Information processing system, information processing apparatus, information processing method, and computer program

Abstract

To suppress a risk of information leakage that may occur due to use of an external service.SOLUTION: An information processing system S according to an embodiment includes a proper noun detection section 103 that detects a proper noun from one or a plurality of sentences when execution of a predetermined process on a processing target sentence including one or a plurality of sentences is instructed, a converted sentence generation section 104 that generates a converted sentence by converting the proper noun detected by the proper noun detection section 103 into another proper noun in a sentence in which a proper noun is detected by the proper noun detection section 103 in the processing target sentence, and a query generation section 105 that generates a query including a processing requesting sentence in which a sentence in which a proper noun is detected by the proper noun detection section 103 in the processing target sentence is replaced with the converted sentence and a sentence in which a proper noun is not detected by the proper noun detection section 103 in the processing target sentence is maintained as it is and an execution request of the predetermined process on the processing requesting sentence.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to an information processing system, an information processing device, an information processing method, and a computer program, and in particular to a technique for reducing the risk of information leakage. [Background technology]

[0002] Generative AI (Generative Artificial Intelligence) is expected to be increasingly used in various fields in the future. Recently, generative AI has been rapidly gaining popularity, and many of these models are natural language processing models. Natural language processing models can, for example, understand natural language created by humans and generate information based on that understanding.

[0003] Specifically, a natural language processing model can understand a query (question) written by a human and generate an answer to the query. A natural language processing model can also recognize the content of natural language speech and generate document data corresponding to the speech.

[0004] Various technologies related to natural language processing have been proposed in the past, and for example, Patent Document 1 discloses a technology for natural language processing of speech. This technology divides speech data into multiple speech segments based on natural language processing, and assigns security-related information to the speech segments. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2023-28391 Summary of the Invention [Problem to be solved by the invention]

[0006] Currently, most of the generative AI available to the general public is managed in data centers owned by companies. When a user uses such a generative AI, the user sends a query to the data center from a terminal device operated by the user, for example, and receives an answer to the query from the generative AI in the data center.

[0007] However, some users have raised concerns about information leakage regarding the above-mentioned usage. Specifically, some users are concerned that confidential information may be leaked via a data center if the query contains confidential information.

[0008] In response to these concerns, companies that manage generative AI generally take strict security measures, but there are still a certain number of users who are hesitant to use external services such as generative AI without any restrictions.

[0009] The present invention has been made in consideration of the above background, and aims to provide an information processing system, an information processing device, an information processing method, and a computer program that can reduce the risk of information leakage that may arise from the use of external services. [Means for solving the problem]

[0010] The present invention relates to the following aspects.

[0011] a proper noun detection unit that detects proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation unit that generates a converted sentence by converting the proper noun detected by the proper noun detection unit into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection unit; an information processing system comprising: a query generation unit that generates a query including a processing request sentence that replaces sentences in which proper nouns have been detected by the proper noun detection unit among the sentences to be processed with the converted sentences, and leaves sentences in which proper nouns have not been detected by the proper noun detection unit among the sentences to be processed as they are, and a request to execute the specified processing on the processing request sentence.

[0012] a proper noun detection unit that detects proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation unit that generates a converted sentence by converting the proper noun detected by the proper noun detection unit into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection unit; An information processing device comprising: a query generation unit that generates a query including a processing request sentence that replaces sentences in which proper nouns have been detected by the proper noun detection unit among the sentences to be processed with the converted sentences, and leaves sentences in which proper nouns have not been detected by the proper noun detection unit among the sentences to be processed as they are, and a request to execute the specified processing on the processing request sentence.

[0013] a proper noun detection step of detecting proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation step of generating a converted sentence by converting the proper noun detected by the proper noun detection step into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection step; an information processing method comprising: a query generation step of generating a query including a processing request sentence that replaces a sentence in which a proper noun is detected by the proper noun detection step among the processing target sentences with the converted sentence, and that leaves as it is a sentence in which a proper noun is not detected by the proper noun detection step among the processing target sentences; and a request to execute the specified processing on the processing request sentence.

[0014] a proper noun detection step of detecting proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation step of generating a converted sentence by converting the proper noun detected by the proper noun detection step into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection step; A computer program that causes a computer to execute a query generation step of generating a query including a processing request sentence that replaces sentences in which proper nouns are detected by the proper noun detection step among the processing target sentences with the converted sentences, and that leaves sentences in which proper nouns are not detected by the proper noun detection step among the processing target sentences as they are, and a request to execute the specified processing on the processing request sentence. [Effects of the Invention]

[0015] According to the present invention, it is possible to reduce the risk of information leakage that may occur when using external services. [Brief explanation of the drawings]

[0016] [Figure 1] 1 is a diagram schematically illustrating a configuration of an information processing system according to a first embodiment. [Figure 2] 2 is a block diagram showing the functional configuration of each unit constituting the information processing system shown in FIG. 1. FIG. [Figure 3] 10 is a flowchart illustrating the flow of an ambiguity check process performed in the information processing system shown in FIG. [Figure 4] 10 is a flowchart illustrating the flow of a logic check process performed in the information processing system shown in FIG. [Figure 5] 2 is a diagram specifically explaining the ambiguity check process performed in the information processing system shown in FIG. 1 and the conversion process of proper nouns in a sentence performed during the ambiguity check process. FIG. [Figure 6] FIG. 6 is a diagram showing the processing result of the ambiguity check processing described in FIG. 5. [Figure 7] 2 is a diagram specifically explaining the logic check process performed in the information processing system shown in FIG. 1 and the conversion process of proper nouns in a sentence performed during the logic check process. [Figure 8] FIG. 8 is a diagram showing the processing results of the logic check processing described in FIG. 7. [Figure 9] FIG. 10 is a diagram schematically illustrating a configuration of an information processing system according to a second embodiment. [Figure 10] FIG. 10 is a diagram schematically illustrating a configuration of an information processing system according to a third embodiment. [Figure 11] FIG. 2 is a diagram illustrating a hardware configuration of an information processing device that can constitute a terminal device and a server that constitute the information processing system according to each embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0017] Each embodiment will be described below.

[0018] <<First embodiment>> <Outline of information processing system configuration> 1 shows a schematic configuration of an information processing system S according to the first embodiment. The information processing system S includes a terminal device 10 as an information processing device, a first server 21, and a second server 22.

[0019] The terminal device 10, the first server 21, and the second server 22 are connected to each other via a network 40 so as to be able to transmit and receive information to and from each other. The network 40 is a communication network including, for example, the Internet, a Wide Area Network (WAN), a Local Area Network (LAN), etc. The terminal device 10, the first server 21, and the second server 22 are connected to the network 40 wirelessly or via a wired connection, and can communicate with external devices. The functional configuration of the information processing system S will be described in detail below.

[0020] <Functional configuration of information processing system> 2 is a block diagram showing the functional configuration of each part (terminal device 10, first server 21, and second server 22) that constitutes the information processing system S. The information processing system S performs predetermined processing on a processing target text that includes one or more sentences that are generated or stored in the terminal device 10. Specifically, the predetermined processing performed by the information processing system S is a processing that checks ambiguity in a text (ambiguity check processing) and a processing that detects whether or not two or more contradictory sentences exist among a plurality of sentences (logic check processing).

[0021] The text handled by the information processing system S refers to information for communicating intentions expressed in natural language and that can be handled (displayed, stored, etc.) on a computer. The predetermined processing performed by the information processing system S is not limited to the two examples mentioned above, and the predetermined processing may be translation, summarization, proofreading, extraction of incorrect information, extraction of detailed information, etc. of the text.

[0022] In the information processing system S according to this embodiment, when a user makes a request (instruction) to execute the ambiguity check process for a sentence on the terminal device 10, a process of converting some or all of the proper nouns in the sentence is performed. In this embodiment, this conversion process is performed in cooperation between the terminal device 10 and the first server 21. Thereafter, the terminal device 10 transmits a request to execute the ambiguity check process to the second server 22, and the target of the process at this time is set to the sentence on which the above-mentioned conversion has been performed (hereinafter, may be referred to as the converted sentence). The second server 22 then derives a response to the request to execute the ambiguity check process and transmits it to the terminal device 10. The logic check process is also performed in the same procedure as the above-mentioned ambiguity check.

[0023] As described above, the information processing system S converts proper nouns in a sentence to be processed into other proper nouns, and then requests an external device (in this example, the second server 22) to execute the predetermined processing. This enables a user to request an external device to execute predetermined processing on the sentence while ensuring the confidentiality of the sentence stored in, for example, the terminal device 10. Each unit of the information processing system S, particularly the functional configuration of each unit, will be described in detail below.

[0024] (Terminal Device) The terminal device 10 is a device operated by a user, such as a personal computer, a smartphone, a tablet, etc. The terminal device 10 includes an ambiguous expression detection and presentation unit 101, a logic check unit 102, a proper noun detection unit 103, a converted sentence generation unit 104, a query generation unit 105, and a query transmission unit 106.

[0025] The ambiguous expression detection and presentation unit 101 performs processing to detect at least one of ambiguous terms and phrases in a processing target sentence including one or more sentences that have been subjected to ambiguity check processing. When performing the ambiguity check processing, the ambiguous expression detection and presentation unit 101 generates a request to execute ambiguity check processing on the processing target sentence in response to an operation from a user.

[0026] The logic check unit 102 compares multiple sentences included in a processing target sentence that is the subject of logic check processing, and performs processing to check (verify) whether or not two or more sentences that are in a contradictory (logically contradictory) relationship exist among the multiple sentences. In other words, a processing target sentence that is the subject of logic check processing includes at least two or more sentences. When performing logic check processing, the logic check unit 102 generates a request to execute logic check processing to verify the existence of contradictory sentences in response to instructions from a user.

[0027] When a request to execute an ambiguity check process or a logic check process is generated as described above, the proper noun detection unit 103 detects proper nouns from one or more sentences in the processing target sentence that is the target of the ambiguity check process or the logic check process. In this embodiment, the proper noun detection unit 103 detects proper nouns based on information in a proper noun database that holds information on multiple proper nouns. The proper noun detection unit 103 then transmits the proper noun detection result to the converted sentence generation unit 104.

[0028] The proper noun database may be stored within the terminal device 10 or in an external device. The proper noun detection unit 103 detects proper nouns by, for example, extracting multiple unit terms from a sentence to be detected for proper nouns using morphological analysis and verifying whether the extracted terms exist in the proper noun database. Note that proper nouns include personal names, organization names (company names, association names, group names, etc.), country names, place names, song titles, copyrighted work titles, trademark names, and specific, unambiguous service names. Note that proper nouns may be registered arbitrarily in the proper noun database. Furthermore, the proper noun detection unit 103 may detect proper nouns in a sentence using a natural language processing model instead of using the proper noun database. However, in this case, it is preferable to use a natural language processing model stored within the terminal device 10. Furthermore, if the confidentiality of the information can be sufficiently ensured, it may also be possible to use a natural language processing model from an external device.

[0029] In this embodiment, as an example, the user interface is configured so that the user specifies part or all of a sentence in document data created in advance and then requests an ambiguity check process. Therefore, in this embodiment, the sentences to be processed in the ambiguity check process are one or more sentences that the user specifies from the document data before requesting execution of the ambiguity check process. Then, the one or more sentences to be processed in the ambiguity check process specified from the document data are the sentences in which proper nouns are detected. Note that document data means electronic data containing at least one sentence.

[0030] In addition, in this embodiment, as an example, the user interface is configured so that the user inputs (selects) multiple document data created in advance and then issues a request for logic check processing. In this embodiment, the target of the logic check processing is the logic between a sentence included in one of the multiple document data and a sentence included in the other document data. Therefore, in this embodiment, the target sentences to be processed in the logic check processing are all of the multiple sentences included in the multiple document data. All of the sentences included in the multiple document data are the sentences to be detected for proper nouns. Note that the target sentences to be processed in the logic check processing do not have to be determined as in the above example, and may be, for example, two or more arbitrarily selected sentences.

[0031] The converted sentence generation unit 104 acquires other proper nouns to replace the proper nouns based on the detection results of the proper nouns sent from the proper noun detection unit 103. The converted sentence generation unit 104 also extracts sentences in which proper nouns have been detected from the sentences to be processed, and generates converted sentences by converting the proper nouns detected by the proper noun detection unit 103 in the extracted sentences into the other acquired proper nouns. The converted sentence generation unit 104 then transmits the generated converted sentences to the query generation unit 105.

[0032] In this embodiment, the converted sentence generation unit 104 acquires the other proper nouns to be used in generating the converted sentences through natural language processing using a learning model. Specifically, the converted sentence generation unit 104 requests the first learning model 211 in the first server 21 to present other proper nouns to replace the proper nouns detected by the proper noun detection unit 103. However, the converted sentence generation unit 104 does not have to acquire the other proper nouns using a learning model. For example, the converted sentence generation unit 104 may randomly acquire other proper nouns from a proper noun database using random numbers or the like.

[0033] Furthermore, in this embodiment, when generating converted sentences, the converted sentence generation unit 104 converts only person names, only organization names, or only person names and organization names detected by the proper noun detection unit 103 into other proper nouns. The proper noun detection unit 103 may also detect types of proper nouns other than person names and organization names. Here, converting an excessively large number of types of proper nouns when generating converted sentences may complicate processing and increase the processing load. Therefore, in this embodiment, only proper nouns for which confidentiality is considered highly desirable are targeted for conversion. However, the types of proper nouns targeted for conversion by the converted sentence generation unit 104 are arbitrary, and for example, all detected proper nouns may be converted.

[0034] Note that the proper noun detection unit 103 may detect multiple identical proper nouns from one or more sentences. In this case, the converted sentence generation unit 104 acquires other proper nouns that are common to the multiple identical proper nouns. Therefore, when the proper noun detection unit 103 detects multiple identical proper nouns from one or more sentences, in this embodiment, the converted sentence generation unit 104 makes a presentation request to the first learning model 211 to present one other common proper noun in place of the multiple detected identical proper nouns.

[0035] The query generation unit 105 receives the converted sentence and the entire processing target sentence that is the target of the ambiguity check process or the logic check process from the converted sentence generation unit 104, and generates a query that includes a processing request sentence in which the converted sentence generated by the converted sentence generation unit 104 is reflected in the processing target sentence, and a request to execute a predetermined process on the processing request sentence (a request to execute the ambiguity check process or the logic check process). In detail, the query generation unit 105 replaces sentences in which proper nouns have been detected by the proper noun detection unit 103 from the processing target sentence that has been the target of the ambiguity check process or the logic check process with the converted sentence, and generates a processing request sentence that leaves as it is sentences in which proper nouns have not been detected by the proper noun detection unit 103 from the processing target sentence, and links this to the request to execute the predetermined process.

[0036] More specifically, the query related to the request to execute the ambiguity check process includes a processing request sentence obtained by converting part or all of the sentences in the processing target sentence designated by the user before the request to execute the ambiguity check process into converted sentences in accordance with the result of the proper noun detection unit 103, and a request (command) to execute the ambiguity check process on the processing request sentence. In this embodiment, the query related to the request to execute the logic check process includes a processing request sentence obtained by converting part or all of the sentences in the processing target sentence, which is all the sentences in multiple document data input before the request to execute the logic check process, into converted sentences in accordance with the result of the proper noun detection unit 103, and a request (command) to execute the logic check process on the processing request sentence.

[0037] The query sending unit 106 receives a query from the query generation unit 105 and sends the query to the second server 22. The second server 22 stores a second learning model 221. In this embodiment, natural language processing by the second learning model 221 derives a response to a request to execute an ambiguity check process and a response to a request to execute a logic check process. After receiving the query, the second server 22 sends the response to the request to execute an ambiguity check process or the response to the request to execute a logic check process to the terminal device 10. Here, in this embodiment, the learning model (second learning model 221) to which the query is sent is different from the learning model (first learning model 211) used by the converted sentence generation unit 104. This improves the confidentiality of sentences sent from the terminal device 10 to the outside.

[0038] The response to the request to execute the ambiguity check process derived by the second server 22 includes information on ambiguous terms and / or phrases, or information indicating that such ambiguity does not exist. In this embodiment, the response to the request to execute the ambiguity check process may also include information on at least one of one or more corrective terms, one or more corrective phrases, and one or more modifiers that correct the ambiguity of the expression that has been determined to be an ambiguous term and / or phrase (response regarding suggested corrective expressions).

[0039] When the response from the second server 22 includes information on at least one of ambiguous terms and phrases in the sentence, the ambiguous expression detection and presentation unit 101 presents at least one of the relevant terms and phrases, for example, on a display of the terminal device 10. The presentation manner by the ambiguous expression detection and presentation unit 101 is not particularly limited, but may be, for example, a manner in which the ambiguous terms and / or phrases are displayed in a space different from the display space of the sentence in the document data on the display, and text notifying that the ambiguity exists. Furthermore, together with the display of the text, the ambiguous terms and / or phrases in the sentence in the document data on the display may be highlighted.

[0040] Furthermore, when a response regarding the proposed modified expression is also transmitted, the ambiguous expression detection and presentation unit 101 presents information corresponding to the response on the display of the terminal device 10. The response regarding the proposed modified expression here refers to information such as a modified term and a modified phrase (alternative expression) for an ambiguous term and / or phrase, or the addition of a modifier. A modifier is a concept that includes a noun, a collocation (phrase) of a noun and a particle, an adjective, and an adverb.

[0041] Furthermore, the response to the request to execute the logic check process derived by the second server 22 includes information on multiple sentences that may be contradictory to each other and information indicating the existence of multiple sentences that may be contradictory to each other. A contradictory relationship refers to a state in which the existence of mutually exclusive facts is recognized between two sentences. Simple examples of contradictions include the relationship between "A is B" and "A is not B" and the relationship between "raise A" and "lower A." In this embodiment, the logic check process is performed by the second learning model 221 of the second server 22 as described above. The second learning model 221 can detect contradictions, for example, by generating and classifying sentence vectors.

[0042] When the response from the second server 22 includes information indicating the existence of two or more sentences that are in a contradictory relationship, the logic check unit 102 presents, for example, a warning on, for example, the display of the terminal device 10. The manner in which the information is presented by the logic check unit 102 is not particularly limited, but for example, it may be such that text indicating the existence of multiple sentences that may be contradictory is displayed in a space on the display different from the display space for the sentences in each of the multiple document data.

[0043] Note that when the ambiguous expression detection and presentation unit 101 displays a response to the request to execute the ambiguity check process, for example, on a display, the converted sentence is not displayed. Similarly, when the logic check unit 102 displays a response to the request to execute the logic check process, for example, on a display, the converted sentence is not displayed.

[0044] (First server) The first server 21 is a device that presents other proper nouns that can be used in place of the proper nouns detected from one or more sentences by the proper noun detection unit 103 of the terminal device 10. In this embodiment, the first server 21 includes a first learning model 211, which presents other proper nouns that can be used in place of the proper nouns based on the information of the proper nouns that are transmitted.

[0045] More specifically, the first server 21 receives a request from the converted sentence generation unit 104 of the terminal device 10 to present another proper noun in place of the proper noun detected by the proper noun detection unit 103 of the terminal device 10. Upon receiving the request to present the proper noun, the first server 21 acquires the other proper noun using the first learning model 211 and transmits it to the converted sentence generation unit 104.

[0046] In this embodiment, the first learning model 211 is a natural language processing model. Therefore, the first learning model 211 can understand sentences. In this case, various forms of a request to the first learning model 211 to present other proper nouns are conceivable, and for example, a natural language question such as, "Please present a (virtual) proper noun that is different from the 'detected proper noun (e.g., a person's name)'" may be used. When the first learning model 211 is a natural language processing model, it is preferably a BERT (Bidirectional Encoder Representations from Transformers) natural learning model. Note that the first server 21 receives only information about proper nouns from the converted sentence generation unit 104, and therefore cannot grasp the entire content of the sentence, including the proper noun. Therefore, the confidentiality of the sentences handled by the terminal device 10 is ensured.

[0047] Furthermore, in generating converted sentences, the converted sentence generation unit 104 in this embodiment converts only the person's name, or only the organization name, or only the person's name and organization name, among the proper nouns detected by the proper noun detection unit 103, into other proper nouns. Therefore, depending on the detection result of the proper noun detection unit 103, the converted sentence generation unit 104 requests the first learning model 211 to present another name in place of the person's name, or another organization name in place of the organization name, or a person's name and organization name in place of each of the person's name and organization name.

[0048] The first learning model 211 may be a general-purpose generation AI managed in a data center owned by a company. The first learning model 211 may also be a learning model created by a user and stored in the first server 21, or may be a learning model created by an operator of the information processing system S, which is different from the user and the company that owns the data center.

[0049] (Second server) The second server 22 is a device that receives a query regarding a request to execute an ambiguity check process or a query regarding a request to execute a logic check process from the terminal device 10 and presents a response to the request. The second server 22 includes a second learning model 221. The response to the request is derived from the second learning model 221.

[0050] The second learning model 221 is a natural language processing model that functions to understand sentences included in processing request sentences provided from the terminal device 10 and to verify ambiguity and logical contradictions. The second learning model 221 is preferably a natural learning model of BERT (Bidirectional Encoder Representations from Transformers).

[0051] The second learning model 221 may be a general-purpose generation AI managed in a data center owned by a company. The first learning model 211 may be a learning model created by a user and stored in the first server 21, or may be a learning model created by an operator of the information processing system S, which is different from the user and the company that owns the data center.

[0052] The second learning model 221 uses, for example, training text as input data and pre-trains learning data in which at least one of the ambiguous terms and phrases in the training text is used as correct answer data. As a result, when the second learning model 221 receives a text as input data, it can derive at least one of the ambiguous terms and phrases as an answer and provide it to the terminal device 10. Note that the ambiguous terms and phrases derived by the first learning model 211 are basically predetermined by pre-training.

[0053] In addition, the second learning model 221 pre-learns learning data in which, for example, training sentences are used as input data and corrected sentences obtained by correcting the training sentences are used as correct answer data. As a result, when the second learning model 221 receives a sentence as input data, it can derive a corrected sentence as an answer. In more detail, in this embodiment, the second learning model 221 provides the terminal device 10 with information on differences and changes between the sentences used as input data and the corrected sentences, such as corrective terms, corrective phrases, or the addition of modifiers to ambiguous terms and / or phrases.

[0054] Furthermore, the second learning model 221 uses, for example, first training sentences and second training sentences different from the first training sentences as input data, and pre-trains learning data in which tags indicating logical agreement (consistency) or contradiction (divergence) between the first training sentences and the second training sentences are used as correct answer data. As a result, when the second learning model 221 receives multiple sentences as input data, it can derive contradictions between the sentences and provide them to the terminal device 10.

[0055] In this embodiment, the second learning model 221 can perform the following functions: 1) detection of ambiguous terms and phrases; 2) deriving information such as adding corrective terms, corrective phrases, and modifiers to such ambiguous terms and / or phrases; and 3) verifying logical consistency (integrity) between sentences. However, these multiple processing functions may be implemented in separate learning models. In this case, for example, a learning model having the above functions 1) and 2) may be stored in the second server 22, and a learning model having the above function 3) may be stored in another server.

[0056] <Processing by information processing system> An example of processing by the information processing system S will be described below. Fig. 3 is a flowchart illustrating the flow of the ambiguity check processing performed by the information processing system S. Fig. 4 is a flowchart illustrating the flow of the logic check processing performed by the information processing system S. First, the ambiguity check processing will be described with reference to Fig. 3.

[0057] (Ambiguous check processing) 3, a flowchart showing the processing of the terminal device 10 is shown on the left side, and a flowchart showing the processing of the second server 22 is shown on the right side. The processing of the terminal device 10 starts with the launch of a dedicated browser that accepts instructions for the ambiguity check processing and the logic check processing. The second server 22 is always running, and advances the processing in conjunction with the terminal device 10. Although a flowchart related to the processing of the first server 21 is not shown, the first server 21 is always running, and advances the processing in conjunction with the terminal device 10.

[0058] In the terminal device 10, first, in step S101, input of document data is monitored. When document data is input (YES in step S101), in step S102, it is monitored whether or not a sentence has been specified by the user in the document data. If document data is not input in step S101 (NO in step S101), as shown by the transition "A" in the figure, the process proceeds to step S109, where an end instruction or the occurrence of another event is monitored. If an end instruction or the occurrence of another event is not received in step S109, input of document data is monitored again in step S101.

[0059] In step S102, it is monitored whether the user has specified all or part of the text included in the document data. The manner of specifying the text monitored in step S102 may be, for example, a manner in which the user specifies all or part of the text included in the document data to be displayed on a display using a mouse, a touch panel, or the like.

[0060] If a sentence is specified in step S102 (YES in step S102), then in step S103 it is monitored whether or not an instruction for ambiguity check processing has been given by the user. The sentence specified in step S102 corresponds to the sentence to be processed. If no sentence is specified in step S102 (NO in step S102), then the process proceeds to step S109, where an end instruction or the occurrence of another event is monitored. If no end instruction or the occurrence of another event is found in step S109, then the process returns to step S101 and the specification of the sentence is monitored again in step S102.

[0061] When the instruction to perform the ambiguity check process is confirmed in step S103 (YES in step S103), a process of detecting proper nouns in the sentence specified in step S102 (proper noun detection step) is performed in step S104. This proper noun detection process is performed by the proper noun detection unit 103 in the terminal device 10. The proper noun detection unit 103 then provides the proper noun detection results to the converted sentence generation unit 104. The proper noun detection unit 103 detects proper nouns by, for example, extracting multiple single-unit terms from the sentence in which proper nouns are to be detected by morphological analysis, and verifying whether the extracted terms exist in a proper noun database, although the detection method is not particularly limited.

[0062] Next, in step S105, a sentence in which a proper noun is detected is extracted from the sentences specified in step S102, and a converted sentence is generated by converting the detected proper noun in the sentence to another proper noun (converted sentence generation step). The converted sentence is generated by cooperation between the converted sentence generation unit 104 of the terminal device 10 and the first server 21. In this embodiment, the converted sentence generation unit 104 requests the first learning model 211 of the first server 21 to present another proper noun to replace the proper noun detected by the proper noun detection unit 103.

[0063] Then, the converted sentence generation unit 104 receives from the first learning model 211 a proposal of another proper noun to replace the proper noun detected by the proper noun detection unit 103, and then generates a converted sentence using the received other proper noun. The converted sentence generation unit 104 then sends the converted sentence to the query generation unit 105. To explain an example of a converted sentence, for example, if the proper noun "President Hashimoto" is detected by the proper noun detection unit 103, the first learning model 211 is requested to propose another name to replace "President Hashimoto." For example, if the first learning model 211 presents the name "President Kawakami," the name "President Kawakami" is used in place of "President Hashimoto" in the converted sentence. If the instruction is not confirmed in step S103 (NO in step S103), the process proceeds to step S109, where an end instruction or the occurrence of another event is monitored. If there is no end instruction or occurrence of another event in step S109, the process passes through steps S101 and S102 and returns to step S103 to monitor for an instruction.

[0064] Next, in step S106, the query generation unit 105 generates a query including a processing request sentence in which the converted sentence generated by the converted sentence generation unit 104 is reflected in the processing target sentence, and a request to execute a predetermined process on the processing request sentence (query generation step). Thereafter, the query sending unit 106 sends the query generated by the converted sentence generation unit 104 to the second server 22 (query sending step). Here, the query generated by the query generation unit 105 includes a processing request sentence in which part or all of the sentence in the processing target sentence specified by the user before the request to execute the ambiguity check process is converted into a converted sentence according to the result of the proper noun detection unit 103, and a request (command) to execute the ambiguity check process on the processing request sentence.

[0065] When the query is transmitted to the second server 22 in step S106, the second server 22, which has been waiting to receive the query from the terminal device 10 in step S121, confirms the reception and proceeds to step S122. Then, in step S122, the second learning model 221 derives the result of the ambiguity check process.

[0066] The result of the ambiguity check process derived by the second learning model 221 corresponds to a response to a request to execute the ambiguity check process. In this embodiment, the response may include at least one of information about ambiguous terms and / or phrases in the sentence, information indicating their absence, and information such as a corrective term for the ambiguous term and / or phrase or the addition of a modifier.

[0067] When the result (answer) of the ambiguity check process is derived in step S122, in step S123, the second server 22 transmits information about the result of the ambiguity check process to the terminal device 10. After the transmission process is performed in step S123, the second server 22 waits to receive a query in step S121.

[0068] Returning to the processing on the terminal device 10 side, when information on the result of the ambiguity check processing is transmitted to the terminal device 10 in step S123, the ambiguous expression detection and presentation unit 101 receives the information in step S107.

[0069] Then, in step S108, the ambiguous expression detection and presentation unit 101 presents, for example, on a display, as a response to the request to execute the ambiguity check process, at least one of an ambiguous term and / or phrase, one or more corrective terms, one or more corrective phrases, and one or more modifiers that correct the ambiguity of the ambiguous term and / or phrase. If there are no ambiguous terms and phrases, the ambiguous expression detection and presentation unit 101 presents a message to that effect.

[0070] Thereafter, in step S109, the occurrence of an end instruction or other event is monitored. If no end instruction or other event has occurred, the process returns to step S101 and the specification of the sentence is monitored again in step S102. If the occurrence of an end instruction or other event is confirmed, the ambiguity check process ends.

[0071] (Logical check processing) Next, the logic check process will be described with reference to Fig. 4. In Fig. 4, a flowchart showing the process of the terminal device 10 is shown on the left side, and a flowchart showing the process of the second server 22 is shown on the right side.

[0072] In the terminal device 10, first, in step S201, the input of multiple document data items associated with each other is monitored. When document data items are input (YES in step S201), in step S202, it is monitored whether or not a logic check process instruction has been issued by the user. All of the multiple sentences in the multiple document data items input in step S201 correspond to the sentence to be processed. If no document data items are input in step S201 (NO in step S201), as shown by the transition "B" in the figure, the process proceeds to step S208, where an end instruction or the occurrence of another event is monitored. If no end instruction or the occurrence of another event is detected in step S208, the input of document data items is monitored again in step S201.

[0073] If an instruction to perform a logic check process is confirmed in step S202 (YES in step S202), then in step S203, a process to detect proper nouns in the sentences in the document data input in step S201 (proper noun detection step) is performed. This proper noun detection process is performed by the proper noun detection unit 103 in the terminal device 10. The proper noun detection unit 103 then provides the proper noun detection result to the converted sentence generation unit 104. The process by the proper noun detection unit 103 is performed in the same manner as in step S104. If the instruction is not confirmed in step S202 (NO in step S202), the process proceeds to step S208, where an end instruction or the occurrence of another event is monitored. If an end instruction or the occurrence of another event is not confirmed in step S208, the process returns to step S201 and the instruction is monitored again in step S202.

[0074] Next, in step S204, a sentence in which a proper noun is detected is extracted from the multiple sentences in the multiple document data input in step S201, and the detected proper noun in the sentence is converted into another proper noun to generate a converted sentence (converted sentence generation step). The converted sentence is generated by cooperation between the converted sentence generation unit 104 of the terminal device 10 and the first server 21. In this embodiment, the converted sentence generation unit 104 requests the first learning model 211 of the first server 21 to present another proper noun to replace the proper noun detected by the proper noun detection unit 103.

[0075] Then, the converted sentence generation unit 104 receives from the first learning model 211 a presentation of another proper noun to replace the proper noun detected by the proper noun detection unit 103, and then generates a converted sentence using the received other proper noun.The converted sentence generation unit 104 then sends the converted sentence to the query generation unit 105.

[0076] Next, in step S205, the query generation unit 105 generates a query including a processing request sentence in which the converted sentence generated by the converted sentence generation unit 104 is reflected in the processing target sentence, and a request to execute a predetermined process on the processing request sentence (query generation step). More specifically, the query generated by the query generation unit 105 here includes a processing request sentence in which part or all of the sentences in the processing target sentence, which is all sentences in multiple document data input before the request to execute the logic check process, are converted into converted sentences according to the results of the proper noun detection unit 103, and a request (command) to execute a logic check process on the processing request sentence. Thereafter, the query transmission unit 106 transmits the query generated by the converted sentence generation unit 104 to the second server 22 (query transmission step).

[0077] When the query is sent to the second server 22 in step S205, the second server 22, which has been waiting to receive the query from the terminal device 10 in step S221, confirms the reception (YES in step S221) and proceeds to step S222. Then, in step S222, the second learning model 221 derives the result of a logic check process that verifies the existence of contradictory sentences among the multiple mutually related document data received.

[0078] In more detail, the logic check unit 102 in this embodiment uses the second learning model 221 to compare a sentence included in one of a plurality of mutually associated document data stored in the terminal device 10 with a sentence included in the other document data, and detects whether or not there is a contradictory relationship between the sentence included in one of the plurality of document data and the sentence included in the other document data. However, the processing by the logic check unit 102 is not limited to comparing sentences between a plurality of document data. For example, the processing by the logic check unit 102 may be an aspect in which documents that are in a contradictory relationship are detected by comparing a plurality of documents within a single document data or a plurality of documents specified unrelated to the document data.

[0079] The result of the logic check process derived by the second learning model 221 corresponds to a response to a request to execute the logic check process. In this embodiment, the response includes information on multiple sentences that may be contradictory to each other and information indicating the existence of multiple sentences that may be contradictory to each other.

[0080] When the result of the logic check process is derived in step S222, in step S223, the second server 22 transmits information regarding the result of the logic check process to the terminal device 10. After the transmission process is performed in step S223, the second server 22 waits to receive a query in step S221.

[0081] Returning to the processing on the terminal device 10 side, when information regarding the result of the logic check processing is transmitted to the terminal device 10 in step S223, the logic check unit 102 receives information regarding the verification result (answer) of the logic check processing in step S206.

[0082] Then, in step S207, the logic check unit 102 presents information corresponding to the response to the request to execute the logic check process, for example, on a display. Here, for example, if a potentially contradictory sentence is detected, a notification to that effect may be displayed on the display, and the corresponding sentence may be highlighted.

[0083] Thereafter, in step S208, the occurrence of an end instruction or other event is monitored. If no end instruction or other event has occurred, the input of document data is monitored again in step S201, and if the occurrence of an end instruction or other event is confirmed, the logic check process ends.

[0084] (Specific image of the process) Hereinafter, a specific image of the ambiguity check process and the logic check process in this embodiment will be described with reference to FIGS.

[0085] First, Fig. 5 is a diagram specifically explaining the ambiguity check process. Fig. 5(A) shows an image of the interface displayed on the display of the terminal device 10 when the ambiguity check process is executed. Reference numeral 51 in Fig. 5(A) indicates the title of the input document data ("Minutes" in the illustrated example). Below the title 51, the text contained in the document data is displayed. The box indicated by reference numeral 52 indicates that a portion of the text has been specified by the user. When such a specification is made, a pop-up 53 is displayed to prompt the execution of the ambiguity check process. Then, when the pop-up 53 is operated with, for example, a hand or a cursor 54, a request to execute the ambiguity check process is generated.

[0086] FIG. 5(B) is a conceptual diagram of the generation of converted sentences performed during the execution of the ambiguity check process. FIG. 5(B) shows that three proper nouns, designated by reference numerals 55 to 57, were detected in the sentence specified in FIG. 5(A). Reference numeral 58 indicates the converted sentence. In converted sentence 58, the personal names and corporate names among the proper nouns detected in the sentence specified in FIG. 5(A) have been converted into other proper nouns. Specifically, in converted sentence 58, the corporate name "Company S (reference numeral 55)," which is a proper noun, has been converted into "Company T." The personal name "President Hashimoto (reference numeral 56)," which is a proper noun, has been converted into "President Kawakami." "Company T" and "President Kawakami" are information proposed by the first learning model 211.

[0087] 5(C) is a conceptual diagram of a query 59 related to a request to execute an ambiguity check process. The query 59 includes, as query contents, a request to execute an ambiguity check process (ambiguity check request) and a processing request sentence including the converted sentence 58 shown in FIG. 5(B) as the target sentence of the request. Such a query is transmitted to the second server 22.

[0088] Fig. 6 shows how a response to a request to execute an ambiguity check process based on query 59 is displayed on a display. In Fig. 6, under the title "Feedback" 61, the response to the ambiguity check for the sentence is displayed, which includes an ambiguous term and / or phrase, one or more corrective terms, one or more corrective phrases, and modifiers that correct the ambiguity of the term and / or phrase. In Fig. 6, information indicated by reference numerals 62 and 64 corresponds to information on an ambiguous term and / or phrase. Information indicated by reference numeral 63 corresponds to one or more corrective terms, one or more corrective phrases, and modifiers that correct the ambiguity.

[0089] 7A and 7B are diagrams specifically explaining the logic check process. Fig. 7A shows an image of an interface displayed on the display of the terminal device 10 when the logic check process is executed. Reference numeral 71 in Fig. 7A indicates the title of the input document data (in the illustrated example, "New Business Development, B Corporation"). As an example, the multiple document data are associated by the title 71. Below the title 71, the text contained in the document data, the date it was created, etc. are displayed.

[0090] More specifically, in column 72, sentences 72A and 72B regarding decisions in the first document data created on October 30, 2023 are displayed. In column 73, sentence 73A regarding decisions in the second document data created on March 30, 2024 is displayed. The interface also shows a logic check execution button 74. Operating the logic check execution button 74 generates a request to execute a logic check process.

[0091] Figure 7(B) is a conceptual diagram of the generation of converted sentences performed during the execution of the logic check process. The left side of Figure 7(B) shows sentence 72B in the first document data shown in Figure 7(A). It is shown that three proper nouns, designated by symbols 75a, 75b, and 75c, were detected in sentence 72B. Symbol 77 shows the converted sentence of sentence 72B. In converted sentence 77, the proper nouns, namely, the person's name and the organization name, detected in sentence 72B, have been converted into other proper nouns.

[0092] The right side of Figure 7(B) shows sentence 73A in the second document data shown in Figure 7(A). It is shown that three proper nouns, designated by reference numerals 75d, 75e, and 75f, were detected in sentence 73A. Reference numeral 78 shows the converted sentence of sentence 73A. In converted sentence 78, the proper nouns, namely, the person's name and the organization's name, detected in sentence 73A, have been converted to other proper nouns. It is noteworthy here that when a common proper noun is detected between sentences 72B and 73A belonging to different document data, the other proper noun that replaces the common proper noun is also common. Specifically, "Company B" detected between sentences 72B and 73A is converted to the same other proper noun, "Company F," in the converted sentence.

[0093] 7(C) is a conceptual diagram of a query 79 relating to a request to execute a logic check process. Query 79 includes, as query content, a request to execute a logic check process (logic check request) and a processing request sentence including converted document data including converted document data including converted document 77 shown in FIG. 7(B) as the target sentence of the request, and converted document data including converted document 78. Such a query is sent to the second server 22.

[0094] Figure 8 shows the display of a response to a request to execute a logic check process based on a query 79. In Figure 8, a notification 80 indicating the existence of a contradictory sentence is displayed at the top of the interface. Furthermore, the contradictory sentence is highlighted by being enclosed in a dashed-dotted line.

[0095] The information processing system S according to the first embodiment described above comprises a proper noun detection unit 103 that detects proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including one or more sentences; a converted sentence generation unit 104 that generates a converted sentence by converting the proper noun detected by the proper noun detection unit 103 into another proper noun in a sentence of the processing target sentence in which a proper noun has been detected by the proper noun detection unit 103, into another proper noun; and a query generation unit 105 that generates a query including a processing request sentence that replaces a sentence of the processing target sentence in which a proper noun has been detected by the proper noun detection unit 103 with a converted sentence and leaves as is a sentence of the processing target sentence in which a proper noun has not been detected by the proper noun detection unit 103, and a request to execute a predetermined process on the processing request sentence.

[0096] In such a configuration, a process of converting proper nouns in a document on which a predetermined process (in this example, an ambiguity check process or a logic check process) is to be performed is performed, and then a request for the predetermined process is made to an external device (in this example, the second server 22), thereby making it possible to request an external device to perform a predetermined process on a text while ensuring the confidentiality of the text held by a user in the terminal device 10. This makes it possible to reduce the risk of information leakage that may occur when using an external service such as a generation AI.

[0097] <<Second embodiment>> 9 is a diagram showing a schematic configuration of an information processing system Sr1 according to the second embodiment. In the following embodiments, components common to those of the first embodiment are assigned the same reference numerals, and redundant explanations will be omitted.

[0098] The information processing system Sr1 shown in Fig. 9 includes a terminal device 10 and a first server 21, but does not include a second server 22. The first server 21 has a first learning model 211 and a second learning model 221. With such a configuration, the system is simplified.

[0099] <<Third embodiment>> FIG. 10 is a diagram illustrating a schematic configuration of an information processing system Sr2 according to a third embodiment. The information processing system Sr2 includes a terminal device 10, a first server 21, a second server 22, and a third server 100. The terminal device 10 does not include a proper noun detection unit 103, a converted sentence generation unit 104, a query generation unit 105, or a query transmission unit 106. On the other hand, the third server 100 includes the proper noun detection unit 103, the converted sentence generation unit 104, the query generation unit 105, and the query transmission unit 106. In this configuration, the third server 100 receives a request from the terminal device 10 to execute an ambiguity check process or a logic check process, and then generates the converted sentence and acquires a response to the execution request. This configuration reduces the processing load and memory usage of the terminal device 10.

[0100] <<Example of hardware configuration of information processing device>> 11 is a diagram showing the hardware configuration of an information processing device 400 that can constitute the terminal device 10, the first server 21, the second server 22, and the third server 100. The information processing device 400 includes a processor 401, a ROM 402, a RAM 403, a communication interface 404, an operation device 405, and a display device 406, which are interconnected by a bus.

[0101] The processor 401 may be a CPU, a GPU, or a computing device that combines a CPU and a GPU. The processor 401 executes, for example, a computer program stored in a ROM 402 on a RAM 403. The computer program is a program that can realize, for example, each of the above-mentioned functional configurations of the terminal device 10, the first server 21, the second server 22, and / or the third server 100. The computer program may be realized not by a single program, but by a combination of multiple programs or scripts.

[0102] The ROM 402 is a non-transitory computer-readable storage medium that stores computer programs, data required for executing the computer programs, and data generated by executing the computer programs. These computer programs and data are read into the RAM 403 when the computer programs are executed. The ROM 402 may be, for example, but is not limited to, a hard disk, an optical disk, a flash memory, or a magnetic tape.

[0103] The computer program may be pre-installed in the information processing device 400 or may be stored in a non-transitory computer-readable storage medium such as a CD-ROM. The computer program may also be uploaded to the Internet. The RAM 403 may be, for example, a DRAM or an SRAM, but is not limited to these. The information processing device 400 may further include a storage device that can read and write various types of information.

[0104] The communication interface 404 is a circuit for the terminal device 10, the first server 21, the second server 22, and / or the third server 100 to communicate wirelessly or wired with external devices. The operation device 405 is a circuit for inputting operation signals from input devices such as a keyboard, a mouse, and a touch panel to the terminal device 10, the first server 21, the second server 22, and / or the third server 100. The display device 406 displays data stored in, for example, the terminal device 10, the first server 21, the second server 22, and / or the third server 100. The display device 406 may be, for example, an LCD (liquid crystal display), an organic electroluminescence display, or the like.

[0105] Furthermore, the information processing device 400 may be configured as a single device, or may be configured as a system made up of multiple computer devices connected to each other.

[0106] The above-described embodiment shows only one example for realizing the present invention, and the present invention can be implemented in various other forms. For example, various modifications, substitutions, omissions, or combinations thereof are possible without departing from the spirit of the present invention. Such modifications, substitutions, omissions, etc. are also included within the scope of the present invention, as well as the scope of the inventions described in the claims and their equivalents. [Explanation of symbols]

[0107] S, Sr1, Sr2...Information processing system 101...Ambiguous expression detection and presentation unit 102...Logical check section 103...Proper noun detection unit 104...Conversion sentence generation unit 105...Query generation unit 106...Query sending unit 21...First server 22...Second server 211…First Learning Model 221...Second Learning Model 40…Network

Claims

1. a proper noun detection unit that detects proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation unit that generates a converted sentence by converting the proper noun detected by the proper noun detection unit into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection unit; an information processing system comprising: a query generation unit that generates a query including a processing request sentence that replaces sentences in which proper nouns have been detected by the proper noun detection unit among the sentences to be processed with the converted sentences, and leaves sentences in which proper nouns have not been detected by the proper noun detection unit among the sentences to be processed as they are, and a request to execute the specified processing on the processing request sentence.

2. The information processing system according to claim 1 , wherein the converted sentence generation unit converts only person names, only organization names, or only person names and organization names among the proper nouns detected by the proper noun detection unit into other proper nouns.

3. The information processing system according to claim 1 , wherein the proper noun detection unit detects the proper noun based on information in a proper noun database that holds information on a plurality of proper nouns.

4. The information processing system according to claim 1 , wherein the converted sentence generation unit acquires the other proper nouns through natural language processing using a learning model.

5. a query sending unit that sends the query to a learning model that performs natural language processing; The information processing system according to claim 4 , wherein the learning model to which the query is sent is different from the learning model used by the converted sentence generation unit.

6. 2. The information processing system according to claim 1, wherein the predetermined process is a process of detecting at least one of ambiguous terms or phrases in the process request text.

7. 2. The information processing system according to claim 1, wherein the predetermined process is a process of detecting whether or not two or more sentences in the process request sentence are in a contradictory relationship.

8. a proper noun detection unit that detects proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation unit that generates a converted sentence by converting the proper noun detected by the proper noun detection unit into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection unit; An information processing device comprising: a query generation unit that generates a query including a processing request sentence that replaces sentences in which proper nouns have been detected by the proper noun detection unit among the sentences to be processed with the converted sentences, and leaves sentences in which proper nouns have not been detected by the proper noun detection unit among the sentences to be processed as they are, and a request to execute the specified processing on the processing request sentence.

9. a proper noun detection step of detecting proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation step of generating a converted sentence by converting the proper noun detected by the proper noun detection step into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection step; an information processing method comprising: a query generation step of generating a query including a processing request sentence that replaces a sentence in which a proper noun is detected by the proper noun detection step among the processing target sentences with the converted sentence, and that leaves as it is a sentence in which a proper noun is not detected by the proper noun detection step among the processing target sentences; and a request to execute the specified processing on the processing request sentence.

10. a proper noun detection step of detecting proper nouns from one or more sentences when an instruction is given to execute a predetermined process on a processing target sentence including the one or more sentences; a converted sentence generation step of generating a converted sentence by converting the proper noun detected by the proper noun detection step into another proper noun in a sentence from among the sentences to be processed in which the proper noun has been detected by the proper noun detection step; A computer program that causes a computer to execute a query generation step of generating a query including a processing request sentence that replaces sentences in which proper nouns are detected by the proper noun detection step among the processing target sentences with the converted sentences, and that leaves sentences in which proper nouns are not detected by the proper noun detection step among the processing target sentences as they are, and a request to execute the specified processing on the processing request sentence.

Images