JP7074785B2 - Ambiguous part correction support device and method - Google Patents

Description

The present invention relates to an ambiguous part correction support device and a method, for example, correcting a part or all of a sentence including numerically ambiguous contents in a software-related document (hereinafter, this is referred to as an ambiguous part) to clear contents. It is suitable for application to an ambiguous part correction support device that supports correction work.

Conventionally, when a defect occurs in a past product, the manufacturer responds by referring to the specifications and instruction manual of the product, as well as technical documents such as development emails, design memos, and / or design materials. Or, when developing a new model of an existing product, it is widely practiced to refer to the above technical document of the existing product.

In this case, if numerically ambiguous expressions are used in the technical document, there is a problem that the reader cannot obtain a sufficient understanding. For example, in the sentence "When the system slows down, it is necessary to increase the amount of memory more", the meaning of "more" is ambiguous, and it is unclear how much memory should be increased concretely. be.

For this reason, when such an expression exists in a technical document, the reader needs to search for "more" specific numerical values from other related technical documents, which is complicated and time-consuming. There was a problem. In addition, even if such a specific numerical value can be found, if multiple specific numerical values are detected, it is difficult for a young engineer without technical knowledge to determine which specific numerical value is an appropriate value. There was also the problem.

The following Patent Document 1 discloses a technique for automatically finding conditions included in requirement specifications described in natural language, comprehensively searching for possible conditions, and presenting missing condition clauses. ing.

Japanese Unexamined Patent Publication No. 2015-14827

By the way, regarding the ambiguity of the technical document as described above, a veteran engineer with some knowledge should search for specific numerical values from other related technical documents and correct the ambiguity so as to clarify the ambiguity. Is desirable. However, even a veteran technician has a problem that such correction work requires a lot of labor and time.

The present invention has been made in consideration of the above points, and an object of the present invention is to propose an ambiguity correction support device and a method capable of facilitating and optimizing the correction work for correcting the contents of ambiguities in a document to clear contents. It is a thing.

In order to solve such a problem, in the present invention, a verb expression having ambiguity in an ambiguity correction support device that supports a correction operation for correcting a numerically ambiguous ambiguity in a first document so as to clarify it. A combination of multiple specific subjects and the predicate verb, each of which is pre-stored as an ambiguous word and whose content is ambiguous without a numerical value indicating the degree of action represented by the predicate verb. A storage device that is pre-stored as an ambiguous syntax, and a multi-condition extractor that extracts part or all of a sentence described in If-Then format as a multi-condition sentence from a plurality of second documents. From the first document, a part of a sentence or sentence containing any of the ambiguous words stored in the storage device, or the subject and the subject constituting any of the ambiguous syntax stored in the storage device. An ambiguous part extraction unit that extracts a sentence or a part of a sentence including a combination of pre-descriptive word verbs as the ambiguous part, and the ambiguous part and the subject of the multiconditional sentences extracted from a plurality of the second documents. The correction candidate extraction unit that extracts the multi-condition sentence that matches the ambiguity as a correction candidate of the ambiguity, the condition of occurrence of the action represented by the verb included in the ambiguity, and each of the corrections extracted for the ambiguity. A correction candidate scoring unit that assigns a score to each of the correction candidates based on the degree of similarity of the candidate to the occurrence condition, and a display that displays each of the correction candidates in a display form according to the given score. I tried to provide a part.

Further, the present invention is an ambiguity correction support method executed by an ambiguity correction support device that supports a correction work for correcting a numerically ambiguous ambiguity in a first document so as to clarify it. In the ambiguity correction support device, a plurality of specific words of an ambiguous expression are stored in advance as ambiguity words, and the content is not accompanied by a numerical value indicating the degree of operation expressed by a predicate verb. A part of a sentence described in If-Then format from a plurality of second documents, each having a storage device in which a combination of a plurality of specific subjects and the predicate verbs to be ambiguous are pre-stored as an ambiguous syntax . Or a part of a sentence or sentence containing any of the ambiguous words stored in the storage device from the first step of extracting all of them as a multi-condition sentence, or the storage device. A second step of extracting a sentence or a part of a sentence including a combination of the subject and the predescriptive word verb that constitutes any of the ambiguities stored in the ambiguity as the ambiguity, and a plurality of the second documents. The third step of extracting the multi-conditional sentence whose subject matches the ambiguous part of the multi-conditional sentence extracted from is extracted as a correction candidate of the ambiguous part, and the verb included in the ambiguous part represent. A fourth step of assigning a score to each of the correction candidates based on the degree of similarity between the operation occurrence condition and the occurrence condition of each of the correction candidates extracted for the ambiguity, and each of the correction candidates. , A fifth step of displaying in a display form according to the given score is provided.

According to the ambiguity correction support device and method of the present invention, as a numerical value to be added as a modifier of the predicate verb to the ambiguity in the first document based on the display mode of each displayed correction candidate, the second The most appropriate correction candidate numbers can be selected and determined from the entire document.

According to the present invention, it is possible to facilitate and optimize the correction work for correcting an ambiguity in a document into a clear expression.

It is a block diagram which shows the whole structure of the ambiguity correction support system by this embodiment. It is a figure which shows the configuration example of a multi-condition table. It is a figure which shows the structural example of an ambiguous word table. It is a figure which shows the composition example of an ambiguous syntax table. It is a figure which shows the structural example of the weight table. It is a figure which shows the structural example of the same syntax judgment table. It is a figure which shows the structural example of a product name table. It is a figure which shows the configuration example of the obscure part presentation screen. It is a figure which shows the configuration example of the ambiguity correction support screen. It is a figure which shows the configuration example of the correction candidate detail screen. It is a flowchart which shows the process procedure of the multi-condition table creation process. It is a flowchart which shows the processing procedure of the ambiguity correction support processing. It is a flowchart which shows the processing procedure of the ambiguous word extraction processing. It is a flowchart which shows the processing procedure of the ambiguous syntax extraction processing. It is a flowchart which shows the processing procedure of the correction candidate extraction processing. It is a flowchart which shows the processing procedure of the correction candidate scoring process. It is a flowchart which shows the processing procedure of the selection correction candidate learning process. Similarly, it is a flowchart which shows the processing procedure of the correction part determination process.

Hereinafter, one embodiment of the present invention will be described in detail with reference to the drawings.

(1) Configuration of Ambiguous Part Correction Support System According to the Present Embodiment In FIG. 1, 1 shows an ambiguous part correction support system according to the present embodiment as a whole. The ambiguity correction support system 1 includes an ambiguity correction support device 3 and a file server 4 connected via a network 2, and one or a plurality of clients 5.

The ambiguity correction support device 3 includes a CPU (Central Processing Unit) 11, a main memory 12, an auxiliary storage device 13, a network adapter 14, an input / output unit 15, and an input device 16 connected to each other via an internal bus 10. It is composed of a general-purpose computer device including a display device 17.

The CPU 11 is a processor that controls the operation of the entire ambiguity correction support device 3. Further, the main memory 12 is composed of, for example, a volatile semiconductor memory and is used as a work memory of the CPU 11.

The auxiliary storage device 13 is composed of a large-capacity non-volatile storage medium such as a hard disk device, SSD (Solid State Drive), or flash memory, and stores various programs and necessary data. The program stored in the auxiliary storage device 13 is loaded into the main memory 12, and the CPU 11 executes this program to perform various processes as a whole of the ambiguity correction support device 3 as described later.

The network adapter 14 is composed of, for example, a NIC (Network Interface Card) and performs protocol control during communication with the file server 4 and the client 5 via the network 2.

The input / output unit 15 is an interface to the input device 16 and the display device 17, and converts a command given from the input device 16 into a signal format inside the ambiguity correction support device 3 according to a user operation, and supports image information. It is converted into a signal format to be output and output to the display device 17.

The input device 16 is a device used for a user to perform a necessary operation, and is composed of, for example, a mouse or a keyboard. The input device 16 outputs a command corresponding to the user operation to the input / output unit 15. Further, the display device 17 is composed of, for example, a liquid crystal display or an organic EL (Electro Luminescence) display, and displays a screen or an image corresponding to the image information given by the input / output unit 15.

The file server 4 is a general-purpose server device equipped with a file system function for storing and managing document files FL, and is a CPU 21, an auxiliary storage device 22, a network adapter 23, and an input / output unit connected via an internal bus 20. 24 is provided with an input device 25 and a display device 26.

The CPU 21, auxiliary storage device 22, network adapter 23, input / output unit 24, input device 25, and display device 26 are the CPU 11 of the ambiguity correction support device 3, the auxiliary storage device 13, the network adapter 14, and the input / output unit 15, respectively. Since it has the same functions and configurations as the input device 16 and the display device 17, detailed description here will be omitted.

In addition to various programs, the auxiliary storage device 22 of the file server 4 stores data of each document file FL, file information FLI for each document file FL, and the like. The file information includes meta information of the document such as the version of the product corresponding to the document based on the corresponding document file FL, the creation date and creator information of the document, and the product name of the product to which the document is related. Information included.

The client 5 is a computer device used by a user to read / write a document file FL to / from a file server 4 or to request an ambiguity correction support device 3 to support an ambiguity correction work in a desired document.

The client 5 includes a CPU 31, a main memory 32, a network adapter 33, an input / output unit 34, and an input device 35 and a display device 36, which are connected to each other via an internal bus 30.

The CPU 31, main memory 32, network adapter 33, input / output unit 34, input device 35, and display device 36 are the CPU 11, main memory 12, network adapter 14, input / output unit 15, and input device of the ambiguity correction support device 3, respectively. Since it has the same functions and configurations as those of 16 and the display device 17, detailed description here will be omitted.

(2) Ambiguous part correction support function Next, the ambiguous part correction support function mounted on the ambiguous part correction support device 3 will be described. This ambiguity correction support function numerically describes the numerically ambiguous expression part (ambiguous part) in the target document (hereinafter referred to as the target document) designated by the user via the client 5. It is a function that supports the correction work that the user corrects so as to clarify the situation.

In the following, the document based on the document file FL stored in the file server 4 (hereinafter referred to as a stored document) and the target document specified by the user are emails exchanged at the development stage of the software product. And design memos or design materials. In addition, the user who executes the correction work to correct the ambiguity of the target document so as to clarify it shall be a veteran engineer who has some knowledge.

As a means for realizing the ambiguity correction support function as described above, the auxiliary storage device 13 of the ambiguity correction support device 3 includes a multi-condition table creation program 40, an ambiguity word extraction program 41, and an ambiguity syntax extraction program 42. Correction candidate extraction program 43, correction candidate scoring program 44, correction part / correction content display program 45, selection correction candidate learning program 46, correction result display program 47 and similar correction part determination program 48, multi-condition table 50, ambiguous word. A table 51, an ambiguous syntax table 52, a weight table 53, the same syntax determination table 54, and a product name table 55 are stored.

The multi-condition table creation program 40 performs a dependency analysis on each sentence in the stored document, thereby performing "multi-condition" for a part or all of the sentences described in "If-Then format" from each stored document. It is a program having a function of extracting as a "sentence" and registering necessary information about the extracted multi-conditional sentence in the multi-condition table 50.

The "If-Then format" here means "subject", "verb", "numerical value" indicating the degree of action represented by the verb, and "occurrence condition" of the action represented by the verb, which are often used in software-related documents. It is an expression format that includes "(~ time, ~ hour)", for example, "Increase memory to 4GB or more when starting for the first time", "Increase memory to 16GB when KBE-500 error occurs", etc. The sentence of corresponds to this. In the former example, "when starting for the first time" corresponds to the occurrence condition, "memory" corresponds to the subject, "4GB or more" corresponds to the numerical value, and "increase" corresponds to the verb. "" Corresponds to the occurrence condition, "memory" corresponds to the subject, "16GB to" corresponds to the numerical value, and "increase" corresponds to the verb.

The ambiguous word extraction program 41 is a sentence or a part of a sentence including a word of an ambiguous expression (hereinafter, referred to as an ambiguous word) having ambiguity such as "more", "less", "long term" and "a little". It is a program that has a function to extract from the target document as an ambiguous part.

Further, the ambiguous syntax extraction program 42 is a sentence or sentence whose content is ambiguous unless it is accompanied by a numerical value (for example, "to 4GB") indicating the degree of operation expressed by a predicate verb, such as "increase memory". It is a program that has a function to extract a part from the target document as an ambiguous part.

The correction candidate extraction program 43 should be used as a reference when correcting the ambiguity into a numerically clear description for each ambiguity in the target document extracted by the ambiguity word extraction program 41 or the ambiguity syntax extraction program 42. This is a program having a function of extracting a multi-condition sentence from the multi-condition table 50 as a correction candidate. Specifically, the correction candidate extraction program 43 identifies all the multi-conditional sentences whose subject matches the subject of the ambiguous part extracted from the target document from among some multi-conditional sentences registered in the multi-condition table 50. Then, each of these multi-conditional statements is extracted as a correction candidate for the ambiguous part.

For example, the ambiguous part in the target document is "uses a lot of memory", and in the multi-condition table 50, there is a multi-condition statement "increase the memory to 4GB or more when starting for the first time" and "KBE-500 error". When the multi-conditional statement "Increase the memory to 16GB when it occurs" is stored, the correction candidate extraction program 43 extracts these multi-conditional statements as correction candidates of the ambiguous part.

The correction candidate scoring program 44 uses the numerical value included in the correction candidate as a modifier of the predicate verb of the corresponding ambiguity for each correction candidate extracted for each ambiguity in the target document by the correction candidate extraction program 43. It is a program having a function of calculating a score indicating the appropriateness of addition (hereinafter, this is referred to as a correction candidate score) and assigning the calculated correction candidate score to the correction candidate.

The correction part / correction content display program 45 includes information on each correction candidate (corresponding multi-condition statement information stored in the multi-condition table 50) of the ambiguous part specified by the user in the target document, and these correction candidates. Based on the correction candidate score of FIG. 9, the program has a function of displaying the ambiguous portion correction support screen 70 described later for FIG. 9 and the correction candidate detail screen 100 described later for FIG. 10 on the display device 17. The user can use the ambiguity correction support screen 70 and the correction candidate detail screen 100 to determine and select which correction candidate numerical value is used to clarify the content of the corresponding ambiguity in the target document. , The ambiguous part can be actually corrected by using the numerical value of the selected correction candidate.

The selection correction candidate learning unit 46 extracts the filtering conditions of the correction candidates for which information should be displayed on the ambiguity correction support screen 70 set by the user on the ambiguity correction support screen 70 (FIG. 9), and FIG. Is a program having a function of registering in the weight table 53, which will be described later. The information registered in the weight table 53 is subsequently used when calculating the correction candidate score of the correction candidate.

Further, the correction result display unit 47 is a program having a function of displaying the work result (that is, the ambiguous part after correction) of the correction work performed for each ambiguity in the target document on the display device 17. The correction result display unit 47 corrects the ambiguity that was not corrected by the user in the target document by using the correction candidate having the highest correction candidate score (that is, the correction candidate is a modifier of the predicate verb). Correct to add as).

Further, similarly, the correction part determination unit 48 performs a part or all of the sentence of the same syntax (combination of noun and predicative verb) as each corrected ambiguity part in the target document from each stored document stored in the file server 4. Is a program having a function of extracting as an ambiguous part of a stored document and presenting a correction candidate of the ambiguous part to a user.

On the other hand, the multi-condition table 50 is a table used for storing and managing necessary information regarding the above-mentioned multi-condition statement extracted from the stored document by the multi-condition table creation program 40 as described above, and is used in FIG. As shown in, the generation condition column 50A, the subject column 50B, the numerical value column 50C, the operation column 50D, the original text column 50E, the product name column 50F, the version column 50G, the creation date column 50H, and the creator column 50I are provided. .. In the multi-condition table 50, one row corresponds to one multi-condition statement extracted from the stored document.

Then, in the original text column 50E, a multi-conditional sentence extracted from the stored document is stored. Further, the subject column 50B stores the phrase of the part corresponding to the subject in the multi-conditional sentence, and the action column 50D stores the phrase of the part corresponding to the predicate verb in the multi-conditional sentence. Further, the numerical value column 50C stores a numerical clause indicating the degree of the action represented by the predicate verb in the multi-conditional sentence, and the generation condition column 50A stores the action represented by the predicate verb in the corresponding multi-conditional sentence. Occurrence condition is stored.

For example, as shown in FIG. 2, when the multi-conditional sentence stored in the original text column 50E is "Increase the memory to 4 GB or more at the first startup", "Memory" is displayed in the subject column 50B. , "Increase" is stored in the operation column 50D, "4GB or more" is stored in the numerical value column 50C, and "when the first startup is performed" is stored in the generation condition column 50A.

Further, in the product name column 50F and the version column 50G, the product name and version (product version) of the product related to the stored document from which the multi-conditional sentence is extracted are stored, and in the creation date column 50H and the creator column 50I, respectively. , The creation date and creator of the stored document are stored, respectively. The product name, version, creation date, and creator information are acquired by the multi-condition table creation program 40 from the file information FLI (FIG. 1) of the corresponding stored document stored in the file server 4 and are multi-condition tables. Registered in 50.

Therefore, in the case of the example of FIG. 2, the stored document including the original text "Increase the memory to 4 GB or more at the first startup" is a document related to the software version "7.0.0" with the product name "Product1". Yes, it is shown in "2010/07/05" that it was created by a person named "A".

The ambiguous word table 51 is a table used for storing and retaining pre-registered ambiguous words, and as shown in FIG. 3, "more", "less", "long-term", "a little", etc. Ambiguous word is stored.

The ambiguous word table 51 may be manually created by the user, or may be automatically created by the ambiguous part correction support device 3. In the latter case, the user gives a noun such as "disk", "memory", and "communication speed" to the ambiguity correction support device 3, and the ambiguity correction support device 3 is stored in the file server 4. The noun may be searched for in the stored document, and the word of the adjective expression dependent on the noun may be extracted as an ambiguous word and registered in the ambiguous word table 51.

Further, the ambiguous syntax table 52 is a table used for storing and holding a pre-registered ambiguous syntax, and is configured to include a subject column 52A and a predicate verb column 52B as shown in FIG. In the ambiguous syntax table 52, one row corresponds to one noun constituting the ambiguous syntax. The subject column 52A stores the subject in the corresponding ambiguous syntax, and the predicate verb column 52B stores the predicate verb in the ambiguous syntax.

The ambiguity syntax table 52 may be manually created by the user, or may be automatically created by the ambiguity correction support device 3. In the latter case, the user gives a noun as a seed to the ambiguity correction support device 3, and the ambiguity correction support device 3 searches for such a noun in each stored document stored in the file server 4, and is related to the noun. The combination of the received predicative verb and its noun may be registered in the ambiguous syntax table 52 as an ambiguous syntax.

The weight table 53 is a table used for storing and managing the filtering conditions specified by the user when narrowing down the correction candidates using the ambiguity correction support screen 70 described later for FIG. 9 displayed on the display device 17. Is.

In practice, in the ambiguity correction support device 3, the user can change the content of each correction candidate of the ambiguity target at that time in the target document to the ambiguity correction support screen 70 of FIG. 9 and the correction candidate detail screen of FIG. When confirming each using 100, in order to narrow down the desired correction candidates from these correction candidates, the product name and version of the product to which the stored document from which the correction candidate is extracted is related, the creation date of the stored document, and the date of creation of the stored document and The creator and the document type (email, design memo, design material, etc.) of the stored document can be set as filtering conditions. The weight table 53 is a table for storing and managing the filtering conditions set in this way.

As shown in FIG. 5, the weight table 53 includes an identification ID column 53A, an ambiguous sentence ID column 53B, a first product name column 53C, a first version column 53D, a first creation date column 53E, and a first. It has a creator column 53F, a document type column 53G, a correction source sentence ID column 53H, a second product name column 53I, a second version column 53J, a second creation date column 53K, and a second creator column 53L. It is composed. In the weight table 53, one row corresponds to the content of each filtering condition set for one ambiguity in the target document.

The identification ID column 53A stores the identification ID assigned to the information in the row in the weight table 53, and the ambiguous sentence ID column 53B is assigned to each sentence in the target document. Among the identification IDs (document IDs) unique to the text, the text ID of the text including the ambiguous part where the ambiguous expression has been corrected is stored.

Further, in the first product name column 53C, "1" is stored as the corresponding filtering condition when the product name of the product related to the stored document of the extraction source of the correction candidate is set, and the filtering condition is stored. If such a product name is not set, "0" is stored.

Similarly, in the first version column 53D, "1" is stored as the corresponding filtering condition when the version of the product related to the stored document of the extraction source of the correction candidate is set, and the filtering condition is stored. If such a version is not set, "0" is stored.

Further, in the first creation date column 53E, if the creation date of the stored document of the extraction source of the correction candidate is set as the corresponding filtering condition, "1" is stored, and the creation date is set as the filtering condition. If it is not set, "0" is stored.

Further, in the first creator column 53F, if the creator of the stored document of the extraction source of the correction candidate is set as the corresponding filtering condition, "1" is stored, and the creator is stored as the filtering condition. If it is not set, "0" is stored.

Further, in the document type column 53G, "1" is stored when the document type of the stored document from which the correction candidate is extracted is set as the corresponding filtering condition, and the document type is set as the filtering condition. If not, "0" is stored.

In the correction source sentence ID column 53H, the sentence ID of the sentence including the correction candidate in the accumulated document of the extraction source of the finally selected correction candidate (hereinafter, this is referred to as the final selection correction candidate extraction source document) is stored. Will be done.

Further, the product name of the product to which the final selection correction candidate extraction source document is related is stored in the second product name column 53I, and the product version of the product is stored in the second version column 53J. Further, the creation date of the final selection correction candidate extraction source document is stored in the second creation date column 53K, and the creator of the final selection correction candidate extraction source document is stored in the second creator column 53L.

Therefore, in the case of the example of FIG. 5, when extracting an appropriate correction candidate for the ambiguous part of the sentence to which the sentence ID "145987234" is given in the target document, the user uses the ambiguous part correction support screen 70 (FIG. 9). "Product name" and "Version" are set as filtering conditions, and the text ID of the text in the accumulated document including the correction candidate to which the numerical value is finally applied to this ambiguous part is "076342456", and the final selection correction The product name of the product to which the candidate extraction source document is related is "Product1", the product version is "10.0.0", the creation date of the final selection correction candidate extraction source document is "2010/05/17", and the creator is " It is shown that it was "A".

The same syntax determination table 54 is a table used for storing and managing the syntax (ambiguous syntax) of the ambiguous part in the target document corrected by the user, and as shown in FIG. 6, the subject column 54A and the predicate It is configured with a verb column 54B. In the same syntax determination table 54, one row corresponds to the syntax of one ambiguity corrected by the user. The subject column 54A stores the subject in the corresponding ambiguous syntax, and the predicate verb column 54B stores the predicate verb in the ambiguous syntax.

The product name table 55 is a table used for managing the version range of products sold so far, and includes a number column 55A, a product name column 55B, and a version range column 55C as shown in FIG. 7. It is composed of. In the product name table 55, one row corresponds to one product name.

The identification number assigned to each product name in the product name table 55 is stored in the number column 55A, and the product name of the corresponding product is stored in the product name column 55B. Further, the version range of the product is stored in the version range column 55C. Therefore, in the case of the example of FIG. 7, it is shown that the version range of the product with the product name "Product1" to which the identification number "1" is given in the product name table is "3.0.0 to 9.2.0". ing.

The product name table 55 may be manually created by the user in advance, or may be automatically created by the ambiguity correction support device 3. In the latter case, the ambiguity correction support device 3 extracts the product name and version of the product to which the stored document is related from the document file information FLI (FIG. 1) of each stored document, and based on the extracted information. The product name of each product up to and the version range thereof may be detected and registered in the product name table 55.

(3) Configuration of various screens FIG. 8 shows an ambiguous part correction support device 3 (FIG. 1) by the correction part / correction content display program 45 (FIG. 1) based on the processing result of the correction candidate scoring program 44 (FIG. 1). An example of the configuration of the ambiguous portion presentation screen 60 displayed on the display device 17 (FIG. 1) is shown. On the ambiguity presentation screen 60, the text of the target document is displayed, and the ambiguity in the text extracted by the ambiguity word extraction program 41 (FIG. 1) or the ambiguity syntax extraction program 42 (FIG. 1) is colored or It is highlighted by a method such as marking. Note that FIG. 8 shows an example in which the ambiguity of the text is marked.

Then, when one highlighted ambiguity portion in the ambiguity portion presentation screen 60 is designated by clicking, the ambiguity portion correction support screen 70 shown in FIG. 9 is displayed on the ambiguity portion presentation screen 60. The ambiguity correction support screen 70 is a screen that supports correction work for numerically clarifying the contents of the ambiguity designated by the user on the ambiguity presentation screen 60 (hereinafter, this is referred to as a user-specified ambiguity). , The graph display filtering area 71 and the filtering parameter area 72 are provided.

The graph display filtering area 71 is composed of a three-dimensional graph display area 80 and a graph display adjustment area 81. Then, in the three-dimensional graph display area 80, three-dimensional coordinates (hereinafter, the correction candidate score is the z-axis, the correction candidate numerical value is the x-axis, and the item of the filtering condition selected in the filtering parameter setting area 72 described later is the y-axis). , This is called a three-dimensional graph) 83 is displayed. Note that FIG. 9 is an example in which the item of the filtering condition selected in the filtering parameter setting area 72 is “version”. In this case, a scale is displayed on the x-axis within the range of the minimum value to the maximum value of the "numerical value" included in each correction candidate extracted for the user-specified ambiguity.

Further, in the three-dimensional graph 83, the rectangular parallelepiped block 84 corresponding to each correction candidate extracted by the correction candidate extraction program 43 (FIG. 1) for the user-specified ambiguity is displayed at the corresponding xy coordinate position. .. At this time, when there are a plurality of correction candidates having the same values in the x-axis direction and the y-axis direction, the blocks 84 corresponding to the correction candidates are stacked and displayed at the same xy coordinate position.

Further, each block 84 is displayed at a height corresponding to the correction candidate score given by the correction candidate scoring program 44 (FIG. 1) for the corresponding correction candidate. As a result, in the three-dimensional graph 83, the user can visually and instantly visually determine which block 84 the "numerical value" of the correction candidate corresponding to which block 84 is appropriate based on the number of stacked blocks 84 and the height of each block 84. It is made to be able to judge.

Further, in the graph display adjustment area 81, a correction candidate score range adjustment slide bar 85, a description sentence number adjustment slide bar 86, and a document type (email, design memo, and design material) of each accumulated document from which each corresponding correction candidate is extracted are provided. A character string 87 representing each type name of (such as) and a check box 88 associated with each of these document types (hereinafter, this is referred to as a document type selection check box) 88 are displayed.

Then, the user slides the correction candidate score range adjustment slide bar 85 to display the corresponding block 84 from the plurality of correction candidates extracted for the user-specified ambiguity in the three-dimensional graph 83. The range of correction candidate scores can be set. Therefore, in the three-dimensional graph 84, only the block 84 corresponding to the correction candidate to which the correction candidate score within the range set by using the correction candidate score range adjustment slide bar 85 is given is displayed, and the block 84 is out of the range. The block 84 corresponding to the correction candidate to which the correction candidate score is given is not displayed.

Further, the user can set the minimum number of stacked blocks 84 to be displayed in the three-dimensional graph 83 by operating the slide bar 86 for adjusting the number of described sentences. The "stacked number" here refers to the number of blocks 84 stacked at the same xy coordinates. Therefore, for example, when the number of stacked items is set to "1", the block 84 corresponding to all the corresponding correction candidates is displayed in the three-dimensional graph 83, while the number of stacked items is set to "2". When set, among the corresponding correction candidates, the block 84 of the correction candidate in which the block 84 is displayed at the same xy coordinate position does not exist is not displayed. Similarly, when the number of stacks is set to "3", among the corresponding correction candidates, the block 84 is displayed at the same xy coordinate position. The block 84 of the correction candidate does not have two or more other correction candidates. Is not displayed.

Further, the user clicks the desired document type selection check box 88 and sets the display form of the document type selection check box 88 to the state in which the corresponding document type is selected (the state filled in black). Therefore, it is possible to select the document type of the stored document from which the correction candidate is extracted, in which the corresponding block 84 should be displayed in the three-dimensional graph 83 from the plurality of correction candidates extracted for the user-specified ambiguity. For example, in the case of the example of FIG. 9, since only "mail (msg)", "design memo (.txt)" and "design document (.doc / docx)" are selected as the document types, the three-dimensional graph 83 Of the plurality of correction candidates extracted for the user-specified ambiguity, only the block 84 corresponding to each correction candidate extracted from the accumulated document of these document types is displayed.

On the other hand, the filtering parameter area 72 is divided so as to correspond to each item of the meta information of the corresponding stored document acquired for the correction candidate. In the case of the present embodiment, there are four items as these items, "product name", "version", "creation date", and "creator" as shown in FIG. 2. Therefore, in FIG. 9, the filtering parameter area is shown. 72 shows an example in which each of these four items (hereinafter, referred to as a filtering condition item) is divided into four region portions 72A to 72D.

Then, in the area portion 72A corresponding to the "product name" in the filtering parameter area 72, the product name of each product to which each stored document of the extraction source of the plurality of correction candidates extracted for the user-specified ambiguity is related. A character string 90 representing each of these product names and a check box 91 associated with each of these product names are displayed. Then, the user clicks the desired check box 91 from these check boxes 91 and sets the display form of the check box 91 to the state in which the corresponding product name is selected (the state filled in black). , The product name of the product related to the stored document of the extraction source from which the correction candidates for which the corresponding block 84 should be displayed in the three-dimensional graph 83 are extracted from the plurality of correction candidates extracted for the user-specified ambiguity. It can be set as a filtering condition. In FIG. 9, the product names of the three products "Product1", "Product2" and "Product4" were detected as such product names, and the product name "Product1" was set as the filtering condition from these product names. An example is shown.

Further, the version range setting slide bar 92 is displayed in the area portion 72B corresponding to the “version” in the filtering parameter area 72. Then, the user uses this version range setting slide bar 92 to block the corresponding block from the corresponding product version range of each stored document of the extraction source of the plurality of correction candidates extracted for the user-specified ambiguity. It is possible to specify the version range of the product corresponding to the stored document of the extraction source of the correction candidate to be displayed in the three-dimensional graph 83. In FIG. 9, since the version range of the product corresponding to each stored document of the extraction source of the plurality of correction candidates extracted for the user-specified ambiguous part was "V1.0.0" to "V10.0.0", this version range. An example is shown in which the scale of the slide bar is shown so that the version range of each stored document of the extraction source of the correction candidate whose corresponding block 84 should be displayed in the three-dimensional graph 83 can be specified.

Further, the creation date range setting slide bar 93 is displayed in the region portion 72C corresponding to the “creation date” in the filtering parameter region 72. Then, the user uses the creation date range setting slide bar 93 to three-dimensionally select the corresponding block 84 from the creation dates of each stored document of the extraction source of the plurality of correction candidates extracted for the user-specified ambiguity. It is possible to specify the range of the creation date of each stored document of the extraction source of the correction candidate to be displayed in the graph 83. In FIG. 9, since the creation date of each stored document of the extraction source of the plurality of correction candidates extracted for the user-specified ambiguous part was all after "2000", the corresponding blocks 84 within this range are shown in the three-dimensional graph 83. An example is shown in which the scale of the creation date range setting slide bar 93 is shown so that the creation date range of each stored document of the extraction source of the correction candidate to be displayed can be specified.

Further, in the area portion 72D corresponding to the "creator" in the filtering parameter area 72, a character string 94 representing the name of the creator of each stored document from which a plurality of correction candidates extracted for the user-specified ambiguity are extracted. And the check box 95 associated with each of these names are displayed. Then, the user clicks the check box 95 corresponding to the desired creator from among the names of these creators, and selects the corresponding creator for the display form of the check box 95 (the state filled in black). By setting, the creator of the storage document of the extraction source from which the correction candidates for which the corresponding block 84 should be displayed in the three-dimensional graph 83 is extracted from the plurality of correction candidates extracted for the user-specified ambiguous part is selected. It can be set as a filtering condition. FIG. 9 shows an example in which four creators "A", "B", "C", and "D" are detected as such creators, and all four creators are set as filtering conditions. Has been done.

Further, in the area portions 72A to 72D corresponding to the filtering condition items of "product name", "version", "creation date" and "creator" of the filtering parameter area 72, check boxes 96A for setting the Y-axis are respectively. ~ 96D is displayed. Then, the user clicks the check boxes 96A to 96D corresponding to one desired filtering condition item from these check boxes 96A to 96D, and selects the item corresponding to the display form (the state filled in black). ), The filtering condition item can be set on the y-axis of the three-dimensional graph 83. The example of FIG. 9 shows an example in which the item of "version" is set on the y-axis of the three-dimensional graph 83.

Further, in the area portions 72A to 72D corresponding to each of the filtering condition items of "product name", "version", "creation date" and "creator" of the filtering parameter area 72, check boxes for questionnaires (hereinafter referred to as "check boxes") are used. This is called a questionnaire check box) 97A to 97D are displayed. Then, the user clicks the questionnaire check boxes 97A to 97D corresponding to the filtering condition item particularly referred to when correcting the user-specified ambiguity from the questionnaire check boxes 97A to 97D to correspond to the display state. By setting the filtering condition item to the selected state (filled in black), especially when correcting the user-specified ambiguous part from a plurality of correction candidates extracted for the user-specified ambiguous part as described later. The referenced filtering condition items can be learned by the ambiguity correction support device 3 (FIG. 1).

The conditions set in the graph display adjustment area 81 of the graph display filtering area 71 and the filtering condition items of "product name", "version", "creation date", and "creator" of the filtering parameter area 72, respectively. Each filtering condition set in the corresponding area portions 72A to 72D is an AND condition, and only the block 84 corresponding to the correction candidate satisfying all the conditions is displayed in the three-dimensional graph 83.

On the other hand, in the ambiguity correction support device 3, among the blocks 84 displayed in the three-dimensional graph 83 of the ambiguity correction support screen 70, the desired block 84 is clicked and selected, as shown in FIG. The correction candidate detail screen 100 can be superimposed on the ambiguous part correction support screen 70.

Then, on the correction candidate detail screen 100, detailed information of the correction candidate corresponding to the block (hereinafter, this is referred to as a selection block) 84 selected by the user in the three-dimensional graph 83 of the ambiguity correction support screen 70 is displayed. The stored correction candidate detailed information table 101 is displayed.

In practice, the correction candidate detailed information table 101 includes the extraction source document ID column 101A, the occurrence condition column 101B, the subject column 101C, the numerical value column 101D, the operation column 101E, the original text column 101F, the product name column 101G, the version column 101H, and the creation date. It is composed of a column 101I, a creator column 101J, and a correction candidate score column 101K.

Then, in the extraction source document ID column 101A, a selection block is selected from among the identification IDs (hereinafter, referred to as sentence IDs) assigned to all the sentences in all the stored documents stored in the file server 4. The sentence ID assigned to the sentence including the correction candidate in the accumulated document in which the correction candidate corresponding to 84 is extracted is displayed.

Further, in the generation condition column 101B, the subject column 101C, the numerical value column 101D, the operation column 101E, the original text column 101F, the product name column 101G, the version column 101H, the creation date column 101I and the creator column 101J, the above-mentioned multi-conditions with respect to FIG. Occurrence condition column 50A, subject column 50B, numerical column 50C, operation column 50D, original text column 50E, product name column 50F, version column 50G, creation date column 50H and creation of the row of correction candidate corresponding to the selection block 84 in the table 50. The same information as the information stored in each of the person columns 50I is displayed. Further, in the correction candidate score column 101K, the correction candidate score calculated by the correction candidate scoring program 44 (FIG. 1) for the correction candidate is displayed.

At this time, in the correction candidate detailed information table 101, among the product name column 101G, the version column 101H, the creation date column 101I, and the creator column 101J, the "product name" in the filtering parameter area 72 of the ambiguous part correction support screen 70, " For the filtering condition items in which the questionnaire check boxes 97A to 97D in the area parts 72A to 72D corresponding to the respective filtering condition items of "version", "creation date", and "creator" are set to be filled in black. The character strings displayed in the corresponding columns 101G to 101I are highlighted in red or the like.

Further, on the upper side of the correction candidate detailed information table 101 of the correction candidate detail screen 100, a character string 102 "determine the numerical value of this correction candidate as the correction content of the ambiguous part" and a check box (hereinafter, this is selected as a correction candidate). (Called a check box) 103 is displayed. Then, the user designates the numerical value of the correction candidate whose details are displayed in the correction candidate detailed information table 101 (the numerical value displayed in the numerical value column 101D) as the numerical value to be added as the modifier of the predicative verb in the user-specified ambiguity. In this case, the correction candidate selection check box 103 is clicked to set the correction candidate selection check box 103 to a valid state (the correction candidate selection check box 103 is filled).

In this case, when the user clicks the "x" mark 104 displayed on the upper right of the correction candidate detail screen 100 and closes the correction candidate detail screen 100, the user specifies an appropriate position in the user-specified ambiguity in the target document. The user-specified ambiguity is corrected by adding the numerical value displayed in the numerical value column 101D of the correction candidate detailed information table 101 of the correction candidate detail screen 100 as a modifier of the predicate verb.

It is also possible to close the correction candidate detail screen 100 by clicking the "x" mark 104 without setting the correction candidate selection check box 103 in the valid state, but in this case, such correction is not performed. ..

(4) Various Processes Related to the Document Correction Function Next, specific processing contents of various processes executed by the ambiguity correction support device 3 related to the document correction function will be described. In the following, the processing subject of various processes will be described as a "program", but it goes without saying that the CPU 11 (FIG. 1) actually executes the processes based on the "program".

(4-1) Multi-condition table creation process FIG. 11 shows specific processing contents of the multi-condition table creation process that is periodically executed by the multi-condition table creation program 40 (FIG. 1). The multi-condition table creation program 40 periodically creates the latest multi-condition table 50 (FIG. 2) according to the processing procedure shown in FIG.

In practice, when the multi-condition table creation program 40 starts this multi-condition table creation process, it first selects one unprocessed stored document from the stored documents stored in the file server 4 (step S2 and subsequent steps) ( S1), the document file FL (FIG. 1) of the selected stored document (hereinafter, this is referred to as a selected stored document) is read from the file server 4 (S2). Further, the multi-condition table creation program 40 acquires the file information FLI (FIG. 1) of the selected storage document by reading it from the file server 4 (S3).

Subsequently, the multi-condition table creation program 40 sets the value of the variable i to 1 (S4), and executes the dependency analysis for the i-th sentence from the beginning among the sentences constituting the selected storage document (S5). ).

Then, when the multi-conditional statement can be detected by the dependency analysis, the multi-condition table creation program 40 registers necessary information regarding the detected multi-conditional sentence in the multi-condition table 50 (S6). When registering this multi-condition statement in the multi-condition table 50, information such as the product version supported by the selective storage document, the creation date and creator of the selective storage document, and the product name to which the selective storage document is related is obtained. Is acquired from the file information FLI of the selected storage document acquired in step S3.

Next, the multi-condition table creation program 40 determines whether or not there is a sentence for which the dependency analysis in step S5 has not been executed among the sentences constituting the selected storage document (S7). Then, when the multi-condition table creation program 40 obtains an affirmative result in this determination, the value of the variable i is incremented (increased by 1) (S8).

Further, the multi-condition table creation program 40 returns to step S5, and then repeats the processes of steps S5 to S8 until a positive result is obtained in step S7. By this iterative process, the receiving analysis is executed for all the sentences constituting the selected storage document, and when there is a multi-conditional sentence in the sentence, the necessary information about the multi-conditional sentence is sequentially registered in the multi-condition table 50.

Then, when the multi-condition table creation program 40 eventually obtains a negative result in step S7 by completing the dependency analysis for all the sentences constituting the selected storage document, is there an unprocessed storage document in step S2 and subsequent steps? It is determined whether or not (S9).

Then, when the multi-condition table creation program 40 obtains an affirmative result in this determination, it returns to step S1, and then, while sequentially switching the stored document selected in step S1 to another unprocessed stored document, steps S1 to step S1 to step S1. The process of S9 is repeated.

Then, when the multi-condition table creation program 40 eventually obtains a negative result in step S9 by completing the dependency analysis for all the sentences of all the stored documents, the multi-condition table creation process ends.

(4-2) Ambiguous part correction support process (4-2-1) Overall processing flow On the other hand, FIG. 12 shows a series of operations executed by the ambiguous part correction support device 3 in relation to the ambiguous part correction support function. (Hereinafter, this is referred to as an ambiguous part correction support process). The ambiguity correction support device 3 supports the correction work of correcting the ambiguity in the target document to a clear description according to the processing procedure shown in FIG.

In practice, when the target document is specified by the user in the ambiguity correction support device 3, the ambiguity correction support process shown in FIG. 12 is started, and first, the ambiguity word extraction program 41 (FIG. 1) starts the ambiguity word table. With reference to 51 (FIG. 3), an ambiguous word extraction process for extracting an ambiguous word in the target document is executed (S10). The part including the ambiguous word extracted at this time is the ambiguous part in the target document.

Subsequently, the correction candidate extraction program 41 (FIG. 1) refers to the ambiguous syntax table 52 (FIG. 4) and extracts an ambiguous syntax in which any ambiguous syntax in the target document matches the subject and the predicative verb. The extraction process is executed (S11). The part extracted at this time is also an ambiguous part in the target document.

Next, the correction candidate extraction program 42 (FIG. 1) sets a numerical value indicating the degree of operation represented by the predicate verb of the ambiguity for each of the ambiguities in the target document detected in steps S10 and S11, respectively. The correction candidate extraction process for extracting the correction candidates including the correction candidates from the multi-condition table 50 is executed (S12).

After that, the correction candidate scoring processing program 44 (FIG. 1) executes a correction candidate scoring process for calculating and assigning a correction candidate score to each correction candidate extracted in step S12 (S13).

Further, after that, the correction part / correction content display program 45 (FIG. 1) displays the above-mentioned ambiguity part presentation screen 60 for FIG. 8 in which the ambiguity part in the target document detected in step S10 and step S11 is highlighted. , The above-mentioned ambiguous part correction support screen 70 with respect to FIG. 9 and the above-mentioned correction candidate detail screen 100 with respect to FIG. 10 are sequentially displayed according to a user operation, and a correction part / correction content display process is executed (S14).

Subsequently, each filtering condition used by the selective correction candidate learning program 46 (FIG. 1) when narrowing down the correction candidates desired by the user on the ambiguous part correction support screen 70 (FIG. 9) for each ambiguous part in the target document. The selection correction candidate learning process for registering the information corresponding to the above in the weight table 53 (FIG. 5) is executed (S15).

Next, the correction result display program 47 (FIG. 1) performs each ambiguity portion of the target document according to the correction work performed by the user using the ambiguity portion correction support screen 70 (FIG. 9) and the correction candidate detail screen 100 (FIG. 10). Is executed (S16) to display the corrected text of the target document corrected by each of the above on the correction result screen (not shown) having the same configuration as that of FIG. 8 (S16). In the correction result screen, for the ambiguous part of the target document that has not been corrected by the user, the numerical value of the correction candidate having the highest correction candidate score among the correction candidates extracted for the ambiguous part is the predicate verb. It is displayed in a state where it is automatically corrected so that it can be added as a modifier.

Further, similarly, the correction part determination program 48 presents to the user a sentence having the same syntax (ambiguous syntax) as the ambiguous part corrected by the user in the stored document in the file server 4, and the correction candidate of the sentence is the same as described above. The correction portion determination process is executed in the same manner as presented in (S17). Then, after this, the ambiguity correction support process ends.

(4-2-2) Ambiguity word extraction process FIG. 13 shows specific processing contents of the above-mentioned ambiguous word extraction process executed by the ambiguous word extraction program 41 in step S10 of the above-mentioned ambiguous part correction support process with respect to FIG. Is shown.

The ambiguous word extraction program 41 starts the ambiguous word extraction process shown in FIG. 13 when the ambiguous part correction support process proceeds to step S10, and first converts the format format of the target document into the plain text format (S20).

Subsequently, the ambiguous word extraction program 41 selects one unprocessed ambiguous word from the ambiguous words registered in the ambiguous word table 51 (S21), and the selected ambiguous word (hereinafter, this). Is called a selected ambiguous word) in the target document (S22).

Next, the ambiguous word extraction program 41 determines whether or not the selected ambiguous word could be detected by the search in step S22 (S23). Then, when the ambiguous word extraction program 41 obtains a negative result in this determination, the process proceeds to step S25. Further, when the ambiguous word extraction program 41 obtains an affirmative result in the judgment of step S23, the ambiguous word extraction program 41 stores the sentence IDs of all the sentences in the target document in which the selected ambiguous word is detected by the search in step S22 (S24).

Next, the ambiguous word extraction program 41 determines whether or not there is an unprocessed ambiguous word in the ambiguous words registered in the ambiguous word table 51 after step S22 (S25). Then, when the ambiguous word extraction program 41 obtains an affirmative result in this determination, it returns to step S21, and after that, the ambiguous word selected in step S21 is not included in step S22 or later among the ambiguous words registered in the ambiguous word table 51. The processing of steps S21 to S25 is repeated while sequentially switching to other ambiguous words in the processing. By this iterative process, each part where any of the ambiguous words registered in the ambiguous word table 51 in the target document appears is extracted as an ambiguous part, and the sentence ID of the sentence including each of these ambiguous parts is stored.

Then, the ambiguous word extraction program 41 obtains a negative result in step S25 by completely detecting the sentence IDs of all the sentences in the target document in which any of the ambiguous words registered in the ambiguous word table 51 appears. This ambiguous word extraction process ends.

(4-2-3) Ambiguous Syntax Extraction Process On the other hand, FIG. 14 shows a specific example of the above-mentioned ambiguous syntax extraction process executed by the ambiguous syntax extraction program 42 in step S11 of the above-mentioned ambiguous part correction support process for FIG. Indicates the processing content.

When the ambiguity correction support process proceeds to step S11, the ambiguity syntax extraction program 42 starts the ambiguity syntax extraction process shown in FIG. 14, and first, from the ambiguity syntax registered in the ambiguity syntax table 52, steps S31 and subsequent steps are performed. Selects one unprocessed ambiguity syntax (S30), and searches the target document for sentences containing the same subject and predicate verb as the selected ambiguity syntax (hereinafter referred to as the selected ambiguity syntax) (S31).

Next, the ambiguous syntax extraction program 42 determines whether or not a sentence containing the same subject and predicate verb as the selected ambiguous syntax could be detected by the search in step S31 (S32). Then, the ambiguous syntax extraction program 42 proceeds to step S34 when a negative result is obtained by the determination in step S32. Further, when the ambiguous syntax extraction program 42 obtains an affirmative result in the judgment of step S32, it stores the sentence IDs of all the sentences in the target sentence including the same subject and predicate verb as the selected ambiguous syntax detected in the search in step S31. (S33).

Next, the ambiguous syntax extraction program 42 determines whether or not there is an unprocessed ambiguous syntax in the ambiguous syntax registered in the ambiguous syntax table 52 after step S31 (S34). Then, when the ambiguous syntax extraction program 42 obtains an affirmative result in this determination, it returns to step S30, and after that, the ambiguous syntax selected in step S30 is not included in step S31 or later among the ambiguous syntax registered in the ambiguous syntax table 52. The processing of steps S30 to S34 is repeated while sequentially switching to another ambiguous syntax of the processing. By this iterative process, parts in the target sentence containing the same subject and predicate verb as any of the ambiguous syntaxes registered in the ambiguous syntax table 52 in the target document are sequentially detected as ambiguous parts, and each of these ambiguous parts is included. The sentence ID of the sentence is sequentially stored.

Then, the ambiguous syntax extraction program 42 eventually detects all sentences in the target document containing the same subject and predicate verb as any of the ambiguous constructs registered in the ambiguous syntax table 52, and finishes storing the sentence ID in the step. When a negative result is obtained in S34, this ambiguous syntax extraction process ends.

(4-2-4) Correction Candidate Extraction Processing FIG. 15 shows specific processing contents of the above-mentioned correction candidate extraction processing executed by the correction candidate extraction program 43 in step S12 of the above-mentioned ambiguity correction support processing with respect to FIG. Is shown.

The correction candidate extraction program 43 starts the correction candidate extraction process shown in FIG. 15 when the ambiguity correction support process proceeds to step S12, and first stores the sentence ID in step S24 of FIG. 13 or step S33 of FIG. Among the ambiguous parts in the target document, one unprocessed ambiguous part after step S41 is selected (S40).

Subsequently, the correction candidate extraction program 43 reads the ambiguity portion selected in step S40 (hereinafter, this is referred to as a selection ambiguity portion in the description of FIG. 15) (S41).

Next, the correction candidate extraction program 43 selects one multi-condition statement that has not been processed after step S43 from the multi-condition statements registered in the multi-condition table 50 (FIG. 2) (S42), and in step S43. It is determined whether or not the subject of the read selection ambiguity portion includes the subject of the multi-conditional sentence selected in step S42 (hereinafter, this is referred to as a selected multi-conditional sentence) (S43).

Then, when the correction candidate extraction program 43 obtains a negative result in this determination, the process proceeds to step S45. Further, when the correction candidate extraction program 43 obtains an affirmative result in the judgment of step S43, the selected multi-conditional sentence is added to the correction candidate of the selected ambiguous part (S44), and after that, the multi-conditional sentence unprocessed after step S43. Is determined whether or not is present in the multi-condition table 50 (S45).

When the correction candidate extraction program 43 obtains an affirmative result in this determination, it returns to step S42, and then, while sequentially switching the multi-conditional sentence selected in step S42 to another multi-conditional sentence that has not been processed in step S43 and thereafter, the step S42 to S45 are repeated in the same manner as described above. By this iterative process, all the correction candidates of the selection ambiguity part are extracted.

Then, the correction candidate extraction program 43 eventually finishes executing the processes of steps S42 to S44 for all the multi-condition statements registered in the multi-condition table 50 with respect to the selection ambiguity at that time, so that the negative result in step S45. Is obtained, and it is determined whether or not there is an unprocessed ambiguity in the target document in which the sentence ID of the corresponding sentence is stored in step S24 of FIG. 13 or step S33 of FIG. (S46).

When the correction candidate extraction program 43 obtains a negative result in this determination, it returns to step S40, and then, the target document in which the position of the ambiguous part selected in step S40 is stored in step S24 of FIG. 13 or step S33 of FIG. The processing of steps S40 to S46 is repeated while sequentially switching to other unprocessed ambiguities in the other ambiguities in step S41 and subsequent steps. By this iterative process, correction candidates for each ambiguous part in the target document in which the sentence ID is stored in step S24 of FIG. 13 or step S33 of FIG. 14 are extracted.

Then, when the correction candidate extraction program 43 obtains a negative result in step S46 by completing the extraction of all the correction candidates for these ambiguous parts, the correction candidate extraction process ends.

(4-2-5) Correction Candidate Scoring Process FIG. 16 shows a specific example of the above-mentioned correction candidate scoring process executed by the correction candidate scoring program 44 in step S13 of the above-mentioned ambiguity correction support process for FIG. The processing content is shown.

The correction candidate scoring program 44 starts the correction candidate scoring process shown in FIG. 16 when the ambiguity correction support process proceeds to step S13, and first, the corresponding sentence in step S24 of FIG. 13 or step S33 of FIG. Among the ambiguous parts in the target document in which the sentence ID of is stored, one unprocessed ambiguous part is selected from step S51 onward (S50).

Subsequently, the correction candidate scoring program 44 is among the correction candidates extracted by the correction candidate extraction program 43 for the ambiguity selected in step S50 (hereinafter, this is referred to as a selection ambiguity in the description of FIG. 16). Step S52 and subsequent steps select one unprocessed correction candidate (S51).

Further, the correction candidate scoring program 44 acquires from the multi-condition table 50 the occurrence conditions in the multi-condition sentence corresponding to the correction candidates selected in step S51 (hereinafter, this is referred to as a selective correction candidate in the description of FIG. 15). (S52), a similarity score representing the degree of similarity between the occurrence condition of the event described in the selection ambiguity and the occurrence condition acquired in step S52 is calculated (S53).

のように、第１及び第２の数値ベクトルの内積を第１及び第２の数値ベクトルのベクトル長の乗算結果で除算することによりコサイン類似度を算出し、これをかかる類似度スコアとする。なおコサイン類似度は、０～1.0の範囲の値として算出され、値が1.0に近いほどより類似していることを意味する。 Specifically, the correction candidate scoring program 44 includes the occurrence condition of the event described in the selection ambiguity (the first occurrence condition) and the occurrence condition acquired in step S52 (the second occurrence condition). Is converted into a numerical vector with doc2vec, respectively, and based on the numerical vector of the first generation condition (referred to as the first numerical vector) and the numerical vector of the second generation condition (referred to as the second numerical vector), the following is performed. formula

As described above, the cosine similarity is calculated by dividing the inner product of the first and second numerical vectors by the multiplication result of the vector lengths of the first and second numerical vectors, and this is used as the similarity score. The cosine similarity is calculated as a value in the range of 0 to 1.0, and the closer the value is to 1.0, the more similar it is.

Further, the correction candidate scoring program 44 is the same product indicating whether or not the product name of the product corresponding to the document including the selection ambiguity (that is, the target document) and the product name of the product corresponding to the selection correction candidate match. Calculate the score (S54).

のように、重みテーブル５３（図５）における第１の製品名欄５３（図５）に「１」が設定されているエントリの合計個数sumに１を加算した値を定数0.01に乗算するようにして同一製品スコアを算出する。 Specifically, the correction candidate scoring program 44 first identifies the product name of the product corresponding to the target document from the description contents of the entire target document, and sets the product name of the product corresponding to the selection correction candidate in a multi-condition table. It is obtained from the product name column 50F (FIG. 2) of the corresponding line of 50, and it is determined whether or not these product names match. Then, when these product names match, the correction candidate scoring program 44 has the following equation.

In the weight table 53 (FIG. 5), the value obtained by adding 1 to the total number of entries in which "1" is set in the first product name column 53 (FIG. 5) is multiplied by the constant 0.01. And calculate the same product score.

Further, the correction candidate scoring program 44 is a product corresponding to a product version corresponding to the target document and a stored document (hereinafter referred to as a corresponding stored document) in which a multi-conditional sentence corresponding to the selected correction candidate is extracted. A score representing the difference from the version is calculated as a version difference score (S55).

Specifically, the correction candidate scoring program 44 first acquires the version of the product corresponding to the target document (hereinafter referred to as the target document product version) from the corresponding file information FLI (FIG. 1) and selects it. The product version corresponding to the stored document including the correction candidate (hereinafter referred to as the corresponding stored document product version) is obtained from the file information FLI (Fig. 1) of the stored document, and these target document product versions and the corresponding stored document are obtained. The version difference of the product version is calculated by the following formula.

In equation (3), the "latest version value" is a value obtained by arranging each integer representing the latest version of the product common to both the target document product version and the corresponding stored document product version as it is. The "oldest version value" is a value obtained by arranging each integer representing the oldest version of the product as it is. The "latest version value" and "oldest version value" are calculated by using the latest version and the oldest version values of the product acquired from the product name table 55 (FIG. 7), respectively.

For example, the "oldest version" is defined as "7.0.1" when the latest version of the product that corresponds to both the target document product version and the corresponding stored document product version is "7.0.1". The latest version is "701" (= 7 x 10 ² + 0 x 10 ¹ + 2) in which the integers "7", "0" and "1" are arranged as they are, and the oldest version of the product is "1.0. In the case of "2", "102" (= 1 x 10 ² + 0 x 10 ¹ + 2), in which "1", "0" and "2", which are the integers constituting "1.0.2", are arranged as they are, is the oldest. The version.

のように、重みテーブル５３における第１のバージョン欄５３Ｄ（図５）に「１」が設定されているエントリの合計個数sumに１を加算した値をバージョン差に乗算するようにしてバージョン差スコアを算出する。 Then, the correction candidate scoring program 44 has the following equation based on the calculated version difference.

As shown in the above, the version difference score is obtained by multiplying the version difference by the value obtained by adding 1 to the total number of entries sum in which "1" is set in the first version column 53D (FIG. 5) in the weight table 53. Is calculated.

In addition, the correction candidate scoring program 44 calculates a score according to the closeness between the creation date of the target document and the creation date of the corresponding accumulated document as the creation date score (S56).

Specifically, the correction candidate scoring program 44 acquires the creation date of the target document (hereinafter referred to as the target document creation date) from the meta information of the target document, and creates a stored document including the selection correction candidate. The date (hereinafter referred to as the corresponding storage document creation date) is acquired from the file information FLI of the relevant storage document, and the difference in the number of days between the target document creation date and the corresponding first and the corresponding storage document creation date (hereinafter referred to as the corresponding storage document creation date). This is called the creation date difference) is calculated by the following formula.

In equation (5), the "oldest stored document creation date" is the oldest stored document creation date among the stored document creation dates stored in the file server 4, and the "latest stored document creation date" is used. "Date" is the creation date of the latest accumulated document among the creation dates of the accumulated documents accumulated in the file server.

のように、重みテーブル５０における第１の作成日欄５３Ｅに「１」が設定されているエントリの合計個数sumに１を加算した値を作成日差に乗算するようにして作成日スコアを算出する。 Then, the correction candidate scoring program 44 has the following equation based on the calculated creation date difference.

As shown in the above, the creation date score is calculated by multiplying the creation date difference by the value obtained by adding 1 to the total number sum of the entries in which "1" is set in the first creation date column 53E in the weight table 50. do.

Further, the correction candidate scoring program 44 calculates a score according to the reliability of the creator of the target document as the creator reliability score (S57).

のように、重みテーブル５０のエントリのうち、当該作成者の名前が第２の作成者欄５３Ｌ（図５）に格納され、かつ第１の作成者欄５３Ｆ（図５）に「１」が格納されているエントリの合計数sumに定数0.001を乗算するようにしてその作成者の信頼作成者スコアを算出する。 Specifically, the correction candidate scoring program 44 acquires the creator of the stored document including the selected correction candidate from the file information FLI of the stored document, and uses the following equation.

As in the entry of the weight table 50, the name of the creator is stored in the second creator column 53L (FIG. 5), and "1" is stored in the first creator column 53F (FIG. 5). Calculate the trust creator score of the creator by multiplying the total number of stored entries sum by the constant 0.001.

Next, the correction candidate scoring program 44 selects and corrects based on the similarity score, the same product score, the version difference score, the creation date score, and the creator reliability score calculated in steps S52 to S57 as described above. The candidate correction candidate score is calculated (S58).

In the present embodiment, the correction candidate scoring program 44 uses the sum of the similarity score, the same product score, the version difference score, the creation date score, and the creator reliability score as the selection correction candidate score. calculate. In this case, according to the calculation method of each score as described above, the similarity score is the largest value, and the other same product score, version difference score, creation date score, and creator reliability score are smaller than this. Therefore, the value of the similarity score (that is, the degree of similarity of the occurrence conditions) greatly contributes to the correction candidate score of the selection correction candidate.

Subsequently, the correction candidate scoring program 44 sorts these correction candidates in descending order of the correction candidate scores based on the correction candidate scores calculated for each correction candidate in the selected ambiguity portion (S59). Then, the correction candidate scoring program 44 then determines whether or not there are unprocessed correction candidates in steps S52 and subsequent steps among the correction candidates extracted by the correction candidate extraction program 43 for the selected ambiguity (S60). ..

When the correction candidate scoring program 44 obtains an affirmative result in this determination, the correction candidate returns to step S51, and then the correction candidates selected in step S51 are the correction candidates extracted for the selection ambiguity by the correction candidate extraction program 53. Then, the processes of steps S51 to S60 are repeated while the steps S52 and subsequent steps sequentially switch to other unprocessed correction candidates. By this iterative processing, the correction candidate scores of all the correction candidates extracted for the selection ambiguity by the positive candidate extraction program 43 are calculated, and these correction candidates are sorted in descending order according to the correction candidate scores.

Then, the correction candidate scoring program 44 eventually obtains a negative result in step S60 by completing the scoring for all the correction candidates extracted for the selection ambiguity by the positive candidate extraction program 43 and the sorting of these correction candidates. It is determined whether or not the processing after step S51 has been executed for all the ambiguous parts in the target document (S61).

When the correction candidate scoring program 44 obtains an affirmative result in this determination, it returns to step S50, and then, while sequentially switching the ambiguous part selected in step S50 to another unprocessed ambiguous part in step S51 and subsequent steps, the step The processes of S50 to S61 are repeated. By this iterative process, the above-mentioned process is executed for each ambiguous part in the target document.

Then, when the correction candidate scoring program 44 eventually obtains a negative result in step S61 by completing the above-mentioned processing for each ambiguous part in the target document, the correction candidate scoring process ends.

(4-2-6) Selective Correction Candidate Learning Process FIG. 17 shows the above-mentioned selective correction candidate learning executed by the selective correction candidate learning program 46 (FIG. 1) in step S15 of the above-mentioned ambiguity correction support process for FIG. The specific processing contents of the processing are shown.

The selection correction candidate learning program 46 corrects this correction in a state in which the correction candidate selection check box 103 is valid (the correction candidate selection check box 103 is filled) on the correction candidate detail screen 100 described above with respect to FIG. When the candidate detail screen 100 is closed, the selection candidate correction learning process shown in FIG. 17 is started.

Then, in the selection correction candidate learning program 46, first, the graph display adjustment area 81 (FIG. 9) of the graph display filtering area 71 (FIG. 9) of the ambiguity correction support screen 70 (FIG. 9) when the correction candidate detail screen 100 is closed. The document type selected in 9) and the setting contents of each filtering condition set in the filtering parameter area 72 (FIG. 9) are imported (S70).

Subsequently, the selection correction candidate learning program 46 secures one unused row in the weight table 50, and stores necessary information in each column of the row based on the information fetched in step S70 (S71). ). Then, the selection correction candidate learning program 46 ends the selection correction candidate learning process.

(4-2-7) Similar correction location determination process FIG. 18 shows the above-mentioned similar correction location determination executed by the same correction location determination program 48 (FIG. 1) in step S17 of the above-mentioned ambiguous location correction support process for FIG. The specific processing contents of the processing are shown.

Similarly, when the ambiguous part correction support process proceeds to step S17, the corrected part determination program 48 starts the same corrected part determination process shown in FIG. 18, and first performs a dependency analysis for each corrected ambiguous part in the target document. Each is executed (S80).

Subsequently, similarly, the correction part determination program 48 extracts the ambiguity syntax for each corrected ambiguity part in the target document based on the analysis result of the dependency analysis in step S80, and all the extracted ambiguity syntaxes have the same syntax. It is stored in the determination table 54 (S81). As the method for extracting the ambiguity syntax for each ambiguity in step S, the same extraction method as the ambiguity syntax extraction process described above for FIG. 14 can be used.

Next, similarly, the correction part determination program 48 is a sentence (that is, an ambiguous syntax) having the same ambiguous syntax as any of the ambiguous syntax stored in the same syntax determination table 54 from each stored document stored in the file server 4 (FIG. 1). And a sentence with the same subject and predicative verb) (S82).

Then, similarly, the correction portion determination program 48 executes the same processing as in steps S12 to S16 of the above-mentioned ambiguity correction support processing (FIG. 12) for each sentence extracted in step S82 (S83). As a result, the user can perform correction work for clarifying each sentence extracted in step S.

Then, when the correction work for each sentence extracted in step S81 is completed, the correction portion determination program 83 ends the same correction portion determination process.

(5) Effect of the present embodiment As described above, the ambiguity correction support device 3 of the present embodiment extracts the multi-condition sentence from the stored document stored in the file server 4 and creates the multi-condition table 50. At the same time, an ambiguity is extracted from the target document, and a multi-condition sentence whose subject matches the extracted ambiguity is extracted from the multi-condition table 50 as a correction candidate for the ambiguity. A correction candidate score is given to each correction candidate based on the degree of similarity with the occurrence condition, and the block 84 having a height corresponding to the correction candidate score is displayed in the three-dimensional graph 83 of the correction location / correction content display screen 70. ..

Therefore, the user selects and determines correction candidates to be applied to the corresponding ambiguity in the target document based on the number of stacked blocks 84 displayed in the three-dimensional graph 83 and the height of each block 84. be able to. Therefore, according to the present ambiguity correction support device 3, the most appropriate numerical value is selected and determined from the entire related accumulated documents as the numerical value to be added as a modifier of the predicative verb to the ambiguity portion in the target document. Therefore, it is possible to facilitate and optimize the correction work for numerically clarifying the description of the ambiguous part in the target document.

(6) Other Embodiments In the above-described embodiment, on the ambiguity correction support screen 70, the correction candidate scores of the correction candidates extracted for the corresponding ambiguities in the target document are converted into the corresponding blocks 84. However, the present invention is not limited to this, and the correction candidate score of each correction candidate may be expressed by other display modes such as color, size, and density. good. Further, the correction candidate score of each correction candidate may be displayed in association with the block 84.

Further, in the above-described embodiment, the case where the ambiguity correction support device 3 is provided separately from the file server 4 has been described, but the present invention is not limited to this, and the ambiguity correction support device is not limited to this. The function of 3 may be installed.

Further, in the above-described embodiment, the case where the target document or the stored document is an email, a design memo, a design document, or the like exchanged at the development stage of the software product has been described, but the present invention is not limited to this, and the target document is not limited to this. The present invention can also be applied when the stored document is a technical document other than a software product or a document of another type.

Further, in the above-described embodiment, as a multi-condition extraction unit, each of a multi-condition extraction unit extracts a part including a numerical value indicating the degree of the subject, a predicate verb, the degree of the predicate verb, and a condition for generating an action represented by the verb from a plurality of accumulated documents. Of the multi-condition table creation program 40 of the above, the ambiguous word extraction program 41 and the ambiguous syntax extraction program 42 as the ambiguous part extraction unit for extracting the ambiguous part in the target document, and the multi-condition sentence extracted from a plurality of accumulated documents. A correction candidate extraction program 43 as a correction candidate extraction unit that extracts a multi-condition sentence whose subject matches the ambiguous part as a correction candidate of the ambiguous part, and a condition for generating an action represented by a verb included in the ambiguous part. A correction candidate scoring program 44 as a correction candidate scoring unit that assigns a correction candidate score to each correction candidate based on the degree of similarity of each correction candidate extracted for the ambiguity to the above-mentioned occurrence condition, and each correction. The correction part / correction content display program 45 as a display unit that displays the candidates in a display format according to the given correction candidate score, and the correction candidate displayed by the display unit for the target ambiguous part in the target document. From the selection correction candidate learning program 46 as a correction candidate learning unit that saves the filtering conditions set by the user to narrow down the correction candidates, the correction result display program 47, and the corrections in the target document from each stored document. A similar correction part determination program as a similar correction part determination unit that extracts a part or all of a document having the same syntax as the ambiguous part as an ambiguous part in the accumulated document and presents a correction candidate for the ambiguous part to the user. The case where 48 and 48 are mounted on one computer device (ambiguity correction support device 2) has been described, but the present invention is not limited to this, and these programs are distributed and arranged in a plurality of computer devices. These programs may cooperate with each other to execute the same processes as in the present embodiment.

The present invention can be widely applied to an ambiguity correction support device having various configurations that support a correction operation for correcting an ambiguity description in a document into a clear description.

1 …… Ambiguous part correction support system 3 …… Ambiguous part correction device 4 …… File server 5 …… Client, 11 …… CPU, 16 …… Input device, 17 …… Display device, 40 …… Multi-condition Table creation program, 41 ... Ambiguous word extraction program, 42 ... Ambiguous syntax extraction program, 43 ... Correction candidate extraction program, 44 ... Correction candidate scoring program, 45 ... Correction part / correction content display program, 46 ... … Selective correction candidate learning program, 47 …… Correction result display program, 48 …… Similar correction part judgment program, 50 …… Multi-condition table, 51 …… Ambiguous word table, 52 …… Ambiguous syntax table, 53 …… Weight table , 54 ... Same syntax judgment table, 55 ... Product name table, 60 ... Ambiguous part presentation screen, 70 ... Ambiguous part correction support screen, 83 ... 3D graph, 84 ... Block, 100 ... Correction candidate Details screen, FL ... document file, FLI ... file information.

Claims (10)

In the ambiguity correction support device that supports the correction work to correct the numerically ambiguous ambiguity in the first document so as to clarify it.
Multiple specific words in an ambiguous adjective are pre-stored as ambiguous words, and the content is ambiguous without a numerical value representing the degree of action represented by the predicate verb. And a storage device in which the combination of the predicate verbs is pre-stored as an ambiguous syntax, respectively.
A multi-condition extraction unit that extracts a part or all of a sentence described in If-Then format as a multi-condition sentence from a plurality of second documents.
From the first document, a part of a sentence or sentence containing any of the ambiguous words stored in the storage device, or the subject and the subject constituting any of the ambiguous syntax stored in the storage device. An ambiguity extraction unit that extracts a sentence or a part of a sentence including a combination of pre-descriptive words and verbs as the ambiguity.
A correction candidate extraction unit that extracts the multi-conditional sentence whose subject matches the ambiguous part of the multi-conditional sentence extracted from the plurality of second documents as correction candidates of the ambiguous part, respectively.
Correction candidates that give a score to each of the correction candidates based on the degree of similarity between the occurrence condition of the action represented by the verb included in the ambiguity and the occurrence condition of each of the correction candidates extracted for the ambiguity. With the scoring section,
An ambiguity correction support device including a display unit that displays each of the correction candidates in a display form corresponding to the given score. A correction candidate learning unit that saves filtering conditions set by the user to narrow down the correction candidates from the correction candidates displayed by the display unit for the ambiguous portion targeted in the first document. Further prepare
The correction candidate scoring unit
The ambiguity correction support device according to claim 1, wherein the score is given to each of the correction candidates by using the setting contents of the filtering conditions saved by the correction candidate learning unit. From each of the second documents, a part or all of a document having the same syntax as the corrected ambiguity in the first document is extracted as an ambiguity in the second document, and the ambiguity is addressed. The ambiguous part correction support device according to claim 1, further comprising a correction part determination unit as well as presenting the correction candidate to the user. The display unit is
A first axis associated with a numerical value included in each of the correction candidates extracted by the correction candidate extraction unit for the ambiguity target in the first document corresponding to each of the correction candidates. Corresponds to the three-dimensional coordinates having the second axis associated with the item of the filtering condition specified by the user for narrowing down the correction candidates and the third axis associated with the score. The ambiguity correction support device according to claim 1, wherein each block of the display form corresponding to the score of the correction candidate is displayed. The display unit is
The ambiguity correction support device according to claim 4 , wherein when any one of the blocks is designated by the user, detailed information of the correction candidate corresponding to the designated block is displayed. .. It is an ambiguity correction support method performed by an ambiguity correction support device that supports correction work for correcting numerically ambiguous ambiguities in a first document.
The ambiguity correction support device is
Multiple specific words in an ambiguous adjective are pre-stored as ambiguous words, and the content is ambiguous without a numerical value representing the degree of action represented by the predicate verb. And each combination of the predicate verbs has a storage device pre-stored as an ambiguous syntax.
The first step of extracting a part or all of a sentence described in If-Then format as a multi-condition sentence from a plurality of second documents, and
From the first document, a part of a sentence or sentence containing any of the ambiguous words stored in the storage device, or the subject and the subject constituting any of the ambiguous syntax stored in the storage device. The second step of extracting a sentence or a part of a sentence including a combination of pre-descriptive word verbs as the ambiguity, and
A third step of extracting the multi-conditional sentence whose subject matches the ambiguous part of the multi-conditional sentence extracted from the plurality of second documents as a correction candidate of the ambiguous part, respectively.
A fourth method of assigning a score to each of the correction candidates based on the degree of similarity between the condition of occurrence of the action represented by the verb included in the ambiguity and the condition of occurrence of each of the correction candidates extracted for the ambiguity. Steps and
A method for supporting ambiguity correction, which comprises a fifth step of displaying each of the correction candidates in a display form corresponding to the given score. The filtering conditions set by the user to narrow down the correction candidates from the correction candidates displayed for the ambiguous part targeted in the first document are saved.
In the fourth step,
The ambiguity correction support method according to claim 6 , wherein the score is given to each of the correction candidates by using the saved setting contents of the filtering condition. From each of the second documents, a part or all of a document having the same syntax as the corrected ambiguity in the first document is extracted as an ambiguity in the second document, and the ambiguity is addressed. The ambiguity correction support method according to claim 6 , further comprising a sixth step of presenting the correction candidate to the user. In the fifth step,
The first axis associated with the numerical value included in each of the correction candidates extracted for the ambiguity targeted in the first document corresponding to each of the correction candidates, and the correction candidate are narrowed down. The score of the corresponding correction candidate on a three-dimensional coordinate having a second axis associated with the item of the filtering condition specified by the user and a third axis associated with the score. The ambiguity correction support method according to claim 6 , wherein each of the blocks in the display form corresponding to the above is displayed. In the fifth step,
The ambiguity correction support method according to claim 9 , wherein when any one of the blocks is designated by the user, detailed information of the correction candidate corresponding to the designated block is displayed. ..

Images