JP2004102764A - Conversation expression generating apparatus and conversation expression generating program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the degree of understanding of readers and listeners by automatically generating conversation expressions based on described monologue sentences. <P>SOLUTION: This conversation expression generating apparatus A comprises a monologue text acquisition means acquiring a monologue text, a pretreatment means generating one or more simple texts by dividing the acquired monologue text, a sentence end treatment means analyzing the sentence end expression of the generated simple text and allowing the sentence end expression to correspond to either of a plurality of sentence end expression patterns, a comment selection means selecting one comment text corresponding to the simple sentence text allowed to correspond to the any one sentence end expression pattern from a plurality of comment texts set as expressions to respond to the sentence end expression patterns while being related thereto, and a means generating and outputting a conversation text formed of the simple sentence text and the comment text by inserting the selected comment text after the simple sentence text. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a conversation expression generation device capable of suitably generating a conversational expression based on a monologue expression, and a program therefor.
[0002]
[Prior art]
At present, some means of providing information via communication means such as the Internet are accompanied by images, but text communication in Web pages, text communication such as e-mail, chat, electronic bulletin boards, etc. are mainly used. It is. In particular, it can be said that many of the sentences in the Web page are monologous sentences unilaterally described by a single writer. Also, in text communication such as e-mail, in the description (speech) of one writer (speaker), the coordination of a dialogue (conversation) in which another person inserts a question is a face-to-face dialogue in which face-to-face conversations are actually faced. It is considered that each utterance is generally long and has a monolog-like property that is fragmented for each utterance because it is more difficult than in the case of. As described above, at present, a large amount of information stored as monologous sentences exists on the Internet. However, monologue sentences are not a means of presenting information for everyone, and conversational sentences that express important parts in a question-answer format are generally more friendly than specialized monologue sentences, The level of understanding also tends to be high. Here, a conversational form is a form in which multiple speakers stack up conversations, as typified by an interview article or a script of a TV program.A conversational form is compared with a monologue form. Although it is difficult to make objective and strictly structured narratives, it has the advantage of being able to provide concise information focusing on important parts, Because it is the most common means of exchanging information for communication, it is easier to understand than monologous sentences, and it is easier to understand the conversation partner and listeners. Taking this as an example of a TV broadcast news program, it is more familiar to introduce two or more personalities in a conversational form than to announcer reading a news manuscript alone for everyday topics. .
[0003]
As a method of using the conversational expression, for example, by having the same character agents perform a conversation or skit related to the page in response to the user's Web browsing, the user can understand the Web page consistently. There has been an attempt to make Agneta & Frida to have a property (Non-Patent Document 1). An agent (Gerd & Matze) that generates a conversation in accordance with each attitude and accuracy with respect to a team based on game information of a live soccer relay is also considered (Non-Patent Document 2). Further, an agent salon that allows a conversation between agents generated by applying simple rules to personal information of exhibition participants has been considered (Non-Patent Document 3).
[0004]
[Non-Patent Document 1] Christina Hook (H "o" ok, K.) and three other authors, "Handling of Hidden Lutheran Views on Interfaces: Evaluation of Agneta & Fleida System (Dealing with the Lurking Lutheran viewon)" Interfaces: Evaluation of the Agneta and Frida system ", (Sitges, Spain), Workshop on Behavioral Planning Tools for Biosynthetic Characters (the workshop, Beyond Life Insurance Company , P125-136
[Non-Patent Document 2] Elizabeth André (Andr'e, E.), and one other author, "Expression by Performance: Use of Multiple Life Characters in a Knowledge-Based Presentation System (Presenting Through Performing: On the Use of Multiple Lifelike) Characters in Knowledge-Based Presentation Systems), (USA), The Second International Conference on Intelligent User Interfaces (IUI 2000, IUI 2000). 1-8
[Non-Patent Document 3] Kado, Mase, "Agent Salon: Encouragement of Engagement and Dialogue Using Personal Agents'Talking" Transactions of the Institute of Electronics, Information and Communication Engineers, Vol. 84, DI, No. 8, 2001, p1231-1243
[0005]
[Problems to be solved by the invention]
However, the utterance content used in Agneta & Frida is predetermined, and does not generate a new conversational sentence. In Gerd & Matze, the utterance is in response to discrete events related to the movement of the ball in soccer, such as passes and intercepts, and a conversational expression is formed from a monologous sentence describing events that are not restricted in the field. Cannot be generated. Further, since the items used for conversation in the agent salon are discrete items such as the visit history of the individual and the evaluation data for the exhibition, the items correspond to only limited events like Gerd & Matze.
[0006]
Therefore, in any of the conventional modes, it is impossible to generate an easily understandable and familiar conversational expression based on an arbitrary monologue-like sentence.
[0007]
Accordingly, the present invention has been made in view of the above-described problems, and has as its main object to convert any monologue-like sentence, whose field is not limited, into a more familiar and understandable conversational expression. And
[0008]
[Means for Solving the Problems]
That is, as shown in FIG. 1, the conversation expression generation device A1 according to the first embodiment of the present invention generates a conversation expression based on a monolog text composed of monolog texts, and stores the monolog text. A monologue text acquisition unit 1 for acquiring from the obtained monologue text storage unit MTD, a preprocessing unit 2 for dividing the acquired monologue text into a single sentence format to generate one or more single sentence texts, and analyzing a sentence end expression of the generated single sentence texts The end-of-sentence processing means 3 for associating the end-of-sentence expression with one of a plurality of end-of-sentence expression patterns set in advance, and a plurality of comment texts set as expressions corresponding to the end-of-sentence expression patterns in response to the end-of-sentence expression patterns Sentence associated with any one of the end-of-sentence expression patterns from the comment storage unit CMD A comment selecting means 4 for selecting one comment text corresponding to the text, a conversation expression generating means 5 for inserting a selected comment text after the simple text and generating a conversation text composed of the simple text and the comment text; And a conversation text output unit 6 for outputting the conversation text. The computer has the above function by operating a computer based on the program for the conversation expression generation device A1. The same applies to each of the conversation expression generation devices described below in that they function based on dedicated programs.
[0009]
Here, a monolog-like text is a text unilaterally described by a single writer as described above, and a monolog text means text data composed of such a text. The simple sentence text means a text consisting of only one sentence ending at a period. The monologue text storage unit MTD may store the monologue text as a database, or may store the input monologue text in a temporary memory.
[0010]
With such a conversation expression generation device A1, a monologue text that is generally difficult to understand can be automatically converted into a conversation text composed of a conversational expression that is easier to understand. It is possible to reduce the burden on the user for understanding information. In particular, unlike systems that use conventional conversation styles, monologue text is monoculturalized, and by paying attention to the end-of-sentence expression, appropriate conversation text is generated. It is completely different from conventional ones in that it can easily handle monologous sentences that have no restrictions in newly created fields, instead of using sentences as it is or only achieving conversational expressions in fixed fields. And it has a wide range of applications.
[0011]
As shown in FIG. 2, the conversation expression generation device A2 according to the second aspect of the present invention further includes a keyword acquisition unit 7 for acquiring a keyword input by a user, in addition to the configuration of the conversation expression generation device A1. The monologue text obtaining means 1 obtains one or more monologue texts corresponding to the keywords from the monologue text storage unit MTD, and the preprocessing means 2 generates a single sentence text for each of the obtained monologue texts. It is configured to generate.
[0012]
With such a configuration, if the user inputs a keyword of interest, the conversation text generated from the corresponding monologue text is output, so that the user can deepen the understanding of the topic of interest.
[0013]
In such conversational expression generation devices A1 and A2, the monologue text storage unit MTD stores opinion texts posted on the electronic bulletin board by user input as monologue texts. If a monologue text, which is an opinion sentence, is obtained, various people can easily understand the opinion.
[0014]
As shown in FIG. 3, the conversation expression generation device A3 according to the third embodiment of the present invention includes two or more preset speaker agents in addition to the configuration of the conversation expression generation device A1 or A2. On the other hand, further comprising speaker determination means 8 for associating one speaker agent as a reader of a single sentence text in the output conversation text and associating another speaker agent as a reader of a comment text. is there. In the conversation expression generation device A3, the keyword acquisition means 7 shown by a broken line in FIG. 3 is not an indispensable function, and can select whether or not to provide this function as an option. Also, in each of the conversation expression generation devices shown in FIGS. 4 to 7 described below, each means indicated by a broken line indicates that the device or program is an optional function.
[0015]
In such a conversation expression generation device A3, the conversation text in the form of conversation is composed of a simple sentence text and a comment text, and these are distinguished by different speaker agents as readers. In the case where the simple text and the comment text are displayed in characters on the display to be used, by simultaneously displaying the images of any of the speaker agents associated with the simple text or the comment text, two or more people can actually talk. Since it is possible to simulate the state as if the user is interacting, it is extremely easy for the user who is the listener to be familiar with, and the degree of understanding can be improved.
[0016]
A conversation expression generation device A4 according to a fourth embodiment of the present invention, which further improves the effect of the conversation expression generation device A3, includes a speaker determination unit in addition to the configuration of the conversation expression generation device A3, as shown in FIG. Further, there is provided a voice output means 9 for outputting a single sentence text or a comment text corresponding to each of the speaker agents determined in 8 with a different voice. That is, by outputting the single sentence text and the comment text with different voices as if different speaker agents were speaking, the listener's understandability was further improved.
[0017]
The conversation expression generation device A5 according to the fifth aspect of the present invention, which further improves the effect of the conversation expression generation device A4, has a voice output in addition to the configuration of the conversation expression generation device A4 as shown in FIG. Further provided is an animation processing means 10 for adding and outputting an animation operation for moving at least a mouth of an image of the speaker agent to each speaker agent in response to the voice of the simple sentence text or comment text output by the means 9. Things. That is, by operating the image of the speaker agent in addition to the voice data of the single sentence text or the comment text as if the speaker agent is actually speaking, the reality becomes more realistic than in the case of the conversation expression generation device A4. It is possible to proceed with the conversation in an appropriate manner.
[0018]
Further, the conversation expression generation device A6 according to the sixth aspect of the present invention, which has an auxiliary function for any one of the conversation expression generation devices A3, A4, and A5, includes a corresponding single sentence text or comment text together with each speaker agent. Is further provided as character data output means 11 for outputting the character data which can be displayed on the screen. With such a configuration, since the simple sentence text and the comment text output as character data can be displayed on the display in association with each speaker agent, even a user who is deaf can understand the contents. By outputting the information together with the audio data and the animation image, the general user can easily receive the information from the eyes and ears, and can easily understand the contents.
[0019]
Further, it is conceivable that the monologue text accompanies the image data. In this case, the conversation expression generation device A7 according to the seventh aspect of the present invention includes the above-described conversation expression generation devices A1, A2, A3, In addition to any one of A4, A5, and A6, an image data processing unit 12 that acquires and outputs the image data from the monolog text storage unit MTD may be further provided. That is, the output image data can be used as an aid for understanding the conversation text.
[0020]
In particular, in any one of the conversation expression generation devices A3, A4, A5, and A6, a main caster agent serving as a facilitator of the conversation expression is set as one of the speaker agents. When the main caster agent is determined as the reader of the comment text, the conversation expression can be smoothly performed. If the monologue text is composed of a text part which is an essential part of the content and a title part indicating the outline, the main caster agent is set as a reader of the title part by the speaker determining means 8. It may be configured to determine. If a comment text indicating the start of the conversational expression is stored in the comment storage unit CMT, the comment selecting means 4 selects the comment text indicating the start when there is no other single sentence text before the single sentence text. Also, the flow of the conversation can be made smoother by configuring the speaker selection means 8 to associate the comment text with the title part in association with the main caster agent. On the other hand, when one or more announcer agents are set as speaker agents different from the main caster agent, if the speaker determination means 8 determines the announcer agent as a reader of the main body, the main Role assignment with caster agents can be clarified.
[0021]
In any one of the conversation expression generation devices A1 to A7 described above, in order to make the flow of the plurality of conversation expressions based on the plurality of monolog texts smooth, the comment selecting means 4 uses Recognize the final simple sentence text and set the expression to be connected to the next monologue text in association with the end-of-sentence expression pattern in the final simple sentence text, and from the comment text stored in the comment storage to the end-of-sentence expression of the final simple sentence text Preferably, it is configured to select the corresponding comment text.
[0022]
The end-of-sentence expression can be classified into some patterns. In other words, the end-of-sentence expression pattern includes at least a phenomenon description format indicating that the phenomenon is described and a message format indicating that it is a report, and the comment text includes a comment corresponding to the question format corresponding to the phenomenon description format. At least the text and the comment text corresponding to the expected sentence format corresponding to the message format are included, and the sentence end processing unit 3 associates the sentence end expression of the simple sentence text with either the phenomenon description format or the message format, and responds to it. Then, it is preferable that the comment selecting means 4 selects one of the comment texts in the question sentence format or the expected sentence format. In addition, a plurality of comment texts in the question sentence format and the expected sentence format are set, and one of the comment texts is selected by the comment selection means 4 from the plurality of comment texts. In addition, variations can be provided so that conversation is not monotonous.
[0023]
By the way, recently, an interactive broadcasting system for a community called a public opinion channel (hereinafter, referred to as “POC”) is being developed. The POC processes opinion sentence posted by the community members to the other members on the electronic bulletin board, adds conversation generation processing to the opinion sentence, and communicates with the main caster agent who is a virtual speaker. The opinion sentence is broadcasted to community members in the form of an opinion introduction program through conversation of an announcer agent. Therefore, the conversational expression generation devices A1 to A7 according to the present invention can be used as a monologue text storage unit that stores a monologue text input to an electronic bulletin board that can be used by a plurality of users, and the comment storage unit. It has an application that can be applied to an interactive broadcasting system such as a POC that can broadcast a conversation expression text generated based on the input monolog text, and generates a conversation text based on the input monolog text; By outputting the conversation text in a broadcastable manner, it can play a very important role in opinion introduction broadcasting for communities such as POC.
[0024]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0025]
This embodiment is a conversation expression generation apparatus applied to the above-mentioned public opinion channel (hereinafter referred to as “POC”), as shown in FIG. In this embodiment, the conversation expression generation device A7 is used. Hereinafter, the conversation expression generation device A7 is referred to as a POC caster A7. The POC provides a client computer CC such as a personal computer, a PDA or a mobile phone used by each user U who is a community member, an electronic bulletin board that the user U can access from the client computer CC to the client computer CC, and a client computer. It is basically composed of a POC server PS and a POC caster A7 for storing the opinion sentence of the user U posted from the CC, and these client computers CC, POC server PS and POC caster A7 can perform two-way communication through the Internet IN. It is connected to the. The POC caster A7 also has a function as a broadcast client for broadcasting a conversational broadcast generated based on the opinion sentence posted by the user U to the client computer CC. The broadcast can be viewed on the screen of the client computer CC.
[0026]
First, the internal device configuration of each device will be described. The POC server PS is configured by a general-purpose server computer, and has a database server function and a Web server function. Among them, the Web server provides a home page and an electronic bulletin board that can be browsed from the client computer CC. Further, the database server has a function as a monologue text storage unit MTD for storing opinion sentences input to the electronic bulletin board. On the other hand, the POC caster A7 is constructed by a general server computer or personal computer, and as shown in FIG. 9, a CPU 101, an internal memory 102, a storage device 103 such as a hard disk, and an input device such as a keyboard and a mouse. An internal device 104 includes an output device 105 such as a display and a speaker, various communication interfaces 106, and the like. Note that the database device 107 may be further included as an internal device or may be included as an external device. Then, the program recorded in the storage device 103 is read into the internal memory 102 in accordance with the instruction of the CPU 101, and necessary data and the like are read from the database device 107 as appropriate. By performing communication, the POC caster A7 operates. When the POC caster A7 needs to input information or output a screen display or the like, the input device 104 or the output device 105 is appropriately used. Further, in this embodiment, the POC server PS and the POC caster A7 are shown as separate computers capable of bidirectional communication through the Internet IN, but they may be connected by a dedicated communication line, or may be connected to a single computer. It may be realized by a computer. Furthermore, the client computer CC is a general personal computer, PDA, mobile phone, or the like, as described above. Here, at least the function of connecting to the Internet IN, the input / output function of characters and images, the display, etc. And an audio output function such as a speaker.
[0027]
Next, the function of the POC caster A7 will be described. In the POC caster A7, the internal device and the external device cooperate according to the instruction of the CPU 101 based on the conversation expression generation program, and as shown in FIG. Means 1, preprocessing means 2, sentence end processing means 3, comment selection means 4, conversation expression generation means 5, conversation text output means 6, keyword acquisition means 7, speaker determination means 8, voice output means 9, animation processing means 10 , Character data output means 11 and image data processing means 12. By installing a program for operating each of these units in the computer, the computer functions as the POC caster A7. Note that this program is implemented by causing a computer to read a program recorded on a recording medium such as a CD-ROM, or by downloading the program to a computer via the Internet IN or the like. In the present embodiment, in particular, a TSS system (manufactured by Toshiba Corporation) is used for the speech synthesis of agents (main caster agent MA and announcer agent AA), which are virtual speakers who have a conversation of opinion introduction, and images of those agents are used. Uses a photo face character creation system (Sharp Corporation) incorporated in a POC caster A7, but other products having the same function as these can also be used.
[0028]
The database 107 has a function as a comment storage unit CMD. Here, an example of the stored comment is shown in FIG. In this example, the sentence end expression pattern of the opinion sentence posted by the user U is roughly classified into two types, one of which is further classified into three types, and a plurality of comments corresponding to each are prepared. More specifically, as shown in the left column of the figure, the end-of-sentence expression is roughly divided into a phenomenon description form that "states a phenomenon" and a "hearing form" that indicates that it is a hearsay. Furthermore, the phenomena are described in the form of "there is a form" that describes the phenomenon, the "form" that describes the phenomenon ("current" and "current"), and the "form" that describes the phenomenon. (Past / Past progress type)]. Examples of these end-of-sentence expressions include those shown in the columns in FIG. In other words, first, in the expression examples of "the form that describes the phenomenon", there is "is,""is,""is," and "was." And the like. Examples of expressions of “aspect phrasing“ phenomenon ”that describes the phenomenon (present and present progressive form)” include “has.”, “Has.”, “Has become popular.” No. Examples of expressions for the aspect word “was” (past / past progressive form) describing the phenomenon include “was.” And “was.” Examples of the expression of "hearsay format" include "It seems to be.", "It seems to be.", "It is called.", "It is said that it is." Such an expression example is based on the fact that the appearance of the sentence at the time of introducing a thing is analyzed, and as a result, it has been found that the sentence is classified into four types of sentence end expression patterns as described above. The comment text to be inserted after each sentence end expression pattern is as shown in the right column of FIG. In other words, the “form” that describes the phenomenon and the “form” that describes the phenomenon (“current” and “current”) have “detailed question text (current content)” as “what kind of Comment text such as "What is it?", "Tell me more", "What is it?", "What is it?" In addition, "Aspect words that describe the phenomenon""(past and past progressive forms)" include "How was it?" And "How was it?" ., Etc. are associated with each other. Further, the "hearsay format" is associated with comment text such as "what is it?", "What is it?", "What is it?" The expression examples and comment text examples described above are merely examples, and may include other examples. Although not shown, an appropriate identifier is given to each pattern of each comment text and to each comment text so as to be distinguished from other comment texts.
[0029]
However, in the database 107 functioning as the comment storage unit CMD, in addition to the above-described comment text, for example, comment text such as “Yes.” Comment texts and conversations, such as "What do you think?", Which are inserted between opinion sentences that are virtually read out from the output and are used to connect topics and call other users, are inserted between opinion sentences. Comment texts, such as "I'm talking about," indicating the start, are also stored singly or plurally.
[0030]
Opinions sent by the user U on his / her client computer CC and transmitted are stored in the POC server PS. Each of these opinion sentences is a monologue composed of monologous sentences described by the user U by himself. It is text. FIG. 11 shows an example of the opinion sentence. As shown in the figure, the opinion sentence OPT is composed of a title part OPH described in the “title” column and a text part OPM described in the “text” column. Statement is described. Each opinion sentence OPT is managed by a unique identifier that is distinguished from other opinion sentences OPT. In the present embodiment, the opinion sentence is converted into the XML (extensible Markup Language) format in the POC server PS, but may be of any other format. Further, a related image OPI composed of a moving image or a still image may be attached to each opinion sentence.
[0031]
It is also assumed that two types of speaker agents, a main caster agent and an announcer agent, are set in advance. In other words, it is assumed that a face character image created by the photo face character creation system and a synthetic voice created by the TSS system are set in association with each of the main caster agent and the announcer agent. Here, the announcer agent is set as a speaker who reads out the body part OPM of the original opinion sentence in the conversation text created based on the opinion sentence. On the other hand, it is assumed that the main caster agent is set to read the title OPH and comment text.
[0032]
Hereinafter, an operation example of the POC caster A7 will be described using the opinion sentence example shown in FIG. 11, the conversation text example shown in FIG. 12, the flowchart example shown in FIG. 13, the screen example shown in FIG.
[0033]
First, it is assumed that a large number of opinion sentences from the user U as shown in FIG. 11 are stored in the POC server PS as monolog texts, for example. That is, the user U posts his / her opinion using the electronic bulletin board provided by the POC server PS. Further, the POC server PS or the POC caster A7 transmits and displays a screen displaying a keyword input field to the client computer CC of the user U who wants to view the introduction of the opinion sentence. Is input from the client computer CC to the POC caster A7.
[0034]
When the POC caster A7 acquires the keyword transmitted from the client computer CC (FIG. 13; step S1), the POC caster A7 searches the POC server PS for an opinion sentence (monologue text) OPT related to the acquired keyword (step S2). ). In this search, an appropriate method such as a search of only the title part OPH, a full-text search of the title part OPH and the body part OPM, and the like can be adopted. When one or more opinion sentences OPT can be obtained from the POC server PS (step S2a; Yes), the process proceeds to the next step. Here, when there are a plurality of obtained opinion sentences OPT, they are arranged in an appropriate order such as, for example, ascending or descending order of the identifiers, or date order. On the other hand, when there is no opinion sentence OPT corresponding to the acquired keyword (step S2a; No), information to that effect is transmitted to the client computer CC (step S2b).
[0035]
Next, it is determined whether or not all the opinion sentences OPT have been introduced, and if the introduction has been completed (step S3; No), a single sentence text in which one opinion sentence OPT is divided for each period "." (Step S4). Then, the sentence end expression of each simple sentence text is analyzed (step S5), and the comment text corresponding to the sentence end expression pattern corresponding to the sentence end expression of each simple sentence text is extracted from the database 107 (step S6). A conversation text CVT is generated by inserting it after each simple sentence text (step S7). Here, as an example, if the obtained opinion sentence OPT is as shown in FIG. 11, the generated conversation text CVT is as shown in FIG. That is, the body part OPM of the opinion sentence OPT shown in FIG. 11 is divided into four simple sentence texts. First, the title OPH is allocated to the main caster agent MCA. Here, since there is no preceding single sentence text in the opinion sentence OPT before the title part OPH, the comment text indicating the start of the conversation and the description of the title part OPH are combined, and " It is a topic. " In this way, whether or not "First of all ..." can be appropriately changed according to the order of introduction of the opinion sentence. For example, comments such as "Next is ..." or "Last is ..." Text is available. Next, the first sentence of the body part OPM is assigned to the announcer agent ANA. The first sentence end expression corresponds to the "hearing style" of "... yes." Is assigned to the main caster agent MCA. Similarly, the second sentence and the third sentence are allocated to the announcer agent ANA, and comment texts corresponding to the end sentence expressions of the second sentence and the third sentence, such as "Tell me more." And "Yes." Allocated to caster agent MCA. Furthermore, the fourth sentence is also assigned to the announcer agent ANA. Since this fourth sentence is a single sentence text at the end of the opinion sentence OPT, it is an expression calling on another user U. The comment text "What about everyone," which is also connected, is allocated to the main caster agent MCA.
[0036]
Returning to the flowchart shown in FIG. 13, in the conversation text CVT generated, for the title part OPH (Step S8; Yes), the main caster agent MCA is given an utterance operation of the comment text, while the announcer is operated. A rest operation is given to the agent ANA (step S8a). If it is not the title part OPH, that is, if it is the body part OPM (step S8; No), it is determined whether or not the end-of-sentence expression is an expression that states introduction, otherwise (step S8b; No). The announcer agent ANA is given a reading operation of a single sentence text, while the main caster agent MCA is given a resting operation (step S8c). On the other hand, if the sentence end expression is an expression that states an introduction (step S8; Yes), the main caster agent MCA is given a comment text reading operation, while the announcer agent ANA is given a resting operation (step S8a). When an action is given to each of the main caster agent MCA and the announcer agent ANA, an animation corresponding to the action is generated and a voice is synthesized. Here, the animation operation includes an operation in which at least the main caster agent MCA and the announcer agent ANA can move their mouths. However, if the attached image is attached to the opinion sentence OPT as described later, any of the agents There are variations such as making a finger pointing motion or a nodding motion. Further, as described above, if the opinion image OPT includes the attached image OPI (Step S10; Yes), the attached image is transmitted, for example, by being arranged at the center of the transmitted image to be displayed on the client computer CC. The attached image is combined with the image. Thereafter, or when there is no attached image (Step S10; No), the character data of the conversation text CVT is combined in, for example, the lower column of the transmission image (Step S11), and all data is transmitted in a format that can be viewed on the client computer CC. (Step S12). As a result of the transmission, the image displayed on the display of the client computer CC is, for example, as shown in FIGS. 14 to 22. From the speaker of the client computer CC, the respective voices of the main caster agent MCA and the announcer agent ANA are output. Is output.
[0037]
The screen displayed on the display of the client computer CC and the sound output from the speaker will be described. In FIGS. 14 to 22, the attached image is displayed at the center of the screen, and the image of the main caster agent MCA is displayed on the left side. The right side shows a state where an image of the announcer agent ANA is arranged. Then, in accordance with the conversation text CVT shown in FIG. 12, the operation of each agent and the voice output for reading out the comment or opinion sentence, and the display of the character data accompanying this voice output (the lower part of the screen) are performed. First, as shown in FIG. 14, the main caster agent MCA performs an operation of reading out a comment "First step is ..." in which the title part OPH of the opinion sentence and the comment text indicating the start of the conversation are synthesized, and the voice is reproduced. Output and display the character data of this comment in the lower column of the screen. Next, as shown in FIG. 15, the announcer agent ANA performs an operation of reading out the first sentence “Ise-Shima no...” Of the body part OPM, outputs its voice, and displays the character data of this comment in the lower column of the screen. Is displayed. At this time, the announcer agent ANA is caused to perform an animation operation indicating the attached image OPI. Next, as shown in FIG. 16, the speaker takes the place of the main caster agent MCA, performs an operation of reading out the comment "What is it?", Outputs the voice, and outputs the voice of the comment in the lower column of the screen. Display character data. Further, as shown in FIG. 17, the speaker takes the place of the announcer agent ANA, performs the operation of reading out the second sentence "Large facility sightseeing ..." of the body part OPM, outputs the voice, and outputs the voice at the bottom of the screen. Column displays the character data of this comment. Subsequently, as shown in FIG. 18, the main caster agent MCA reads out the comment "Tell me more.", Outputs its voice, and displays the character data of the comment in the lower column of the screen. Next, as shown in FIG. 19, the speaker changes to the announcer agent ANA, performs the operation of reading out the third sentence “Beautiful scenery and so on” of the body part OPM, outputs the voice, and outputs the voice to the lower part of the screen. Displays the character data of this comment. On the other hand, as shown in FIG. 20, the main caster agent MCA performs an operation of reading out the comment "Yes.", Outputs its voice, and displays the character data of this comment in the lower column of the screen. Then, as shown in FIG. 21, the announcer agent ANA performs an operation of reading out the final sentence “Would you like to join together?” In the body part OPM, and outputs the voice thereof, and the text of this comment is displayed in the lower column of the screen. Display data. Finally, as shown in FIG. 22, the main caster agent MCA performs an operation of reading out the comment "What are you doing?", Outputs the voice, and displays the character data of the comment in the lower column of the screen. Leads to the introduction of the next opinion statement OPT.
[0038]
That is, when each of the above steps is completed, the process returns to step S3 to process the next opinion sentence OPT. Then, when the introduction of all the opinion sentences OPT is completed (step S3; Yes), the process for the initially acquired rank keyword is completed. Steps S9 to S11 do not necessarily have to be performed in this order, and the order may be appropriately changed.
[0039]
As described above, in the opinion sentence posted by the user U, the intention of the target opinion sentence is guessed from the sentence end expression which is a surface clue of the opinion sentence, and the comment text is inserted and the addition synthesis process is performed. Thus, the other user U can view the opinion sentence in a conversation format. Therefore, even if the original information presentation is an opinion sentence which is a monologous sentence, the information can be provided to the user U who views the information in a manner that is more familiar and reduces the burden on understanding.
[0040]
Note that the present invention is not limited to the above embodiment. For example, it is possible to increase or decrease the pattern of the sentence end expression of the divided single sentence text, and to appropriately increase or decrease the number of expression examples corresponding to each pattern and the corresponding comment text. In addition, since the text posted on the electronic bulletin board is not a formal text, the title part and the text part may be connected as a single text in the posted opinion sentence. It is also possible to perform processing such as normalizing a sentence starting with a case particle such as "", "", "" or "" or a symbol such as "...". In addition, the specific configuration of each unit is not limited to the above embodiment, and various modifications can be made without departing from the spirit of the present invention. Further, the generated conversation text is not limited to two, but may be a conversation sentence of three or more. Further, the present invention can be applied to fields or systems other than POC.
[0041]
【The invention's effect】
According to the present invention, as described in detail above, through a process of dividing a monologue text, which is a sentence written by one person, into a single sentence, into a conversation format that can reduce the burden on the understanding required by the listener or reader. It can be converted. That is, by patterning the sentence end expression, which is a surface clue of the monologue text, by guessing the intended purpose of the monologue text and inserting an appropriate comment at the end of the single sentence, It is possible to generate a conversational expression without any restrictions. Therefore, by applying the present invention, it will be very useful for the management of opinion introduction programs in communities where unspecified topics such as POC appear, and for research on conversation.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a functional configuration of a conversation expression generation device according to claim 1 of the present invention.
FIG. 2 is a block diagram showing a functional configuration of a conversation expression generation device according to a second embodiment of the present invention.
FIG. 3 is a block diagram showing a functional configuration of a conversation expression generation device according to claim 4 of the present invention.
FIG. 4 is a block diagram showing a functional configuration of a conversation expression generation device according to claim 5 of the present invention.
FIG. 5 is a block diagram showing a functional configuration of a conversation expression generation device according to claim 6 of the present invention.
FIG. 6 is a block diagram showing a functional configuration of a conversation expression generation device according to claim 7 of the present invention.
FIG. 7 is a block diagram showing a functional configuration of a conversation expression generation device according to claim 8 of the present invention.
FIG. 8 is a schematic view showing a POC system to which an embodiment of the present invention is applied.
FIG. 9 is a schematic internal device configuration diagram of the POC caster of the embodiment.
FIG. 10 is an exemplary view showing an example of internal data of a comment storage unit applied to the embodiment.
FIG. 11 is an exemplary view showing an example of an opinion sentence applied to the embodiment;
FIG. 12 is an exemplary view showing an example of a conversation text generated in the embodiment.
FIG. 13 is a flowchart schematically showing the operation of the embodiment.
FIG. 14 is an exemplary view showing an example of a screen displayed on a client computer in the embodiment.
FIG. 15 is an exemplary view showing an example of a screen displayed on the client computer in the embodiment.
FIG. 16 is an exemplary view showing an example of a screen displayed on the client computer in the embodiment.
FIG. 17 is an exemplary view showing an example of a screen displayed on a client computer in the embodiment.
FIG. 18 is an exemplary view showing an example of a screen displayed on a client computer in the embodiment.
FIG. 19 is an exemplary view showing an example of a screen displayed on the client computer in the embodiment.
FIG. 20 is an exemplary view showing an example of a screen displayed on a client computer in the embodiment.
FIG. 21 is an exemplary view showing an example of a screen displayed on the client computer in the embodiment.
FIG. 22 is an exemplary view showing an example of a screen displayed on the client computer in the embodiment.
[Explanation of symbols]
1. Monologue text acquisition means
2 ... Pretreatment means
3 ... sentence end processing means
4: Comment selection means
5. Conversational expression generation means
6… Conversation text output means
7 ... Keyword acquisition means
8 ... Speaker determination means
9 ... Audio output means
10. Animation processing means
11 ... Character data output means
12 ... Image data processing means
A1, A2, A3, A4, A5, A6, A7 ... Conversational expression generation device
CMD: Comment storage
MTD: Monologue text storage

Claims (32)

It is for generating a conversational expression based on a monologue text composed of monologue-like sentences. The monologue text acquisition means acquires from a monologue text storage unit that stores the monologue text, and the acquired monologue text is divided into a single sentence format. Pre-processing means for generating the above simple sentence text; sentence end processing means for analyzing the sentence end expression of the generated simple sentence text and associating the sentence end expression with one of a plurality of preset sentence end expression patterns; One comment text corresponding to a single sentence text associated with any one of the end-of-sentence expression patterns is stored in a comment storage unit that stores a plurality of comment texts set as expressions corresponding to the expression patterns. A comment selecting means to be selected, and a comment selected after the simple text A conversation expression generating means for generating a conversation text by inserting the instrument text consisting sentence text and comment text conversation representation generator apparatus characterized by comprising; and a conversation text output means for outputting the generated conversation text. The apparatus further includes keyword acquisition means for acquiring a keyword input by the user, wherein the monologue text acquisition means acquires one or more monologue texts corresponding to the keyword from the monologue text storage unit, and the preprocessing means The conversation expression generation device according to claim 1, wherein a single sentence text is generated for each of the obtained monologue texts. The monolog text storage unit stores opinion texts posted on the electronic bulletin board by user input as monolog texts, and the monolog text acquisition means acquires the monolog texts as the opinion texts. 3. The conversation expression generation device according to 1 or 2. A process of associating one speaker agent as a reader of a single sentence text in an output conversation text with another speaker agent as a reader of a comment text, for each of two or more preset speaker agents. 4. The conversation expression generation device according to claim 1, further comprising speaker determination means for performing the following. 5. The conversation expression generation device according to claim 4, further comprising: voice output means for outputting a single sentence text or comment text corresponding to each of the speaker agents determined by the speaker determination means with different voices. An animation processing means for adding and outputting an animation operation for moving at least a mouth out of an image of the speaker agent to each speaker agent in accordance with the voice of the single sentence text or comment text output by the voice output means. The conversation expression generation device according to claim 5, wherein 7. The conversation expression generation device according to claim 4, further comprising character data output means for outputting, with each speaker agent, the corresponding simple sentence text or comment text as character data that can be displayed on a screen. 8. The image processing apparatus according to claim 1, further comprising image data processing means for acquiring and outputting the image data from the monologue text storage unit when the monologue text accompanies the image data. Conversation expression generator. A main caster agent, which is a facilitator of conversational expression, is set as one of the speaker agents, and the speaker determining means determines the main caster agent as a reader of the comment text. The conversation expression generation device according to 6, 7, or 8. The monologue text is composed of a text part which is an essential part of the content and a title part indicating the outline, and the speaker determining means determines the main caster agent as a reader of the title part. Item 9. The conversation expression generation device according to Item 9. A comment text indicating the start of the conversational expression is stored in the comment storage unit, and the comment selecting means selects the comment text indicating the start when there is no other single sentence text before the single sentence text. 11. The conversation expression generation device according to claim 10, wherein the selecting unit associates the comment text with a title part in association with the main caster agent. The monologue text is composed of a text part which is an essential part of the content and a title part indicating the outline, and one or more announcer agents are set as speaker agents different from the main caster agent, 12. The conversation expression generation device according to claim 9, wherein the speaker determination means determines the announcer agent as a reader of the body part. The comment selecting means recognizes the final single sentence text in one monologue text, and sets the comment text stored in the comment storage unit as an expression connected to the next monologue text in association with the end-of-sentence expression pattern in the final single sentence text The conversation expression generation according to claim 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, or 12, wherein a comment text corresponding to the last sentence expression of the final simple sentence text is selected from the following. apparatus. The sentence end expression pattern includes at least a phenomenon description format indicating that the phenomenon is stated and a hearing format indicating that it is a report, and the comment text includes a comment text corresponding to a question format corresponding to the phenomenon description format. At least a comment text corresponding to the expected sentence format corresponding to the message format is included, and the sentence end processing means associates the sentence end expression of the simple sentence text with either the phenomenon description format or the message format, and the comment selecting device correspondingly. The comment text of one of a question text format and a predicted text format is selected in claim 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, or 13. Conversation expression generator. 15. The conversation expression according to claim 14, wherein a plurality of comment texts in a question text format and a predicted text format are respectively set, and the comment selection means selects any one of the comment texts. Generator. The monologue text storage unit stores a monologue text input to an electronic bulletin board that can be used by a plurality of users, and the comment storage unit, and is generated based on the input monologue text. 3. An interactive broadcast system capable of broadcasting a conversation expression text, generating a conversation text based on an input monologue text, and outputting the conversation text in a broadcastable manner. A conversational expression generation device according to any one of claims 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, and 15. A program for causing a computer to function as a conversation expression generation device that generates a conversation expression based on a monolog text composed of monolog texts by operating the computer, the computer comprising: a monolog that stores a monolog text A monologue text acquisition unit that acquires from the text storage unit, a preprocessing unit that divides the acquired monologue text into a single sentence format and generates one or more single sentence texts, analyzes a sentence end expression of the generated simple sentence text, and analyzes the sentence end expression A sentence end processing unit for associating with any one of a plurality of preset sentence end expression patterns; and a comment storage unit for storing a plurality of comment texts set as expressions corresponding to the end of sentence expression patterns in response to the end of sentence expression patterns. Any one of the sentence end expression patterns Comment selecting means for selecting one comment text corresponding to the associated simple text, and conversation expression generating means for inserting the selected comment text after the simple text to generate a conversation text composed of the simple text and the comment text A conversation expression output program for outputting the generated conversation text. The computer further functions as a keyword acquiring unit for acquiring a keyword input by a user.In the monologue text acquiring unit, one or more monologue texts corresponding to the keyword are acquired from a monologue text storage unit. 18. The conversation expression generation program according to claim 17, wherein a single sentence text is generated for each of the acquired monologue texts. A monologue text storage unit for storing opinion texts posted on the electronic bulletin board by user input as monologue texts, wherein the monologue text obtaining means obtains the monologue texts as the opinion texts. Item 19. The conversation expression generation program according to Item 17 or 18. The computer associates one speaker agent as a reader of a single sentence text in the output conversation text with each of two or more preset speaker agents, and another speaker as a reader of a comment text. 20. The conversation expression generation program according to claim 17, further comprising a speaker determination unit that performs a process of associating an agent. 21. The conversation expression generation program according to claim 20, further causing the computer to function as voice output means for outputting a single sentence text or comment text corresponding to each speaker agent determined by the speaker determination means with a different voice. An animation processing means for adding, to the speaker agent, an animation operation for moving at least a mouth out of an image of the speaker agent to each speaker agent in accordance with the voice of a simple sentence text or a comment text output by a voice output means, and outputting the computer; 22. The conversation expression generation program according to claim 21, further comprising: 23. The conversation expression generation program according to claim 19, wherein the computer further causes the computer to function together with each speaker agent as character data output means for outputting a corresponding single sentence text or comment text as character data that can be displayed on a screen. 21. When the monologue text is accompanied by image data, the computer further functions as image data processing means for acquiring and outputting the image data from the monologue text storage unit. 23. A conversation expression generation program according to claim 23. 22. A main caster agent serving as a facilitator of conversational expression is set as one of the speaker agents, and the speaker determining means determines the main caster agent as a reader of the comment text. , 22, 23 or 24. The monologue text is composed of a text part which is an essential part of the content and a title part indicating the outline, and the speaker determination means determines the main caster agent as a reader of the title part. A conversation expression generation program according to claim 25. A comment text indicating the start of the conversational expression is stored in the comment storage unit, and when there is no other single sentence text before the single sentence text in the comment selection means, the comment text indicating the start is selected, and the speaker selection is performed. 27. The conversation expression generation program according to claim 26, wherein the means associates the comment text with a title part in association with the main caster agent. The monologue text is composed of a text part which is an essential part of the content and a title part indicating the outline, and one or more announcer agents are set as speaker agents different from the main caster agent, 28. The conversation expression generation program according to claim 25, wherein the speaker determination means determines the announcer agent as a reader of the body part. The comment selecting means recognizes the final single sentence text in one monologue text, and sets the comment text stored in the comment storage unit as an expression connected to the next monologue text in association with the end-of-sentence expression pattern in the final single sentence text 29. The conversation expression according to claim 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27 or 28, wherein a comment text corresponding to the last sentence expression of the final simple sentence text is selected from the following. Generation program. The sentence end expression pattern includes at least a phenomenon description format indicating that the phenomenon is stated and a hearing format indicating that it is a report, and the comment text includes a comment text corresponding to a question format corresponding to the phenomenon description format. At least a comment text corresponding to the expected sentence format corresponding to the message format is included, and the sentence end expression of the simple sentence text is associated with either the phenomenon description format or the message format in the sentence end processing means, and the comment selection means The comment text of one of a question text format and a predicted text format is selected. 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28 or 29. Conversation expression generator. 31. The conversation according to claim 30, wherein a plurality of comment texts in a question sentence format and a predicted sentence format are respectively set, and one of the comment texts is selected by the comment selection means. Expression generator. The monologue text storage unit stores a monologue text input to an electronic bulletin board that can be used by a plurality of users, and the comment storage unit, and is generated based on the input monologue text. 21. An interactive broadcast system capable of broadcasting conversational expression texts, generating conversational texts based on input monolog texts, and outputting the conversational texts in a broadcastable manner. , 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, or 31.

Images