Text verification support device and text verification support method

Abstract

It assists in verifying correspondences between texts. [Solution] A text verification support device 1 comprising a storage device 12 for storing a first text and a second text, and a calculation device 11 that performs a related portion search process for searching for a portion of the second text related to the first text based on a predetermined algorithm, a generation process for outputting text representing the relationship between the first text and the related portion by inputting the first text and the related portion into a trained model 23 that outputs text corresponding to an input text, and a determination process for outputting information indicating the degree of relationship between the first text and the second text to an output device 14 based on the output text.

Description

Technical Field

[0001] The present invention relates to a text verification support device and a text verification support method.

Background Art

[0002] In software development, it is important to ensure the traceability of the work products at each development stage. For example, in each stage, it is necessary to check whether the requirements described in the requirements specification are sufficiently reflected in the design document. By ensuring traceability, for example, it is possible to prevent omissions or leaks in the software specifications and designs, and to prevent inconsistencies between the two.

[0003] As a method for managing traceability in software development, for example, in Patent Document 1, when a work product created in a certain process of developing a certain system is input, a predetermined keyword is searched from the input work product, and the searched text, and attribute information such as page numbers and file names are extracted. A traceability matrix is created by associating the text, page numbers, file names, and other attribute information extracted from the work products created in other processes of developing the system with the extracted information, and the created traceability matrix is stored in a work product database together with the work products. A work product management device is disclosed.

Prior Art Documents

Patent Documents

[0004] [[ID=2,6]]<0,000020>

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Patent Document 1 envisions detecting texts and other elements corresponding to deliverables through manual keyword searches. However, as software development projects become larger, manually setting complex and numerous keywords becomes cumbersome. Furthermore, even if the same keywords are used, there is a high probability that irrelevant texts will be detected. It also becomes difficult to adequately verify the correspondence between the deliverables and their corresponding original texts, such as whether there are inconsistencies or omissions in the deliverables.

[0006] This invention has been made in view of these circumstances, and its objective is to provide a text verification support device and a text verification support method that can assist in verifying correspondences between texts. [Means for solving the problem]

[0007] One of the present inventions for solving the above problems is a text verification support device comprising a storage device for storing a first text and a second text, a related portion search process for searching for a portion of the second text related to the first text based on a predetermined algorithm, a generation process for outputting text representing the relationship between the first text and the related portion by inputting the first text and the related portion into a trained model that outputs text corresponding to an input text, and a determination process for outputting information indicating the degree of relationship between the first text and the second text to an output device based on the output text. [Effects of the Invention]

[0008] According to the present invention, it is possible to support the verification of correspondences between texts.

[0009] Other issues, configurations, and effects not mentioned above will be clarified by the following description of embodiments for carrying out the invention.

[0010] Other configurations and effects will be clarified by the following description of the embodiments. [Brief explanation of the drawing]

[0011] [Figure 1] This figure shows an example of the hardware and functions of a text verification support device. [Figure 2] This is a flowchart illustrating an example of text validation support processing. [Figure 3] This figure shows an example of a question creation request prompt. [Figure 4] This figure shows an example of a response generation request prompt. [Figure 5] This figure shows an example of a screen that appears when the design document is consistent with the requirements specification document. [Figure 6] This figure shows an example of a screen that appears when the design document does not match the requirements specification document. [Figure 7] This figure shows an example of a screen that appears when the related design document section has deficiencies compared to the requirements section. [Figure 8] This figure shows an example of the traceability matrix that will be output. [Figure 9] This figure shows another example of the traceability matrix that will be output. [Modes for carrying out the invention]

[0012] Embodiments of the present invention will be described in detail below with reference to the drawings.

[0013] The text verification support device in this embodiment is an information processing device that uses a pre-trained model, described later, to assist in verifying the correspondence (consistency, etc.) between a text such as a requirements definition document (first text) and another text created in response to that text (second text).

[0014] Figure 1 shows an example of the hardware and functions of a text verification support device.

[0015] The text verification support device 1 includes an arithmetic device 11 such as a CPU (Central Processing Unit), a storage device 12 such as a RAM (Random Access Memory), a ROM (Read Only Memory), and an SSD (Solid State Drive), an input device 13 such as a keyboard, a mouse, or a touch panel, an output device 14 such as a display or a touch panel, and a communication device 15 composed of a NIC (Network Interface Card), a wireless communication module, a USB (Universal Serial Interface) module, or a serial communication module.

[0016] The text verification support device 1 stores a requirement definition document 21 (the first text), a design document 22 (the second text), and a learned model 23.

[0017] The requirement definition document 21 is a text document related to software under development or already developed. For example, it is a text describing requirements (business, functions, non - functions, etc.) to be realized by the software.

[0018] The design document 22 is a text document of more specific development content of software based on the content of the requirement definition document 21. For example, it includes a screen layout, a list of functions, etc.

[0019] The learned model 23 is a mathematical model that outputs a text (natural - language text) corresponding to the input text (natural - language text). The learned model 23 is, for example, a large - language model (LLM: Large Language Model). The learned model 23 is constructed by, for example, BERT (Bidirectional Encoder Representations from Transformers), XLNet, or GPT (Generative Pre - trained Transformer).

[0020] The text verification support device has various functional units, including a requirements extraction unit 24, a requirements selection unit 25, a design document division unit 26, a related part search unit 27, a problem answer generation unit 28, and a determination unit 29.

[0021] The requirements extraction unit 24 extracts the section describing the requirements (requirements list 30) from the requirements definition document 21.

[0022] The requirements selection unit 25 selects one or more requirements from the requirements list 30 extracted by the requirements extraction unit 24. Hereinafter, the selected requirements will be referred to as the requirements portion 31.

[0023] On the other hand, the design document division unit 26 divides the design document 22 into multiple parts. Hereinafter, each of the divided parts will be referred to as a design document chunk 32.

[0024] The related portion search unit 27 searches for the portion of the design document 22 (specifically, the design document chunk 32) that is related to the requirements definition document 21 (specifically, the requirements portion 31) (hereinafter referred to as the related design document portion 33).

[0025] Specifically, the related portion search unit 27 searches for design document chunks 32 that satisfy the search conditions, based on whether the design document chunks 32 contain keywords extracted from the requirements definition document 21, and identifies them as related design document portions 33. In this embodiment, the related portion search unit 27 identifies the selected design document chunk from among the candidates of multiple design document chunks that satisfy the search conditions as the related design document portion 33.

[0026] The problem answer generation unit 28 outputs text that represents (explains) the relationship between the requirements definition document 21 and the related design document portion 33 by inputting the requirements definition document 21 and the related design document portion 33 into the trained model 23. Specifically, the problem answer generation unit 28 includes a problem generation unit 34 and an answer generation unit 35.

[0027] The problem generation unit 34 inputs the requirements section 31 of the requirements specification document 21 into the trained model 23, and outputs a problem 36 and an answer 38 to problem 36, which have questions asking about the content of the requirements section 31 of the requirements specification document 21. The problem generation unit 34 uses the content of the requirements section 31 of the requirements specification document 21 as the correct answer 37 to problem 36.

[0028] The answer generation unit 35 outputs an answer 38 to problem 36 by inputting problem 36 and related design document portion 33 into the trained model 23. Specifically, the answer generation unit 35 outputs an answer 38 to problem 36 based on related design document portion 33 by inputting a prompt including problem 36 and related design document portion 33 into the trained model 23. If it is not possible to provide an answer based on related design document portion 33, the answer generation unit may also input a prompt to the trained model 23 requesting that it respond accordingly.

[0029] Based on the answer 38 output by the problem answer generation unit 28, the determination unit 29 outputs information 39 to the output device 14 indicating the degree of relationship between the requirements definition document 21 and the design document 22.

[0030] Specifically, the determination unit 29 outputs information 39 to the output device 14 indicating the degree of relationship between the design document 22 and the requirements definition document 21, depending on the extent to which the answer 38 output by the problem answer generation unit 28 conforms to the content of the requirements section 31 of the requirements definition document 21 (correct answer 37).

[0031] The functions of each functional unit of the text verification support device 1 described above are realized by the arithmetic unit 11 of the text verification support device 1 reading programs from the storage device 12. Furthermore, each program can be recorded and distributed, for example, on a portable or fixed recording medium. It should be noted that all or part of each program in the text verification support device 1 may be realized using virtual information processing resources provided using virtualization technology, process space isolation technology, etc., such as a virtual server provided by a cloud system. Also, all or part of these programs may be realized by services provided by a cloud system via an API (Application Programming Interface), etc.

[0032] Next, we will explain the processing performed by the text verification support device 1.

[0033] Figure 2 is a flowchart illustrating an example of a text verification support process performed by the text verification support device 1. The text verification support process is initiated, for example, when a user provides a predetermined input to the text verification support device 1.

[0034] First, the text verification support device 1 accepts the registration of the requirements definition document 21 and the design document 22 (s1). For example, the text verification support device 1 displays a predetermined upload screen and accepts the user's specification of the requirements definition document 21 and the design document 22 to be registered with the text verification support device 1, and stores the specified requirements definition document 21 and design document 22.

[0035] Subsequently, the text verification support device 1 extracts the portion describing the requirements from the requirements definition document 21 and generates a requirements list 30 (s2).

[0036] For example, the text verification support device 1 may input a prompt to the trained model 23 that includes an instruction to extract the requirements definition document 21 and the parts describing the requirements from the requirements definition document 21, thereby extracting each part describing the requirements, and each extracted part may be made into a requirements list 30. Alternatively, for example, the text verification support device 1 may search the requirements definition document 21 and detect predetermined text in the requirements definition document 21, thereby extracting each part describing the requirements as a requirements list 30.

[0037] The text verification support device 1 selects one of the requirements list 30 extracted in s2 to form the requirements section 31 (s3). For example, the text verification support device 1 may display a predetermined selection screen and accept input from the user to select one of the requirements list 30 extracted in s2. Alternatively, for example, the text verification support device 1 may automatically select one of the requirements list 30 according to a predetermined rule.

[0038] Meanwhile, the text verification support device 1 divides the design document 22 into multiple design document chunks 32 (s4). For example, the text verification support device 1 may analyze the design document 22 and divide it sentence by sentence, divide it into units of a predetermined number of characters, or divide it based on a predetermined string (such as a title).

[0039] The text verification support device 1 generates multiple questions (M questions) (s5) that can be answered based on the requirement section 31 selected in s3.

[0040] Specifically, the text verification support device 1 generates a question 36 and its correct answer 37 by inputting a prompt (question creation request prompt) to the trained model 23, which requests that the model create a question 36 that can be answered based on the content of the requirement part 31 selected in s3 (in accordance with the content of the requirement part selected in s3). The generated question 36 is, for example, a sentence. The generated correct answer 37 has a uniquely determined answer and is, for example, a sentence, but is not limited to words or numbers.

[0041] Furthermore, after executing this process, the text verification support device 1 may verify whether it is possible to answer each generated question 36 based on the requirement portion 31 selected in s3 (i.e., whether it is possible to identify the correct answer 37) (s5). If the text verification support device 1 determines that it is not possible to answer based on the content of the requirement portion 31, it will execute the process in s5 again to generate a different question 36.

[0042] For example, the text verification support device 1 inputs a prompt to the trained model 23 that includes the requirement portion 31 selected in s3, the generated question 36, and the correct answer 37, and asks whether the correct answer 37 for question 36 can be derived based on the content of the requirement portion 31. This prompts the device to output information indicating whether or not question 36 can be answered based on the content of the requirement portion 31. Note that the method described here is just one example; for example, the text verification support device 1 may also perform verification based on a criterion such as whether or not the requirement portion 31 contains the text of the correct answer 37.

[0043] Figure 3 shows an example of a question creation request prompt. This question creation request prompt 300 includes a statement 301 requesting the creation of a question-and-answer problem 36 and its correct answer 37, which can be answered based on the content of the requirements section, and a quotation section 302 of the requirements section.

[0044] On the other hand, as shown in Figure 2, the text verification support device 1 searches for the design document chunk 32 (related design document chunk 33) that is related to the requirement chunk 31 selected in s3 from among the design document chunks 32 divided in s4 (s6).

[0045] For example, the text verification support device 1 searches the related design document section 33 according to predetermined search conditions using keyword search or vector search in the requirements section 31.

[0046] In the case of keyword search, for example, the text verification support device 1 extracts one or more keywords from the requirements section 31, either by user specification or automatically. Then, the text verification support device 1 searches for up to N design document chunks 32 that contain those keywords, and these design document chunks 32 become the related design document section 33. In this process, the text verification support device 1 sets a higher priority for each design document chunk 32 the more keywords it contains.

[0047] In the case of vector search, for example, the text verification support device 1 calculates the embedding vector for the requirements section 31 and the embedding vector for each design document chunk 32. Then, the text verification support device 1 identifies up to N design document chunks 32 in descending order of cosine similarity with the embedding vector of the requirements section 31, designating these design document chunks 32 as related design document sections 33, and sets a priority order based on cosine similarity.

[0048] The text verification support device 1 checks whether the related design document section 33 was found as a result of the processing in s5 (s7). If the related design document section 33 was found (s6:YES), the text verification support device 1 executes the processing in s9. On the other hand, if the related design document section 33 was not found (s6:NO), the text verification support device 1 relaxes the search conditions (s8) and repeats the processing in s6 based on the relaxed search conditions.

[0049] In s8, the text verification support device 1 searches for design document chunks 32 with a predetermined number of items (e.g., N+K items) increased by a predetermined number K from the current number (e.g., N items). Additionally, the text verification support device 1 sets keywords similar to or additional keywords to those extracted in s6.

[0050] In s9, the text verification support device 1 repeats the processing in s10-s17 for each related design document portion 33 (design document chunk 32) found in s6 (s9, s18).

[0051] In other words, in s10, the text verification support device 1 answers each of the M problems 36 generated in s5 based on the relevant design document section 33, and outputs the corresponding answers 38.

[0052] Specifically, for each problem 36 generated in s5, the text verification support device 1 generates a prompt (answer generation request prompt) that requests the answer to problem 36 based on the related design document section 33, and requests the answer to be "unable to answer" if it is not possible to answer based on the related design document section 33, and inputs the generated answer generation request prompt into the trained model 23.

[0053] Figure 4 shows an example of an answer generation request prompt. This answer generation request prompt 400 includes wording 401 requesting an answer based on the relevant design document portion 33, wording 402 requesting "Unable to answer" if it is not possible to answer based on the relevant design document portion 33, a quoted portion 403 of the relevant design document portion 33, and a quoted portion 404 of the problem 36 generated in s5.

[0054] Next, as shown in Figure 2, the text verification support device 1 compares the answers 38 to each question 36 with the correct answers 37 generated in s5, and performs processing according to the degree of accuracy (s11-17).

[0055] Specifically, the text verification support device 1 first determines whether each question 36 is correct or incorrect (s11). Specifically, the text verification support device 1 determines that if question 36 and the answer 38 match, then question 36 is correct, and if question 36 and the answer 38 do not match, then question 36 is incorrect.

[0056] The text verification support device 1 determines whether or not there are any incorrect answers among the M questions 36 (s12). If there are incorrect answers among the M questions 36 (s12: YES), the text verification support device 1 executes the process in s13. If there are no incorrect answers among the M questions 36 (s12: NO), the text verification support device 1 executes the process in s14.

[0057] In s13, the text verification support device 1 stores that there is no consistency between the related design document section 33 and the requirements section 31, that is, the design document 22 is not consistent with the requirements definition document 21. After that, the process in s17 is performed.

[0058] In s14, the text verification support device 1 determines whether there is a question 36 among the M questions 36 whose answer 38 is "unanswerable". If there is a question 36 among the M questions 36 whose answer 38 is "unanswerable" (s14: YES), the text verification support device 1 executes the process in s15. If there is no question 36 among the M questions 36 whose answer 38 is "unanswerable" (s14: NO), the text verification support device 1 executes the process in s16.

[0059] In s15, the text verification support device 1 stores that the related design document section 33 has deficiencies compared to the requirements section 31, that is, that the design document 22 has deficiencies compared to the requirements definition document 21. After that, the process in s17 is performed.

[0060] In s16, the text verification support device 1 stores that there is consistency between the related design document section 33 and the requirements section 31, that is, the design document 22 is consistent with the requirements definition document 21. After that, the process in s17 is performed.

[0061] In s17, the text verification support device 1 displays the contents stored in s13, s15, or s16 on the screen.

[0062] Subsequently, the text verification support device 1 checks whether the processes s10-s17 have been executed for all related design document portions 33 (design document chunks 32) (s18). If the processes s10-s17 have been executed for all related design document portions 33 (design document chunks 32), the text verification support device 1 terminates the text verification support process. If there are any related design document portions 33 (design document chunks 32) for which the processes s10-s17 have not been executed, the text verification support device 1 repeats the processes from s10 onward for those related design document portions 33 (design document chunks 32).

[0063] Figure 5 shows an example of a screen displayed when the design document 22 is consistent with the requirements definition document 21. This screen displays a message 501 indicating that the related design document portion 33, or design document chunk 32, is related to the requirements definition document 21 (the requirement portion, "Requirement A").

[0064] Figure 6 shows an example of a screen displayed when the design document 22 is inconsistent with the requirements specification document 21. This screen displays a message 601 indicating that the design document 22 is inconsistent with the requirements specification document 21, along with information 602 (question, answer, correct answer, and correct / incorrect) from problem 36, for which answer 38 was incorrect, which was the basis for this inconsistency.

[0065] Figure 7 shows an example of a screen displayed when the related design document section 33 has deficiencies in the requirements section. This screen displays a message 701 indicating the deficiencies, along with information 702 (question, answer, correct answer, and correct / incorrect) for the problem 36 for which the answer 38 was "unanswerable," corresponding to those deficiencies.

[0066] Furthermore, after completing the above text verification support processing, the text verification support device 1 may output a traceability matrix (a correspondence table between requirements and design).

[0067] Figure 8 shows an example of the traceability matrix that is output. This traceability matrix 800 contains data including each requirement 801 in the requirements definition document 21, the design document chunk 802 (related design document portion 33) of the design document related to that requirement 801, and the degree of relevance 803 between requirement 801 and design document chunk 802. The degree of relevance 803 is set to a value that indicates the degree of relevance, such as the degree of relevance calculated by a predetermined formula based on cosine similarity, or the number of keyword matches.

[0068] Figure 9 shows another example of the traceability matrix that is output. This traceability matrix 900 contains data including each requirement 901 in the requirements definition document 21, the design document 22 or design document chunk 902 (related design document portion 33) related to that requirement 901, the degree of relevance 903 between the requirement 901 and the design document chunk 902, and consistency information 904.

[0069] The consistency information 904 displays a predetermined symbol 905 ("○", etc.) if the text verification support process determines that the design document chunk 32 (related design document portion 33) and the requirements are consistent; information to that effect 906 (for example, the answer 38 and the correct answer 37 for the incorrectly answered question 36) if the text verification support process determines that the design document chunk 32 (related design document portion 33) and the requirements are not consistent; and information to that effect 907 (for example, the answer 38 and the correct answer 37 for the question 36 for which the answer 38 was "unanswerable").

[0070] As described above, the text verification support device 1 of this embodiment searches for the portion of the second text (design document 22) that is related to the first text (relevant design document portion 33), inputs the requirements document 21 and the related design document portion 33 into the trained model 23, outputs text that represents the relationship between the requirements document 21 and the related design document portion 33, and outputs information indicating the degree of relationship between the requirements document 21 and the design document 22 to the output device based on the output text.

[0071] In other words, the text verification support device 1 of this embodiment identifies the relationship between the requirements definition document 21 and the related design document portion 33 by inputting the design document chunk 32 related to the requirements definition document 21 and the requirements definition document 21 into a trained model and outputs the degree of that relationship. This makes it possible to support the verification of the relationship between the design document 22 and the requirements definition document 21, for example, whether the design document 22 appropriately reflects the contents of the requirements definition document 21.

[0072] Thus, the text verification support device 1 of this embodiment can support the verification of correspondences between texts. For example, it eliminates the need for a verifier to manually set a complex and large number of keywords to verify correspondences between texts, as was done in the past.

[0073] Furthermore, the text verification support device 1 of this embodiment outputs a question 36 and its correct answer 37, which have questions about the contents of the requirements definition document 21, when the requirements definition document 21 is input into the trained model 23. When the question 36 and the related design document portion 33 are input into the trained model 23, an answer 38 to the question 36 is output. Depending on the degree to which the output answer 38 conforms to the correct answer 37, information indicating the degree of relationship between the design document 22 and the requirements definition document 21 is output to the output device.

[0074] In this way, by generating a question 36 that asks about the content of the requirements specification document 21, and determining the degree of relationship between the design document 22 and the requirements specification document 21 according to the extent to which the answer 38 to the question 36 conforms to the content of the requirements specification document 21, it is possible to determine whether the related design document portion 33 conforms to the content of the requirements specification document 21. This helps to accurately verify whether the content of one text is consistent with the content of the other text.

[0075] More specifically, the text verification support device 1 of this embodiment inputs an instruction to the trained model 23 to request an answer to question 36 from the relevant design document section 33, and to request an answer to that effect if an answer is not possible. (1) If the outputted answer 38 is consistent with the content of the correct answer 37, it outputs information indicating that there is consistency between the design document 22 and the requirements definition document 21. (2) If the outputted answer 38 is not consistent with the content of the correct answer 37, it outputs information indicating that there is inconsistency between the design document 22 and the requirements definition document 21. (3) If the outputted answer 38 is not possible, it outputs information indicating that the design document 22 has deficiencies compared to the requirements definition document 21.

[0076] In this way, by providing the option "Unable to answer" as the answer 38 to question 36, it is possible to prevent the generation of an answer 38 that is not based on the contents of the requirements specification document 21, and to determine with greater accuracy whether the related design document section 33 conforms to the contents of the requirements specification document 21.

[0077] Furthermore, the text verification support device 1 of this embodiment outputs multiple answers 38 for each of the multiple questions 36 that have been output, by inputting the question 36 and the related design document portion 33 into the trained model 23.The text verification support device 1 then (1) outputs information indicating that there is consistency between the design document 22 and the requirements definition document 21 if each of the multiple answers 38 is in accordance with the content of the correct answer 37, (2) outputs information indicating that there is inconsistency between the design document 22 and the requirements definition document 21 if there is an answer 38 among the multiple answers 38 that is not in accordance with the content of the correct answer 37, and (3) outputs information indicating that the design document 22 has deficiencies compared to the requirements definition document 21 if there is an answer 38 among the multiple answers 38 that cannot be answered (Figures 5-7, 9).

[0078] In this way, by determining whether each of the answers 38 is correct or not, the consistency, inconsistency, and omissions between the design document 22 and the requirements definition document 21 can be determined, allowing for a systematic understanding of the correspondence between the texts.

[0079] Furthermore, the text verification support device 1 of this embodiment divides the design document 22 into multiple parts (design document chunks 32) and searches for design document chunks 32 that satisfy the above search conditions from among the multiple design document chunks 32 as related design document parts 33 associated with the requirements definition document 21 (requirements part 31), based on predetermined search conditions related to the requirements definition document 21 (for example, cosine similarity with requirements part 31, search keywords in requirements part 31).

[0080] In this way, by searching for the related design document portion 33 for each design document chunk 32 obtained by dividing the design document 22, it is possible to support the verification of correspondences between texts without excessively burdening the text verification support device 1 (trained model 23), even when the volume of requirements definition documents 21 and design documents 22 is large.

[0081] Furthermore, if the text verification support device 1 of this embodiment is unable to find a design document chunk 32 that satisfies the above search conditions among the multiple design document chunks 32 as a related design document portion 33, it sets new search conditions that relax the above search conditions and searches for a design document chunk 32 that satisfies the above search conditions among the multiple design document chunks 32 as a related design document portion 33 based on the set search conditions.

[0082] This prevents omissions in verifying the correspondence between the requirements definition document 21 and the design document 22.

[0083] Furthermore, after generating the problem 36 and answer 38, the text verification support device 1 of this embodiment inputs the requirements definition document 21 and the problem 36 into the trained model 23 to determine whether the correct answer 37 for problem 36 can be identified in accordance with the contents of the requirements definition document 21. If it determines that the correct answer 37 for problem 36 cannot be identified in accordance with the contents of the requirements definition document 21, it generates a different problem 36 and its correct answer.

[0084] In this way, by confirming that Problem 36 cannot be answered in accordance with the contents of the Requirements Definition Document 21, it is possible to more accurately verify the correspondence between the Requirements Definition Document 21 and the Design Document 22.

[0085] Furthermore, the text verification support device 1 of this embodiment outputs a traceability matrix by associating the requirements definition document 21, the part of the design document 22 related to the requirements definition document 21 (related design document portion 33), information indicating the relationship between the requirements definition document 21 and the related design document portion 33 (consistency, inconsistency, or omission, etc.), and information indicating the degree of relationship (degree of relevance, etc.) with each other (Figures 8 and 9).

[0086] This allows verifiers to reliably understand the correspondence between texts and their specific content, as well as to verify that the correspondence between texts is correct. This enables them to, for example, check for omissions, omissions, and excesses in specifications and designs within the texts (ensuring traceability), and to reduce the man-hours required at each stage of the process.

[0087] The present invention is not limited to the embodiments described above, and can be implemented using any components without departing from its spirit. The embodiments and modifications described above are merely examples, and the present invention is not limited to these as long as the features of the invention are not impaired. Furthermore, although various embodiments and modifications have been described above, the present invention is not limited to these. Other embodiments conceivable within the scope of the technical idea of ​​the present invention are also included within the scope of the present invention.

[0088] For example, some of the hardware provided in each device of each embodiment may be provided in other devices.

[0089] Furthermore, each program of each device may be provided in other devices, a program may consist of multiple programs, or multiple programs may be integrated into a single program.

[0090] Furthermore, although this embodiment shows examples of verification between texts, such as the requirements definition document 21 and the design document 22, the present invention can also be applied to other texts (where it is desirable that one text and the other text correspond to each other).

[0091] Furthermore, although this embodiment describes a case where the generated question 36 is a question that requires the answer to provide a predetermined word or sentence, the generated question 36 may be of other types (for example, a question that requires the answer to "yes" or "no", or a multiple-choice question).

[0092] Furthermore, in this embodiment, consistency and inconsistency between the requirements definition document 21 and the design document 22 were determined by whether all questions were answered correctly or incorrectly. However, other criteria may be adopted (for example, consistency is determined if more than half of the answers are correct).

[0093] Furthermore, in this embodiment, the relationship between the requirements definition document 21 and the design document 22 is assumed to be either consistent, inconsistent, or incomplete. However, depending on the relationship between the correct answer 37 and the answer 38, other relationships such as contradictions between the requirements definition document 21 and the design document 22 may be identified.

[0094] Furthermore, the partitioning algorithm for the design document 22 and the requirements extraction algorithm from the requirements definition document 21 described in this embodiment are merely examples, and other arbitrary algorithms may be adopted. [Explanation of Symbols]

[0095] 1 Text verification support device, 21 Requirements definition document, 22 Design document, 23 Trained model, 36 Question, 37 Correct answer, 38 Answer

Claims

1. A storage device for storing the first text and the second text, and A related portion search process that searches for a portion of the second text related to the first text based on a predetermined algorithm, A generation process that outputs text representing the relationship between the first text and the related parts by inputting the first text and the related parts into a trained model that outputs text corresponding to the input text, The system includes a arithmetic unit that performs a determination process that outputs information indicating the degree of relationship between the first text and the second text to an output device based on the output text. Text verification support device.

2. The aforementioned computing device is In the above generation process, A problem generation process that outputs a problem having a question that asks about the content of the first text and the correct answer to the problem by inputting the first text into the trained model, The following steps are performed: input the aforementioned problem and the related parts into the trained model to perform an answer generation process that outputs an answer to the aforementioned problem; In the judgment process, information indicating the degree of relationship between the second text and the first text is output to the output device, according to the extent to which the answer output in the answer generation process corresponds to the correct answer to the question output in the question generation process. The text verification support device according to claim 1.

3. The aforementioned computing device is In the aforementioned response generation process, The trained model is instructed to provide an answer to the problem from the relevant section and, if an answer is not possible, to indicate that fact. In the aforementioned determination process, If the answer output in the answer generation process corresponds to the correct answer output in the question generation process, information indicating consistency between the second text and the first text is output to the output device. If the answer output in the answer generation process does not correspond to the correct answer output in the question generation process, information indicating that there is no consistency between the second text and the first text is output to the output device. If the answer output by the answer generation process is not available, the output device outputs information indicating that the second text has missing information compared to the first text. The text verification support device according to claim 2.

4. The aforementioned computing device is In the aforementioned problem generation process, multiple problems and correct answers are output. In the aforementioned answer generation process, for each of the multiple questions, the question and the related parts are input to the trained model to output multiple answers. In the aforementioned determination process, If each of the multiple answers output corresponds to the content of the correct answer output in the problem generation process, information indicating that there is consistency between the second text and the first text is output to the output device. If, among the multiple answers output, there is an answer that does not correspond to the correct answer output in the problem generation process, information indicating that there is no consistency between the second text and the first text is output to the output device. If any of the outputted responses are unanswerable, the output device will output information indicating that the second text has missing information compared to the first text. The text verification support device according to claim 3.

5. The aforementioned computing device is A text splitting process is performed to divide the second text into multiple parts. In the aforementioned related portion search process, based on predetermined search conditions relating to the first text, portions of the plurality of parts that satisfy the search conditions are searched as portions related to the first text. The text verification support device according to claim 1.

6. The aforementioned computing device is In the aforementioned related partial search process, If a portion of the aforementioned multiple parts that satisfies the search criteria cannot be found as a portion related to the first text, a new search criterion is set by relaxing the search criteria, and based on the set search criterion, a portion of the aforementioned multiple parts that satisfies the search criteria is searched as a portion related to the first text. The text verification support device according to claim 5.

7. The aforementioned computing device is After the problem generation process is executed, the first text and the problem are input to the trained model to determine whether the correct answer to the problem can be identified based on the content of the first text. If it is determined that the correct answer to the problem cannot be identified based on the content of the first text, the problem generation process is executed again to generate a different problem and its correct answer. The text verification support device according to claim 2.

8. The aforementioned computing device is The first text, the portion of the second text related to the first text, information indicating the relationship between the first text and the related portion, and information indicating the degree of the relationship are output to the output device in correspondence with each other. The text verification support device according to claim 1.

9. A text verification support method comprising a storage device for storing a first text and a second text, and an arithmetic unit, The aforementioned computing device A related portion search process that searches for a portion of the second text related to the first text based on a predetermined algorithm, A generation process that outputs text representing the relationship between the first text and the related parts by inputting the first text and the related parts into a trained model that outputs text corresponding to the input text, Based on the output text, a determination process is performed to output information indicating the degree of relationship between the first text and the second text to an output device. Text validation support methods.

Images