that constitutes the input information can be added to “1”, ““ new score ”is“ song name 1 ”” can be output as an answer to be output, As a result, answers can be output in a more natural format.
[0018]
The invention according to the present application stores each second morpheme information in advance in association with a plurality of keyword groups, collates the extracted first morpheme information with each keyword group, and forms at least one of the first morpheme information A keyword that matches the morpheme is acquired, and second morpheme information associated with the acquired keyword is searched.
[0019]
According to the invention according to the present application, the conversation control device collates the extracted first morpheme information with each keyword group, and acquires a keyword that matches at least one morpheme constituting the first morpheme information. Since the second morpheme information associated with the keyword group belonging to the acquired keyword can be searched, the developer who develops the conversation control device can obtain a large amount of second morpheme information that approximates the first morpheme information. There is no need to store the conversation control device one by one, and the storage amount of the second morpheme information can be reduced.
[0020]
In the invention according to the present application, when two or more second morpheme information associated with the keyword group is searched, the second morpheme information includes at least one morpheme constituting the first morpheme information. It is characterized by searching for dimorphic information.
[0021]
According to the invention according to the present application, when the conversation control device searches for two or more second morpheme information associated with the keyword group, the first morpheme information is extracted from each second morpheme information. Since the second morpheme information including at least one morpheme that constitutes can be searched, the conversation control device can select a more appropriate first morpheme related to the input information input by the user from each second morpheme information. Dimorphic information can be searched.
[0022]
In addition, since the developer can reduce the second morpheme information to be stored in the conversation control device, the conversation control device can obtain the second morpheme information that approximates the first morpheme information from the few second morpheme information. You can search quickly.
[0023]
The invention according to the present application is that the second morpheme information or answer content stored is the type of topic the user wants, the degree of emotion or statement that the user has in a certain matter, an affirmative sentence, a question sentence, a repulsive sentence, etc. Are described in a language for hierarchically storing them in a database in association with morphemes according to their semantic content. This language is preferably described in DKML (Discourse Knowledge Markup Language).
[0024]
According to the invention according to the present application, the developer who develops the conversation control device hierarchizes the second morpheme information and the like for searching the response content for the input information from the user using the language in the database. Therefore, the conversation control apparatus can acquire the response content for the input information from the database through a hierarchical procedure based on the input information of the user.
[0025]
In other words, the conversation control device has determined and determined the hierarchy of input information from the user (for example, whether it is in a higher concept or a lower concept with respect to the second morpheme information stored in the database). Appropriate answer contents can be acquired from the answer contents accumulated in advance based on the hierarchy.
[0026]
The second morpheme information including the first morpheme information includes not only the case where the second morpheme information includes the first morpheme information as it is, but the second morpheme information includes at least one morpheme constituting the first morpheme information. It also means a case.
[0027]
DETAILED DESCRIPTION OF THE INVENTION
[First embodiment]
(Basic configuration of conversation control system)
A conversation control system according to the present invention will be described with reference to the drawings. FIG. 1 is a schematic configuration diagram of a conversation control system having a conversation control apparatus 1 according to the present embodiment.
[0028]
As shown in the figure, the conversation control device 1 includes an input unit 100, a speech recognition unit 200, a conversation control unit 300, a sentence analysis unit 400, a conversation database 500, an output unit 600, and a speech recognition dictionary storage. Part 700.
[0029]
In the present embodiment, for convenience of explanation, the description is limited to the user's utterance content (this utterance content is a kind of input information), but is not limited to the user's utterance content, and the keyboard. The input information may be input from the above. Therefore, the “utterance content” shown below can be described by replacing “utterance content” with “input information”.
[0030]
Similarly, in the following description, for convenience of explanation, the description will be limited to the “spoken sentence type” (speech type), but is not limited to this “spoken sentence type”, and input from a keyboard or the like. It may be an “input type” indicating the type of input information. Accordingly, the following “speech sentence type” (speech type) can be replaced with “input sentence type”.
[0031]
The input unit 100 is an acquisition unit that acquires input information from a user. In the present embodiment, a microphone, a keyboard, and the like are used. The input unit 100 is character recognition means for specifying a character string corresponding to input information (other than speech) based on input information input from a user.
[0032]
Here, the input information means characters, symbols, voices and the like input through a keyboard or the like. Specifically, the input unit 100 acquires user input information (other than voice), specifies the acquired input information as a character string, and outputs the specified character string to the conversation control unit 300. Further, the input unit 100 that has acquired the utterance content from the user (the utterance content is composed of voice and is a kind of input information) with a microphone or the like, the voice constituting the acquired utterance content is a voice signal. To the voice recognition unit 200.
[0033]
The voice recognition unit 200 is a character recognition unit that identifies a character string corresponding to the utterance content based on the utterance content acquired by the input unit 100. Specifically, the speech recognition unit 200 to which a speech signal is input from the input unit 100 analyzes the input speech signal, and a character string corresponding to the analyzed speech signal is stored in the speech recognition dictionary storage unit 700. The specified character string is output to the conversation control unit 300 as a character string signal. The speech recognition dictionary storage unit 700 stores a dictionary (a, i, e, e, etc.) corresponding to standard audio signals.
[0034]
The sentence analysis unit 400 analyzes the character string input to the conversation control unit 300. In this embodiment, as shown in FIG. 2, the morpheme extraction unit 410, the phrase analysis unit 420, the sentence structure, It has an analysis unit 430, an utterance type determination unit 440, a morpheme database 450, and an utterance type database 460.
[0035]
The morpheme extraction unit 410 is a morpheme extraction unit that extracts, as first morpheme information, each morpheme constituting the minimum unit of the character string based on the character string specified by the speech recognition unit 200.
[0036]
Specifically, the morpheme extraction unit 410 to which a character string is input from the management unit 310 extracts each morpheme from the input character string. Here, in this embodiment, the morpheme means the minimum unit of the word configuration represented in the character string. As the minimum unit of the word configuration, for example, as shown in FIG. Part of speech such as adjectives and verbs. In the present embodiment, each morpheme is expressed as m1, m2,.
[0037]
That is, the morpheme extraction unit 410 collates a character string corresponding to the input character string signal with a morpheme group including nouns, adjectives, verbs, and the like stored in advance in the morpheme database 450, and from among the character strings. Each morpheme (m1, m2,...) That matches the morpheme group is extracted, and each extracted morpheme is output to the phrase analysis unit 420 as an extraction signal.
[0038]
The phrase analysis unit 420 is a conversion unit that converts each morpheme into a phrase format based on each morpheme extracted by the morpheme extraction unit 410. Specifically, the phrase analysis unit 420 to which the extraction signal is input from the morpheme extraction unit 410 uses the morphemes corresponding to the input extraction signal to combine them into a phrase format.
[0039]
Here, in this embodiment, the phrase format is a sentence in which the independent grammar or one or more attached words are attached to the independent grammar in the Japanese grammar, or a character string that does not destroy the meaning of the Japanese grammar. Means a sentence that is separated as finely as possible. This clause is expressed as p1, p2,... Pk in this embodiment.
[0040]
That is, as shown in FIG. 4, the phrase analysis unit 420 extracts and extracts the dependency elements (for example, gahahahahaha ...) of each morpheme based on each morpheme corresponding to the input extraction signal. Each morpheme is grouped into each clause based on the dependency element. The phrase analysis unit 420 that summarizes each morpheme into each phrase includes a sentence structure analysis unit 430 and an utterance type determination unit 440 using sentence pattern information including each phrase that combines each morpheme and each morpheme constituting each phrase as a sentence pattern signal. Output to.
[0041]
The sentence structure analysis unit 430 is a classification unit that classifies each morpheme of the first morpheme information segmented by the phrase analysis unit 420 into attributes such as subject case and target case. Specifically, the sentence structure analysis unit 430, to which the sentence pattern signal is input from the phrase analysis unit 420, determines each morpheme included in the phrase based on each morpheme corresponding to the input sentence pattern signal and the phrase composed of each morpheme. Determine the “case composition” (attribute).
[0042]
Here, the “case structure” means a case (attribute) indicating a substantial concept in the clause. In the present embodiment, for example, a subject (subject) that represents a subject / subject, Object meaning (object case), action meaning action / verb, time meaning time (consisting of tense, mood, aspect), location meaning location, etc. In the present embodiment, each morpheme associated with the “case” (case configuration) of the three elements of the subject, the object, and the action in the phrase is set as the first morpheme information.
[0043]
That is, as shown in FIG. 5, for example, when the dependency element of each morpheme is “” or “is”, the sentence structure analysis unit 430 applies the subject (subject) to the morpheme before the dependency element. Or a leading character). In addition, the sentence structure analysis unit 430
For example, if the dependency element of each morpheme is “NO” or “NO”, it is determined that the morpheme before the dependency element is an object (target).
[0044]
Further, the sentence structure analysis unit 430, for example, when the dependency element of each morpheme is “Yes”, the morpheme preceding the dependency element is an action (predicate; this predicate is composed of a verb, an adjective, and the like). It is determined that
[0045]
The sentence structure analysis unit 430 that has determined the “case structure” of each morpheme that constitutes each clause determines the range of topics (topics) to be described later based on the first morpheme information associated with the determined “case structure”. A topic search command signal for specifying is output to the topic search unit 320.
[0046]
The utterance type determination unit 440 is a type specifying unit that specifies an utterance type indicating the type of utterance content based on the phrase specified by the phrase analysis unit 420. Specifically, “speech sentence type” (speech type) is determined based on each morpheme corresponding to the sentence pattern signal input from the phrase analysis unit 420 and a phrase composed of each morpheme.
[0047]
In this embodiment, as shown in FIG. 6, the “spoken sentence type” is a statement sentence (D; Declaration), an impression sentence (I; Impression), a conditional sentence (C; Condition), and a result sentence ( E; Effect, time sentence (T; Time), location sentence (L; Location), rebound sentence (N; Negation), affirmative sentence (A; Answer), question sentence (Q; Question) It is.
[0048]
The statement sentence means a sentence composed of a user's opinion or idea, and in this embodiment, as shown in FIG. 6, for example, a sentence such as “I like Sato” can be cited. An impression sentence means the sentence which consists of an impression which a user holds. A place sentence means a sentence made up of place elements.
[0049]
A result sentence means a sentence composed of sentences including a result element for a topic. A time sentence means a sentence composed of sentences including temporal elements related to a topic.
[0050]
The conditional sentence means a sentence composed of sentences including elements such as a premise of a topic, a condition and a reason why the topic is established, when one utterance is regarded as a topic. The repulsive sentence means a sentence composed of a sentence including an element that repels a user's utterance partner. An example sentence for each “spoken sentence type” is as shown in FIG.
[0051]
That is, the utterance type determination unit 440 collates each phrase with each dictionary stored in the utterance type database 460 based on each phrase corresponding to the input sentence pattern signal, and from each phrase, Sentence elements related to the dictionary (see FIG. 7) are extracted. The utterance type determination unit 440 that extracts sentence elements related to each dictionary from each phrase determines “spoken sentence type” based on the extracted sentence elements. The sentence element means a type for identifying the type of character string, and the sentence element in the present embodiment includes the above-described definition phrase (which means “-”).
[0052]
Here, as shown in FIG. 7, the utterance type database 460 includes a definition expression example dictionary having a dictionary related to a definition phrase (for example,), an affirmative phrase (for example, approval, sympathy, ping pong). positive cases dictionary with a dictionary related to, a result clause (for example, so, so) the results representation case dictionary with a dictionary related to, greeting clause (for example, Hello) greeting dictionary with a dictionary related to, negative It consists of negative case dictionaries with dictionaries related to phrases (for example, not stupid, opposite), and each dictionary is associated with a “spoken sentence type”.
[0053]
As a result, the utterance type determination unit 440 collates the phrase with each dictionary stored in the utterance type database 460, extracts the sentence element related to each dictionary from the phrase, and associates it with the extracted sentence element. By referring to the type of determination, the “spoken sentence type” can be determined (see FIG. 7).
[0054]
The utterance type determination unit 440 outputs an answer search command signal for causing the corresponding user to search for a specific answer sentence to the answer sentence search unit 330 based on an instruction from the topic search unit 320 described later.
[0055]
The conversation database 500 stores a plurality of second morpheme information indicating a morpheme composed of one character, a plurality of character strings, or a combination thereof, and a plurality of reply contents to the user with respect to the utterance contents in association with each other in advance. Means.
[0056]
The conversation database 500 is also an answer storage unit that associates a plurality of answer contents with the second morpheme information, and stores in advance an answer type indicating the type of each answer content in association with each answer content. Furthermore, the conversation database 500 is also an answer storage means for classifying and storing each element constituting the second morpheme information into attributes such as each subject consisting of a main case and a target case consisting of a target case.
[0057]
As shown in FIG. 8, in this embodiment, the conversation database 500 is roughly divided into a discourse range (discourse) that means a relevance range for the content of the user's utterance, and a discourse range. It is composed of topics (topics) that mean a range that is most closely related to the content that the user utters. As shown in the figure, “discourse range” is positioned as a superordinate concept of “topic” in the present embodiment.
[0058]
Each discourse range can be configured to have a hierarchical structure as shown in FIG. As shown in the figure, for example, a higher level discourse range (entertainment) for a certain discourse range (movie) is positioned in the upper hierarchical structure, and a lower level discourse range (movie) for the discourse range (movie). The attribute (movie) can be located in the lower hierarchical structure. That is, in the present embodiment, each discourse range can be arranged at a hierarchical position where the relationship between the higher concept, the lower concept, the synonym, and the antonym is clear in relation to other discourse ranges.
[0059]
As described above, the discourse range is composed of topics, and in this embodiment, for example, if the discourse range is an A movie name, it includes a plurality of topics related to “A movie name”. .
[0060]
This topic means a morpheme composed of a single character, a plurality of character strings, or a combination thereof, that is, each morpheme constituting speech content that will be uttered by the user. Each morpheme is associated with a subject (subject), an object (target case), and an action “case” (attribute). In this embodiment, each morpheme associated with these three elements is expressed as a topic title (this topic title corresponds to a subordinate concept of “topic”) (second morpheme information). .
[0061]
The topic title is not limited to each morpheme associated with the above three elements, but other “cases”, that is, a time meaning time (consisting of mood, tense, aspect, etc.), Each morpheme may be associated with a location meaning a place, a condition meaning a condition, an impression meaning an impression, an effect meaning a result, and the like.
[0062]
In this embodiment, the topic title (second morpheme information) is stored in advance in the conversation database 500, and is distinguished from the first morpheme information (derived from the content spoken by the user). It is what is done.
[0063]
For example, if the talk range is “A movie name”, as shown in FIG. 10, the subject title is subject (A movie name), object (director), action (great) {this is “A movie name” The director of the name is composed of "meaning great".
[0064]
If there is no morpheme associated with “case composition” (subject, object, action, etc.) among the topic titles, “*” is indicated for the portion in the present embodiment.
[0065]
For example, {A What is a movie name? } Is converted into a topic title (subject; object; action). }, “A movie name” can be specified as the subject, but “object” and “action” are not elements of the sentence, so the topic title is “subject” (A movie name) ); No “object” (*); no “action” (*) (see FIG. 10).
[0066]
The answer sentence means an answer sentence (answer contents) to be answered to the user. In this embodiment, the answer sentence is associated with each topic title (second morpheme information). It is stored in advance in the conversation database 500 (see FIG. 8). In this embodiment, as shown in FIG. 11, the answer sentence is a statement sentence (D; Declaration) or an impression sentence (I; Impression) in order to make an answer corresponding to the type of utterance sentence uttered by the user. , Conditional sentence (C; Condition), result sentence (E; Effect), time sentence (T; Time), location sentence (L; Location), negative sentence (N; Negation), positive sentence (A; Answer), question It shall be classified into a type (answer type) such as a sentence (Q).
[0067]
That is, as shown in FIG. 12, each answer sentence is associated with, for example, a discourse range (Sato) {subordinate concept; home run, superordinate concept; grass baseball, synonym; panda Sato, Sato player, panda} and each topic title. It has been.
[0068]
As shown in the figure, for example, the topic title 1-1 is {(Sato; *; I like): this consists of the order of (subject; object; action) as described above. If this order is the same below, the answer sentence 1-1 corresponding to the topic title 1-1 is (DA; statement affirmative sentence ("I also like Sato")), (IA ; Sentence affirmative sentence "I like Sato very much"), (CA; Condition affirmative sentence "Sato's home run is very impressive"), (EA; Result affirmative sentence "I always watch Sato's games on TV Masu "), (TA; time affirmative sentence" I actually like it from the five-batt continual refrain at Koshien "), (LA; place affirmative sentence" I like the serious face when standing at the batting " ), (NA; repulsion affirmation "I don't want to talk to people who don't like Sato, goodbye").
[0069]
In the present embodiment, the conversation control unit 300 includes a management unit 310, a topic search unit 320, and an answer sentence search unit 330, as shown in FIG.
[0070]
The management unit 310 controls the entire conversation control unit 300. Specifically, the management unit 310 to which a character string is input from the input unit 100 or the speech recognition unit 200 outputs the input character string to the morpheme extraction unit 410. Also, the management unit 310 outputs the answer sentence searched by the answer sentence search unit 330 to the output unit 600.
[0071]
The topic search unit 320 collates the first morpheme information extracted by the phrase analysis unit 420 with each second morpheme information (topic title), and forms the first morpheme information from each second morpheme information. It is the 1st search means which searches the 2nd morpheme information containing. Specifically, the topic search unit 320 to which the topic search command signal is input from the sentence structure analysis unit 430, based on the first morpheme information included in the input topic search command signal, the first morpheme information and the conversation database 500. Is compared with the discourse range group stored in, and the discourse range related to the first morpheme information is searched from the discourse range group.
[0072]
For example, when each morpheme (first morpheme information) that belongs to the “case structure” constituting the utterance sentence uttered by the user is (Sato; *; I like) {I like Sato}, the topic search unit Since “Sato” is included in the “case composition”, 320 compares this “Sato” with the discourse range group, and searches for a discourse range (Sato) that matches “Sato”.
[0073]
Further, the topic search unit 320 that has selected the discourse range related to “case configuration” searches for the “topic title” closest to each morpheme belonging to “case configuration” from among the topic titles belonging to the selected discourse range. The search result is output as a search result signal to the answer sentence search unit 330 and the utterance type determination unit 440.
[0074]
For example, if the “case composition” of the utterance content is (Sato; *; I like) {I like Sato}, the topic search unit 320, as shown in FIG. The morpheme (Sato; *; I like) and the topic titles 1-1 to 1-4 belonging to the discourse range (Sato) are collated, and from each of the topic titles 1-1 to 1-4, “case composition” A topic title 1-1 (Sato; *; likes) that matches (or approximates) each morpheme belonging to (Sato; *; likes) is searched, and the search result signal 330 is used as a search result signal as a search result signal. Output to the utterance type determination unit 440.
[0075]
The utterance type determination unit 440 to which the search result signal is input from the topic search unit 320, based on the input search result signal, an answer search command for searching for a specific answer sentence that answers to the corresponding user. The signal (this answer search command signal includes the determined “sentence type”) is output to the answer sentence search unit 330.
[0076]
The answer sentence search unit 330 is an answer acquisition unit that acquires an answer sentence associated with the second morpheme information based on the second morpheme information (topic title) searched by the topic search unit 320. In addition, the answer sentence search unit 330 collates the utterance type of the identified user with each answer type associated with the second morpheme information based on the second morpheme information searched by the topic search unit 320, The answer type that matches the user's utterance type is searched from each answer type, and an answer sentence associated with the answer type is acquired based on the searched answer type (second search means, answer Acquisition means).
[0077]
Specifically, the answer sentence search unit 330, to which the search result signal from the topic search unit 320 and the answer search command signal from the utterance type determination unit 440 are input, the topic title (search result corresponding to the input search result signal). Based on the second morpheme information) and the “utterance sentence type” (utterance type) corresponding to the answer search command signal, from among the answer sentence groups (respective answer contents) associated with the topic title , Search for an answer sentence having an answer type (this answer type means “type of answer sentence” shown in FIG. 11) that matches the “sentence sentence type” (DA, IA, CA, etc.).
[0078]
For example, if the topic title (second morpheme information) corresponding to the search result is the topic title 1-1 (Sato; *; I like) shown in FIG. The answer type that matches the “type of utterance” (for example, DA: utterance type) determined by the utterance type determination unit 440 from among the answer sentences 1-1 (DA, IA, CA, etc.) associated with 1 An answer sentence 1-1 consisting of (DA) (DA; (I also like Sato) is retrieved), and the retrieved answer sentence is output to the management unit 310 as an answer sentence signal.
[0079]
The management unit 310 to which the answer sentence signal is input from the answer sentence search unit 330 outputs the input answer sentence signal to the output unit 600. The output unit 600 is an output unit that outputs the answer sentence acquired by the answer sentence search unit 330. In the present embodiment, for example, a speaker, a display, and the like can be given. Specifically, the output unit 600 to which an answer sentence signal is input from the management unit 310 outputs an answer sentence {for example, I also like Sato} corresponding to the input answer sentence signal.
[0080]
(Conversation control method using conversation control device)
The conversation control method by the conversation control apparatus 1 having the above configuration can be implemented by the following procedure. FIG. 13 is a flowchart showing the procedure of the conversation control method according to the present embodiment.
[0081]
First, the input unit 100 performs a step of acquiring the utterance content from the user (S100). Specifically, the input unit 100 acquires the voice that constitutes the utterance content of the user, and outputs the acquired voice to the voice recognition unit 200 as a voice signal. Further, the input unit 100 identifies a character string corresponding to the input information (other than the voice) based on the input information (other than the voice) input from the user, and the conversation control unit uses the identified character string as a character string signal. Output to 300.
[0082]
Next, the voice recognition unit 200 performs a step of specifying a character string corresponding to the utterance content based on the utterance content acquired by the input unit 100 (S102). Specifically, the speech recognition unit 200 to which a speech signal is input from the input unit 100 analyzes the input speech signal, and a character string corresponding to the analyzed speech signal is stored in the speech recognition dictionary storage unit 700. The specified character string is output to the conversation control unit 300 as a character string signal.
[0083]
Next, the morpheme extraction unit 410 performs a step of extracting each morpheme constituting the minimum unit of the character string based on the character string specified by the speech recognition unit 200 (S103).
[0084]
Specifically, the morpheme extraction unit 410 to which the character string signal is input from the management unit 310 includes a character string corresponding to the input character string signal and nouns, adjectives, verbs, and the like stored in advance in the morpheme database 450. The morpheme group is collated, each morpheme (m1, m2,...) Matching the morpheme group is extracted from the character string, and each extracted morpheme is output to the phrase analysis unit 420 as an extraction signal.
[0085]
Then, the phrase analysis unit 420 collects each morpheme into a phrase format based on each morpheme extracted by the morpheme extraction unit 410 (S104). Specifically, the phrase analysis unit 420 to which the extraction signal is input from the morpheme extraction unit 410, as shown in FIG. 4, is based on each morpheme corresponding to the input extraction signal (for example, a dependency element of each morpheme (for example, Is extracted, and morphemes are grouped into clauses based on the extracted dependency elements. In the present embodiment, the first morpheme information means each morpheme belonging to one phrase.
[0086]
The phrase analysis unit 420 that summarizes each morpheme into each phrase includes a sentence structure analysis unit 430 and an utterance type determination unit 440 using sentence pattern information including each phrase that combines each morpheme and each morpheme constituting each phrase as a sentence pattern signal. Output to.
[0087]
Thereafter, the sentence structure analysis unit 430 performs a step of classifying each morpheme of the first morpheme information segmented by the phrase analysis unit 420 into attributes such as subject case and target case (S105). Specifically, the sentence structure analysis unit 430 to which the sentence pattern signal is input from the phrase analysis unit 420, based on each morpheme corresponding to the input sentence pattern signal and a phrase composed of each morpheme, Determine the “case composition”.
[0088]
That is, as shown in FIG. 5, the sentence structure analysis unit 430, for example, when the dependency element of each morpheme in the phrase is “or” or “is”, the morpheme before the dependency element is the subject. (Subject or subject). In addition, for example, when the dependency element of each morpheme in the phrase is “NO” or “NO”, the sentence structure analysis unit 430 determines that the morpheme before the dependency element is an object (target). To do.
[0089]
Further, for example, when the dependency element of each morpheme in the clause is “Yes”, the sentence structure analysis unit 430 determines that the morpheme before the dependency element is an action (predicate; this predicate is a verb, adjective, etc.). It is determined.
[0090]
The sentence structure analysis unit 430 that has determined the “case structure” of each morpheme that constitutes each clause determines the range of topics (topics) to be described later based on the first morpheme information associated with the determined “case structure”. A topic search command signal for specifying is output to the topic search unit 320.
[0091]
Next, the utterance type determination unit 440 performs a step of specifying an utterance type indicating the type of utterance content based on the phrase specified by the phrase analysis unit 420 (S106). Specifically, the utterance type determination unit 440 sets the “spoken sentence type” (utterance type) based on each morpheme corresponding to the sentence pattern signal input from the phrase analysis unit 420 and a phrase composed of each morpheme. judge.
[0092]
That is, the utterance type determination unit 440 collates each phrase with each dictionary stored in the utterance type database 460 based on each phrase corresponding to the input sentence pattern signal, and from each phrase, Extract sentence elements related to the dictionary. The utterance type determination unit 440 that extracts sentence elements related to each dictionary from each phrase determines “spoken sentence type” (utterance type) based on the extracted sentence elements.
[0093]
The utterance type determination unit 440 outputs an answer search command signal for causing the corresponding user to search for a specific answer sentence to the answer sentence search unit 330 based on an instruction from the topic search unit 320 described later.
[0094]
Next, the topic search unit 320 collates the first morpheme information extracted by the phrase analysis unit 420 with each second morpheme information, and includes the morpheme constituting the first morpheme information from each second morpheme information. A step of searching for second morpheme information (topic title) is performed (S107).
[0095]
Specifically, the topic search unit 320 to which the topic search command signal is input from the sentence structure analysis unit 430, based on the first morpheme information included in the input topic search command signal, the first morpheme information and the conversation database 500. Is compared with the discourse range group stored in, and the discourse range related to the first morpheme information is searched from the discourse range group.
[0096]
For example, when each morpheme (first morpheme information) that belongs to the “case structure” constituting the utterance sentence uttered by the user is (Sato; *; I like) {I like Sato}, the topic search unit Since “Sato” is included in the “case composition”, 320 compares this “Sato” with the discourse range group, and searches for a discourse range (Sato) that matches “Sato”.
[0097]
Further, the topic search unit 320 that has selected the discourse range related to “case configuration” searches for the “topic title” closest to each morpheme belonging to “case configuration” from among the topic titles belonging to the selected discourse range. The search result is output as a search result signal to the answer sentence search unit 330 and the utterance type determination unit 440.
[0098]
For example, when “case composition” is (Sato; *; I like) {I like Sato}, the topic search unit 320, as shown in FIG. 12, each morpheme (Sato, which belongs to the above “case composition”). ; *; I like) and each topic title 1-1 to 1-4 belonging to the discourse range (Sato), and each morpheme belonging to “case composition” from among the topic titles 1-1 to 1-4 A topic title 1-1 (Sato; *; likes) that matches (or approximates) (Sato; *; likes) is searched, and the search result signal 330 is used as a search result signal and an utterance type determination unit. Output to 440.
[0099]
The utterance type determination unit 440, to which the search result signal is input from the topic search unit 320, based on the input search result signal, answers search command signal for causing the corresponding user to search for a specific answer sentence (this answer) The search command signal also includes the determined “sentence sentence type”) to the answer sentence search unit 330.
[0100]
Next, the answer sentence search unit 330, based on the second morpheme information (topic title) searched by the topic search unit 320, the identified utterance type of the user and each answer type associated with the second morpheme information, Are searched, a response type matching the user's utterance type is searched from among the response types, and an answer sentence associated with the searched response type is obtained (S108).
[0101]
Specifically, the answer sentence search unit 330 to which the search result signal is input from the topic search unit 320 and the answer search command signal is input from the utterance type determination unit 440 is the topic title (second second) corresponding to the input search result signal. Morphological information) and the “speech sentence type” (utterance type) corresponding to the answer search command signal, from the answer sentence group (each answer content) associated with the topic title, An answer sentence having an answer type (this answer type means "type of answer sentence" shown in FIG. 11) matching "type" (DA, IA, CA, etc.) is searched.
[0102]
For example, when the topic title corresponding to the search result is the topic title 1-1 (Sato; *; I like) shown in FIG. 12, the answer sentence search unit 330 is associated with the topic title 1-1. Of the reply sentences 1-1 (DA, IA, CA, etc.), the reply composed of the reply type (DA) that matches the “spoken sentence type” (for example, DA; utterance type) determined by the utterance type determination unit 440. The sentence 1-1 (DA; (I also like Sato) is searched), and the searched answer sentence is output to the management unit 310 as an answer sentence signal.
[0103]
Then, the management unit 310 to which the answer sentence signal is input from the answer sentence search unit 330 outputs the input answer sentence signal to the output unit 600. Thereafter, the output unit 600 to which the response sentence signal is input from the management unit 310 outputs an answer sentence {for example, I also like Sato} corresponding to the input response sentence signal (S109).
[0104]
(Operation and effect of conversation control device and conversation control method)
According to the invention according to the present application having the above-described configuration, the topic search unit 320 includes a second morpheme (element constituting the utterance content of the user) constituting the first morpheme information from the second morpheme information. Since the morpheme information is searched, and the answer content associated with the second morpheme information can be acquired based on the searched second morpheme information, the topic search unit 320 includes each morpheme constituting the utterance content of the user. Based on the meaning space (meaning grasped from the character string made up of each morpheme), the answer contents created in advance based on the meaning space can be obtained, It is possible to acquire the answer contents more suitable for the utterance contents than to obtain the answer contents associated with the keyword using the entire utterance contents as a keyword.
[0105]
Further, since the conversation control device 1 can search for the optimum answer content suitable for the user's utterance content based on each morpheme constituting the user's utterance content, development for developing the conversation control device 1 The user only needs to create the answer contents based on the semantic space constructed by each morpheme and the search function for searching the answer contents. Artificial intelligence and neural network to analyze the user's utterance contents There is no need to build a program consisting of a network or the like.
[0106]
Furthermore, since the topic search unit 320 searches for the second morpheme information including the first morpheme information, it is not necessary to search for the second morpheme information that completely matches the utterance content of the user, and the conversation control device 1 is developed. The developer who does not need to memorize a huge amount of answer contents corresponding to the utterance contents that will be uttered by the user in advance, and can reduce the capacity of the storage unit.
[0107]
Furthermore, the answer sentence search unit 330 searches the answer type (description, affirmation, location, repulsion, etc.) associated with each second morpheme information for an answer type that matches the user's utterance type, and performs a search. Since the answer contents associated with the answer type can be acquired based on the answer type, the answer sentence search unit 330 can utter the utterance types constituting the conversation contents of the user, for example, the user simply gives an opinion. Response content that matches the type of utterance from multiple response content based on what the user has, what the user has to say, what the user describes the location element, etc. A more optimal answer can be given to the corresponding user.
[0108]
(Example of change)
In addition, this invention is not limited to the said embodiment, The change as shown below can be added.
[0109]
(First change example)
In this modification, the conversation database 500 may be an element storage unit that stores a plurality of pieces of element information indicating the entire set group including a plurality of morpheme sets in association with the set group. Further, the morpheme extraction unit 410 collates the morpheme extracted from the character string with each set group, selects a set group including the extracted morpheme from each set group, and selects an element associated with the selected set group. Information may be extracted as first morpheme information.
[0110]
As shown in FIG. 14, some morphemes included in the character string uttered by the user are similar. For example, as shown in FIG. 14, if the element information indicating the entire group is “gift”, the “gift” is similar to a present, a gift, a year-end gift, a mid-year gift, a celebration, etc. (group) Therefore, when there is a morpheme similar to “gift” (such as the present), the morpheme extraction unit 410 can handle the similar morpheme as “gift”.
[0111]
That is, for example, when the morpheme extracted from the character string is “present”, the morpheme extraction unit 410 has “present” as the element information representing “present” as shown in FIG. "Gift" can be replaced with "Gift".
[0112]
This allows the morpheme extraction unit 410 to sort out morphemes that are similar to each other, so that the developer who develops the conversation control device can perform the first corresponding to the semantic space that is grasped from the mutually similar first morpheme information. It is not necessary to create reply contents related to the bimorphic information and the second morpheme information one by one, and as a result, the amount of data stored in the storage unit can be reduced.
[0113]
(Second modified example)
As shown in FIG. 15, in the present modification example, the topic search unit 320 may include a ratio calculation unit 321 and a selection unit 322.
[0114]
The ratio calculation unit 321 collates the first morpheme information extracted by the morpheme extraction unit 410 with each second morpheme information (topic title), and for each second morpheme information, the first morpheme information is first compared with the second morpheme information. It is a calculation means for calculating the proportion of morpheme information.
[0115]
Specifically, the ratio calculation unit 321 to which the topic search command signal is input from the sentence structure analysis unit 430, based on the first morpheme information included in the input topic search command signal, as shown in FIG. One morpheme information is compared with each topic title (second morpheme information) belonging to the discourse range stored in the conversation database 500, and the first morpheme information occupies in each topic title for each topic title. Calculate the percentage.
[0116]
For example, as shown in FIG. 15, when the first morpheme information constituting the utterance sentence uttered by the user is (Sato; *; I like) {I like Sato}, the ratio calculation unit 321 Each morpheme (Sato; *; I like) belonging to the “case structure” is matched with each morpheme (Sato; *; I like) included in the topic title. The ratio that each morpheme (Sato; *; I like) belonging to “composition” is included is calculated to be 100%. The ratio calculation unit 321 performs these calculations for each topic title, and outputs each calculated ratio to the selection unit 322 as a ratio signal.
[0117]
The selection unit 322 is a selection unit that selects one second morpheme information from each second morpheme information according to the size of each rate calculated by the rate calculation unit 321 for each second morpheme information. is there.
[0118]
Specifically, the selection unit 322 to which the ratio signal is input from the ratio calculation unit 321 selects from among the ratios included in the input ratio signal (element of “case configuration” / element of “topic title” × 100). For example, a topic title with a high ratio is selected (see FIG. 16). The selection unit 322 that has selected the topic title with a high ratio outputs the selected topic title to the answer sentence search unit 330 and the utterance type determination unit 440 as a search result signal. The answer sentence search unit 330 acquires an answer sentence associated with the topic title based on the topic title selected by the selection unit 322.
[0119]
Accordingly, the ratio calculation unit 321 calculates the ratio of the first morpheme information to the second morpheme information for each second morpheme information, and the selection unit 322 is calculated for each second morpheme information. Since one second morpheme information can be selected from each second morpheme information according to the size of each ratio, the selection unit 322, for example, selects the first morpheme information (the user's utterance content). If the second morpheme information with a large proportion of the second morpheme information can be acquired from a plurality of second morpheme information groups, the first one that follows the semantic space composed of the first morpheme information. Bimorphic information can be acquired more accurately.
[0120]
As a result, the answer sentence search unit 330 can acquire the answer sentence associated with the second morpheme information based on the second morpheme information acquired by the selection unit 322. Can obtain the best answer sentence.
[0121]
Moreover, since the selection part 322 can select a topic title with a high ratio calculated by the ratio calculation part 321 from a plurality of topic titles, it belongs to the “case configuration” included in the user's utterance sentence. Even if each morpheme and each topic title stored in the conversation database 500 do not completely match, it is possible to acquire a topic title closely related to each morpheme belonging to “case composition”.
[0122]
As a result, since the selection unit 322 acquires topic titles closely related to “case configuration”, the developer who develops the conversation control apparatus 1 stores topic titles that completely match “case configuration” in the conversation database 500 one by one. Therefore, the capacity of the conversation database 500 can be reduced.
[0123]
The ratio calculation unit 321 collates each morpheme of the first morpheme information belonging to each classified attribute and each morpheme of each second morpheme information belonging to each attribute stored in advance for each attribute, It may be a first search means for searching for second morpheme information including each morpheme of the first morpheme information in at least one attribute from the second morpheme information.
[0124]
Specifically, the ratio calculation unit 321 to which the topic search command signal is input, for each “case” (subject; object; action) of the “case configuration” included in the input topic search command signal. Morphemes of the first morpheme information belonging to "" and each morpheme belonging to "case" of the topic title consisting of the same "case" as "case structure" are collated, and the morphemes constituting each "case" are the same It is determined whether or not.
[0125]
For example, as illustrated in FIG. 17, when the “case” morpheme of “case configuration” is (dog; person; bite) {dog bites a person}, the ratio calculation unit 321 displays those morphemes. Match "dog", "person", and "chewed" with the morpheme "dog", "person", and "chewed" of the topic title consisting of the same "case" that constitutes the morpheme Then, among the morphemes “dog”, “people”, and “chewing” that make up the topic title, the “case composition” morpheme “dog” consisting of the same “case” corresponding to each morpheme, The ratio of “people” and “biting” is calculated (100%).
[0126]
If the element constituting the topic title is (person; dog; bite) {person bites the dog}, the ratio calculation unit 321 uses the same procedure as described above, and the morphemes belonging to the two cases are calculated. Since they are different, the degree of coincidence for each “case” between the morpheme constituting “case composition” and “topic title” is calculated to be 33% (see FIG. 17).
[0127]
The ratio calculation unit 321 that has calculated the ratio selects a topic title with a high ratio from each ratio, and outputs the selected topic title to the answer sentence search unit 330 and the utterance type determination unit 440 as a search result signal.
[0128]
As a result, the ratio calculation unit 321 displays each morpheme of the first morpheme information belonging to each classified “case configuration” (subject case, target case, etc.) and a pre-stored topic title for each “case”. Since the second morpheme information including each morpheme of the first morpheme information in at least one “case” can be searched from the topic titles, the ratio calculation unit 321 is different from the normal word order. If the utterance content is composed of things, for example, “a person bites a dog”, the morpheme of the subject case is “person” and the morpheme of the subject case is “dog”. The matching second morpheme information can be searched, and the reply content {"really?" Or "I don't know the meaning"} associated with the second morpheme information can be acquired.
[0129]
That is, since the ratio calculation unit 321 can identify utterance contents that are difficult to identify, for example, “a person bites a dog” and “a dog bites a person”, an answer that is more suitable for the identified utterance contents, For the former, for example, “Really?”, For the latter, for example, “Are you okay?”
[0130]
(Third change example)
As shown in FIG. 18, the present modification example is not limited to the above-described embodiment and each modification example, and the communication unit 800 in the conversation control devices 1 a and 1 b and the communication unit via the communication network 1000. A communication unit 900 for transmitting / receiving data to / from 800, conversation databases 500b to 500d connected to the communication unit 900, and servers 2a to 2c may be provided (conversation control system).
[0131]
Here, the communication network 1000 means a communication network that transmits and receives data. In the present embodiment, for example, the Internet is used.
[0132]
In this modified example, the conversation control devices 1a and 1b, the conversation databases 500b to 500d, and the servers 2a to 2c are limited for convenience. However, the present invention is not limited to this, and another conversation database is provided. Also good. The servers 2a to 2c store contents similar to those stored in the conversation databases 500a to 500d.
[0133]
Thereby, the conversation control unit 300 not only includes the conversation database 500a arranged inside the conversation control device 1a but also the other conversation control device 1b, the conversation databases 500b to 500d, and the server 2a to the communication database 1000. 2c can be referred to, for example, it is possible to search the conversation database 500a for the discourse range related to each morpheme (first morpheme information) belonging to “case structure” included in the topic search command signal. Even if it is not possible, by referring to the other conversation control device 1b, the conversation databases 500b to 500d, and the servers 2a to 2c, it is possible to search the discourse range related to the first morpheme information, It is possible to search for an answer sentence more suitable for the utterance sentence.
[0134]
[program]
The contents described in the conversation control system and the conversation control method can be realized by executing a dedicated program for using a predetermined program language in a general-purpose computer such as a personal computer.
[0135]
Here, as the programming language, in this embodiment, the topic requested by the user, the degree of emotion held for the user in a certain matter, or the type of statement, affirmative sentence, question sentence, repulsive sentence, etc. The language for hierarchically storing in the database in association with the morpheme, for example, DKML (Discourse Knowledge Markup Language) developed by the present inventors, and other C languages.
[0136]
That is, the conversation control apparatus 1 stores data (stored information such as second morpheme information, answer text, answer type, set group, element information, etc.) stored in each of the conversation databases 500a to 500d, and other parts as DKML ( It can be realized by executing a program that uses the stored information and the like constructed by Discourse Knowledge Markup Language).
[0137]
According to such a program according to the present embodiment, each morpheme constituting the utterance content of the user is identified, the semantic content grasped from each identified morpheme is analyzed, and associated with the analyzed semantic content. A conversation control device, a conversation control system, and a conversation control method that have the effect of being able to output the optimum answer contents corresponding to the user's utterance contents by outputting the answer contents prepared in advance. It can be easily realized by a general purpose computer.
[0138]
Furthermore, communication between the communication unit 800 and the communication unit 900 may be performed by transmitting and receiving data using a protocol such as DKML via a communication network. Thereby, for example, when there is no answer content suitable for the user's utterance content in the conversation control device, the conversation control device 1 is suitable for the user's utterance content through the communication network 1000 according to the convention such as DKML. It is possible to search for the response content (described in DKML or the like) and acquire the searched response content.
[0139]
The program can be recorded on a recording medium. As shown in FIG. 19, examples of the recording medium include a hard disk 1100, a floppy (registered trademark) disk 1200, a compact disk 1300, an IC chip 1400, and a cassette tape 1500. According to the recording medium on which such a program is recorded, the program can be easily stored, transported, sold, and the like.
[0140]
[Second Embodiment]
(Basic configuration of conversation control system)
A second embodiment of the present invention will be described with reference to the drawings. FIG. 20 is a diagram showing an internal structure of the conversation control system according to the present embodiment. As shown in the figure, the conversation control device 1 in the conversation control system is substantially the same as the internal structure of the conversation control device 1 in the first embodiment, but is different in that it has a morpheme addition unit 340. Since it is the same as the structure of 1st embodiment and a modified example except this difference, description about structures other than a difference is abbreviate | omitted.
[0141]
In the first embodiment, the conversation control device 1 has been described with respect to the process of acquiring the optimum answer content corresponding to the input information based on the input information from the user. However, in the present embodiment, the conversation control device 1 A process for adding the morpheme constituting the input information to the optimum answer content corresponding to the input information based on the input information from the user will be described. The specific explanation is as follows.
[0142]
The morpheme adding unit 340 configures at least the first morpheme information extracted by the morpheme extraction unit 410 or the second morpheme information searched by the topic search unit 320 based on the answer content acquired by the answer sentence search unit 330. This is a morpheme adding means for adding one morpheme to the response content.
[0143]
In addition, the morpheme adding unit 340 retrieves (acquires) the morpheme belonging to at least one attribute classified by the phrase analysis unit 420 (this morpheme belongs to the first morpheme information) by the reply sentence retrieval unit 330. It is also a morpheme adding means for adding to the contents. Furthermore, the morpheme adding unit 340 is also an answer adding unit that adds a morpheme belonging to at least one attribute constituting the second morpheme information searched by the topic search unit 320 to the answer content.
[0144]
Specifically, the morpheme adding unit 340 to which the response content from the response text search unit 330 and the first morpheme information extracted by the morpheme extraction unit 410 are input is based on the input response content and the first morpheme information. Thus, at least one morpheme constituting the first morpheme information is added to the response content.
[0145]
In addition, the morpheme adding unit 340 to which the response content from the response text search unit 330 and the second morpheme information from the topic search unit 320 are input, based on the input response content and the second morpheme information. At least one morpheme constituting the information is added to the response content.
[0146]
The morpheme adding unit 340 that adds the morpheme to the answer content outputs the answer content added with the morpheme to the management unit 310 as the added answer content. The management unit 310 that has received the additional response content from the morpheme adding unit 340 outputs the input additional response content to the output unit 600.
[0147]
For example, when the input information from the user is {Tell me the new singer name of B singer}, the morpheme extraction unit 410 includes “B singer”, “new syllabary name”, “tell me” among the input information. Are extracted as first morpheme information. The answer sentence search unit 330 acquires the answer content associated with the second morpheme information based on the second morpheme information that matches the first morpheme information searched by the topic search unit 320, The morpheme “B singer” and “new title” constituting the first morpheme information are added.
[0148]
The phrase analysis unit 420 classifies each morpheme belonging to the first morpheme information extracted by the morpheme extraction unit 410 into attributes such as subject case (subject), target case (object), action / verb (action). . The first morpheme information classified into each attribute is (B singer; new score; tell me) {this is the order of subject, object, action}. Since the operations of the morpheme extraction unit 410 and the phrase analysis unit 420 here are the same as those in the first embodiment, a detailed description of the operation is omitted.
[0149]
Based on the first morpheme information (B singer; new score; tell me) classified into each attribute, the topic search unit 320 searches for second morpheme information that matches the first morpheme information from each second morpheme information. get. Based on the acquired second morpheme information, the answer sentence search unit 330 acquires the response content associated with the second morpheme information from the conversation database 500, and the acquired response content includes the first morpheme information ( Or at least one morpheme belonging to each attribute constituting the second morpheme information).
[0150]
Here, as shown in FIG. 21, the conversation database 500 stores a plurality of answer contents as an answer contents table. As shown in the figure, in this answer content table, in this embodiment, a singer name (A singer, B singer, C singer ...), and a new song sung on a commercially available CD ( Phrase 1, Phrase 2, Phrase 3 ...), the singer's office (Office 1, Office 2, Office 3 ...), the scheduled date for the concert performed by the singer (July 1, 2002,・ ・), Ranking of new music (1st place, 10th place, 17th place ...) etc. are included.
[0151]
Specifically, when each morpheme constituting the first morpheme information or the second morpheme information is (B singer; new score; teach), the answer sentence search unit 330 selects the subject morpheme from the morpheme. The morpheme “B singer” and the object morpheme “new music” are extracted. The answer sentence search unit 330 may extract one of the morphemes belonging to the subject or the object from the morphemes constituting the first morpheme information or the second morpheme information.
The answer sentence search unit 330 that extracted the subject morpheme “B singer” and the object morpheme “new music” refers to the answer content table stored in the conversation database 500 based on each extracted morpheme, and FIG. As shown in FIG. 3, the answer content “Song Title 2” (shaded portion shown in FIG. 4) corresponding to the morpheme “B Singer” and the morpheme “New Title” is acquired. The response sentence search unit 330 that has acquired the response content “song name 2” outputs the acquired response content “song name 2” to the morpheme adding unit 340.
[0152]
The morpheme adding unit 340 to which the response content “song name 2” is input from the response text search unit 330 is based on the input response content “song name 2” and includes at least one of the first morpheme information and the second morpheme information. Add morphemes.
[0153]
Specifically, the morpheme adding unit 340 receives, for example, the morpheme “B singer” and the morpheme “new score” belonging to the subject and the object constituting the first morpheme information or the second morpheme information, as shown in FIG. Is added to the response content “Song Title 2”. The morpheme adding unit 340 can add each morpheme “B singer” and “new score” to the input response content “song name 2”, so that “B singer's new score is song name 2” is output as an output response can do.
[0154]
(Conversation control method using conversation control device)
The conversation control method by the conversation control apparatus 1 having the above configuration can be implemented by the following procedure. The procedure up to acquiring the reply content can be performed in the same procedure as the procedure (S101 to S109) in the first embodiment. For this reason, in the conversation control method described below, the flow after the response content is acquired will be mainly described.
[0155]
The morpheme adding unit 340 forms at least the first morpheme information extracted by the morpheme extraction unit 410 or the second morpheme information searched by the topic search unit 320 based on the answer content acquired by the answer sentence search unit 330. A step of adding one morpheme to the response content is performed.
[0156]
Specifically, the morpheme adding unit 340 to which the response content from the response text search unit 330 and the first morpheme information extracted by the morpheme extraction unit 410 are input is based on the input response content and the first morpheme information. Thus, at least one morpheme constituting the first morpheme information is added to the response content.
[0157]
In addition, the morpheme adding unit 340 to which the response content from the response text search unit 330 and the second morpheme information from the topic search unit 320 are input, based on the input response content and the second morpheme information. At least one morpheme constituting the information is added to the response content.
[0158]
The morpheme adding unit 340 that adds the morpheme to the answer content outputs the answer content added with the morpheme to the management unit 310 as the added answer content. The management unit 310 that has received the additional response content from the morpheme adding unit 340 outputs the input additional response content to the output unit 600.
[0159]
For example, when the input information from the user is {Tell me the new singer name of B singer}, the morpheme extraction unit 410 includes “B singer”, “new syllabary name”, “tell me” among the input information. Are extracted as first morpheme information. The answer sentence search unit 330 acquires the answer content associated with the second morpheme information based on the second morpheme information that matches the first morpheme information searched by the topic search unit 320, The morpheme “B singer” and “new title” constituting the first morpheme information are added.
[0160]
The phrase analysis unit 420 classifies each morpheme belonging to the first morpheme information extracted by the morpheme extraction unit 410 into attributes such as subject case (subject), target case (object), action / verb (action). . The first morpheme information classified into each attribute is (B singer; new score; tell me) {this is the order of subject, object, action}.
[0161]
And the topic search part 320 is based on the 1st morpheme information (B singer; new score; tell me) classified into each attribute, and the 2nd morpheme which corresponds with 1st morpheme information from each 2nd morpheme information. Get information. Based on the acquired second morpheme information, the answer sentence search unit 330 acquires the response content associated with the second morpheme information from the conversation database 500, and the acquired response content includes the first morpheme information ( Or at least one morpheme belonging to each attribute constituting the second morpheme information).
[0162]
Specifically, when each morpheme constituting the first morpheme information or the second morpheme information is (B singer; new score; teach), the answer sentence search unit 330 selects the subject morpheme from the morpheme. The morpheme “B singer” and the object morpheme “new music” are extracted.
[0163]
Then, the answer sentence search unit 330 that extracts the subject morpheme “B singer” and the object morpheme “new score” refers to the answer content table stored in the conversation database 500 based on each extracted morpheme, As shown in FIG. 21, the response content “song name 2” (the hatched portion shown in the figure) corresponding to the morpheme “B singer” and the morpheme “new musical title” is acquired. The response sentence search unit 330 that has acquired the response content “song name 2” outputs the acquired response content “song name 2” to the morpheme adding unit 340.
[0164]
After that, the morpheme adding unit 340 to which the response content “song name 2” is input from the response text search unit 330 forms at least the first morpheme information or the second morpheme information based on the input response content “song name 2”. Add one morpheme.
[0165]
Specifically, the morpheme adding unit 340 receives, for example, the morpheme “B singer” and the morpheme “new score” belonging to the subject and the object constituting the first morpheme information or the second morpheme information, as shown in FIG. Is added to the response content “Song Title 2”. The morpheme adding unit 340 can add each morpheme “B singer” and “new score” to the input response content “song name 2”, so that “B singer's new score is song name 2” is output as an output response can do.
[0166]
(Operation and effect of conversation control device and conversation control method)
According to the invention according to this application, the answer sentence search unit 330 acquires the answer content associated with the second morpheme information based on the second morpheme information searched by the topic search unit 320, and adds the morpheme. Since the unit 340 can add at least one morpheme constituting the first morpheme information or the second morpheme information to the answer content based on the answer content acquired by the answer text search unit 330, the morpheme adding unit 340 Can output the response content corresponding to the input information in a natural format.
[0167]
For example, when the user inputs input information “Is there a new record for A singer?”, The answer sentence search unit 330 acquires the answer content “song name 1” corresponding to the input information, and the morpheme addition unit 340 can add the morpheme “new singer's score” constituting the input information to the reply content “song name 1” acquired by the reply sentence search unit 330. "" Is "Song name 1""can be output.
[0168]
As a result, since the answer that directly answers the input information input by the user is output from the conversation control device 1, the user can feel as if the user is having a conversation with another user. be able to.
[0169]
Moreover, the topic search unit 320 searches the second morpheme information from each second morpheme information, and searches for second morpheme information including the morpheme constituting the first morpheme information (elements constituting the input information from the user). Since the unit 330 can acquire the content of the answer associated with the second morpheme information based on the second morpheme information searched by the topic search unit 320, the answer sentence search unit 330 receives the input from the user. Based on each morpheme that constitutes information, it is possible to obtain the answer content created in advance based on the semantic space constructed by each morpheme (meaning content grasped from the character string made up of each morpheme). It is possible to acquire the answer contents more suitable for the input information than to acquire the answer contents associated with the keyword using “whole” of the input information as a keyword.
[0170]
Further, the answer sentence search unit 330 can search for the answer content most suitable for the input information of the user based on each morpheme constituting the input information from the user. A person only needs to create an answer content based on a semantic space constructed from each morpheme and a search function for searching the answer content, artificial intelligence for analyzing input information from the user, There is no need to construct a program consisting of a neural network or the like.
[0171]
Furthermore, the topic search unit 320 does not search for the second morpheme information that completely matches the input information from the user, but searches for the second morpheme information including the first morpheme information. The developer who develops need not store a large amount of answer contents corresponding to input information that will be input from the user in advance, and can reduce the capacity of the storage unit.
[0172]
The phrase analysis unit 420 classifies each morpheme constituting the extracted first morpheme information into attributes such as a subject case consisting of a main case and a target case consisting of a target, and a morpheme addition unit 340 includes a phrase analysis unit Since the morpheme belonging to at least one “attribute” classified in 420 can be added to the answer content, the conversation control apparatus 1 supports the input information based on the answer content to which the morpheme belonging to a certain attribute is added. Can be output in a more natural form, that is, in an interactive form.
[0173]
For example, when the user inputs input information “Is there a new singer A”, the conversation control device acquires the response content “Song Name 1” corresponding to the input information, and this response content “Song Name”. Since only the morpheme “new score” (target morpheme) that constitutes the input information can be added to “1”, ““ new score ”is“ song name 1 ”” can be output as an answer to be output, As a result, it can be output in a more natural format suitable for the conversation content.
[0174]
(Example of change)
In addition, this invention is not limited to said 2nd embodiment, The topic search part 320 can also perform the following functions. The topic search unit 320 collates the first morpheme information extracted by the morpheme extraction unit 410 with each keyword group, acquires a keyword that matches at least one morpheme constituting the first morpheme information, and sets the acquired keyword It is also a first search means for searching for the associated second morpheme information.
[0175]
Further, when searching for two or more second morpheme information associated with a keyword, the topic search unit 320 includes at least one morpheme constituting the first morpheme information from each second morpheme information. It is also a first search means for searching for second morpheme information.
[0176]
Here, as shown in FIG. 23, the second morpheme information has a keyword group associated with any one morpheme constituting the second morpheme information in the present embodiment. A plurality of morpheme information is stored in the conversation database 500.
[0177]
Specifically, the topic search unit 320, for example, if the first morpheme information extracted by the morpheme extraction unit 410 is (B singer; new score; tell me) {tell B singer's new score} Each morpheme of one morpheme information is matched with each keyword group ("general singer", "new singing", etc.) shown in the figure, and a keyword "B singer" that matches the morpheme "B singer" is obtained.
[0178]
The topic search unit 320 that has acquired the keyword “B singer” that matches the morpheme “B singer” uses the second morpheme information (general singer) associated with the keyword “B singer” based on the acquired keyword “B singer”. Select new music; tell me) (see FIG. 23).
[0179]
If the topic search unit 320 has acquired the keyword “B singer” that matches the morpheme “B singer”, the second morpheme information associated with the acquired keyword “B singer” is two as shown in FIG. In some cases, of these two second morpheme information (general singer; new score; tell me), (general singer; sing new song; tell me), the first morpheme information (B singer; new score; tell me) ), The second morpheme information (general singer; new score; tell me) close to each morpheme is selected. This selection is performed depending on whether at least one morpheme constituting the first morpheme information is included in the second morpheme information.
[0180]
According to the invention according to this embodiment, the topic search unit 320 collates the extracted first morpheme information with each keyword group, and matches the keyword with at least one morpheme constituting the first morpheme information. Since the second morpheme information associated with the acquired keyword can be searched, the developer who develops the conversation control device 1 can talk to the second morpheme information that approximates the first morpheme information. It is not necessary to store the information in the control device one by one, and the storage amount of the second morpheme information can be reduced.
[0181]
In addition, when the topic search unit 320 searches for two or more second morpheme information associated with the keyword, the topic search unit 320 includes at least one morpheme constituting the first morpheme information from each second morpheme information. Since the second morpheme information can be searched, the topic search unit 320 searches the second morpheme information for more appropriate second morpheme information related to the input information input from the user. Can do.
[0182]
Furthermore, since the developer can reduce the second morpheme information stored in the conversation database 500 by associating the keyword group with the second morpheme information, the topic search unit 320 can select the second morpheme information. The second morpheme information approximate to the first morpheme information can be searched quickly.
[0183]
【The invention's effect】
As described above, according to the present invention, the conversation control device identifies each morpheme constituting the input information input from the user, analyzes the semantic content grasped from each identified morpheme, and performs analysis. By acquiring pre-prepared answer contents that are associated with the meaning contents, the user can obtain the optimum answer contents corresponding to the input information from the user, and input even if each answer contents stored in advance is reduced. The optimum answer content corresponding to the information can be acquired, and the obtained answer content can be output in a natural format closer to the interactive format.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a schematic configuration of a conversation control system according to a first embodiment.
FIG. 2 is a block diagram showing an internal structure of a conversation control unit and a sentence analysis unit in the first embodiment.
FIG. 3 is a diagram showing the contents of each morpheme extracted by a morpheme extraction unit in the first embodiment.
FIG. 4 is a diagram showing the contents of each phrase extracted by a phrase analysis unit in the first embodiment.
FIG. 5 is a diagram showing the contents of “case” specified by the sentence structure analysis unit in the first embodiment.
FIG. 6 is a diagram showing an “uttered sentence type” specified by an utterance type determining unit in the first embodiment.
FIG. 7 is a diagram showing the contents of each dictionary stored in the utterance type database in the first embodiment.
FIG. 8 is a diagram showing the contents of a hierarchical structure built inside the conversation database in the first embodiment.
FIG. 9 is a diagram showing a detailed relationship of a hierarchical structure built inside the conversation database in the first embodiment.
FIG. 10 is a diagram showing the content of a “topic title” constructed within the conversation database in the first embodiment.
FIG. 11 is a diagram showing the content of “answer sentence type” associated with “topic title” built inside the conversation database in the first embodiment.
FIG. 12 is a diagram showing the contents of “topic title” and “answer sentence” belonging to “discourse range” built inside the conversation database in the first embodiment.
FIG. 13 is a flowchart showing a procedure of a conversation control method according to the first embodiment.
FIG. 14 is a diagram showing utterance contents organized by a morpheme extraction unit in the first modification.
FIG. 15 is a diagram illustrating an internal configuration of a topic search unit in a second modified example.
FIG. 16 is a diagram illustrating a state in which the ratio calculation unit in the second modification collates each morpheme belonging to “case configuration” and each “topic title” for each “topic title”.
FIG. 17 is a diagram illustrating a state in which the ratio calculation unit according to the second modification collates each morpheme belonging to “case configuration” and each morpheme belonging to “topic title” for each “case”.
FIG. 18 is a diagram showing a schematic configuration of a conversation control system in a third modified example.
FIG. 19 is a diagram showing a recording medium for storing a program in the first embodiment.
FIG. 20 is a block diagram showing a schematic configuration of a conversation control system according to a second embodiment.
FIG. 21 is a diagram showing the contents of an answer content table stored in a conversation database in the second embodiment.
FIG. 22 is a diagram illustrating a state in which a morpheme adding unit adds a morpheme to an answer content in the second embodiment.
23. Until the topic search unit in the modification collates the first morpheme information with each second morpheme information and selects second morpheme information that approximates the first morpheme information from each second morpheme information. FIG.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Conversation control apparatus, 100 ... Input part, 200 ... Speech recognition part, 300 ... Conversation control part, 310 ... Management part, 320 ... Topic search part, 321 ... Ratio calculation part, 322 ... Selection part, 330 ... Answer sentence search , 400 ... Sentence analysis unit, 410 ... morpheme extraction unit, 420 ... sentence analysis unit, 430 ... sentence structure analysis unit, 440 ... speech type determination unit, 450 ... morpheme database, 460 ... speech type database, 500 ... conversation database, DESCRIPTION OF SYMBOLS 600 ... Output part, 700 ... Voice recognition dictionary memory | storage part, 800 ... Communication part, 900 ... Communication part, 1000 ... Communication network, 1100 ... Hard disk, 1200 ... Floppy (registered trademark) disk, 1300 ... Compact disk, 1400 ... IC chip 1500 ... cassette tape

Claims (14)

Character recognition means for identifying a character string indicating the input information based on the input information input by the user;
Based on the character string specified by the character recognition means, morpheme extraction means for extracting at least one morpheme constituting the minimum unit of the character string as first morpheme information;
Answer storage means for storing a plurality of second morpheme information indicating a morpheme composed of one character, a plurality of character strings, or a combination thereof, and a response content to the user corresponding to the input information in advance ,
The first morpheme information extracted by the morpheme extraction means and the second morpheme information are collated, and the second morpheme includes the second morpheme constituting the first morpheme information. A first search means for searching for morpheme information;
Based on the second morpheme information searched by the first search unit, an answer acquisition unit that acquires the response content associated with the second morpheme information;
At least one of the first morpheme information extracted by the morpheme extraction unit or the second morpheme information searched by the first search unit based on the response content acquired by the response acquisition unit A conversation control apparatus comprising: morpheme adding means for adding a morpheme to the answer content. The conversation control device according to claim 1,
Classification means for classifying each morpheme of the first morpheme information extracted by the morpheme extraction unit into each attribute such as a subject case consisting of a main case and a target case consisting of a target;
The conversation control apparatus, wherein the morpheme adding unit adds the morpheme belonging to at least one attribute classified by the classification unit to the answer content acquired by the answer acquiring unit. The conversation control device according to claim 1,
Each morpheme constituting the second morpheme information is classified in advance for each attribute such as the subject case and the target case and stored in advance in the answer storage unit,
The morpheme adding unit is configured to acquire the morpheme belonging to at least one of the attributes constituting the second morpheme information based on the second morpheme information searched by the first search unit by the answer acquisition unit. A conversation control apparatus characterized by being added to the answer content. The conversation control device according to claim 1,
Each of the second morpheme information is associated with a plurality of keyword groups,
The first search unit collates the first morpheme information extracted by the morpheme extraction unit with each keyword group, and acquires the keyword that matches at least one morpheme constituting the first morpheme information. A conversation control device that searches for the second morpheme information associated with the acquired keyword. The conversation control device according to claim 4,
When the first search means searches for two or more second morpheme information, the second search unit includes at least one of the second morpheme information including at least one morpheme constituting the first morpheme information. A conversation control device characterized by retrieving morpheme information. The conversation control device according to claim 1,
The second morpheme information or the answer content stored in the answer storage means is a topic requested by the user, a feeling level or a statement sentence to the user in a certain matter, an affirmative sentence, a question sentence, a repulsive sentence The conversation control device is characterized in that it is described in a language for hierarchically storing it in the database in association with the morpheme according to its semantic content. The conversation control device according to claim 6,
The conversation control apparatus according to claim 1, wherein the language is DKML. A conversation control method in a computer that functions as character recognition means, morpheme extraction means, first search means, answer acquisition means and morpheme addition means ,
A plurality of second morpheme information indicating a morpheme composed of one character, a plurality of character strings, or a combination thereof, and a response content to the user corresponding to the input information are stored in a plurality of response storage means in association with each other in advance. Steps,
A step of identifying a character string indicating the input information based on the input information input from the user by the character recognition means;
Morpheme extraction means, based on the character string specified by the character recognition means, to extract at least one morpheme constituting the minimum unit of the character string as first morpheme information;
The first search means collates the first morpheme information extracted by the morpheme extraction means with the second morpheme information, and configures the first morpheme information from the second morpheme information. Retrieving the second morpheme information including morphemes;
A step of acquiring an answer content associated with the second morpheme information based on the second morpheme information searched by the first search unit;
The morpheme adding unit configures the first morpheme information extracted by the morpheme extraction unit or the second morpheme information searched by the first search unit based on the response content acquired by the response acquisition unit Adding at least one morpheme to the answer content. 9. The conversation control method in a computer according to claim 8, wherein the computer further functions as a classification means ,
A classifying unit classifying each morpheme of the first morpheme information extracted by the morpheme extracting unit into attributes such as a subject case consisting of a main case and a target case consisting of a target;
Conversation control in a computer, wherein the morpheme adding means includes a step of adding the morpheme belonging to at least one attribute classified by the classifying means to the answer content acquired by the answer acquiring means Method. A conversation control method in a computer according to claim 8,
Each morpheme constituting the second morpheme information is classified in advance for each attribute such as the subject case and the target case and stored in advance in the answer storage unit,
Based on the second morpheme information searched by the first search unit, the morpheme addition unit acquires the morpheme belonging to at least one of the attributes constituting the second morpheme information by the answer acquisition unit. A method for controlling conversation in a computer, further comprising the step of adding to the answer contents. A conversation control method in a computer according to claim 8,
Each of the second morpheme information is associated with a plurality of keyword groups,
The first search unit collates the first morpheme information extracted by the morpheme extraction unit with each of the keyword groups, and acquires the keyword that matches at least one morpheme constituting the first morpheme information. A method for controlling conversation in a computer, comprising the step of retrieving the second morpheme information associated with the acquired keyword. A conversation control method for a computer according to claim 11, comprising:
When the first search means searches for two or more second morpheme information, the second morpheme information includes the second morpheme that includes at least one morpheme constituting the first morpheme information. A conversation control method in a computer comprising a step of retrieving morpheme information. A conversation control method in a computer according to claim 8,
The second morpheme information or the content of the answer stored in the answer storage means is a topic requested by the user, a feeling or a statement sentence to the user in a certain matter, an affirmative sentence, a question sentence, a repulsive sentence A conversation control method in a computer, characterized in that the types are described in a language for hierarchically storing them in a database in association with the morpheme according to the semantic content. A conversation control method for a computer according to claim 13, comprising:
The conversation control method in a computer, wherein the language is DKML.

Images