JPH0793330A - Document correcting device - Google Patents

Abstract

PURPOSE:To improve the efficiency of correcting operation by not making uncertain corrections when parts which are recognized as improper words by a grammatical analysis are automatically corrected together at a time and to enable document corrections which are small in failure in detection and easy for a user to understand by making a display in consideration of the vagueness of the grammatical analysis when sequential corrections are made. CONSTITUTION:A correction object part decision means 14 decides parts meeting conditions of automatic correction by referring to the result of a vagueness deciding means 12 for the vagueness of an input document and the conditions in a correction condition memory 19 and an automatic correcting means 16 automatically substitutes correction candidates for the parts; when the sequential corrections are made, a vague part determining means 31 detects parts where the recognition of an inadequate word is vague and displays them on a display means 22, and the vagueness of the detected parts is eliminated according to a vagueness elimination indication from a vagueness elimination indication input means 23. Consequently, the improper words are corrected in sequence by a sequential correcting means 6.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is a sentence correction for correcting an incorrect portion (a typographical error / missing character, an inappropriate expression, an incomprehensible expression, etc.) based on a grammatical analysis using a word dictionary for an input Japanese sentence. Regarding the device.

[0002]

2. Description of the Related Art In a conventional text correction device, an inappropriate portion is first detected based on a result of grammatical analysis of an input sentence using a word dictionary, and a mark is added to the inappropriate portion. Next, a correction candidate is displayed for each of the improper portions, the user selects a desired one from the candidates, and the sentence is corrected by replacing it. Such a conventional technique is described, for example, in Japanese Patent Laid-Open No. 61-208164, "Display system of Japanese proofreading device".

The method of recognizing an improper portion is mainly a method of detecting a portion that fails in grammar analysis (first method) and a method of detecting a portion whose grammar analysis result matches a specific condition (second method). Method), and a method of registering an incorrect word in a word dictionary in advance and detecting a portion recognized as an incorrect word in a grammar analysis result (third method). Of these, the third method is to register an incorrect word correction method in the word dictionary so that not only the incorrect part can be detected but also the correction can be made. ing.

FIG. 5 is a block diagram showing an example of the configuration of a conventional text correction device.

The conventional text correction device has a text input means 1,
Input sentence memory 2, grammar analysis unit 3, proofreading word dictionary memory 4, analysis result memory 5, sequential correction unit 6, candidate selection instruction input unit 7, working memory 11, corrected sentence memory 17, correction candidate information memory 18 , Display control means 2
1. The display means 22.

The text input means 1 reads a text input by the user and stores the input text in the input text memory 2 together with its position information and the like. The grammar analyzing unit 3 refers to the proofreading word dictionary memory 4 and stores the intermediate result as the working memory 1.
The grammatical analysis is performed while temporarily storing in 1 and the analysis result is stored in the analysis result memory 5. The proofreading word dictionary memory 4 is a word dictionary for performing a normal grammatical analysis, in which proofreading information such as words to be corrected (inappropriate words) and correction candidates thereof is added. The sequential correction means 6 refers to the word dictionary memory 4 with proofreading information and the analysis result memory 5 to search for a correction candidate for the incorrect portion of the input sentence, and to obtain the information of the correction candidate for the incorrect portion, the correction candidate information memory 18. To display on the display means 22 by writing
According to the candidate selection instruction from the candidate selection instruction input means 7, the correction of the input sentence is sequentially corrected. The corrected sentence resulting therefrom is stored in the corrected sentence memory 17. The display control means 21 controls the display means 22 in accordance with the contents of the corrected sentence memory 17 and the correction candidate information memory 18 to display the input sentence, add marks to the improper portion and the corrected portion, display the correction candidate, etc. I do.

A keyboard, a mouse, a character recognition device, a kana-kanji conversion input device, etc. are used for the sentence input means 1, and a keyboard, a mouse, etc. are used for the candidate selection instruction input means 7. An IC memory, a magnetic disk device, a magnetic tape device or the like is used for the input sentence memory 2, the proofreading word dictionary memory 4, the analysis result memory 5, the working memory 11, the corrected sentence memory 17, and the correction candidate information memory 18. . Grammar analysis means 3, sequential correction means 6, display control means 2
1 is a CPU of a computer or the like. A CRT display or the like is used as the text display unit 22.

That is, the conventional correction candidate device sequentially points out an unsuitable portion, shows a correction candidate for the portion, and allows the user to select it (or let the user directly correct the portion). Then, the text was corrected to a proper one. Further, the following conventional techniques have been devised in order to reduce the number of operations required for correction by the user and improve efficiency.

One of them is described in Japanese Unexamined Patent Publication No. Hei 2-140861 "Sentence Making Device", which omits the operation of checking correction candidates one by one for each detected improper portion, It can be corrected with a single key press. The other is described in Japanese Patent Laid-Open No. 3-240160, “Character string replacement method”, and when a certain portion is replaced with a correction candidate, the same error as that portion appearing after that is generated. Appropriate locations are to be replaced all at once.

Although not used in the text correction device, in a general text editing device, the character string A is replaced by the character string B all at once, and the replacement is improper. It is also considered that the character string A can be returned to the original character string later (for example, it is described in Japanese Patent Laid-Open No. 233669/1993, “Document Creation Device”). According to this method, the number of operations required for correction is reduced as a whole if the replacement is inappropriate and there are few places that must be restored.

[0011]

In the conventional sentence correction device as described above, the correction candidate is confirmed for each of the detected high possibility of improperness, and the appropriate one is selected. The operations of selecting and replacing are sequentially required. On the other hand, although measures for reducing the number of operations have been studied in the past, the most significant effect is that the batch replacement as described at the end of the section of the related art is also performed in the text correction device. It is to be realized. That is, if the correction candidate (especially the first candidate) is often an appropriate correction result, it is automatically corrected in a batch first, and when the correction is not appropriate, the operation to restore By doing so, the number of operations as a whole may be reduced and the work efficiency may be improved.

However, if a portion that is unlikely to be an inappropriate portion is corrected or replaced with an uncertain correction candidate in many cases, the number of operations to be undone is increased and the efficiency is reduced. I can't go up. Therefore, it is preferable to perform automatic correction by focusing on a portion that is surely incorrect and the correction method is reliable.

On the other hand, a portion where automatic correction has not been performed will be sequentially checked and corrected in the same manner as in a conventional text correction device. At that time, if there is uncertainty in the location detected as an inappropriate location, it is preferable to present it to the user so that it can be known.

Ambiguity in grammar analysis is a factor that causes the above-mentioned uncertainty. For example, when the grammar analysis can be interpreted in a plurality of ways due to different isomorphisms but different readings, uncertainty arises in the method of detecting and correcting the improper portion. Specifically, suppose there are two possible interpretations of the word "burnt clay", "yaketsuchi" and "shodo." At this time, if it is interpreted as "yaketsuchi", it is recognized as an unsuitable word and the correction to "burnt soil" is correct. On the other hand, if it is interpreted as “shodo”, it is recognized as an incorrect term and the correction to “scorch” is correct.

In such a case, an appropriate interpretation may be possible by a process that understands a deep meaning or context.
With conventional grammar analysis techniques, such processing has not reached a practical level. In addition, sometimes, even when a human interprets the ambiguity, the interpretation may be ambiguous (when multiple interpretations are possible). In such a case, the conventional text correction device forcibly selects one of the interpretations even if there is not enough information for performing an appropriate interpretation. As a result, it is not uncommon for an incorrect interpretation to be taken, resulting in omission of detection and loss of the user's confidence in the text correction device (or the text improper portion detection device).

Therefore, the present invention eliminates the above-mentioned drawbacks of the conventional sentence correction device, and makes an uncertain correction when collectively correcting the portions recognized as inappropriate words by grammar analysis. By not performing it, the efficiency of correction work is improved,
Further, by providing a display in consideration of ambiguity in grammatical analysis when performing sequential correction, it is possible to provide a sentence correction device that is capable of realizing sentence correction that is less likely to be missed and is easy for the user to understand. To aim.

[0017]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention provides the following two types of sentence correction devices.

The sentence correction device according to the first aspect of the present invention is an proofreading memory for storing an input sentence memory for storing an inputted sentence and a word dictionary having proofreading information such as an incorrect word type and a correction candidate. A word dictionary memory with information, a grammar analysis unit for recognizing an incorrect word by performing a grammatical analysis of the input sentence with reference to the proofreading word dictionary memory, and a result of the grammar analysis of the input sentence. An analysis result memory that stores, a working memory that temporarily stores an intermediate result of the grammar analysis, an ambiguity determination unit that determines ambiguity in the grammar analysis of the input sentence with reference to the working memory, A determination result memory that stores the determination result of the ambiguity determination unit, a correction condition memory that stores the condition to be satisfied by the automatic correction target portion, the analysis result memory, and the determination result A correction target location determining means for detecting a location corresponding to the condition of the correction condition memory with reference to a memory, a correction target location memory for storing the location detected by the correction target location determining means, and the calibration information-equipped dictionary memory. An automatic correction unit is provided for referring to the input sentence and automatically correcting the portion stored in the correction target portion memory.

The sentence correction device of the second invention is a proofreading device for storing an input sentence memory for storing an inputted sentence and a word dictionary having proofreading information such as types of incorrect words and correction candidates such as correction candidates. A word dictionary memory with information, a grammar analysis unit for recognizing an incorrect word by performing a grammatical analysis of the input sentence with reference to the proofreading word dictionary memory, and a result of the grammar analysis of the input sentence. An analysis result memory that stores, a working memory that temporarily stores an intermediate result of the grammar analysis, an ambiguity determination unit that determines ambiguity in the grammar analysis of the input sentence with reference to the working memory, A determination result memory that stores the determination result of the ambiguity determination unit, an ambiguity resolution instruction input unit that inputs an instruction to eliminate the ambiguity regarding the recognition of an inappropriate word, and the determination result An ambiguity point determining means for detecting an ambiguity part of the incorrect word in the input sentence with reference to the memory and eliminating the ambiguity of the part by an instruction from the ambiguity resolution instruction input means, and an improper part Candidate selection instruction input means for inputting a correction candidate selection instruction regarding the correction candidate, and the determination of the ambiguous portion determination means and the analysis result memory to detect an inappropriate portion in the input sentence and correct the incorrect portion. It is characterized by further comprising: sequential correction means for retrieving a candidate from the proofreading word dictionary memory and replacing the inadequate portion with a correction candidate designated by the candidate selection designation input means.

[0020]

According to the present invention, first, when the parts that have been identified as incorrect words by the grammar analysis are automatically corrected collectively, uncertain correction is not performed, thereby improving the efficiency of the correction work. The determination as to whether or not the correction is an uncertain correction is performed mainly by checking whether some kind of ambiguity has occurred in the grammar analysis process. For example, when "burnt clay" can be interpreted as "yaketsuchi" or "shodo", it is possible to judge that part is excluded from the target of automatic correction.

Next, regarding the location detected as an inappropriate location, a correction candidate is displayed and the display is performed in consideration of the ambiguity in the grammatical analysis when performing sequential correction such that the user can select it. It realizes a sentence correction that is easy to understand by the user with few omissions. The display considering ambiguity in grammar analysis is, for example,
In the case of the above-mentioned "burnt clay", once it is detected as an incorrect word, a color mark etc. is added, and when the correction candidate for that part is displayed, it is interpreted as "Yaketsuchi" and "Shodo". It is displayed so that the user can understand that there are two ways of interpretation (reading). After allowing the user to select one of the interpretations, the correction candidate corresponding to it is shown. That is, if the user selects “Yaketsuchi”, the correction candidate “burnt soil” is displayed, and if the user selects “Shodo”, the correction candidate “scorched soil” is displayed. In such a case, if one of the interpretations is forcibly selected and the correction candidate is presented, the user will be prompted to make an unintended correction, and the confidence of the text correction device will be lost. But
By adopting the display method as described above, it is possible to obtain a natural understanding of the user.

Further, in the above example, the case where the word can be interpreted as two incorrect words is explained.
In some cases, the street interpretation is possible. For example, there is a case where "scent wheel" can be interpreted as "kyosha" or "kyasha". If you interpret it as "Kyosha," there is no need to correct it with proper words. On the other hand, when it is interpreted as "Kyasha", it is "delicate" and must be corrected. In this case, forcibly interpreting the proper word "Kyosha" would result in omission of detection if it was written as "Kyasha". On the other hand, if priority is always given to the interpretation of the inappropriate word as “chasha”, the missed detection of the incorrect word can be avoided, but the user is strongly given an unnatural impression. Even in such a case, according to the present invention, it is possible to avoid omission of detection and give the user a natural understanding.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing the configuration of a first embodiment of a text correction device according to the present invention. Hereinafter, each component will be described.

The sentence input means 1 is a means for inputting a character string containing kanji and kana of a Japanese sentence. A kana-kanji conversion device using a standard keyboard, an all-character arrangement keyboard, a character recognition device, a voice recognition device, etc. are used. Alternatively,
A magnetic disk device or a magnetic tape device for reading a text file created by a word processor or the like may be used.

The input text memory 2 stores the text input by the text input means 1. An IC memory, a magnetic disk device, a magnetic tape device or the like is used.

The proofreading word dictionary memory 4 stores a word dictionary having proofreading information including an incorrect word type, an incorrect word type, a correction candidate, and the like. An IC memory, a magnetic disk device, a magnetic tape device or the like is used. FIG. 7 shows an example of the contents of the word dictionary memory with proofreading information. In the word dictionary of FIG. 7, a notation 700, a part of speech 701, a reading 702, a proofreading attribute 703, and a correction candidate 704 are registered for each word. Of these, notations 700 and parts-of-speech 70 are commonly registered in word dictionaries used for normal grammar analysis.
The first item is the reading 702 item. In the proofreading word dictionary memory 4 of FIG. 7, in addition to these items, items such as a proofreading attribute 703 and a correction candidate 704 are added.
Further, not only ordinary words (appropriate words) but also words to be corrected (inappropriate words) are registered.

In the example of FIG. 7, "perfect", "scorched earth", "burnt earth" and the like are proper words, and "complete wall" and "burnt earth" are inappropriate words. As for the inappropriate word, an incorrect type such as an incorrect word or an unsuitable word is given as the proofreading attribute 703. In addition to incorrect words and incorrect words that can not be sent, there are various definitions of incorrect categories, such as recommended rewriting words, recommended kana writing words, synonyms, and prohibited words, depending on the target field and application of the text correction device. It is possible (in the present invention, the improper type is not particularly limited). Also,
Correction candidates 704 are also attached to the inappropriate words. When a plurality of correction candidates are considered, a plurality of correction candidates may be registered. On the contrary, even if the word is an incorrect word, the correction candidate may not be registered because the correction candidate cannot be determined. In the present specification, a portion that is not registered in a normal word dictionary, that is, an inappropriate word, and the inappropriate type and correction candidate thereof are collectively referred to as proofreading information. Although FIG. 7 shows an example in which the proofreading information added word dictionary is configured as one file, the present invention limits the structure of the proofreading information added word dictionary to the form shown in FIG. 7. is not. It is also possible to form a plurality of files, such as separating the part of the notation that serves as the search key as an index file, or separating the part of the correction candidates and the entire calibration information into separate files.

The grammar analyzing means 3 refers to the proofreading word dictionary memory 4 to perform grammatical analysis of the sentence stored in the input sentence memory, and recognizes an inappropriate word in the sentence.
A CPU or the like of a computer is used. FIG. 17 is a flowchart showing an example of implementation of the grammar analysis means 3. In FIG. 17, first, in step 170, the word dictionary is searched,
The word candidates that match the partial character strings of the input sentence are extracted. Next, in step 171, combinations of the words retrieved in step 170 that satisfy grammatical constraints are formed. Finally, when a plurality of combinations are formed, in step 172, the most probable one is selected. Various methods have been devised as methods for selecting the most probable ones, such as a method using a semantic constraint and a method using heuristics that a combination with the smallest number of clauses is most likely. For details, see “Memory of Japanese language dictionary and automatic division of Japanese sentence” (Nagao et al., “Information Processing”, Vol. 19, No. 6).
No. 1978), "Language Processing Method for Japanese Sentence Output" (Miyazaki / Oyama, "IPSJ Transactions", Vol. 27, No. 11, 1986), "Solid writing using minimum phrase number method Japan. Morphological Analysis of Words "(Yoshimura et al.," Information Processing Society of Japan ", Vol. 24, No. 1, 1983)," On Proposal of Morphological Analysis by Minimum Connection Cost Method and Evaluation of Complexity "(Hisami &Nitta," IEICE research report "NLC90-
8, 1990) and the like.

The grammar analyzing means 3 of the present invention differs from the usual grammar analyzing means as described in the above-mentioned document.
Proofreading information is added to the word dictionary. This does not change the processing procedure as described with reference to FIG. 17, but differs in that necessary items of the proofreading information are read from the word dictionary in step 170, and information about the incorrect word is obtained in the analysis result. . For details of such grammar analysis performed by referring to a word dictionary with proofreading information, refer to "Proposal of a high-speed editing support system using dedicated hardware" (Takemoto
IPSJ 43rd National Convention Proceedings, 1991)
And "Japanese document proofreading support system St. "WORDS"
(Fukushima et al., Proc. Of the 45th National Convention of IPSJ, 1
992) and the like.

The working memory 11 temporarily stores the intermediate result of the grammar analysis by the grammar analysis means 3. An IC memory, a magnetic disk device, a magnetic tape device or the like is used. Figure 8
An example of the contents of the working memory 11 is shown in FIG. The content shown in FIG. 8 represents the state immediately before the candidate selection step 172 (immediately after the connection verification step 171) in the grammar analysis for the example sentence "My hometown was completely burned in the war." ing. It expresses what part of speech / reading each character string in the input sentence can be interpreted as.In regard to the incorrect word, the incorrect type and correction candidate information are also read from the word dictionary and stored. ing.

The analysis result memory 5 stores the result of the grammar analysis by the grammar analysis means 3 for the sentence stored in the input sentence memory 2. FIG. 9 shows an example of the contents of the analysis result memory 5. The content shown in FIG. 9 is the result of the grammar analysis for the example sentence “My hometown was completely burned in the war.” The candidates are further selected from the state of the working memory 11 in FIG. The interpretation is uniquely determined. The ambiguity determination unit 12 refers to the work memory 11 and determines the ambiguity in the grammar analysis of the grammar analysis unit 3 regarding the sentence stored in the input sentence memory. Computer CP
U or the like is used. FIG. 18 shows a flowchart of an example of implementation of the ambiguity determination means 12. In the flowchart of FIG. 18, regarding the content of the working memory 11 immediately before the candidate selection step of grammar analysis, paying attention to the inappropriate word type read from the word dictionary, and the incorrect word overlaps with other word candidates. When it appears, the analysis information of that portion is written in the determination result memory 13. The reason for paying attention to the incorrect word is that only the incorrect word portion is to be corrected, so that unnecessary information is not stored. However, such saving is not indispensable for realizing this means, and there can be various ways of realizing it depending on the type of condition described in the correction condition memory 19 described later. For example, there is a realization method such as detecting all the places that can be interpreted in a plurality of ways, or detecting only the part having the same word.

The determination result memory 13 is used by the ambiguity determination means 1
The determination result of 2 is stored. An IC memory, a magnetic disk device, a magnetic tape device or the like is used. FIG. 10 shows an example of the contents of the judgment result memory 13. The contents shown in FIG. 10 correspond to the contents of the judgment result memory 13 when the ambiguity judging means 12 is executed by the processing procedure as shown in FIG. 18 with respect to the example of the working memory 11 of FIG. In the determination result memory 13 of FIG. 10, the position and the analysis information are stored with respect to the ambiguity portion.

The correction condition memory 19 stores the conditions to be met by the automatic correction target portion. An IC memory, a magnetic disk device, a magnetic tape device or the like is used. Figure 11
FIG. 6 is a diagram showing an example of a correction condition memory 19. In the example of the correction condition memory 19 of FIG. 11, whether or not the location detected by the ambiguity determining unit 12 is to be the target of automatic correction, and what kind of inappropriate word is to be the target of automatic correction is the target. It is described as 1 if yes, and 0 if not. In the example of FIG. 11, the location detected by the ambiguity determination unit 12 is not a target for automatic correction, and the incorrect type is a wrong term in other locations,
It represents the condition that automatic correction is performed in the case of an unsuitable word that is not sent or a rewriting recommended word. However, the description format of the condition is not limited to such a flag format, and may be a format described as a natural language or a program.

The correction point judging means 14 refers to the analysis result memory 5 and the judgment result memory 13 to refer to the correction condition memory 1
The part corresponding to the condition 9 is detected. Computer C
PU or the like is used. FIG. 19 is a flow chart showing an example of implementation of the correction point determination means 14 in accordance with the example of the correction condition memory 19 of FIG. Step 191 in the flowchart of FIG. 19 corresponds to the condition check of the inappropriate type of FIG. 11, and step 192 corresponds to the condition check of whether or not the analysis ambiguity has occurred in FIG. 11.

The correction target location memory 15 stores the detection location of the correction target location determination means 14. An IC memory, a magnetic disk device, a magnetic tape device or the like is used. Figure 1
FIG. 2 is a diagram showing an example of the contents of the correction target location memory 15. In Fig. 12, the corrected part for the example sentence "My hometown was completely burned by the war." Is shown to be a word of length 2 from the 9th character position (complete wall). There is. If necessary, the incorrect type of the relevant part may be stored.

The automatic correction means 16 refers to the proofreading-added dictionary memory 4 and corrects the sentence stored in the input sentence memory 2 with respect to the portion stored in the correction target portion memory 15 (if there are a plurality of correction candidates, the correction candidate). It is automatically corrected by substituting one candidate). A CPU or the like of a computer is used. FIG. 20 is a flowchart of an implementation example of the automatic correction means 16. In the flowchart of FIG. 20, a correction target portion in the input sentence is found from the correction target portion memory 15, a correction candidate for the correction portion is searched from the proofreading word dictionary memory 4, and the correction portion and the first correction candidate searched are searched. Replacing one candidate. When the end of the input sentence is reached, the corrected sentence is written in the corrected sentence memory 17.

The corrected text memory 17 stores the correction result of the automatic correction means 16 for the text stored in the input text memory 2. An IC memory, a magnetic disk device, a magnetic tape device or the like is used. In the present embodiment, the corrected text memory 17 is a component different from the input text memory 2, but the corrected text memory 17 is not provided and the automatic correction means 1 is provided.
It is also possible for 6 to directly rewrite the input sentence memory 2. FIG. 13 is a diagram showing an example of the contents of the corrected text memory 20. The text output means 20 outputs the corrected contents of the text memory 17. A printer or CRT display is used.

Among the above components, the sentence input means 1, the input sentence memory 2, the grammar analysis means 3, the proofreading word dictionary memory 4, the analysis result memory 5, and the working memory 1
1. The corrected text memory 17 is the same as the constituent elements of the conventional text correction device shown in FIG. Ambiguity determination means 1
2, determination result memory 13, correction target portion determination means 14,
The correction target location memory 15, the automatic correction means 16, and the correction condition memory 19 are new components according to the present invention.

Next, an example of the operation when the sentence "My hometown was completely burned in the war" was input using the sentence input means 1 will be described. The entered text is
It is stored in the input sentence memory 2 as shown in FIG. Figure 6
In the following, the number of characters counted from the beginning as shown in is used to represent the character position. The grammar analyzing means 3 performs grammar analysis with reference to the proofreading word dictionary memory 4 as shown in FIG. The intermediate result of the grammar analysis is temporarily stored in the working memory 11 in a form leaving two possibilities for the interpretation of "burnt clay" as shown in FIG.

The ambiguity determining means 12 determines the working memory 11 in step 181 as shown in the flowchart of FIG.
If it is determined that the 9th word candidate “complete wall” of is an inappropriate word, it is checked in step 182 whether there is a word whose existence section overlaps with “complete wall”. Since there is no word in the working memory 11 in which the existence section overlaps in “Complete wall”, it is determined as NO in step 182 and is the next character position.
Move to the 0th determination process. Similarly, in step 182, 1
When it is determined that the second word candidate “burnt clay” is an inappropriate word, it is checked whether there is a word whose existence section overlaps with “burnt clay”. Since there is a word in the working memory 11 in which the existence section overlaps in the "burnt clay", YES is obtained in step 182.
It was determined that the 12th character position, and the length of 2 "fired clay" had an ambiguity in reading in two ways, "Yaketsuchi" and "Shodo" Judgment result memory 1 that the word corresponds to two types of correction items
3 is stored as shown in FIG.

In the analysis result memory 5, candidates are further selected from the state of the working memory 11 shown in FIG. 8 by the grammar analyzing means 3, and the interpretation is uniquely determined as shown in FIG.

It is assumed that the correction condition memory 19 describes the conditions as shown in FIG. At this time, the correction location determining unit 14 refers to the analysis result memory 5 and the determination result memory 13 to detect the correction location like the correction target location memory 15.

In the correction target portion memory 15, as shown in FIG. 12, it is shown that the correction portion is a word (complete wall) having a length of 2 from the ninth character position.

The automatic correction means 16 refers to the proofreading dictionary memory 4 and is stored in the input sentence memory 2 as shown in FIG.
Is automatically replaced with the first correction candidate “perfect” for the part stored in the correction target part memory 15.

The corrected sentence memory 17 stores the correction result of the sentence stored in the input sentence memory 2 by the automatic correction means 16 as shown in FIG. The text output means 20 outputs the corrected contents of the text memory 17. as a result,
The "complete wall" is automatically corrected to "perfect", but the "burned clay" part is not corrected because the interpretation is ambiguous.

FIG. 2 is a block diagram showing the configuration of the second embodiment of the sentence correction device of the present invention. Correction condition input means 25 is added to the configuration of FIG. 1 as a component.

The correction condition input means 25 is an input means for the user to change the contents stored in the correction condition memory 19. A keyboard or the like is used as the correction condition input means 25. For example, the correction condition memory 1 as shown in FIG.
Regarding 9, if the value of the incorrect term of the incorrect type is rewritten from 1 to 0, the "complete wall" (correction candidate: "perfect") that is detected and corrected in the above operation example will not be automatically corrected. it can.

FIG. 3 is a block diagram showing the configuration of the third embodiment of the sentence correction device of the present invention.

In addition to the structure of the conventional sentence correction device shown in FIG. 5, the ambiguity point determining means 31 and the ambiguity resolution instruction input means 3 are added.
2. The ambiguous location information memory 33 is added as a constituent element. The ambiguous portion determining means 31 detects a portion in the input sentence in which the recognition of an incorrect word is ambiguous by referring to the determination result memory 13, and writes the information of the ambiguous portion in the ambiguous portion information memory 33. Display means 22
, And the ambiguity of the detected position is resolved according to the ambiguity resolution instruction from the ambiguity resolution instruction input means 32. The result is notified to the sequential correction means 6. A CPU of a computer or the like is used as the ambiguous portion determination unit 31. The ambiguity resolving instruction input means 32 is means for inputting an instruction for resolving the ambiguity regarding the recognition of an inappropriate word by the ambiguity determining means 31, and a keyboard, a mouse or the like is used. FIG. 21 shows the ambiguous portion determining means 31.
It is a flow chart which shows an example of realization of. The ambiguous portion information memory 33 is a memory that is temporarily stored to display information about multiple interpretations of the ambiguous portion recognition of the uncorrected word in the input sentence, which is detected by the ambiguous portion determining unit 31, and is an IC memory, A magnetic disk device, a magnetic tape device, etc. are used.

An operation example of the third embodiment will be described with reference to an input sentence "My hometown was completely burned in the war." As described in the first embodiment, the judgment result memory 13 has two interpretations of "yaketsuchi" and "shodo" for the "burnt clay" having a length of 2 from the 12th character position.
It is stored as shown in FIG. 10 that the word corresponding to two kinds of correction items, that is, the word (reading), the incorrect word and the incorrect word. Ambiguous location determining means 31
Detects that the “burned earth” part is a part having ambiguity regarding the recognition of an incorrect word, and writes the information to the ambiguity part information memory 33. As a result, for example, in FIG.
A display such as a) is displayed.

In FIG. 22 (a), there are two readings for "burnt earth", "shodo" and "yaketsuchi", and it is shown that either one can be selected.

The ambiguity resolution instruction input means 32 allows the user to select either "Yaketsuchi" or "shodo" reading. When the user selects the reading of “shodo”, the sequential correction means 6 refers to the proofreading word dictionary memory 4 and refers to the ambiguous part determination means 3 of the input sentence.
"Soil" for those who read "Shodo", which is the result of decision 1.
The correction candidate for “?” Is retrieved, and the correction candidate “scorch” is written in the correction candidate information memory 18 to be displayed on the display means 22. FIG. 22 (b) is a display example of that state. When the user designates "scorched earth" as a correction candidate for "burnt clay" by means of the candidate selection instruction input means 7, the sentence "correction made my hometown completely scorched" in the corrected text memory 17 is displayed. Written. On the other hand, when "Yaketsuchi" is selected by the disambiguation instruction input means 32, the result is as shown in FIG. 22 (b ').

As described above, in the present embodiment, the place where the ambiguity occurs in the interpretation of the inappropriate word can be posted to the user in a natural form. In the above description, the "complete wall" portion, which did not cause ambiguity in the recognition of the incorrect word, is skipped. However, that portion is stored in the analysis result memory 5 as in the conventional sentence correction device. Based on the content, the sequential correction means 6 can detect it as an erroneous term, and can indicate and replace the correction candidate “perfect”.

Similarly, in the third embodiment, an example of processing an input sentence "It is a fragrance wheel" will be described. Ambiguity occurs in the grammatical analysis in the part of "Kosha", and it is assumed that there are candidates to read "Kyosha" and "Kyasha". At this time, the conventional text correction device forcibly decides one interpretation. for that reason,
If you take "Kyosha", it will not be detected because it is a proper word,
If "Kasha" is taken, it will be detected as an erroneous term. The user can determine which interpretation is correct, but from the user's perspective, the user may feel that the text correction device has missed the detection or made an incorrect interpretation. On the contrary,
In the text correction device of this embodiment, as shown in FIG.
Mark "Perfume wheel" once and let the user know that there are two readings, "Kyosha" and "Kyasha", and let the user select. If "Kyosha" is selected, it is not an incorrect word as shown in Fig. 23 (b), so the mark is removed. When "Kyasha" is selected, a correction candidate is shown as an incorrect term as shown in Fig. 23 (b '). . This prevents omission of detection and gives the user a natural understanding.

FIG. 4 is a block diagram showing the configuration of the fourth embodiment of the text correction device of the present invention.

In addition to the combination of the configuration of FIG. 2 and the configuration of FIG. 3, a correction cancellation instruction input means 34, a correction cancellation means 35,
An automatic correction information memory 36 is added as a constituent element.

The correction cancellation instruction input means 34 inputs a correction cancellation instruction. The automatic correction information memory 36 stores the position and analysis information of the portion replaced with the correction candidate by the automatic correction means 16. The correction canceling means 35 is provided in the automatic correction information memory 3
6, the automatically corrected portion of the corrected sentence is returned to the original word. The correction cancellation instruction input means 34 is a keyboard,
A mouse or the like is used. As the correction canceling means 35, a CPU of a computer or the like is used.

The automatic correction information memory 36 is an IC memory,
A magnetic disk device, a magnetic tape device, etc. are used.
These means are for making it possible to revert to the original word at a portion where the result of the automatic correction is inappropriate.

For example, the input sentence “My hometown was completely burned in the war.” As a result of the automatic correction means 16, the corrected sentence memory 17 reads “My hometown was completely burned in the war. It has been remembered as a sentence. At that time, the automatic correction information memory 36 stores that “complete wall” is the original notation before correction for the automatically corrected ninth position. It is possible to display the automatically corrected portion with a mark to that effect and to return to the uncorrected word when an instruction is given to the portion by the correction cancellation instruction input means 34. In the text correction device of the fourth embodiment, first, the constituent elements corresponding to the second embodiment are operated, and the parts that meet the conditions of the automatic correction are replaced and posted, and then the third embodiment is described. It can be used to operate the corresponding components to perform sequential correction of incorrect words at locations that were not automatically corrected. Further, as described above, it is also possible to perform an operation to restore the place where automatic correction was not appropriate.

[0060]

As described above, according to the sentence correction device of the present invention, first, uncertain correction is avoided at the time of automatically correcting collectively the portions recognized as inappropriate words by grammar analysis. As a result, the number of operations to restore the automatically corrected portion later is greatly reduced compared to the conventional one, and the efficiency of the correction work is improved. Furthermore, when making sequential corrections,
Since the display is performed in consideration of the ambiguity in the grammatical analysis, it is possible to realize the sentence correction that is not overlooked and is easy for the user to understand.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of a text correction device of the present invention.

FIG. 2 is a block diagram showing a second embodiment of the text correction device of the present invention.

FIG. 3 is a block diagram showing a third embodiment of the text correction device of the present invention.

FIG. 4 is a block diagram showing a fourth embodiment of the text correction device of the present invention.

FIG. 5 is a block diagram showing an embodiment of a conventional text correction device.

FIG. 6 is a diagram showing an example of contents of an input sentence memory 2.

FIG. 7 is a diagram showing an example of contents of a word dictionary memory 4 with proofreading information.

FIG. 8 is a diagram showing an example of contents of a working memory 11.

9 is a diagram showing an example of contents of an analysis result memory 19. FIG.

FIG. 10 is a diagram showing an example of contents of a determination result memory 13.

11 is a diagram showing an example of contents of a correction condition memory 19. FIG.

12 is a diagram showing an example of the contents of a correction target location memory 15. FIG.

FIG. 13 is a diagram showing an example of the contents of a corrected text memory 20.

FIG. 14 is a diagram showing an example of contents of an ambiguous portion information memory 33.

FIG. 15 is a diagram showing an example of the contents of a correction candidate information memory 18.

16 is a diagram showing an example of contents of an automatic correction information memory 36. FIG.

FIG. 17 is a flowchart showing an example of implementation of grammar analysis means 3.

FIG. 18 is a flowchart showing an implementation example of the ambiguity determination unit 12.

FIG. 19 is a flowchart showing an example of implementation of the correction target portion determination means 14.

FIG. 20 is a flowchart showing an implementation example of the automatic correction means 16.

FIG. 21 is a flow chart showing an example of an ambiguous point determination means 31.

FIG. 22 is a diagram showing a first display example of a portion where there is ambiguity in the recognition of an inappropriate word;

FIG. 23 is a diagram showing a second display example of a portion where there is ambiguity in the recognition of an inappropriate word.

[Explanation of symbols]

 1 sentence input means 2 input sentence memory 3 grammar analysis means 4 word dictionary memory with proofreading information 5 analysis result memory 6 sequential correction means 7 candidate selection instruction input means 11 working memory 12 ambiguity determination means 13 determination result memory 14 correction target location determination Means 15 Correction target location memory 16 Automatic correction means 17 Corrected text memory 18 Correction candidate information memory 19 Correction condition memory 20 Text output means 21 Display control means 25 Correction condition input means 31 Ambiguous location determination means 32 Ambiguous resolution instruction input means 33 Ambiguous Location information memory 34 Correction cancellation instruction input means 35 Correction cancellation means 36 Automatic correction information memory

Claims (2)

[Claims] 1. An input sentence memory for storing an inputted sentence, a word dictionary memory with proofreading information for storing a word dictionary having proofreading information such as an incorrect word type and correction candidate including an incorrect word, Grammar analysis means for recognizing an incorrect word by performing grammatical analysis of the input sentence with reference to a proofreading word dictionary memory, and an analysis result memory for storing the result of the grammar analysis of the input sentence, A working memory for temporarily storing an intermediate result of the grammar analysis, an ambiguity determining unit for determining the ambiguity in the grammar analysis of the input sentence with reference to the working memory, and a determination result of the ambiguity determining unit A determination result memory that stores a condition, a correction condition memory that stores a condition that should be satisfied by an automatic correction target portion, the analysis result memory and the determination result memory, and the correction condition Correction target portion determination means for detecting a portion corresponding to the condition of the memory, correction target portion memory for storing the detected portion of the correction target portion determination means, and the input sentence with reference to the dictionary memory with calibration information And an automatic correction means for automatically correcting a portion stored in the correction target portion memory. 2. An input sentence memory for storing an inputted sentence, a word dictionary memory with proofreading information for storing a word dictionary having proofreading information such as types of incorrect words and correction candidates including incorrect words, Grammar analysis means for recognizing an incorrect word by performing grammatical analysis of the input sentence with reference to a proofreading word dictionary memory, and an analysis result memory for storing the result of the grammar analysis of the input sentence, A working memory for temporarily storing an intermediate result of the grammar analysis, an ambiguity determining unit for determining the ambiguity in the grammar analysis of the input sentence with reference to the working memory, and a determination result of the ambiguity determining unit A decision result memory for storing a decision result memory, an ambiguity resolution instruction input means for inputting an instruction for resolving an ambiguity relating to recognition of an incorrect word, and the decision result memory for inputting the ambiguity. An ambiguity point determining means for detecting an ambiguity in the recognition of an incorrect word in a sentence and eliminating the ambiguity of the ambiguity by the instruction from the ambiguity resolution instruction inputting means, and an instruction for selecting a correction candidate for the incorrectness input The candidate selection instruction inputting means, the determination of the ambiguous portion determining means, and the analysis result memory are referred to detect an inappropriate portion in the input sentence, and a correction candidate for the incorrect portion is detected, and the correction information-related word dictionary A sentence correction device comprising: a sequential correction unit that replaces the inappropriate portion with a correction candidate that is retrieved from a memory and that is designated by the candidate selection designation input unit.