JP2023072321A - Document proofreading support device, document proofreading support method, and document proofreading support program - Google Patents

Abstract

To provide a document proofreading support device, method, and program capable of detecting a part where a semantic error is potentially present in a document.SOLUTION: The document proofreading support device comprises a document analysis unit for determining whether or not a sentence included in a document is an unknown sentence which may include a semantic error according to whether or not the sentence included in the document follows a predetermined rule, an unknown sentence processing unit for estimating to which one of a plurality of components defined according to the type of the sentence the unknown sentence approximates, and a display processing unit for displaying the unknown sentence and a component to which the unknown sentence approximates.SELECTED DRAWING: Figure 1

Description

The present invention relates to a document proofreading support device, a document proofreading support method, and a document proofreading support program.

In companies, a large number of documents of various kinds are created on a daily basis. However, document creators often do not have enough time to review the document. In addition, business documents should be written in a language that accurately conveys the technical content. In recent years, it has become common for a computer to perform such a review.

The document proofreading support device disclosed in Patent Document 1 extracts inappropriate descriptions that do not meet predetermined rules from a document, calculates an expected correction time required for correcting them, and identifies the inappropriate descriptions and their expected corrections. Output time. The document proofreading support apparatus compares the correction time specified by the user with the expected correction time. If the expected correction time is shorter, the document proofreading support device corrects all inappropriate entries. Conversely, if the expected correction time is longer, the document proofreading support apparatus will correct only those inappropriate descriptions with high importance.

JP 2017-41164 A

Manuscripts before proofreading may contain not only grammatical or formal errors such as spelling errors, omissions, and spelling inconsistencies, but also semantic errors. The rules of Patent Literature 1 are for detecting grammatical or formal errors. A separate strategy was required to inform the user where a potential semantic error exists.
Accordingly, an object of the present application is to detect a place where a semantic error potentially exists in a document.

A document proofreading support apparatus according to the present invention includes a document analysis unit that determines whether a sentence contained in a document is an unknown sentence that may contain a semantic error, based on whether the sentence matches a predetermined rule. , an unknown sentence processing unit for estimating to which of a plurality of components defined according to the type of the document the unknown sentence approximates; and a display processing unit for displaying.
Other means are described in the detailed description.

According to the present invention, it is possible to detect a place where a semantic error potentially exists in a document.

1 is a diagram for explaining the configuration of a document proofreading support device; FIG. This is an example document. It is an example of rule information. It is an example of matching sentence information. It is an example of unknown sentence information. It is an example of rule-by-rule calibration time information. It is an example of proofreading time information classified by matching sentence. This is an example of document type/component information. It is an example of sentence space. It is an example of distance information. It is an example of importance information for each component. It is an example of score information. This is an example of score/correction time conversion information. It is an example of calibration time information for each unknown sentence. 4 is a flowchart of a document analysis processing procedure; 4 is a flow chart of a matching sentence processing procedure; 4 is a flowchart of an unknown sentence processing procedure; It is an example of a calibration time display screen.

EMBODIMENT OF THE INVENTION Hereinafter, the form (it is called "this embodiment") for implementing this invention is demonstrated in detail, referring drawings. This embodiment is an example of extracting grammatical or formal errors and semantic errors from business documents. The work of correcting typographical errors or expressions in a manuscript prepared by a document creator for the purpose of printing, bookbinding, etc., and making it into a final manuscript is generally called “proofreading”. This embodiment is also used for purposes other than printing and binding. This embodiment uses the term "proofreading" including "correction" in that case.

(terms, etc.)
A document is an electronic file containing character strings, and is a manuscript before proofreading.
A sentence is one unit separated by a full stop "." in a continuous character string included in a document. In this embodiment, "sentence" and "sentence" are synonymous.
A rule is a specific criterion for detecting grammatical or formal errors from a sentence.

A matching sentence is a sentence that contains a portion that matches the rule. Matching sentences contain grammatical or formal mistakes.
An unknown sentence is a sentence that does not contain a part that matches the rule. Unknown sentences may contain semantic mistakes. The reason for naming the "unknown" sentence is that it is unknown whether or not it contains a semantic error. In that sense, it can be said that the unknown sentence contains the potential proofreading part.
Matching sentences may also contain semantic mistakes. In this embodiment, a matching sentence becomes an unknown sentence once it has been proofread and contains no grammatical or formal errors.

A document type is a category of a document, and includes, for example, "estimate", "patent specification", "report", "specification", "minutes", and "approval".
A component is a description item that a document normally contains, and is defined for each document type. For example, the components of the quotation are "process", "work cost", "travel cost" and "work content".

(Configuration of Document Proofing Support Device)
FIG. 1 is a diagram for explaining the configuration of a document proofreading support apparatus 1. As shown in FIG. The document proofreading support apparatus 1 is a general computer, and includes a central control unit 11 , an input device 12 such as a mouse and a keyboard, an output device 13 such as a display, a main storage device 14 and an auxiliary storage device 15 . These are interconnected by a bus.

The auxiliary storage device 15 stores documents 31, rule information 32, matching sentence information 33, unknown sentence information 34, rule-based proofreading time information 35, matching sentence-based proofreading time information 36, document type/component information 37, distance information 38, It stores component element importance information 39, score information 40, score/proofreading time conversion information 41, unknown sentence proofreading time information 42, and component element estimation model 43 (details will be described later).

Of these, the document 31, the rule information 32, the rule-by-rule proofreading time information 35, the document type/component information 37, the component-by-component importance information 39, the score/proofreading time conversion information 41, and the component element estimation model 43 are is the result of taking in the document proofreading support apparatus 1 into the auxiliary storage device 15. The remaining matching sentence information 33, unknown sentence information 34, matching sentence proofreading time information 36, distance information 38, score information 40, and unknown sentence proofreading time information 42 are created by the document proofreading support apparatus 1 during processing. be.

The document analysis section 21, matching sentence processing section 22, unknown sentence processing section 23 and display processing section 24 in the main memory 14 are programs. The central control unit 11 reads out these programs from the auxiliary storage device 15 and loads them into the main storage device 14, thereby implementing the functions of each program (details will be described later). The auxiliary storage device 15 may be configured independently from the supply and demand adjustment support device 1 (cloud).

(documents)
FIG. 2 is an example of the document 31. As shown in FIG. A document 31 in FIG. 2 is a development-related report by the manufacturer. Document 31 includes sentences SE01 and SE02. Two rules match for the sentence SE01 (reference numerals 51 and 52). Therefore, sentence SE01 is a matching sentence. There is no rule that matches sentence SE02 (reference numeral 53). Therefore, sentence SE02 is an unknown sentence. Note that the rules 51 to 53 are for the purpose of explanation and are not described in the document 31 itself.

(rule information)
FIG. 3 is an example of the rule information 32. As shown in FIG. In the rule information 32, a rule ID (column 101), a rule (column 102), and a degree of importance (column 103) are associated with each other and stored.
A rule ID (column 101) is an identifier that uniquely identifies a rule.
Rules (column 102) are the rules described above.
Importance is relative weight among multiple rules. The user sets the importance within the range of "0<importance≦1".

(matching sentence information)
FIG. 4 is an example of matching sentence information 33 . In the matching sentence information 33, a sentence ID (column 111), a matching sentence (column 112), a rule ID (column 113) and a degree of importance (column 114) are associated with each other and stored.
A sentence ID (column 111) is an identifier that uniquely identifies a sentence, and here identifies a matching sentence.
Matching Sentences (column 112) are the matching sentences described above.
The rule ID (column 113) is the same as the rule in FIG.
The importance (column 114) is the same as the importance in FIG.
The matching sentence information 33 in FIG. 4 includes two records for the sentence SE01. This corresponds to rules 51 and 52 in FIG.

(Unknown sentence information)
FIG. 5 is an example of the unknown sentence information 34. As shown in FIG. In the unknown sentence information 34, the sentence ID (column 121) and the unknown sentence (column 122) are associated with each other and stored.
A sentence ID (column 121) is an identifier that uniquely identifies a sentence, and here identifies an unknown sentence.
The unknown sentence (column 122) is the previously described unknown sentence.
The unknown sentence information 34 in FIG. 5 includes one record for sentence SE02. This corresponds to column 53 (no matching rule) in FIG.

(Calibration time information by rule)
FIG. 6 is an example of the calibration time information 35 by rule. In the rule-by-rule proofreading time information 35, rule IDs (column 131) and proofreading times (column 132) are associated with each other and stored.
The rule ID (column 131) is the same as the rule ID in FIG.
The proof time (column 132) is the time required to proof the errors that match the rule. The user sets the calibration time in seconds based on past cases.

(Proofreading time information for each matching sentence)
FIG. 7 is an example of the matching sentence proofreading time information 36 . In the matching sentence proofreading time information 36, the sentence ID (column 141) and the proofreading time (column 142) are associated with each other and stored.
The sentence ID (column 141) is the same as the sentence ID in FIG.
The proofreading time (column 142) is the same as the proofreading time in FIG. 6, but here, the proofreading time in FIG. 6 is aggregated for each matching sentence. For example, the proofreading time "390" of sentence SE01 is the sum of "360" of rule R02 and "30" of rule R04 in FIG.

(Document type/component information)
FIG. 8 is an example of the document type/component information 37. As shown in FIG. In the document type/component information 37, component 1 (column 152) to component 4 (column 155) are stored in association with the document type (column 151).
The document type (column 151) is the document type described above. A user sets a plurality of document types according to his or her business.
Component 1 (column 152) to component 4 (column 155) are the components described above. The user sets any number of constituent elements for each document type. "KPI" means "Key Performance Indicator".

FIG. 9 is an example of a sentence space. Sentence vectors are defined on the premise that the document proofreading support apparatus 1 uses the sentence space for processing.

(sentence vector)
The document proofreading support device 1 converts one sentence as a character string into one sentence vector. The number of dimensions (the number of elements) of the sentence vector is equal to the number of words in the word dictionary of the sentence's language. And each element of the sentence vector is, for example, the number of times the word appears in that sentence. Now, the word dictionary consists of word a, word b, word c, word d, and word e. In a sentence, word a is 1 time, word b is 0 times, word c is 2 times, and word d is 0 times. , the word e appears once, the sentence vector is "(1, 0, 2, 0, 1)". The sentence vector described here is a very simple example. The document proofreading support apparatus 1 can create a more precise sentence vector that more accurately indicates the semantic features of the sentence by any method.

(sentence space)
The document proofreading support apparatus 1 can draw sentence vectors as points in the sentence space 44 . The number of dimensions of the sentence space 44 is equal to the number of dimensions of the sentence vector. Each axis of sentence space 44 indicates the number of occurrences of a particular word. The document proofreading support apparatus 1 collects a plurality of sample documents (learning data) in which all sentences are known to be grammatically, formally, and semantically correct for each document type, and converts all sentences of each sample document into a sentence vector. , and drawn as “●” in the sentence space 44 .

As a result, the document proofreading support apparatus 1 creates a sentence space 44 for each document type. One "●" in FIG. 9 corresponds to one sentence. The document proofreading support apparatus 1 uses a technique such as the k-means method to classify these ● into clusters. It is empirically known that the clusters 61a-61d often correspond one-to-one to the components of the document type. The sentence space 44 is a space in which sentences as learning data are classified into a plurality of clusters.

The document proofreading support apparatus 1 converts one unknown sentence into a sentence vector and draws it in the sentence space 44 as a circle. Then, a certain circle is classified into one of the clusters 61a to 61d, while another certain circle is not classified into any of the clusters 61a to 61d. The sentence 62a is classified into the cluster 61a and describes the component "process" of the document type "estimate". o Sentence 62b is not classified into any cluster 61a-61d. The sentence cannot be said to describe any component of the quotation, and there is a high possibility that it contains a semantic error (for example, an advertising word that is not suitable for the contents of the quotation). By the way, if at least one ○ is classified in the cluster of all the components of a certain document type, it can be said that the document covers all necessary description items. If there is at least one cluster that is not classified as ○, it can be said that the document lacks its constituent elements (entry items).

(Constituent element estimation model)
When a sentence vector forming a document of a certain document type is input, the constituent element estimation model 43 calculates the distance between the sentence vector (○) in the sentence space 44 and each constituent element (the center of each cluster) of the document type. is a function that outputs the distance of The component estimation model 43 exists for each document type. The constituent element estimation model 43 may also perform a process of converting the unknown sentence into a sentence vector. The document proofreading support apparatus 1 may update the positions and sizes of the clusters 61a to 61d in the sentence space 44 and store them in the auxiliary storage device 15 at any timing using the latest learning data.

(distance information)
FIG. 10 is an example of the distance information 38. As shown in FIG. In the distance information 38, the sentence ID (column 161), the unknown sentence (column 162), the process distance (column 163), the work cost distance (column 164), the travel cost distance (column 165), and the work content distance (column 166). stored in association with each other.
The sentence ID (column 161) is the same as the sentence ID in FIG.
The unknown sentence (column 162) is the same as the unknown sentence of FIG.

The step distance (column 163) is the distance between the unknown sentence (indicated by "o") in the sentence space 44 (FIG. 9) and the center of the cluster 61a. The distance may be Euclidean distance, Mahalanobis distance or some other distance. If this distance is greater than a predetermined threshold (for example, the radius of cluster 61a), it is highly probable that the unknown sentence does not describe at least the component "process" (and so on).
The work cost distance (column 164) is the distance between the unknown sentence in sentence space 44 and the center of cluster 61b.
Travel distance (column 165) is the distance between the unknown sentence in sentence space 44 and the center of cluster 61c.
The work content distance (column 166) is the distance between the unknown sentence in the sentence space 44 and the center of the cluster 61d.

The distance information 38 in FIG. 10 is the distance information 38 for the document type "estimate". If FIG. 10 is the distance information 38 for the document type "patent specification", for example, the process distance, work distance, travel cost distance, and work content distance are the problem distance, solution distance, claim distance, and prior art distance, respectively. change in distance.

(Importance information for each component)
FIG. 11 is an example of the importance level information 39 for each component. In the component-specific importance level information 39, the component (column 171) and the level of importance (column 172) are associated with each other and stored.
The component (column 171) is the component described above.
Importance (column 172) is the relative weight between multiple components. The user sets the importance within the range of "0<importance≦1". The document proofreading support apparatus 1 may automatically set the importance based on the number of characters or keywords in the sentence of each component of the sample document.
The component importance information 39 exists for each document type.

(Score information)
FIG. 12 is an example of the score information 40. As shown in FIG. In the score information 40, the sentence ID (column 181), the unknown sentence (column 182), the process score (column 183), the work cost score (column 184), the travel cost score (column 185), and the work content score (column 186). stored in association with each other.
The sentence ID (column 181) is the same as the sentence ID in FIG.
The unknown sentence (column 182) is the same as the unknown sentence of FIG.

The process score (column 183) is a value obtained by multiplying the process distance shown in FIG. 10 by the degree of importance shown in FIG. 11 corresponding to the process.
The work cost score (column 184) is a value obtained by multiplying the work cost distance in FIG. 10 by the degree of importance in FIG. 11 corresponding to the work cost.
The travel expense score (column 185) is a value obtained by multiplying the travel expense distance in FIG. 10 by the degree of importance in FIG. 11 corresponding to the travel expense.
The work content score (column 186) is a value obtained by multiplying the work content distance in FIG. 10 by the degree of importance in FIG. 11 corresponding to the work content.
Score information 40 also exists for each document type. In the above description, the score is obtained by multiplying the distance by the degree of importance, but this is only an example. Scores may be calculated using addition, exponential calculations, and the like. In short, the greater the distance and the greater the importance, the greater the score.

(Score/correction time conversion information)
FIG. 13 is an example of the score/proofreading time conversion information 41 . In the score/proofreading time conversion information 41, scores (column 191) and proofreading times (column 192) are stored in association with each other.
The score (column 191) is, for example, the "process score" described above, and more generally calculates the importance of the component with respect to the distance between the sentence vector and the center of the cluster of the component. It is a value obtained by (multiplying, etc.).
The proofreading time (column 192) is the time required to proofread the erroneous part corresponding to the score in the unknown sentence. The user sets the calibration time in seconds based on past cases. The document proofreading support apparatus 1 may update the proofreading time based on the time the user actually spent proofreading.

(Calibration time information for each unknown sentence)
FIG. 14 is an example of the calibration time information 42 classified by unknown sentence. In the unknown sentence proofreading time information 42, sentence ID (column 201), unknown sentence (column 202), process score (column 203a), process proofreading time (column 203b), work cost score (column 204a), work cost proofreading Time (column 204b), travel expense score (column 205a), travel expense proofreading time (column 205b), work content score (column 206a), and work content proofreading time (column 206b) are stored in association with each other.

The sentence ID (column 201) is the same as the sentence ID in FIG.
The unknown sentence (column 202) is the same as the unknown sentence in FIG.
The process score (column 203a) is the same as the process score in FIG.
The process calibration time (column 203b) is the calibration time resulting from conversion of the process score by the score/calibration time conversion information 41 (FIG. 13).
The work cost score (column 204a) is the same as the work cost score in FIG.
The work cost proofreading time (column 204b) is the proofreading time resulting from the conversion of the work cost score by the score/proofreading time conversion information 41 .

The travel expense score (column 205a) is the same as the travel expense score in FIG.
The travel expense proofreading time (column 205b) is the proofreading time resulting from the conversion of the travel expense score by the score/proofreading time conversion information 41 .
The work content score (column 206a) is the same as the work content score in FIG.
The work content proofreading time (column 206b) is the proofreading time resulting from conversion of the work content score by the score/proofreading time conversion information 41 .

The unknown sentence proofreading time information 42 stores not only the score but also the proofreading time, so that the user can know how much time it takes to proofread an unknown sentence to which component.

(Processing procedure)
The processing procedure of this embodiment will be described below. There are three procedures: document analysis procedure, matched sentence procedure and unknown sentence procedure.

(Document analysis processing procedure)
FIG. 15 is a flow chart of the document analysis processing procedure.
In step S301, the document analysis unit 21 of the document proofreading support apparatus 1 acquires a document. Specifically, the document analysis unit 21 acquires the document 31 from the outside or from the auxiliary storage device 15 via the input device 12 .

In step S302, the document analysis unit 21 acquires a character string. Specifically, the document analysis unit 21 acquires character strings from the document 31 .
In step S303, the document analysis unit 21 divides the character string into sentences. Specifically, the document analysis unit 21 divides the character string into a plurality of sentences using the period "." as a delimiter. At this time, the document analysis unit 21 may perform morphological analysis (part-of-speech analysis) and dependency analysis between words.

In step S304, the document analysis unit 21 compares sentences and rules. Specifically, first, the document analysis unit 21 acquires any one of the unprocessed sentences.
Second, the document analysis unit 21 compares the sentence with each rule in the rule information 32 (FIG. 3) to identify all rules that match the sentence.
Third, the document analysis unit 21 counts the number of rules specified in the "second" of step S304. The count results are "0", "1", "2", "3", .

In step S305, the document analysis unit 21 determines whether the sentence matches the rule. Specifically, when the count result in the “third” of step S304 is “0” (step S305 “NO”), the document analysis unit 21 proceeds to step S307; otherwise (step S305 “YES”). ”), and the process proceeds to step S306.

In step S306, the document analysis unit 21 registers the matching sentence information 33 (FIG. 4). Specifically, the document analysis unit 21 creates a record for the sentence to be processed in the matching sentence information 33 .

In step S307, the document analysis unit 21 registers the unknown sentence information 34 (FIG. 5). Specifically, the document analysis unit 21 creates a record for the sentence to be processed in the unknown sentence information 34 . If the sentence included in the document 31 does not match the predetermined rule in step S305, the document analysis unit 21 determines in step S307 that the sentence is an unknown sentence that may contain a semantic error.

The document analysis unit 21 repeats the processing after step S304 for each unprocessed sentence, and ends the document analysis processing procedure after step S306 or S307 for the last sentence. When the document analysis processing procedure ends, all the sentences included in the document 31 acquired in step S301 are sorted into matching sentence information 33 (FIG. 4) or unknown sentence information 34 (FIG. 5) and stored. ing.

(Matching sentence processing procedure)
FIG. 16 is a flow chart of the matching sentence processing procedure.
In step S321, the matching sentence processing unit 22 of the document proofreading support apparatus 1 acquires a matching sentence. Specifically, the matching sentence processing unit 22 acquires any unprocessed matching sentence from the matching sentence information 33 (FIG. 4).

In step S322, the matching sentence processing section 22 acquires the proofreading time based on the rule. Specifically, the matching sentence processing unit 22 acquires the proofreading times of all rules matching the sentence obtained in step S321 from the rule-by-rule proofreading time information 35 (FIG. 6).

In step S323, the matching sentence processing unit 22 totals the proofreading time for each sentence. Specifically, the matching sentence processing unit 22 sums up the proofreading times acquired in step S322.

In step S324, the matching sentence processing unit 22 registers in the matching sentence proofreading time information 36 (FIG. 7). Specifically, the matching sentence processing unit 22 creates a record for the sentence to be processed in the matching sentence proofreading time information 36 .
The matching sentence processing unit 22 repeats the processing of steps S321 to S324 for each unprocessed matching sentence. When there are no more unprocessed matching sentences, the matching sentence processing procedure ends.

(Unknown sentence processing procedure)
FIG. 17 is a flow chart of the unknown sentence processing procedure.
In step S341, the unknown sentence processing section 23 of the document proofreading support apparatus 1 acquires an unknown sentence. Specifically, the unknown sentence processing unit 23 acquires any unprocessed unknown sentence from the unknown sentence information 34 (FIG. 5).

At step S342, the unknown sentence processing unit 23 accepts the document type. Specifically, first, the unknown sentence processing unit 23 displays the document 31 acquired in step S301 on the output device 13 .
Second, the unknown sentence processing unit 23 accepts input of the document type by the user via the input device 12 . The user views the document 31 and determines the type of document to be input. For convenience of explanation, it is assumed here that "estimate" has been entered. The unknown sentence processing unit 23 may automatically determine the document type based on, for example, the title of the document 31 without waiting for input by the user.

In step S343, the unknown sentence processing unit 23 creates a sentence vector. Specifically, the unknown sentence processing unit 23 converts the sentence obtained in step S341 into a sentence vector by the method described above.

In step S344, the unknown sentence processing unit 23 creates the sentence space 44. FIG. Specifically, first, the unknown sentence processing unit 23 creates the sentence space 44 in FIG. 9, and creates a plurality of clusters using the sample document of the quotation as learning data (●). Each created cluster corresponds to the component 1 to component 4 of the document type/component information 37 (FIG. 8). A cluster here may be the smallest sphere that envelops all ● classified into that cluster, or a sphere whose center is the center of gravity of all ● and whose radius is the distance from the center of gravity to the furthest ● may be The unknown sentence processing section 23 may complete the processing in advance at any timing.
Second, the unknown sentence processing unit 23 draws the sentence vector (○) created in step S343 in the sentence space 44 .

In step S345, the unknown sentence processing section 23 determines whether or not the unknown sentence contains a component. Specifically, first, the unknown sentence processing unit 23 checks whether or not the ◯ drawn in the “second” of step S344 exists inside any cluster.
Secondly, the unknown sentence processing unit 23 proceeds to step S346 if ◯ exists inside any cluster (step S345 "YES"), otherwise (step S345 "NO"), step S347. proceed to

At step S346, the unknown sentence processing unit 23 sets the score and proofreading time to "0". Specifically, the unknown sentence processing unit 23 assumes that the score and proofreading time of the unknown sentence acquired in step S341 are "0". Here, the unknown sentence processing unit 23 determines that the unknown sentence does not require proofreading because the unknown sentence describes one of the constituent elements normally included in the quotation.

In step S347, the unknown sentence processing section 23 calculates the distance. Specifically, first, the unknown sentence processing unit 23 inputs the sentence vector created in step S343 to the constituent element estimation model 43 for the quotation. Then, the constituent element estimation model 43 outputs the distance between the unknown sentence (○) and the center of each cluster in the sentence space 44 . The unknown sentence processor 23 receives this distance.
Secondly, the unknown sentence processing unit 23 creates a record of the distance information 38 (FIG. 10) based on the distance received in the "first" of step S347. At step S347, the unknown sentence processing unit 23 estimates which of the plurality of components defined according to the type of the document 31 the unknown sentence is similar to.

In step S348, the unknown sentence processing unit 23 calculates a score. Specifically, first, the unknown sentence processing unit 23 multiplies the process distance of the record created in the "second" step S347 by the importance corresponding to the process in FIG. Calculate the score. The unknown sentence processing unit 23 similarly calculates the work cost score, the travel cost score, and the work content score.
Secondly, the unknown sentence processing unit 23 creates a record of the score information 40 (FIG. 12) based on the score calculated in the "first" of step S348.

In step S349, the unknown sentence processing section 23 calculates the calibration time. Specifically, the unknown sentence processing unit 23 applies the score/proofreading time conversion information 41 of FIG. 13 to the process score of the record created in the "second" of step S348 to calculate the process proofreading time. The unknown sentence processing unit 23 similarly calculates the work expense proofreading time, the travel expense proofreading time, and the work content proofreading time.

In step S350, the unknown sentence processing unit 23 registers the unknown sentence-based calibration time information 42 (FIG. 14). Specifically, the unknown sentence processing unit 23 creates a record of the unknown sentence proofreading time information 42 (FIG. 14) based on the scores and proofreading times calculated in steps S346, S348 and S349.

In step S351, the display processing unit 24 of the document proofreading support apparatus 1 displays the proofreading time. Specifically, the display processing unit 24 displays the calibration time display screen 71 (FIG. 18) on the output device 13 using the record created in step S324 and the record created in step S350. After that, the unknown sentence processing procedure is terminated.

FIG. 18 is an example of the calibration time display screen 71. As shown in FIG. The matching sentence column 72 displays the proofreading time and importance of the matching sentence of the document 31 . The proofreading time and importance here are, in principle, displayed for each rule that matches the matching sentence. The unknown sentence column 73 displays the score and proofreading time for the unknown sentence of the document 31 . In principle, the score and proofreading time here are displayed for each unknown sentence and for each component. Suppose now that the user has entered a check mark in the selection column of the record having the matching sentence column 72 and the unknown sentence column 73 . Then, the display processing unit 24 displays the document 31 in the document column 74 and then highlights (for example, underlines) the selected sentence. The document 31 here is different from the document 31 of FIG.

In the document field 74, sentence SE03 is an unknown sentence. The display processing unit 24 attaches a balloon 75 to the sentence SE03. A balloon 75 describes "closest component: process". This indicates that the "process distance" is the shortest among the distances between the sentence SE03 and each component.

In this case, for example, the following are assumed.
・Although the document creator tried to write the sentence SE03 about the process, it is highly likely that the sentence SE03 contained a semantic error as a result of a slight lack of attention.
・There is a high possibility that the document creator wrote sentence SE03 about an event unrelated to any component. It is possible to modify this sentence to describe any component. In that case, considering that the unknown sentence SE03 is closest to the process, the calibration time to calibrate the unknown sentence SE03 to a sentence about the process is the shortest unless the importance of the process is extremely large.

The display processing unit 24 attaches a balloon 76 to the sentence SE21. Balloon 76 describes two rules that match sentence SE21. The display processing unit 24 receives from the user or obtains from the user's schedule information or the like the time that the user (document creator or person in charge of proofreading) can use to proofread the document 31 , and displays it as the available time 77 . The display processing unit 24 displays the time required for proofreading all the sentences included in the document 31 or the sentence corresponding to the input check mark (the sum of the proofreading times described above) as the predicted proofreading time 78 . The display processing unit 24 may store the result of proofreading of the sentence by the user in the document field 74 in the auxiliary storage device 15 .

The display processing unit 24 may display the unknown sentence determined not to require proofreading in step S345 at any location on the proofreading time display screen 71. FIG.

(Effect of this embodiment)
The effects of the document proofreading support apparatus of this embodiment are as follows.
(1) The document proofreading support device can display an unknown sentence that may contain a semantic error and the constituent elements of the document that the unknown sentence is similar to.
(2) The document proofreading support apparatus can quantify the approximation between the unknown sentence and each component of the document as the distance in the sentence space.
(3) The document proofreading support apparatus can accurately estimate the components of an unknown sentence by converting the unknown sentence into a sentence vector.

(4) The document proofreading support apparatus can update the position and size of the cluster by updating the learning data.
(5) The document proofreading support device can accurately identify unknown sentences that do not need to be proofread.
(6) The document proofreading support device can reflect the importance of each component in the distance.
(7) The document proofreading support device can display the time required to proofread an unknown sentence.

In addition, the present invention is not limited to the above-described embodiments, and includes various modifications. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the described configurations. In addition, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Moreover, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

1 Document Proofing Support Device 11 Central Control Device 12 Input Device 13 Output Device 14 Main Storage Device 15 Auxiliary Storage Device 21 Document Analysis Section 22 Matching Sentence Processing Section 23 Unknown Sentence Processing Section 24 Display Processing Section 31 Document 32 Rule Information 33 Matching Sentence Information 34 unknown sentence information 35 proofreading time information by rule 36 proofreading time information by matching sentence 37 document type/component information 38 distance information 39 importance information by component 40 score information 41 score/proofreading time conversion information 42 proofreading time by unknown sentence Information 43 Component estimation model 44 Sentence space 71 Calibration time display screen

Claims (9)

a document analysis unit that determines whether or not the sentence contained in the document is an unknown sentence that may contain a semantic error, based on whether or not the sentence contained in the document matches a predetermined rule;
an unknown sentence processing unit for estimating to which of a plurality of components defined according to the type of the document the unknown sentence approximates;
a display processing unit that displays the unknown sentence and the component that the unknown sentence approximates;
A document proofreading support device comprising: The unknown sentence processing unit
using a component estimation model that takes as input an unknown sentence that does not match the predetermined rule and outputs distances between the unknown sentence and each of the plurality of components;
The document proofreading support device according to claim 1, characterized by: The unknown sentence processing unit
converting the unknown sentence to a sentence vector, inputting the converted sentence vector to the component estimation model, and obtaining the distance from the component estimation model;
3. The document proofreading support device according to claim 2, characterized by: The component estimation model is
Calculating the distance between the converted sentence vector and each of the plurality of clusters in a space in which sentences as learning data are classified into a plurality of clusters;
4. The document proofreading support device according to claim 3, characterized by: The unknown sentence processing unit
determining that there is no need to proofread the unknown sentence if the unknown sentence is classified into one of the plurality of clusters;
5. The document proofreading support device according to claim 4, characterized by: The unknown sentence processing unit
calculating a score for each component based on the distance and the importance defined for each component;
The display processing unit
displaying the calculated score in association with the unknown sentence;
The document proofreading support device according to claim 5, characterized by: The unknown sentence processing unit
Convert the score to the time required for proofreading,
The display processing unit
displaying the converted time in association with the unknown sentence;
7. The document proofreading support device according to claim 6, characterized by: The document analysis unit of the document proofreading support device
determining whether or not the sentence contained in the document is an unknown sentence that may contain a semantic error, depending on whether or not the sentence contained in the document matches a predetermined rule;
The unknown sentence processing unit of the document proofreading support device includes:
estimating to which of a plurality of components defined according to the type of the document the unknown sentence approximates;
The display processing unit of the document proofreading support device includes:
displaying the unknown sentence and the component to which the unknown sentence approximates;
A document proofreading support method for a document proofreading support device characterized by: the computer,
a document analysis unit that determines whether or not the sentence contained in the document is an unknown sentence that may contain a semantic error, based on whether or not the sentence contained in the document matches a predetermined rule;
an unknown sentence processing unit for estimating to which of a plurality of components defined according to the type of the document the unknown sentence approximates;
a display processing unit that displays the unknown sentence and the component that the unknown sentence approximates;
A document proofreading support program to function as

Images