Response language expression output device and method

Abstract

Conventional conversational AI has been unable to visualize and organize the thought structure or reconstruct it for communication with others when the user's utterance is ambiguous or disorganized, making it difficult to use as a decision-making or sharing support tool. [Solution] This invention provides a structure that extracts logical structures such as purpose, decision criteria, and hypotheses from the user's natural language utterances, generates follow-up questions and advice for omissions and ambiguities, and further formats them into report and proposal formats. This realizes a support AI that assists decision-making while converting it into a format that is easy to communicate to others.

Description

Technical Field

[0001] This invention relates to outputting a response language expression that responds to a language expression given by a user (including language expressions such as words spoken by the user, language expressions by voice input from a microphone like sentences, language expressions input by the user using an input device such as a keyboard, words and sentences, language expressions obtained by reading strings described on paper, etc.).

[0002] In particular, this invention relates to a judgment structure support system that assists an AI in structurally grasping statements related to judgments and decision-making that a user is about to make, visualizing and reshaping a thinking structure through an interactive form, and reconstructing it into a reporting / proposal form to others as needed. The "dialog interaction state" and "dialog state" in the present invention refer to a dialog interaction state based on external (non-semantic) feature quantities such as utterance length, utterance interval, word ending, intonation, tempo, etc. (a state that can be seen from the outside when the user interacts with the response language expression output device, for example, the user's appearance, attitude, behavior, activity state, etc.).

Background Art

[0003] In recent years, the introduction of dialogue-type AI (Artificial Intelligence) and chatbots has advanced, and systems for handling FAQ (Frequently Asked Questions), casual conversations, inquiry responses, etc. are widely used. Also, some response optimization technologies using sentiment analysis and intention recognition have been developed. Furthermore, when a generative AI interacts with each of multiple users, there are also those that smooth the dialogue between the user and the generative AI even when the attributes of individual users are different (for example, when the user's country of origin is different) (Patent Document 1).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

[0005] In business settings, users often struggle to organize their thoughts and structure proposals because they haven't yet determined "what they want to decide" or "how to organize their thoughts." In such cases, simple informational responses are insufficient; there is a need for a function that supports the organization and structuring of the user's ambiguous statements, hesitations, and unarticulated decision structures, and then transforms them into a form that can be communicated. Furthermore, it is conceivable that this invention can be used in conjunction with empathetic response control in dialogue AI (e.g., relationship temperature classification system) or a personality style switching structure according to the state of the dialogue (e.g., a dialogue system that adapts to the state of the dialogue) to further improve the naturalness of the dialogue flow and formatting advice.

[0006] When user utterances were disorganized, hesitant, or incomplete as reports or suggestions, conventional conversational AIs faced the following challenges: 1. The content of the conversation was not structured, making it unclear what decisions should be made. 2. Advice and questions remained in a "question-and-answer" format, failing to allow for a shift in perspective or organization. 3. The utterances were not automatically formatted into reports or suggestions for superiors or others. As a result, it was difficult to draw out the user's judgment and communication skills.

[0007] The method described in Reference 1 facilitates smooth dialogue between the user and the generating AI when the generating AI interacts with multiple users, even if the attributes of each user differ, such as having different countries of origin. However, it does not aim to respond as accurately as possible even when the user's statements are ambiguous.

[0008] The purpose of this invention is to enable the system to derive the most accurate possible answer, even if the language input from the user is ambiguous. [Means for solving the problem]

[0009] The response language expression output device according to this invention includes: a language expression input means for inputting a language expression provided by a user; a determination means for determining whether the logical structure of the language expression input from the language expression input means (a term referring to the framework of a text or argument, and an example of a language expression that is a structural element included in a text, such as "Purpose: the intention behind the language expression, what you want to achieve," "Judgment axis: an indicator that serves as a criterion for indecision, what criteria to use for evaluation and selection," "Prerequisites: known situations and facts, the background and conditions for an argument or choice to be valid," "Options: possible actions, alternatives to be considered," "Hypothesis: expected results based on judgment, claims based on predictions or assumptions," etc., is a concept that comprehensively represents such structural elements) that are a response to the input language expression and have a confidence score (a score indicating the reliability of the response language expression generated as a response to the input language expression, whether it meets the user's expectations, whether it is accurate as a response, etc.) is above a threshold; and when the determination means determines that the necessary structural elements are not included, the device outputs to the user. The present invention is characterized by comprising: a preparatory expression generation means that generates a preparatory expression which is at least one of a question or a suggestion (a suggestion about the content of the necessary structural element) that prompts the recipient to answer the content of the necessary structural element (for example, if the structural element is a judgment axis, the content is the answer of what criteria to use for evaluation and selection, for example, the answer when the structural element is asked "What is the judgment axis?", for example, "The judgment axis is budget." is the content of the structural element); a preparatory expression output means that outputs the preparatory expression generated by the preparatory expression generation means; an answer input means that inputs the answer to the preparatory expression output from the preparatory expression output means; a response language expression acquisition means that obtains the response language expression from the answer input from the answer input means and the input language expression (the response language expression may be obtained by a response language expression output device generating it, or by having a device other than the response language expression output device generate the response language expression); and a response language expression output means that outputs the response language expression obtained by the response language expression acquisition means.

[0010] This invention also provides a method suitable for a response language expression output device. Specifically, this method involves a language expression input means inputting a language expression given by a user, a determination means determining whether the logical structure of the language expression input from the language expression input means contains structural elements necessary to generate a response language expression that is a response to the input language expression and has a confidence score equal to or greater than a threshold, a prepared expression generation means generating a prepared expression which is at least one of a question or a suggestion prompting the user to answer with the content of the necessary structural elements, depending on whether the determination means has determined that the necessary structural elements are not included, a prepared expression output means outputting the prepared expression generated by the prepared expression generation means, an answer input means inputting an answer to the prepared expression output from the prepared expression output means, a response language expression acquisition means acquiring the response language expression from the answer input from the answer input means and the input language expression, and a response language expression output means outputting the response language expression acquired by the response language expression acquisition means.

[0011] This invention also provides a program for controlling the computer of a response language expression output device, and a recording medium storing that program.

[0012] The system may further include structural element extraction means for extracting structural elements from the logical structure of the input language expression. In this case, the determination means determines, for example, whether the structural elements extracted by the structural element extraction means include the required structural elements.

[0013] The system may also include a response language expression output stopping means that stops outputting the response language expression until the prepared expression is output from the prepared expression output means. Since the response language expression can be generated considering the answers to the prepared expressions such as questions and suggestions, it is possible to generate relatively accurate responses.

[0014] The system may further include a suggestion output control means that controls the preparation expression output means to output additional preparation expressions between the end of outputting the response language expression generated from the response to the above preparation expression and the input language expression, and the time when the language expression from the user is input again from the language expression input means. Since additional questions, suggestions, etc. are output before the user re-inputs the language expression, the user can input answers to the additional questions, suggestions, etc., and generate a more accurate response language expression that takes the additional questions and suggestions into account.

[0015] The above-mentioned means for acquiring response language expressions may, for example, acquire response language expressions in a style appropriate to the recipient of the response language expression output from the above-mentioned means for acquiring response language expressions. For example, by using a polite style if the recipient is a superior, and a casual and friendly style if it is a colleague, the recipient who receives the response language expression will be able to trust the user's character.

[0016] The system may also include a selection means for selecting the output format of the response language expression, and a format changing means for changing the response language expression acquired by the response language expression acquisition means to an output format corresponding to the output format selected by the selection means. In this case, the response language expression output means outputs the response language expression in the output format changed by the format changing means, for example. Since the output format is suitable for reports and the like, the user does not need to adjust the format of the response language expression.

[0017] The system may further include a memory control means that controls the memory device to store, for each user, at least one of the structural elements included in the logical structure of the response to the above-mentioned preparatory expression and the structural elements included in the logical structure of the above-mentioned input language expression. The response language expression acquisition means acquires a response language expression that places emphasis on the structural elements that the user considers important, based on the structural elements stored in the memory device. A detailed response language expression can be output regarding the content that the user considers important.

[0018] The decision-making support AI system according to this invention uses the user's natural language statements to determine the purpose, It includes a structural analysis means that extracts elements such as judgment criteria, preconditions, options, and hypotheses and organizes them as a logical structure, and an advice generation means that generates advice and questions based on that structure.

[0019] Furthermore, the system may also include document formatting means for reconstructing the logical structure obtained by the structural analysis means into a proposal or report in a format that is suitable for the perspective of the recipient (supervisor, team members, etc.).

[0020] Furthermore, the system may include a history learning means that records the logical structure and advice content as history and learns each user's thinking and expression tendencies to optimize structure extraction and expression suggestions. [Effects of the Invention]

[0021] If the linguistic expression entered by the user does not contain the structural elements necessary to generate a response linguistic expression with a confidence score above a threshold, questions or suggestions will be output to prompt the user to provide the content of those necessary structural elements. By using the answers to these questions and suggestions, a more accurate response linguistic expression can be generated as a response to the entered linguistic expression. Even if there is ambiguity or uncertainty in the linguistic expression entered by the user, the user can make a more appropriate judgment about the entered linguistic expression.

[0022] According to the present invention, even if a user is seeking advice or is unsure of what to say, the AI ​​can decompose and reconstruct the utterance and present and support it as a logical decision structure. The present invention enables the organization and supplementation of the semantic structure of a dialogue through a three-layer structure of structural analysis, advice generation, and formatted output, and converts it into a form of communication for others. It can be implemented independently without external control such as dialogue interaction state classification or personality switching.

[0023] In addition, it becomes possible to automatically convert into "a form to be conveyed to someone" such as reports and proposals, greatly enhancing the user's judgment, structuring ability, and sharing ability. Furthermore, by the AI supplementally asking questions, a process in which the user himself / herself deepens his / her own thinking naturally occurs, and an effect of suppressing discrepancies in understanding within the team and variations in the quality of reports can also be expected.

Brief Explanation of Drawings

[0024] [Figure 1] It is an overview of the response language expression output system. [Figure 2] It is a block diagram showing the electrical configuration of the response language expression output device. [Figure 3] It is an example of a learned model. [Figure 4] It is a flowchart showing the processing procedure of the response language expression output device. [Figure 5] It shows the interaction between the user and the response language expression output device. [Figure 6] It is an example of a structure classification table. [Figure 7] It shows the T-L-C1 space. [Figure 8] It shows the interaction between the user and the response language expression output device. [Figure 9] It is an example of a structure classification table. [Figure 10] It is a flowchart showing a part of the processing procedure of the response language expression output device. [Figure 11] It shows the interaction between the user and the response language expression output device. [Figure 12] It is a flowchart showing a part of the processing procedure of the response language expression output device. [Figure 13] It is an example of output. [Figure 14] It is an example of output. [Figure 15] It is an example of a history table. [Figure 16] It is a flowchart showing the output processing procedure of the response language expression. [Figure 17]Overall structure diagram (Speech → Structural analysis → Advice generation → Reconstruction → Output) [Figure 18] Structural organization step diagram (User utterance → Structural element classification map) [Figure 19] Advice generation and questioning logic flow diagram [Figure 20] Report restructuring step diagram (Structure → Template → Display switching) [Modes for carrying out the invention]

[0025] Figure 1 shows an embodiment of this invention and is an overview of the response language expression output system.

[0026] The response language expression output system comprises a response language expression output device 1 and an AI (Artificial Intelligence) server 20, which can communicate with each other via the internet.

[0027] In this embodiment, the response language expression output device 1 and the AI ​​server 20 communicate to output the response language expression from the response language expression output device 1. However, by performing the classification to obtain the language expression of the normal response in the AI ​​server 20 in the response language expression output device 1, the AI ​​server 20 is not necessarily required.

[0028] Figure 2 is a block diagram showing the electrical configuration of the response language expression output device 1.

[0029] The overall operation of the response language expression output device 1 is controlled by the CPU 2.

[0030] The response language expression output device 1 includes memory 5 for temporarily storing data. The response language expression output device 1 also includes a GPU (Graphics Processing Unit) 3 (an example of a prepared expression generation means), and AI learning and generation are performed by the GPU 3. As will be described later, a display device 4 (an example of a response language expression output means) that displays questions, suggestions, response language expressions, etc., is connected to the GPU 3 via an interface (not shown).

[0031] The response language expression output device 1 includes a PCH (Platform Controller Hub) 6, which controls data communication between the CPU 2 and the communication device 7 that communicates with the AI ​​server 20, a microphone 8 (an example of a language expression input means), a speaker 9 (an example of an output means), a keyboard 10 (an example of a language expression input means), an SSD (Solid State Drive) 11, a CD (Compact Disc) drive 12, etc. The SSD stores various trained models and other data. A CD 13 containing a program that controls the operation of the response language expression output device 1 is inserted into the CD drive 12, and the program is read from the CD 13 and installed in the response language expression output device 1. Alternatively, the program may be received via the internet and installed in the response language expression output device 1.

[0032] Furthermore, the system may be equipped with a scanner (an example of a language expression input means) for reading language expressions written on paper or other materials, so that the language expression is read by the scanner and the response language expression output device 1 recognizes the language expression by OCR (Optical Character Recognition).

[0033] Figure 3 shows the trained models 31 for structural element extraction, 32 for question generation, 33 for proposal generation, 34 for style modification, and 35 for output format modification, all stored in SSD11.

[0034] The pre-trained model 31 for structural element extraction extracts structural elements such as purpose, premises, decision axes, hypotheses, and options from the logical structure of the input linguistic expression. For example, the purpose is extracted from the verb phrase (in order to, want to, object) and conclusion sentence of the input linguistic expression. Premises are extracted by separating them into explicit facts (known information) and implicit assumptions (background knowledge and omitted conditions). Decision axes are extracted by listing words indicating comparison or priority (importance, cost, safety, availability, etc.). Hypotheses are extracted (assumed causality or effect). For example, they are extracted from claims written in the form of cause and effect, or from sentences of the type "if - then -". Options are extracted by checking three types: explicit proposals, implicit alternatives, and excluded options. The pre-trained model 31 for structural element extraction is generated by pre-training on a large amount of training data in which these premises, decision axes, hypotheses, and options are tagged, including the purpose, premises, decision axes, hypotheses, and options. By using the pre-trained model 31 for structural element extraction, structural elements contained in the input language expression are extracted.

[0035] The pre-trained model 32 for question generation generates questions in which the user answers about structural elements not included in the linguistic structure of the input language expression. The pre-trained model 31 for structural element extraction generates questions in which the user answers about structural elements other than those obtained using the pre-trained model 31 for structural element extraction.

[0036] Questions that elicit answers about the structural elements of the objective may include questions such as, "List three higher-level objectives that could be achieved if the user's proposed linguistic expression is successful, and indicate the minimum conditions for each objective to be fulfilled," which create a starting point for inverse questioning by extracting the "conclusion," "proposal," and "action" from the input linguistic expression, defining the question format such as the number, output format, and type of evidence required, specifying the perspective of the respondent (e.g., business manager, legal officer), and requiring a description of the verification methods, such as which data and metrics will be used to confirm the objective.

[0037] Questions that ask about the structural elements of the premises should ask students to list the conditions necessary for the conclusion or proposal in the text to be true, such as, "List five minimum premises necessary for the conclusion to be true. Also, describe the consequences if each premises is not met." For example, if the text states, "Cost reduction is possible," the student should be asked to answer about the premises by saying, "List three external conditions that are implicitly assumed for cost reduction to be possible, and explain the consequences if each of them is not met."

[0038] Questions that ask about the structural elements of the decision-making criteria might ask, for example, "Please list four essential evaluation criteria for evaluating this proposal, assign them weights (total 100), and explain the reasoning behind them." This asks the user to supplement the evaluation criteria necessary for decision-making and to assign weights. For example, if the user's verbal expression does not mention quality or delivery time, the question might ask, "Please assign weights to the four axes of quality, cost, delivery time, and risk, and define the minimum acceptable value for each axis," to get the user to answer about the content of the decision-making criteria.

[0039] Questions that elicit the structural elements of a hypothesis, such as, "Create three hypotheses derived from the user's claims and provide specific verification methods (data, experiments) for each hypothesis," draw out if-then hypotheses and request means of verification. For example, if the user's statement is "The new feature will increase usage," the question might be, "Provide three hypotheses for improving usage and write which log metrics will be used and when to verify them," prompting the user to provide the content of their hypotheses.

[0040] Questions that require users to identify the structural elements of the options should prompt them to consider "alternatives," "combined options," and "rejected options" in relation to the proposal presented by the user. For example, if only one proposal is presented, the question could be phrased as, "Propose three other feasible options and compare their implementation costs and expected effects," prompting the user to provide the content of the options.

[0041] The pre-trained model 33 for proposal generation generates proposals for the content of structural elements that are not included in the linguistic structure of the input linguistic expression. It generates proposals for the content of structural elements other than those obtained using the pre-trained model 31 for structural element extraction.

[0042] In the pre-trained model 33 for proposal generation, the user's answers to questions generated by the pre-trained model 32 for question generation should be used as proposals.

[0043] The pre-trained model 34 for style modification changes the generated response language expression to a style appropriate for the recipient. For example, if the recipient is a superior, the style will be to "state the conclusion first," "clearly state the purpose and main points," and "write in a readable and concise manner," making it easy to judge in a short time. It will also use consistent polite language and objective expressions, selecting polite language that is not redundant, and minimizing subjective evaluations. Furthermore, it will use bullet points, headings, and paragraphs to make it short and easy to read. For example, if the response language expression is "I was ultimately unable to complete the matter requested in July," it will be changed to a polite style such as "The matter you requested in July has not yet been completed. I will report on the status and future action plans," and will also generate information on the current situation, reasons, and countermeasures. If the recipient is a colleague, the style will be concise, clear, and approachable. It will primarily use the "desu / masu" style, maintaining politeness while avoiding overly polite language. For example, instead of "I will report," it will use a simple and easy-to-read expression such as "I will report." Furthermore, considering that colleagues are busy, the conclusion and key points should be stated first. If the recipient is a client, politeness, accuracy, and objective, easy-to-understand language are essential. The writing style should be more formal and careful than that used for reports to colleagues, with the aim of building trust and ensuring smooth business operations. For example, while the basic style is "desu / masu," politeness should be the primary consideration, and honorifics and humble language should be used appropriately. The description should be based on objective facts, clearly conveying objective facts and data without including personal opinions or speculations. The expression should be concise, using simple language so that clients unfamiliar with technical terms and industry customs can understand it.

[0044] The pre-trained model 35 for changing output formats outputs the generated response language expression in different output formats. For example, it can output the response language expression in various formats such as a presentation format, a letter format like a report, an email format, and a CSV (Comma-Separated Values) format.

[0045] With the exception of pre-trained model 31 for structural element extraction, all pre-trained models 32-34 collect data representing multiple language expressions in advance, statistically analyze it along with labels assigned based on self-reports from subjects and annotations by experts, and train the machine learning model using these relationships as training data. This makes it possible to generate questions, suggestions, modify writing style, and change output formats even for unknown inputs. Of course, even without using all or part of pre-trained models 31-35, structural element extraction, question generation, suggestion generation, writing style modification, and output format modification can be achieved by creating templates in advance and applying them.

[0046] Figure 4 is a flowchart showing the processing procedure of the response language expression output device 1, Figure 5 shows the interaction between the user and the response language expression output device 1, and Figure 6 is an example of a structural classification table.

[0047] Referring to Figure 4, the user's response language expression is input to the response language expression output device via the keyboard 10 (an example of a language expression input means) (step 41). For example, the user inputs the language expression "Responding to Company A is the top priority, but I'm unsure whether it's okay to proceed with the current policy regarding Company A..." (see Figure 5) to the response language expression output device 1.

[0048] When a linguistic expression is input to the response linguistic expression output device 1, the GPU 3 (an example of a structural element extraction means) uses a trained model 31 for structural element extraction to extract structural elements from the logical structure of the input linguistic expression (step 42). Suppose that from the input linguistic elements, "judgment on whether to proceed" and "Company A's response is the highest priority" are extracted as the target structural element and the premise structural element, while other structural elements such as judgment axes, hypotheses, and choices are not extracted. Then, a structural classification table like the one shown in Figure 6 is generated. This structural classification table may be displayed on the display screen of the display device 4. The user will know what information is needed.

[0049] Once structural elements are extracted, the CPU2 (an example of a determination means) determines whether any of the extracted structural elements are necessary to generate a response language expression with a confidence score equal to or greater than a predetermined threshold (step 43). For example, suppose that in order to generate a response language expression with a confidence score equal to or greater than a predetermined threshold, at least two structural elements other than the objective are required from the five structural elements of objective, premise, decision axis, hypothesis, and choice. Then, it is determined whether a total of three structural elements, including the objective and at least two other structural elements, have not been extracted. If simply extracting three structural elements is sufficient to generate a response language expression with a confidence score equal to or greater than a predetermined threshold, then it is sufficient to extract those three structural elements even if the objective structural element is not included. In this embodiment, if the number of extracted structural elements is three or more, it is assumed that a response language expression with a confidence score equal to or greater than a predetermined threshold can be generated if three or more structural elements other than the objective structural element are extracted. However, it is also possible to determine whether the necessary structural elements for generating a response language expression with a confidence score equal to or greater than a predetermined threshold have been extracted by other means.

[0050] If the necessary structural elements are not included (YES in step 43), the necessary structural elements are determined (step 44). If at least two structural elements are needed in addition to the objective (the same applies even if at least three structural elements are needed), then at least one structural element from the missing decision axes, hypotheses, and options is required. Any of these decision axes, hypotheses, and options would suffice, but here the objective is "determining whether to proceed" and the decision axis structural element is missing, so the decision axis structural element is determined to be the necessary structural element.

[0051] CPU2 causes GPU3 (an example of a prepared expression generation means) to generate a question or suggestion to elicit a response containing the necessary components of the decision axis (step 45). For example, the question "What is the decision axis: delivery time, quality, or cost?" (see Figure 5) is generated and output from the response language expression output device 1 (display, audio output, etc.) (step 46). For example, it is displayed on the display screen of the display device 4 (an example of a prepared expression output means) and audio is output from the speaker 9 (an example of a prepared expression output means). If the necessary structural elements are not included, CPU2 (an example of a response language expression output stopping means) will stop outputting the response language expression until a question or suggestion is output.

[0052] When the user recognizes a question output from the response language expression output device 1, the user inputs an answer to that question from the microphone 8 (an example of an answer input means) or keyboard 10 (an example of an answer input means) to the response language expression output device 1 (YES in step 47). For example, the user inputs the language expression "About the delivery date." (see Figure 5) as an answer to the response language expression output device 1.

[0053] Then, GPU3 is controlled by CPU2 to extract structural elements from the input response (step 48). In this case, the delivery date is extracted as a structural element of the decision axis.

[0054] Since the structural elements of the judgment axis have been obtained, the extracted structural elements are included in the structural elements necessary for the confidence score of the generated response language expression to be above the threshold (NO in step 43). Therefore, the language expression entered by the user in step 41 ("Responding to Company A is the top priority, but I'm not sure if it's okay to proceed with this policy for Company A...") and the answer to the question entered by the user in step 48 ("Regarding the deadline.") along with the extracted structural elements are sent from the response language expression output device 1 to the AI ​​server 20.

[0055] The AI ​​server 20 generates a response language expression from the received language expression and transmits it to the response language expression output device 1. The response language expression transmitted from the AI ​​server 20 is received by the response language expression output device 1 via the communication device 7, and the response language expression is acquired by the CPU 2 of the response language expression output device 1 (which is an example of a response language expression acquisition means) (step 49). Alternatively, the GPU 3 may be used to generate the response language expression and acquire it.

[0056] The acquired response language expression is output from the response language expression output device 1 (step 50). For example, "In that case, the deadline will be met, so there is no problem in proceeding with the current plan to Company A." is output. By recognizing the response language expression, the user can confirm that the current plan is correct. Even if ambiguous language expressions are input, the system can support the user's decision-making.

[0057] In the above embodiment, questions are generated because it is determined that structural elements of the decision axis are necessary. However, it is also possible to generate and output both questions and suggestions, such as a suggestion like, "I think you should prioritize the delivery date," or "I think you should prioritize the delivery date. Or would it be better to prioritize quality or cost?"

[0058] In this embodiment, the structural shaping process can be executed based on the mutual relationship (conditional coupling structure) of a plurality of feature quantities, such as the semantic feature quantity of the input dialogue (an example of the input language expression), the speaking interval Δt, the relational temperature score T, the pitch change rate Pe at the end of a sentence, and the logical connection degree L (the quantification of semantic consistency) of the speech content. (Here, the relational temperature T is an index of the dialogue interaction intensity or an index indicating emotional stability calculated from non-semantic features (such as word endings, intonation, speaking intervals, etc.) included in the user's speech. For example, it is represented by a score between 0 and 1, and it detects whether the index of dialogue interaction intensity is high based on the word sense of the input language expression, the words at the end of the sentence, the intonation, etc., and the higher the score, the higher it is. Also, the dialogue interaction state signal M is an internal signal representing the user's dialogue state or mental tendency, and it may be generated in the response language expression output device 1 in this embodiment, or the dialogue state signal classified in another configuration may be used.) When at least two or more of these feature quantities simultaneously satisfy a predetermined condition, the intended structure of the dialogue is determined to be "unorganized" or "unconvincing" (for example, the confidence score is determined to be below the threshold, and it is determined that the necessary structural elements are not included (step 43), and the structural shaping unit (CPU2, GPU3) automatically activates the reconstruction process. Thus, when using non-semantic features, which are parts of language expressions excluding content words with meaning in addition to content words with meaning, and the score representing those non-semantic features is below the threshold, etc., it may be determined that the language expression from the user lacks confidence, is ambiguous, etc., and the confidence score is determined to be below the threshold, and it may be determined that the necessary structural elements are not included.)

[0059] As an example of the structural shaping determination, the following conditions 1 to 3 are listed. T < Tt and L < Lt ··· Condition 1 Δt > Δtt and the logical consistency rate C1 of the speaking order is less than the threshold Ct ··· Condition 2 When T ≈ Tt and the word-ending polarity score Ep (an index based on the non-semantic feature quantity of the end-of-sentence expression) is in the neutral range ··· Condition 3 Here, the relational temperature threshold Tt, the logical connectivity threshold Lt, the utterance interval threshold Δtt, and the consistency rate threshold Ct are dynamic thresholds that are sequentially updated based on user history, dialogue history, and the results of dialogue interaction state determination (for example, results obtained by determining the user's dialogue state from the endings and intonation included in the user's linguistic expressions). This allows for the autonomous optimization of structural shaping determination according to each user's thinking characteristics and expression tendencies.

[0060] When a formatting trigger occurs due to the fulfillment of conditions 1-3, this embodiment extracts the main elements of the dialogue (purpose, decision axis, options, hypothesis, reason), structures them hierarchically, and presents them again. During the presentation, the relational temperature T and the dialogue interaction state signal M (for example, a dialogue state signal obtained in a different configuration or another invention may be used) are referenced to adjust the output format to minimize the user's cognitive load. This configuration allows the user to structurally overview their own content and smoothly form judgments, while controlling the process to avoid placing an unnecessary burden on their speech.

[0061] The above conditions and thresholds are merely examples, and the present invention is not limited to them. Structural shaping and comprehension support can be performed using any combination of non-semantic features, semantic features, or contextual relational quantities. Furthermore, since the condition-linked structure is dynamically learned and updated, it is optimized through continuous interaction with the user and can be applied to various application forms such as dialogue support AI, educational support AI, and decision support AI.

[0062] Figure 7 shows a structure shaping determination structure based on the correlation between relational temperature T, speech logical connection degree L, and speech sequence consistency rate C1.

[0063] The T-axis represents the "dialogue interaction intensity index (score from 0 to 1)" calculated from non-semantic features (endings of words, tone, intervals between utterances, intonation, etc.) included in the user's utterances. The L-axis indicates the logical consistency of dialogue sentences with scores between 0 and 1, with lower scores indicating a state of confusion. The C1-axis indicates the consistency rate of sentence order with scores between 0 and 1. Based on the interrelationships of each feature, the structure shaping unit is classified into three regions: R1 (consistent region), R2 (partially consistent region), and R3 (reconstruction region). In the R3 region, the relational temperature T is low, and the logical connectivity L and consistency rate C1 are below predetermined thresholds (conditions 1 to 3), thus satisfying the conditions. This triggers the structure shaping signal S_struct, and the reconstruction flag F_reform is turned ON. With this configuration, the system analyzes both the content structure of the dialogue and the dialogue interaction intensity index calculated from non-semantic features included in the utterances in a conditionally linked manner, allowing for stepwise structure shaping until the user is satisfied.

[0064] Figures 8 and 9 show examples of modified forms; Figure 8 shows the interaction between the user and the response language expression output device 1, and Figure 9 shows the structural classification table.

[0065] Suppose the user inputs the phrase, "I'm unsure whether we should proceed with the current policy regarding Company A..." (Figure 4, Step 41), and only the target structural element is extracted, while other structural elements are not extracted (see Figure 9).

[0066] For example, if it is determined that structural elements of premises and decision axes are necessary, the question "What are your decision axes: delivery time, quality, or cost?" is generated to obtain an answer regarding the decision axes, and the suggestion "It might also be good to add customer preferences as a premise" is generated to obtain an answer regarding the premises. These questions and suggestions are output from the response language expression output device 1, and when the user recognizes them, the user inputs "I don't know whether it's delivery time or quality," and "That's right, I forgot about customer preferences," into the response language expression output device 1 as answers to these questions and suggestions. As described above, the response language expression "In that case, why don't you ask your supervisor for their opinion and have them tell you what to prioritize?" is obtained and output from the response language expression output device 1. In this way, even when the user has vague anxieties, it is possible to suggest a direction for the user and alleviate their anxieties.

[0067] Figures 10 and 11 illustrate other embodiments; Figure 10 is a flowchart showing part of the processing procedure of the response language expression output device 1, and Figure 11 shows the interaction between the user and the response language expression output device 1.

[0068] As shown in step 43 of Figure 4, if the extracted structural elements are found to be present in the generated response language expression so that its confidence score is above a threshold (NO in step 43), the response language expression is acquired and output by the response language expression output device 1 (step 49). If no further language expression from the user is input to the response language expression output device 1 (NO in step 51), it is determined whether to add a question or suggestion to the user (step 52). For example, if the input language expression contains two structural elements, it is determined that the structural elements necessary for the generated response language expression to have a confidence score above a threshold (first threshold) are present. However, in order to obtain a response language expression that is more suitable for responding to the user's language expression, it is better to know more structural elements, so it is decided to add a question or suggestion (YES in step 52) (a question or suggestion is added so that the confidence score is above a second threshold, which is higher than the first threshold). Furthermore, if a certain amount of time has elapsed since the output of the response language expression, and the user has not yet input the next language expression to the response language expression output device 1, it is possible that the user is confused. Therefore, in order to obtain a response language expression that is more appropriate to the user's language expression, it is decided that a question or suggestion should be added to understand the structural elements further (YES in step 52).

[0069] Additional questions or suggestions (which may be both, or if the output in step 46 of Figure 4 is a question, then it may be a suggestion (or a question), and if the output in step 46 of Figure 4 is a suggestion, then it may be a question (or a suggestion)) are generated by the GPU3 (which is an example of a prepared expression output control means) (step 53) and output from the response language expression output device 1 (step 54).

[0070] When the user's response to the outputted additional question or suggestion is input to the response language expression output device 1 (YES in step 55), structural elements are extracted from the input response (step 56). Then, the newly extracted structural elements, the already extracted structural elements, and the language expression input by the user are sent to the AI ​​server 20, where a response language expression is generated. As described above, the response language expression generated in the AI ​​server 20 is sent to the response language expression output device 1 and received, thereby acquiring the response language expression in the response language expression output device 1 (step 49). The newly acquired response language expression is more accurate.

[0071] Referring to Figure 11, let's assume the initial language expression input by the user was "I'm unsure whether we should proceed with the current policy for Company A..." (Figure 4, Step 41). Since this input language expression does not contain the necessary structural elements for the confidence score of the generated response language expression to be above the threshold (NO in Figure 4, Step 43), the response language expression output device 1 outputs a question (or suggestion) to the user: "Which of the following is your decision-making criterion: delivery date, quality, or cost?" (Figure 4, Step 46). The user's answer to the question is "Regarding the delivery date" (YES in Figure 4, Step 47), and the response language expression "In that case, the delivery date will be met, so I think it's fine to proceed with the current policy for Company A." is obtained (Figure 10, Step 49) and output from the response language expression output device 1 (Figure 10, Step 50).

[0072] Here, we assume that a suggestion is generated that says, "Also, as a premise, adding the customer's intentions might help clarify the options." (Figure 10, Step 53). When this suggestion is output (Figure 10, Step 54), the user responds with, "That's right, I forgot about the customer's intentions." (YES in Figure 10, Step 55), and the response language expression "In that case, why don't you ask your supervisor for their opinion and tell them what to prioritize?" is output to the user (Figure 10, Step 50). This allows for the output of response language expressions that are more appropriate to the user.

[0073] Figures 12 to 14 show modified examples. Figure 12 corresponds to Figure 4 and is a flowchart showing part of the processing procedure of the response language expression output device 1, while Figures 13 and 14 show examples of output, respectively.

[0074] Once the response language expression is obtained as described above (step 49), the response language expression output device 1 outputs a question to the user asking who the recipient (the person to whom the output response language expression is being shared) is (step 61). The user answers with the recipient, and when that answer is input to the response language expression output device 1, the response language expression is modified to a style appropriate to the recipient (step 62). This style modification is also generated by the GPU 3 using the trained style modification model 34.

[0075] Next, a question asking for the output format of the response language expression is output to the user from the response language expression output device 1 (step 63). The user is asked what output format they want, such as a report format like a word-processed letter, a presentation format, or an email format. The user selects the output format using the keyboard 10, microphone 8 (an example of a selection means), etc., and when the output format answered by the user is input to the response language expression output device 1, the GPU 3 (an example of an output format changing means) uses the trained model 35 for output format changing to change the output format as needed to have the output format of the answer, and a response language expression is generated. The response language expression having the output format of the answer is output from the response language expression output device 1 (step 50).

[0076] Figure 13 shows an example of a response language expression output in a letter-like report format.

[0077] Report 70 includes the report title 71, the name of the recipient 72, the date, the user's name 73, the specific case name 74, and the content (corresponding to the response language expression) 75. The report title 71, the name of the recipient 72, the date, the user's name 73, and the specific case name 74 are pre-entered into the response language expression output device 1.

[0078] Since the response language expression is output in the format of Report 70, the user does not need to reformat the response language expression into the report format.

[0079] Figure 14 shows an example of a response language expression output in presentation format.

[0080] Slides 80A and 80B are shown.

[0081] Slide 80A is the cover page and contains the title 81, date, and the user's name 82. Slide 80B is the second slide and contains the generated response language expression 83, modified for presentation purposes.

[0082] Since the response language is output in a presentation format, users do not need to convert the response language back into a presentation format.

[0083] Figures 15 and 16 show examples of modified examples. Figure 15 shows an example of a table representing the history of extracted structural elements. This history table is stored for each user in the SSD 11 of the response language expression output device 1 by the CPU 2 (which is an example of a memory control means).

[0084] The history table stores structural elements (purpose, premise, decision criteria, hypothesis, options) extracted from linguistic expressions entered by the user, as well as structural elements extracted from answers to questions or suggestions, each corresponding to the date and time of extraction.

[0085] For example, from the initially entered linguistic expression, the structural elements of the objective (judgment on whether to proceed) and the structural elements of the premise (Company A's response is the highest priority) are extracted, but the linguistic elements of other judgment axes, hypotheses, and options are not extracted. In contrast, the user is output with questions (or suggestions) that ask for answers regarding the content of the structural elements of the judgment axes, and the structural element of the judgment axis (delivery date) is obtained. The structural elements of the objective or premise are already obtained from the entered linguistic expression and are therefore considered already obtained, and the structural elements of hypotheses and options are not reported because no questions (or suggestions) that ask for answers regarding them are output. The same applies to the structural elements extracted from other linguistic expressions and answers.

[0086] Referring to the history table shown in Figure 15, it can be seen that the input language expressions often contain structural elements of purpose and assumptions, indicating that emphasis is placed on purpose and assumptions. Therefore, it is possible to generate response language expressions in which the answers to the structural elements of purpose and assumptions are also important.

[0087] Referring to Figure 16, a table representing the user's history is read (step 91). From the contents of the structural elements stored in that table, structural elements that the user considers important are found and weighted accordingly (step 92). For example, in the example shown in Figure 15, if the weight (importance) of the objective and premise structural elements is set to 1, then other structural elements are set to 0.5, for example, and response language expressions are generated in which the answers to the objective and premise structural elements are important. This makes it possible to obtain relatively detailed response language expressions for the structural elements that the user considers important.

[0088] (Main components) 1. Speech acquisition unit 2. Structural Analysis Module (Extraction of objectives / judgment criteria / hypotheses, etc.) 3. Advice generation module (omissions, question corrections, structural assistance) 4. Formatting Module (Conversion to Report / Proposal Format) 5. Output section (text, diagrams, summary) 6. History Recording and Learning Module

[0089] This invention provides an AI dialogue system that supports users by breaking down their judgments, doubts, and reports into the following three-layer structure. • Layer for organizing thought structure: Decompose user statements into logical structures such as "purpose, premise, decision criteria, conditions, options, and hypotheses." • Speech generation and proposal shaping layer: Based on that structure, organize from a different perspective, fill in gaps, ask follow-up questions, and offer advice. • Report Restructuring and Sharing Support Layer: Supports rephrasing, illustrating, and converting reports into formatted reports for communication with superiors and others. This allows users' own judgments to be "structurally verified and refined," and then converted into a "format for communication with others," thereby improving their judgment and communication skills.

[0090] (Components) This invention supports the shaping and sharing of a user's thought structure through the following components. 1. Speech acquisition unit Receives user's natural language speech (text / voice compatible). 2. Structural Analysis Module Classify statements into structural elements such as purpose, premise, judgment criteria, options, and hypotheses, and construct a logical map. 3. Advice Generation Module Based on the organized structure, it assesses omissions, contradictions, and the strength of hypotheses, and generates supplementary questions and additional advice on perspectives. 4. Shaping Module Based on the structure, it automatically formats the output into formats such as supervisor reports, proposal documents, and slide layout drafts. 5. Output Section Responses are presented in user-selectable formats, including text, tables, diagrams, and summaries. 6. History Recording and Learning Module By accumulating past decision structures and advice responses, we improve individual optimization and re-referencing.

[0091] In this embodiment, the user interacts with the AI ​​through Slack, a web application, or a voice input device. The spoken content is passed to a structural analysis module in real time, and the structure, such as "what is the purpose?" and "what decision criteria are being used?", is presented on the spot in the form of a visualization map or a table. Through the presentation of the structure in a way that makes the user feel "that's what I wanted to say," the AI ​​provides supplementary advice and clarification questions. If necessary, output suggestions such as "this format is good for telling your boss" or "this structure is good for explaining with a diagram" are automatically generated, allowing the AI ​​to present output that matches the user's purpose. This process is continuously optimized by accumulating historical data.

[0092] (Supplementary embodiment: Output format conversion function) The decision structure formed by the structural organization layer and the advice generation layer is reconfigured by the output format conversion unit according to its intended use. In this invention, each structured element (purpose, decision axis, hypothesis, supplementary perspective, etc.) can be automatically converted into different formats such as reports, proposals, and slides. For example, in report output, the hypothesis and decision axis are reconfigured as text in chronological and logical order and converted into a report format such as "background → consideration process → reason for decision → next action". Furthermore, when outputting as proposal material, each element is associated with "slide headings," "bullet points," and "supplementary fields," and presented in a visually easy-to-understand outline format. This realizes a function that consistently supports not only structured thinking but also the development of output in practical settings.

[0093] (Example of an embodiment) Example 1: A scene where a young employee consults with their superior. User: "I'm a little unsure whether we should proceed with the current plan..." → AI: "What are the assumptions behind this decision? Is the basis for the decision delivery date or quality?" → Structure presentation → Organization → Automatic generation of report document for superior Example 2: A scenario where you want to structure a proposal. User: "I need to create a proposal by the end of this week, so I want to finalize the main points." → AI: "Do you have a hypothesis? Is your objective internal approval or customer acquisition?" → The proposal slide structure will be output along with the structure map.

[0094] (Supplementary example: Structural linkage) The decision-making structure support function of the present invention, when combined with response style control structures in dialogue AI (e.g., relational temperature classification module, dialogue interaction state selection module, etc.), makes it possible to adjust the tone, endings, and pauses of questions and advice according to the context. This integrates thinking support and expression support, resulting in more natural and reliable dialogue support.

[0095] (Industrial applicability) This invention is applicable as an AI system that supports the visualization, advice, and shaping of decision-making structures in areas such as conversational AI, business decision support, educational support, proposal creation support, report organization in medical and nursing care settings, thought organization assistance at home, and issue organization in customer support, and has broad industrial applicability.

[0096] Conventional conversational AIs primarily focus on "informational responses" or "emotional empathy responses," lacking the functionality to deconstruct, organize, and reconstruct the user's decision-making structure. While some decision-making support AIs have features such as option evaluation, these mainly involve numerical evaluation and quantitative condition optimization, lacking a structure that supports the user's contextual and subjective judgment axes and hypothesis formation abilities. This invention, based on a three-layer model of structure extraction, advice, and shared transformation, possesses exceptional originality and inventiveness in its ability to generate a state where "structural thinking and communication are possible" even from meaningless consultations or disorganized thoughts.

[0097] Figure 17 is a block diagram showing the overall configuration of one embodiment of the decision structure support AI system according to the present invention.

[0098] This system receives user-generated information such as consultations, decisions, doubts, and proposal preparations in natural language, and has the function of structurally organizing, formatting, and supporting this information, and reconstructing it into a format that can be reported and shared with others as needed. User input (voice or text) is received by the speech acquisition unit (1 in the diagram), and then broken down and classified into logical elements such as "purpose," "premise," "decision axis," "hypothesis," and "options" by the structural analysis module (2).

[0099] Based on the obtained structure, the advice generation module (3) activates to detect missing elements, ambiguities in hypotheses, biases in judgment criteria, etc., and generates follow-up questions and advice (presentation of alternative perspectives). Subsequently, the organized structure and supplementary information are restructured by the formatting module (4) into a style and format that is easy for the user to share with others, such as a supervisor report, proposal, or summary slide, and presented from the output unit (5) in an appropriate form such as text, tables, or diagrams.

[0100] Furthermore, this system is equipped with a history recording and learning module (6) that records each user's thinking tendencies, expression style, and decision structure patterns, and utilizes them for reuse and optimization.

[0101] As shown in this figure, the configuration of the present invention is not merely a question-answering type, but is designed to structurally guarantee a series of processes that support, expand, shape, and share the user's judgment and thought structure.

[0102] Figure 18 is a structural organization step diagram showing the process of structurally organizing and classifying a user's natural language utterances in the decision structure support AI system according to the present invention.

[0103] When a user makes a free-form utterance to the AI ​​regarding consultation or decision-making, the utterance is processed by the structural analysis module. For example, if a user says, "I'm not sure if I should proceed with the current plan...", the system will not treat this input as a simple string, but will break it down and organize it into structural elements as follows.

[0104] • Purpose: The underlying intention behind this utterance (e.g., "decision to proceed"). • Premise: Known situation / facts (e.g., "Responding to Company A is the top priority") • Decision-making criteria: Indicators that serve as the basis for indecision (e.g., "Delivery time vs. Quality") • Hypothesis: Assumed outcome based on judgment (e.g., "Prioritizing quality may lead to delivery delays") • Options: Possible alternative actions (e.g., "Adjust the process," "Consult with your supervisor," etc.)

[0105] These structured elements are presented to the user as a structure map or tabular format, serving as a foundation for visualizing their own thought processes.

[0106] Figure 18 shows the overall process of separating, extracting, and visualizing meaning and structure rather than simply recording utterances, illustrating an example of the operation of the module that plays a central role in the "treating judgment as structure" function of the present invention.

[0107] Figure 19 is an advice generation logic diagram showing the processing flow when generating advice based on structural analysis results in the decision structure support AI system according to the present invention.

[0108] As shown in Figure 18, after the logical structure, including purpose, decision criteria, and hypotheses, is extracted from the user's utterance, the advice generation module shown in this figure is activated to evaluate whether there is any excess or deficiency of information or any inconsistencies in the structure.

[0109] First, when the structural analysis results are passed to the advice generation module, (1) a "gap detection" process is performed to identify elements that are not explicitly stated among the constituent elements (e.g., unclear decision criteria, omitted hypotheses, etc.). Next, (2) a "question generation" process is activated, generating questions that prompt the user to consider different perspectives. For example, supplementary questions such as "What are your decision criteria: delivery time, quality, or cost?" or "Do you have any hypotheses?" are included. Furthermore, (3) a "perspective addition / alternative structure presentation" process is executed, and supplementary advice is presented to broaden the user's perspective based on the decision structures of others and past cases for similar structures. An example of this is, "Adding 'customer expectations' to the premise might change your options."

[0110] In this way, the advice generation module helps deepen the user's thinking and improve the completeness of their judgment through "structural questioning" and "redesigning perspectives." This diagram is an important structural diagram that shows that the conversational AI can not only provide answers but also intervene in the user's logical structure itself in a supportive and editorial manner.

[0111] Figure 20 is a report reconstruction flowchart showing the processing procedure for reconstructing the extracted logical structure into a writing style and structure suitable for sharing and communicating with superiors and relevant parties, based on the judgment structure support AI system according to the present invention.

[0112] In this invention, based on structural elements (purpose, decision criteria, hypotheses, options, supplementary perspectives, etc.) extracted from the user's utterance, it is possible to automatically reconstruct and output reports and proposals according to their intended use by selecting a writing style, structure, and level of detail that is appropriate for the shared audience (e.g., superiors, team members, customers) and purpose (e.g., consultation, approval, status report).

[0113] The flow shown in this diagram consists of the following four steps. Step 1. Obtaining a structural map: Based on the analysis results in Figure 2, the constituent elements (issues, judgments, objectives, etc.) are obtained. Step 2: Select a template: Choose a writing style template (report, proposal, request, etc.) that suits the recipient and purpose. Step 3. Expression transformation process: Formatting the constituent elements to create a natural context, endings, and word order. Step 4. Output Format Conversion: Depending on user selection, the output format can be converted to text, report sheet (table format), slide layout (diagram), etc.

[0114] For example, if there is a situation where "prioritizing quality may lead to delays in delivery, so we would like to request a priority decision," • Report format (text) Subject: Consultation regarding the decision on the progress of the case Text: Currently, we are prioritizing quality, but this is creating a trade-off with deadlines. If you can clarify your priorities, we will proceed with resource adjustments. • Slide format (illustrations) Slide Heading: Consultation on Project Progress Decisions Bullet points: Concerns about delivery dates due to quality prioritization / Need for prioritization decisions / Preparation for resource adjustments

[0115] Thus, the present invention goes beyond mere advice generation and possesses the function of transforming and outputting a well-structured thought process in a form that can be understood by others. Figure 4 is a core diagram that visually demonstrates how the present invention can be applied in practical applications as a "reporting and proposal support AI," going beyond a "thought support AI," and clearly illustrates a part of a new intellectual support technology that simultaneously achieves the reconstruction of judgment structures and the improvement of explainability. [Explanation of symbols]

[0116] 1: Response language expression output device, 2: CPU, 3: GPU, 4: Display device, 5: Memory, 7: Communication device, 8: Microphone, 9: Speaker, 10: Keyboard, 12: CD drive, 20: AI server, 31: Trained model for structural element extraction, 32: Trained model for question generation, 33: Trained model for proposal generation, 34: Trained model for style modification, 35: Trained model for output format modification

Claims

1. A language expression input means that inputs language expressions provided by the user. A structural element extraction means extracts structural elements that indicate purpose, premise, decision axis, hypothesis, and options from the logical structure of the linguistic expression input from the above-mentioned linguistic expression input means. A determination means that compares structural elements defined as necessary for generating a response language expression with structural elements extracted by the structural element extraction means, and determines which structural elements among the defined structural elements are not included in the extracted structural elements. A preparatory expression generation means that generates a preparatory expression which is at least one of a question or a suggestion for obtaining the content of a structural element that is not included, depending on the structural element that is not included as determined by the above determination means, A preparation expression output means that outputs the preparation expression generated by the above preparation expression generation means, A response input means for inputting a response to a prepared expression output from the above prepared expression output means, A response language expression acquisition means that generates a logical structure that complements the structural elements based on the answer entered from the above response input means and the input language expression, and generates or acquires a response language expression based on the said logical structure, and A response language expression output means that outputs the response language expression obtained in the above-mentioned response language expression acquisition means, A response language expression output device equipped with the following features.

2. A response language expression output stop means stops outputting the response language expression until the preparation expression is output from the preparation expression output means. An output device for a response language expression according to claim 1, comprising:

3. A preparation expression output control means controls the preparation expression output means to output an additional preparation expression from the preparation expression output means between the end of outputting the response language expression obtained from the response to the above preparation expression and the above input language expression, and the time when a language expression from the user is input again from the language expression input means. The response language expression output device according to claim 1, further comprising:

4. The above means for obtaining the response language expression is: The response language expression output from the above-mentioned response language expression output means is obtained in a style appropriate to the sharing partner. The response language expression output device according to claim 1.

5. A selection means for selecting the output format of the response language expression, and The response language expression acquired by the response language expression acquisition means is provided with a format changing means that changes the response language expression to an output format corresponding to the output format selected by the selection means. The above-mentioned response language expression output means is The above format modification means outputs a response language expression of the modified output format. The response language expression output device according to claim 1.

6. The device further comprises a memory control means for controlling the memory to store for each user at least one of the structural elements included in the logical structure of the response to the above-mentioned prepared expression and the structural elements included in the logical structure of the above-mentioned input language expression. The above means for obtaining the response language expression is: Based on the structural elements stored in the above memory device, a response language expression is obtained that places emphasis on the structural elements that the user considers important. The response language expression output device according to claim 1.

7. The language expression input means inputs language expressions provided by the user, The structural element extraction means extracts structural elements representing the purpose, premise, decision axis, hypothesis, and options from the logical structure of the linguistic expression input from the linguistic expression input means. The determination means compares the structural elements defined as necessary for generating the response language expression with the structural elements extracted by the structural element extraction means, and determines which of the defined structural elements are not included in the extracted structural elements. The preparatory expression generation means generates a preparatory expression which is at least one of a question or a suggestion for obtaining the content of a structural element that is not included, according to the structural element that is not included, as determined by the determination means. The preparation expression output means outputs the preparation expression generated by the preparation expression generation means, The response input means inputs the response to the prepared expression output from the prepared expression output means, The response language expression acquisition means generates a logical structure that complements the structural elements based on the response input from the response input means and the input language expression, and generates or acquires a response language expression based on the logical structure. The response language expression output means outputs the response language expression acquired by the response language expression acquisition means. A method for outputting a response language representation.

8. A program that controls the computer of a response language expression output device, and is readable by this computer. The user provides a linguistic expression as input. From the logical structure of the input linguistic expression, structural elements representing the purpose, premise, decision criteria, hypothesis, and options are extracted. The system compares the structural elements defined as necessary for generating the response language expression with the extracted structural elements, and determines which of the defined structural elements are not included in the extracted structural elements. For structural elements that are determined not to be included in the extracted structural elements, generate at least one of the following questions or suggestions to obtain the content of the missing structural elements: Output the generated preparation expression, The user is prompted to input a response to the outputted prepared expression. Based on the input response and the input language expression, a logical structure is generated that complements the above structural elements, and a response language expression is generated or obtained based on that logical structure. A program that controls a response language expression output device to output the acquired response language expression.

9. A recording medium storing the program described in claim 8.

Images