Teacher data generation device, teacher data generation method, and teacher data generation program

Abstract

This teacher data generation device includes a search program generation processing unit, a search processing unit, and a teacher data generation unit. The search program generation processing unit generates a search program from a Pad diagram of a lower document. The search processing unit searches for the Pad diagram of the lower document by using the search program by using a search key sentence in a text document of an upper document. The teacher data generation unit generates teacher data representing a correspondence relationship between the upper document and the lower document.

Description

Teacher data generation device, teacher data generation method, and teacher data generation program

[0001] The present invention relates to a teacher data generation device, a teacher data generation method, and a teacher data generation program.

[0002] When creating business documents such as governance documents that describe goals and how to achieve them, it is necessary to check whether the methods for achieving the goals are described. If the goals are written in a higher-level document and the methods for achieving them are written in a lower-level document, the multiple documents will need to be compared to check.

[0003] Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova: "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding", Proceedings of NAACL-HLT 2019, pages 4171-4186. Pretrained Japanese BERT models released.<URL:https: / / www.nlp.ecei.tohoku.ac.jp / news-release / 3284 / >

[0004] To reduce the effort required for checking, a method using machine learning (Non-Patent Documents 1 and 2) that associates sentences in a document may be considered.

[0005] However, when using machine learning, training data is required that shows the correspondence between sentences in higher-level documents and sentences in lower-level documents. Creating training data requires labeling the implication relationships between sentences in higher-level documents and sentences in lower-level documents, but this labeling process is time-consuming. In addition, the basis for labeling decisions is subjective, and the thought process is unclear.

[0006] The present invention has been made in light of the above-mentioned circumstances, and its purpose is to provide a teacher data generation device, a teacher data generation method, and a teacher data generation program that make it easy to create teacher data.

[0007] One aspect of the present invention is a teacher data generation device that generates teacher data from a text document and Pad diagram of a higher-level document and a text document and Pad diagram of a lower-level document. The teacher data generation device has a search program generation processing unit, a search processing unit, and a teacher data generation unit. The search program generation processing unit generates a search program from the Pad diagram of the lower-level document. The search processing unit uses a search key sentence in the text document of the higher-level document to search the Pad diagram of the lower-level document using the search program. The teacher data generation unit generates teacher data that represents the correspondence between the higher-level document and the lower-level document.

[0008] One aspect of the present invention is a method for generating training data from a text document and a Pad diagram of a higher-level document and a text document and a Pad diagram of a lower-level document. The training data generation method includes the steps of generating a search program from the Pad diagram of the lower-level document, searching the Pad diagram of the lower-level document using the search program by utilizing a search key sentence in the text document of the higher-level document, and generating training data that indicates the correspondence between the higher-level document and the lower-level document.

[0009] One aspect of the present invention is a training data generation program that causes a computer having a processor and a storage device to execute at least some of the functions of the components of the training data generation device.

[0010] According to the present invention, a teacher data generation device, a teacher data generation method, and a teacher data generation program are provided that facilitate the creation of teacher data.

[0011] FIG. 1 is a block diagram illustrating the functional configuration of a teacher data generation device according to an embodiment. FIG. 2 is a diagram illustrating a notation method for Pad diagrams. FIG. 3 is a diagram illustrating an example of a Pad diagram. FIG. 4 is a diagram illustrating a data conversion table from Pad diagram elements to a search program. FIG. 5 is a flowchart of a search process and a teacher data generation process executed by the search processing unit and teacher data generation unit of FIG. 1. FIG. 6 is a diagram illustrating the processing of the flowchart of FIG. 5. FIG. 7 is a diagram illustrating a text document of a higher-level document and a path of a Pad diagram in an embodiment of the search process and teacher data generation process. FIG. 8 is a diagram illustrating a text document of a lower-level document and a Pad diagram in an embodiment of the search process and teacher data generation process. FIG. 9 is a diagram illustrating a path of a Pad diagram of a higher-level document, a search path of a Pad diagram of a lower-level document, and an output of a teacher data generation device in an embodiment of the search process and teacher data generation process. FIG. 10 is a block diagram illustrating the hardware configuration of a teacher data generation device according to an embodiment.

[0012] The present invention will be described below with reference to the accompanying drawings. The present invention relates to a training data generation device that generates training data representing the correspondence between a higher-level document and a lower-level document from a text document and a Pad diagram of the higher-level document and a text document and a Pad diagram of the lower-level document in a business document such as a governance document.

[0013] (Functional Configuration) First, the functional configuration of the teacher data generation device 10 according to the embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the functional configuration of the teacher data generation device 10 according to the embodiment.

[0014] The teacher data generation device 10 includes a search program generation processing unit 11, a search processing unit 12, and a teacher data generation unit 13.

[0015] The teacher data generation device 10 receives as input data a text document and a Pad diagram of a higher-level document of a business document such as a governance document, and a text document and a Pad diagram of a lower-level document.

[0016] A text document is a document created using text. Text documents are organized hierarchically. For example, a higher-level text document describes a goal, and a lower-level text document describes how to achieve that goal.

[0017] A Pad diagram is a diagram that represents the outline of a text document using a hierarchical structure and conditional branching. Pad diagrams are also called problem analysis diagrams and are one of the methods for describing structured programming. Structured programming is a method for describing a program structurally so that what the program is doing can be visually understood. Pad diagrams are a method for practicing structured programming, in which the process to be described is broken down into the order and conditions, and the process is detailed and specified for each unit, describing it in stages.

[0018] Here, the notation of a Pad diagram will be explained with reference to FIG. 2. The elements that make up a Pad diagram are concatenated processing and branch processing. Concatenated processing means that processing is performed in order. The concatenated processing shown in FIG. 2 means that first, processing A shown at the top is performed, and after processing A is completed, processing B shown at the bottom is performed. Branch processing means that the processing to be performed changes depending on the condition. Branch processing 1 shown in FIG. 2 means that processing A is performed if condition P is true. Branch processing 2 shown in FIG. 2 means that processing A is performed if condition P is true, and processing B is performed if condition P is false. Branch processing 3 shown in FIG. 2 means that processing A is performed if the value of condition P is value 1, processing B is performed if the value of condition P is value 2, and processing C is performed if the value of condition P is value 3. Branch processing 3 has three branches, but it may have more branches.

[0019] An example of a Pad diagram is shown in Figure 3. The depth of branching is shown to the right, and the length of processing is shown downward. A Pad diagram shows the outline of a text document with a hierarchical structure and conditional branching. Processing is executed from top to bottom and from left to right according to conditions.

[0020] The search program generation processing unit 11 generates a search program from the Pad diagram of the upper document and the Pad diagram of the lower document, and provides the generated search program to the search processing unit 12.

[0021] 4 is a diagram showing a data conversion table from Pad diagram elements to a search program. This diagram illustrates conversion from Pad diagram elements to C language. The search program generation processing unit 11 converts the Pad diagram elements according to this data conversion table.

[0022] The search program generation processing unit 11 converts the concatenation process into the following code.

[0023]

[0024] The search program generation processing unit 11 converts the branch process 1 into the following code.

[0025]

[0026] The search program generation processing unit 11 converts the branch processing 2 into the following code.

[0027]

[0028] The search program generation processing unit 11 converts the branch processing 3 into the following code.

[0029]

[0030] The search processing unit 12 uses a search program to search for key sentences (including words, phrases, clauses, etc.) in the text document of the higher-level document to search the Pad diagram of the lower-level document. The key sentences are, for example, the lowest-level sentences (including words, phrases, clauses, etc.) in the hierarchically organized text document.

[0031] First, the search processing unit 12 obtains a path from the root to the search key sentence in the Pad diagram of the parent document. The path from the root to the search key sentence can be obtained by tracing the processes in the Pad diagram backwards, from the search key sentence to the root. Furthermore, the search processing unit 12 obtains information about the path to the search key sentence. The path information includes sentences (including words, phrases, clauses, etc.) to be processed in each process represented by a square in the Pad diagram of the parent document included in the path. Note that each process represented by a square is a process for searching for the sentence to be processed and a process for obtaining the sentence to be processed. For example, the path information is a character string formed by combining sentences to be processed in each process included in the path.

[0032] Next, the search processing unit 12 uses the path information to search the Pad diagram of the lower document using a search program. The search processing unit 12 searches the Pad diagram of the lower document according to the hierarchical structure of the Pad diagram of the lower document and conditional branching based on the path information in the Pad diagram of the upper document. In the branching process, the search processing unit 12 searches for a character string included in the sentence to be processed in the branch destination process by searching for a partial match with the character string that is the path information, and determines the branch destination according to the search result.

[0033] The search processing unit 12 provides information on the path of the Pad diagram of the higher-level document and information on the search results to the training data generation unit 13. The information on the search results is information on the search path searched in the Pad diagram of the lower-level document. The search path information includes sentences (including words, phrases, clauses, etc.) to be processed in processing the Pad diagram of the lower-level document included in the search path. The search path information further includes corresponding sentences (including words, phrases, clauses, etc.) in the text document of the lower-level document that correspond to the search path information. The search path information may be either a sentence included in an element of the Pad diagram of the lower-level document included in the search path, or a corresponding sentence in the text document of the lower-level document that corresponds to the search path information.

[0034] The training data generation unit 13 generates training data that represents the correspondence between a higher-level document and a lower-level document. The training data generation unit 13 combines path information in the Pad diagram of the higher-level document with search path information in the Pad diagram of the lower-level document. Furthermore, the training data generation unit 13 uses a machine learning algorithm to estimate the implication relationship between the path information in the Pad diagram of the higher-level document and the search path information in the Pad diagram of the lower-level document, and assigns implication labels that represent the implication relationship to the combined path information in the Pad diagram of the higher-level document and the search path information in the Pad diagram of the lower-level document. The training data generation unit 13 outputs the path information in the Pad diagram of the higher-level document and the search path information in the Pad diagram of the lower-level document to which implication labels have been assigned as training data.

[0035] For example, the path information of the Pad diagram of the higher-level document is a search key sentence, and the search path information of the Pad diagram of the lower-level document is a corresponding sentence in the text document of the lower-level document that corresponds to the search path information. In this case, the training data generation unit 13 outputs data in which implication labels are assigned to the key sentence in the text document of the lower-level document and the corresponding sentence in the text document of the lower-level document as training data.

[0036] (Search Processing and Teacher Data Generation Processing) Next, the search processing and teacher data generation processing executed by the search processing unit 12 and teacher data generation unit 13 will be described with reference to Fig. 5 and Fig. 6. Fig. 5 is a flowchart of the search processing and teacher data generation processing executed by the search processing unit 12 and teacher data generation unit 13. Fig. 6 is a diagram illustrating the processing of the flowchart in Fig. 5.

[0037] In step S1, the search processing unit 12 acquires a key sentence V (including a word, phrase, clause, etc.) for a search in a text document of a higher-level document. The key sentence V for a search is, for example, the lowest-level sentence in a hierarchically organized text document.

[0038] In step S2, the search processing unit 12 acquires the Pad drawing P1 of the upper document.

[0039] In step S3, the search processing unit 12 traces the process of the Pad diagram P1 of the upper document backward from the key sentence of the search to the root, and acquires the path from the root to the key sentence V of the search in the Pad diagram P1 of the upper document.

[0040] In step S4, the search processing unit 12 acquires information about a path p1 from the root to the search key sentence V. The information about the path p1 includes the search key sentence V and sentences (including words, phrases, clauses, etc.) to be processed in the processing of the Pad diagram of the upper document included in the path p1. For example, the path information is a character string formed by combining sentences to be processed in the processing included in the path.

[0041] In step S5, the search processing unit 12 acquires the Pad image P2 of the lower-level document.

[0042] In step S6, the search processing unit 12 uses the information in path p1 to search the Pad drawing P2 of the lower document using a search program. Specifically, the search processing unit 12 searches the Pad drawing P2 of the lower document while determining whether the information in path p1 and the branching conditions in the Pad drawing P2 of the lower document are partially identical. That is, the search processing unit 12 proceeds with the search along the processing of the Pad drawing P2 of the lower document, which processes a sentence that partially matches the information in path p1. In the branching processing, the search processing unit 12 searches for a character string contained in the sentence to be processed in the branch destination processing by partially matching the character string that is the path information, and determines the branch destination according to the search results. As a result of the search, the search processing unit 12 obtains a path p2 from the root to the lowest-level processing K1 in the Pad drawing P2 of the lower document.

[0043] In step S7, the search processing unit 12 acquires information about the search path p2. The information about the search path p2 includes character strings contained in sentences (including words, phrases, clauses, etc.) to be processed in the processing of the Pad diagrams of the subordinate documents included in the search path p2. For example, the information about the search path p2 is a character string formed by combining sentences to be processed in the processing of the Pad diagrams of the subordinate documents included in the search path p2.

[0044] In step S8, the search processing unit 12 acquires the corresponding sentence K2 in the text document of the lower-level document that corresponds to the process K1 in the Pad drawing P2 of the lower-level document as information on the search path p2.

[0045] In step S9, the teacher data generation unit 13 combines information on path p1 in the Pad diagram P1 of the higher-level document with information on search path p2 in the Pad diagram P2 of the lower-level document to obtain a pair of information on path p1 and information on search path p2. For example, the teacher data generation unit 13 obtains a pair of a search key sentence V and a corresponding sentence K2. Furthermore, the teacher data generation unit 13 uses a machine learning algorithm to estimate the entailment relationship between the pair of the search key sentence V and the corresponding sentence K2, and assigns an entailment label representing the entailment relationship to the pair of the search key sentence V and the corresponding sentence K2. The teacher data generation unit 13 outputs the pair of the search key sentence V and the corresponding sentence K2 to which the entailment label has been assigned as teacher data.

[0046] (One Example of Search Processing and Training Data Generation Processing) An example of search processing and training data generation processing will be described below with reference to Figs. 7 to 9. Fig. 7 is a diagram showing a text document of a higher-level document and a path of a Pad diagram according to this example. Fig. 8 is a diagram showing a text document of a lower-level document and a Pad diagram according to this example. Fig. 9 is a diagram showing a path of a Pad diagram of a higher-level document, a search path of a Pad diagram of a lower-level document, and the output of a training data generation device according to this example.

[0047] As shown in FIG. 7 , the text document of the parent document includes the following sentences organized hierarchically: "Information Security Regulations," "Safety Measures," "Information System," "When Information is Stored," and "Setting Appropriate Access Authorities to Prevent Information Leaks and False Information from Being Found Due to Unwanted Viewing." Correspondingly, the path of the Pad diagram of the parent document is a series of processes that connect "Information Security Regulations," "Safety Measures," "Information System," "When Information is Stored," and "Setting Appropriate Access Authorities to Prevent Information Leaks and False Information from Being Found Due to Unwanted Viewing." The search processing unit 12 acquires the sentences to be processed that are written within the boxes. For example, the search processing unit 12 acquires the character string "Information Security Regulations / Safety Measures / Information System / When Information is Stored / Setting Appropriate Access Authorities to Prevent Information Leaks and False Information from Being Found Due to Unwanted Viewing" by combining these sentences with a delimiter " / " as path information of the Pad diagram of the parent document.

[0048] As shown in Figure 8, the text document of the lower document includes the following sentences organized hierarchically: "Security Implementation Standards: Safety Measures for Management Information and Information Devices," "Safety Measures for Management Information and Information Devices," "1. Safety Measures for Storing Classified Management Information," "1.1. Media," "1.1.1. Paper Media," "1.1.1. Electronic Media (CD, DVD, etc.)," ​​"1.2. Information Devices," "1.2.1. Client Terminal," "1.2.2. Server," and "1.2.3. External Storage Device."

[0049] In response to this, the search processing unit 12 performs a "Scene" branching process following the processing of the "Management Information / Information Device Security Measures Frame" according to the Pad diagram of the lower document. In the "Scene" branching process, the branching condition is whether the location is when information is stored, when information is used, or when information device is installed, and the branch destination is determined to be one of the processes for "Security measures when information is stored," "Security measures when information is used," or "Security measures when information device is installed."

[0050] The search processing unit 12 performs branching processing for "information storage item" following "safety measures when storing information." In the branching processing for "information storage item," the branching condition is whether the information storage item is a medium or an information device, and the branch destination is determined to be either a "medium" or an "information device" branching process. If there is no information and a decision cannot be made, the next hierarchical level processing is performed.

[0051] In the "medium" branching process, the search processing unit 12 determines whether the medium is a paper medium or an electronic medium (CD, DVD, etc.) as the branching condition, and determines the branch destination to be either a "paper medium" or an "electronic medium (CD, DVD, etc.)" process.

[0052] In the branching process for "information device," the search processing unit 12 determines the branch destination as branching process for "client terminal type," processing for "server," or processing for "external storage device," using the branching condition as whether the information device is a client terminal, a server, or an external storage device.

[0053] In the branching process for "client terminal type," the search processing unit 12 determines the branch destination to either the "smartphone" or "PC" process, using whether the client terminal type is a smartphone or a PC as the branching condition.

[0054] The lower-level document has a corresponding statement for processing "paper media" that reads "locking management within the management area," a corresponding statement for processing "electronic media (CDs, DVDs, etc.)" that reads "encryption settings," a corresponding statement for processing "smartphones" that reads "authentication settings," a corresponding statement for processing "PCs" that reads "wire lock installation," and a corresponding statement for processing "servers" that reads "access permission management settings."

[0055] 9, the path of the Pad diagram of the upper document is a series of processes that are connected in order from "information security regulations," "safety measures," "information system," "when information is stored," and "setting appropriate access permissions to prevent information leaks and input of incorrect information due to unnecessary viewing." The search processing unit 12 acquires, as path information of the Pad diagram of the upper document, the character string "information security regulations / safety measures / information system / when information is stored / setting appropriate access permissions to prevent information leaks and input of incorrect information due to unnecessary viewing," which is obtained by combining these sentences with the delimiter " / ."

[0056] The search processing unit 12 searches for the Pad diagram of the subordinate document using the path information of the Pad diagram of this superior document. In the search, for a conditional branch, the branch destination is determined by determining whether the character string corresponding to the conditional branch partially matches the character string that is the path information of the Pad diagram of the superior document. As a result, the search processing unit 12 obtains, as a search path, the processes and conditional branches that are connected to the "Management Information / Information Device Security Measures Frame," "Scene," "Safety Measures When Information Is Stored," "Information Storage Item," "Information Device," and "Server," which are surrounded by thick lines.

[0057] The search processing unit 12 also acquires character strings corresponding to the processing and conditional branching of the search path as information on the search path. For example, in response to the path information of the Pad diagram of the higher-level document, the search processing unit 12 acquires the character string "Management information / information device safety measures frame / Scene / Safety measures when storing information / Information storage item / Information device / Server" as information on the search path of the Pad diagram of the lower-level document, which is obtained by combining character strings corresponding to the processing and conditional branching of the search path with a delimiter " / ".

[0058] Furthermore, the search processing unit 12 acquires the character string "access authority management setting," which corresponds to the lowest-level process, from the text document of the lower document as search path information. That is, the search processing unit 12 ultimately acquires the character string "management information / information device safety measure frame / scene / safety measures when storing information / information storage item / information device / server / access authority management setting" as search path information.

[0059] The search processing unit 12 provides the teacher data generation unit 13 with the path information of the Pad diagram of the higher-level document, i.e., the string "Information security regulations / Safety measures / Information system / When storing information / Set appropriate access rights to prevent information leakage due to unnecessary viewing and the input of incorrect information." and the search path information of the Pad diagram of the lower-level document, i.e., the string "Safety measures frame for management information and information devices / Scene / Safety measures when storing information / Information storage items / Information devices / Server / Access right management settings."

[0060] The teacher data generation unit 13 combines the path information of the Pad diagram of the higher-level document with the search path information of the Pad diagram of the lower-level document. Furthermore, the teacher data generation unit 13 uses a machine learning algorithm to estimate the implication relationship between the path information of the Pad diagram of the higher-level document and the search path information of the Pad diagram of the lower-level document, and assigns implication labels representing the implication relationship to the path information of the Pad diagram of the higher-level document and the search path information of the Pad diagram of the lower-level document. The teacher data generation unit 13 outputs, as teacher data, character strings in which implication labels have been assigned to "Safety measures for management information and information devices / Scene / Safety measures when storing information / Information storage item / Information device / Server" and "Safety measures for management information and information devices / Scene / Safety measures when storing information / Information storage item / Information device / Server / Access permission management settings."

[0061] (Hardware Configuration) Next, a description will be given of the hardware configuration of the teacher data generation device 10. Here, an example in which the teacher data generation device 10 is configured by a computer 20 will be described.

[0062] 10 shows the hardware configuration of the computer 20 that constitutes the teacher data generation device 10. The computer 20 has a processor 21, a read-only memory (ROM) 22, a random access memory (RAM) 23, an auxiliary storage device 24, and an input / output interface 25.

[0063] The processor 21 , ROM 22 , RAM 23 , auxiliary storage device 24 , and input / output interface 25 are electrically connected to one another via a bus 26 , and exchange data via the bus 26 .

[0064] The processor 21 is configured by a general-purpose hardware processor including, for example, a CPU (Central Processing Unit), a GPU (Graphical Processing Unit), etc. The processor 21 controls the ROM 22, the RAM 23, the auxiliary storage device 24, and the input / output interface 25 as a whole.

[0065] The ROM 22 is a non-volatile memory that constitutes part of the main storage device. The ROM 22 non-temporarily stores a startup program required when the processor 21 is started. The processor 21 starts up by executing the program in the ROM 22. The ROM 22 is configured, for example, by an EPROM (Erasable Programmable Read Only Memory), and stores various startup settings in addition to the startup program.

[0066] The RAM 23 is a volatile memory that constitutes part of the main storage device. The RAM 23 temporarily stores programs required for processing by the processor 21 and data required for executing the programs. The processor 21 executes the programs in the RAM 23 to perform calculations on the data in the RAM 23 and store the calculation results in the RAM 23.

[0067] The auxiliary storage device 24 is configured by a non-volatile memory such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The auxiliary storage device 24 non-temporarily stores programs to be executed by the processor 21 and data required for executing the programs. The processor 21 loads the programs and data in the auxiliary storage device 24 into the RAM 23 and executes the programs to perform various functions.

[0068] The input / output interface 25 is connected to an external input device 31, an output device 32, etc., and enables input of information from the input device 31 and output of information to the output device 32. For example, the input / output interface 25 may be a wired interface or a wireless interface. The wired interface includes a port to which a device is connected, etc. The wireless interface includes Bluetooth (registered trademark), Wi-Fi (registered trademark), etc.

[0069] The input device 31 may include a keyboard, a mouse, a touch panel, a receiving device, a disk drive, etc. The input device 31 is not limited to these and may include any other input device. The output device 32 may include a display, a transmitting device, a disk drive, etc. The output device 32 is not limited to these and may include any other output device. The input device 31 and the output device 32 may be configured as an input / output device 33 that has the functions of both the input device 31 and the output device 32.

[0070] The input data, that is, the upper-level document / lower-level document correspondence matrix file, the supplementary learning data matrix file, and the ratio data file, are input to the search program generation processing unit 11 via the input device 31 .

[0071] The program non-temporarily stored in the auxiliary storage device 24 is provided to the computer 20, for example, via a recording medium 34 on which the program is non-temporarily recorded and which can be read by the computer 20. Such a recording medium 34 is called a non-temporarily computer-readable recording medium. Non-temporarily computer-readable recording media include disks such as flexible disks, optical disks (CD-ROM, CD-R, DVD-ROM, DVD-R, etc.), and magneto-optical disks (MO, etc.), as well as semiconductor memories.

[0072] The programs non-temporarily stored in the auxiliary storage device 24 include a teacher data generation program. The teacher data generation program is a program that causes the computer 20 to execute at least some of the functions of the components of the teacher data generation device 10, namely, the search program generation processing unit 11, the search processing unit 12, and the teacher data generation unit 13.

[0073] If the recording medium 34 is a disk, the program non-temporarily stored in the auxiliary storage device 24 is read into the auxiliary storage device 24 via a disk drive, which is the input device 31, and the input / output interface 25, or if the recording medium 34 is a semiconductor memory, via a port, which is the input / output interface 25, and non-temporarily stored therein. Alternatively, the program may be stored in a server on a network, downloaded from the server, and non-temporarily stored in the auxiliary storage device 24.

[0074] When the computer 20 starts up, the processor 21 executes a program in the ROM 22 and loads and starts the OS into the RAM 23. Under control of the OS, the processor 21 monitors input instructions, connections to external devices, and the like. Under control of the OS, the processor 21 also sets up a program area and a data area in the RAM 23. In response to an instruction to start the teacher data generation device 10, the processor 21 loads the teacher data generation program from the auxiliary storage device 24 into the program area of ​​the RAM 23 and loads data required for executing the teacher data generation program from the auxiliary storage device 24 into the data area of ​​the RAM 23. The processor 21 calculates data in the data area according to the teacher data generation program and writes the calculation results to the data area. Through these operations, the processor 21, RAM 23, auxiliary storage device 24, input / output interface 25, and bus 26 work together to execute at least some of the functions of the components of the teacher data generation device 10, namely, the search program generation processing unit 11, the search processing unit 12, and the teacher data generation unit 13.

[0075] (Effect) In the embodiment, the search program generation processing unit 11 generates a search program from the Pad diagram of the lower-level document, the search processing unit 12 uses the search key sentence in the text document of the upper-level document to search the Pad diagram of the lower-level document using the search program generated by the search program generation processing unit 11, and the training data generation unit infers an implication relationship from the path information of the Pad diagram of the upper-level document and the search path information of the Pad diagram of the lower-level document, and generates training data in which an implication label indicating the implication relationship is assigned to the path information of the Pad diagram of the upper-level document and the search path information of the Pad diagram of the lower-level document.

[0076] This eliminates the need for the laborious task of labeling implications to create training data. Furthermore, since the estimation of implication relationships is performed according to the hierarchical structure and conditional branching of the Pad diagrams of lower-level documents, estimation can be performed taking the context into account. This allows for appropriate implication relationships to be estimated. Finally, a technology is provided that makes it easy to create training data.

[0077] The present invention is not limited to the above-described embodiments, and various modifications can be made in the implementation stage without departing from the spirit of the invention. Furthermore, the embodiments may be implemented in appropriate combinations, in which case the combined effects can be obtained. Furthermore, the above-described embodiments include various inventions, and various inventions can be extracted by combining selected elements from the disclosed elements. For example, if the problem can be solved and the desired effect can be obtained even if some elements are deleted from all elements shown in the embodiments, the configuration from which these elements are deleted can be extracted as an invention.

[0078] 10... Teacher data generation device 11... Search program generation processing unit 12... Search processing unit 13... Teacher data generation unit 20... Computer 21... Processor 22... ROM 23... RAM 24... Auxiliary storage device 25... Input / output interface 26... Bus 31... Input device 32... Output device 33... Input / output device 34... Recording medium

Claims

1. A teacher data generation device that generates teacher data from a text document and Pad diagram of a higher-level document and a text document and Pad diagram of a lower-level document, comprising: a search program generation processing unit that generates a search program from the Pad diagram of the lower-level document; a search processing unit that uses a search key sentence in the text document of the higher-level document to search the Pad diagram of the lower-level document using the search program; and a teacher data generation unit that generates teacher data that represents the correspondence between the higher-level document and the lower-level document.

2. The teacher data generation device according to claim 1, wherein the search processing unit obtains information on a path from the root to the key sentence in the Pad diagram of the higher-level document, and uses the information on the path to search the Pad diagram of the lower-level document.

3. The teaching data generating device according to claim 2, wherein the search processing unit searches the Pad diagram of the lower document according to the hierarchical structure and conditional branching of the Pad diagram of the lower document based on the information of the path in the Pad diagram of the higher-level document.

4. The teacher data generating device according to claim 3, wherein the teacher data generating unit combines the information of the path with information of the search results and outputs the combined information as the teacher data.

5. The teacher data generation device of claim 4, wherein the teacher data generation unit further combines the information of the path and the information of the search results with corresponding sentences in the text document of the lower-level document that correspond to the information of the search results, and outputs the combined information as the teacher data.

6. The teacher data generating device of claim 4, wherein the teacher data generation unit estimates an implication relationship between the information of the path and the information of the search results, assigns an implication label representing the implication relationship to the information of the path and the information of the search results, and outputs the information as the teacher data.

7. A method for generating teacher data from a text document and Pad diagram of a higher-level document and a text document and Pad diagram of a lower-level document, comprising the steps of: generating a search program from the Pad diagram of the lower-level document; searching the Pad diagram of the lower-level document using the search program by utilizing a search key sentence in the text document of the higher-level document; and generating teacher data that represents the correspondence between the higher-level document and the lower-level document.

8. A teacher data generation program that causes a computer having a processor and a storage device to execute at least some of the functions of the components of the teacher data generation device described in any one of claims 1 to 6.

Images