JPH10207889A - Document proofing device - Google Patents

Abstract

(57) [Summary] To provide a document proofreading apparatus that reduces excessive indication of correct candidates for an erroneous part (a part that may have an error). A morphological analysis unit converts a text into a word string. The error part detector 200 detects an error part from the word string. The correct answer candidate developing unit 300 generates a correct answer candidate group corresponding to the error part. Correct answer candidate verification unit 4
00 selects a correct answer candidate from the correct answer candidate group and outputs a verified correct answer candidate group including the selected correct answer candidates. Correct answer candidate verifying section 400 uses code 4
10 to 440. The correct answer probability giving unit 410 gives the occurrence probability to each correct candidate with reference to the word occurrence probability database 440. The error probability calculation unit 420 calculates the error probability of the correct answer candidate based on the occurrence probability. The error candidate selection unit 430 narrows down the correct candidates based on the error probability of each correct candidate.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention reduces the work of a user to proofread a document with respect to document data input by a user or obtained as an electronic medium in a text processing apparatus, and greatly improves the efficiency of document proofreading. The present invention relates to a document proofreading device that can be improved.

[0002]

2. Description of the Related Art As a conventional error indication technique, a morphological analysis is performed and an unregistered word part in a result is indicated. Points out words with homonyms. And the like.

In the case of a function for pointing out an unregistered word, if there is a word with an incorrect spelling, it is highly likely that the word will be an unregistered word. .
Similarly, a word location where a homonym exists is pointed out as a location where an operation error is likely to occur during kana-kanji conversion. The user makes the right or wrong judgment for each case in the user.

As another means, when a specific word string is detected after morphological analysis, it is recognized as an error and pointed out. For example, a part-of-speech sequence called noun + verb is checked, or an error occurs when there is a one-character kanji word. In addition, katakana / kanji character strings are arranged in phonetic order,
In some cases, a word with a slight spelling of the same word is located next to the spelling to make it easier to detect the sway of the spelling.

Further, after a new error candidate is detected,
There is also a system that creates a plurality of hypotheses that estimate the content of an error, tests the hypotheses by means such as matching with a compound word or the like, and presents only the likely hypotheses that have survived.

[0006]

The function of pointing out an unregistered word or a word having a homonymous word cannot be determined to be incorrect.
It points out where errors may exist.
However, regarding the indication of an unregistered word, the cause of the unregistered word may be the existence of an original unregistered word that a proper noun or the like does not exist in the dictionary other than the spelling error. Regarding the indication of words with homonyms,
Just because there are many mistakes does not necessarily mean that there are mistakes. For this reason, in the above method, all the points pointed out are not true errors (there are many over points)
That is the most problematic.

In the method of detecting an error using a specific part-of-speech sequence, the target of the error to be handled is very limited.
The problem is that many errors in the text are not detected.
In addition, the method of sorting and showing katakana words and kanji words to the user has a problem that the work to be performed by the user is large and the efficiency of the proofreading work is not improved much.

Further, in a method of generating hypotheses and leaving only probable ones by a test, it is important to give a correct evaluation to each generated hypothesis.
In this case, there is no problem that the original unregistered word is simply pointed out that it is not listed in the dictionary, but there is a problem of fluctuation in evaluation. For example, when the word corresponding to the original notation in the text is not in the dictionary, the relative evaluation is lower than other hypotheses, and there is a possibility that the word may be indicated even when the target portion is correct.

In a general document proofreading support system, in order to increase the accuracy of pointing out errors, it is necessary to narrow down the types of errors to be targeted. A lot of indications (excessive indications) for parts that are not original errors are mixed. In order to cope with this, a part that generates many original spelling candidates in consideration of the possibility of typographical errors existing in the input text (correct candidate expansion) and matching it with the contents of the dictionary The applicant of the present invention has already proposed a document proofreading support system in which the part to be verified (correct word search) is made independent, but the verification ability is weak, and there are still many points of excess. The present invention has been made in view of these points, and uses statistical data and dictionary information to provide an error probability of a correct answer candidate generated at the time of developing the correct answer candidate (correct correct candidate It is an object of the present invention to provide a document proofreading apparatus for obtaining a partial word or a word string).

[0010]

According to a first aspect of the present invention, there is provided a document proofreading apparatus for converting a text to be input into a word string, and a morphological analysis unit for extracting a possible error portion from the word string obtained as a result of the morphological analysis. An error part detection unit to be extracted, a correct candidate expansion unit for generating a correct candidate for the error-possible part extracted by the error part extraction unit, and one or more obtained as a result of expansion of the correct candidate expansion unit A document proofreading apparatus comprising a correct answer candidate verifying unit that verifies each correct candidate and narrows down to only correct correct candidates, wherein the correct candidate verifying unit includes a database regarding the occurrence probability of a word or a word string. Referring to the database, an occurrence probability assigning unit that outputs an occurrence probability of a word or a word string required to calculate an error probability of a correct answer candidate, and an output that is output from the occurrence probability assigning unit. Based on the occurrence probability of a word or word string, a test of the probability of each correct answer candidate is performed, and an error probability calculation unit that gives an error probability to each correct candidate, and an error probability calculation unit assigns each correct candidate to each correct candidate. An error candidate selecting unit for selecting a correct answer candidate having a predetermined threshold or more with reference to the error probability.

According to a second aspect of the present invention, there is provided the document proofreading apparatus according to the first aspect, wherein the error probability calculation unit calculates an error based on a ratio between an occurrence probability of an error-possible portion existing in the text and an occurrence probability of a correct candidate. It is characterized by calculating a probability.

According to a third aspect of the present invention, there is provided the document proofreading apparatus according to the first aspect, wherein the error probability calculation unit includes an occurrence probability that each correct candidate word independently occurs and an occurrence probability as a word string in a context in the text. The error probability is calculated for each correct candidate with reference to the ratio to the probability.

According to a fourth aspect of the present invention, in the document proofreading apparatus of the first aspect, the error probability calculation unit calculates a co-occurrence probability that each correct candidate co-occurs with a word group to be tested. The error probability is calculated based on the similarity between the obtained co-occurrence probability pattern and the co-occurrence probability pattern in which the error-probable part in the text co-occurs with the word group to be tested. is there.

According to a fifth aspect of the present invention, there is provided the document proofreading apparatus according to the first aspect, wherein the occurrence probability assigning unit includes a word dictionary with priority information in the expanded group having priority information in the expanded group; A relative occurrence probability calculation unit that calculates an occurrence probability for the word or the word string based on the priority information in the group of the word dictionaries corresponding to the input word or word string. It is.

According to a sixth aspect of the present invention, there is provided the document proofreading apparatus according to the first, second, third, fourth, or fifth aspect, wherein the correct answer candidate developing unit includes a read word dictionary and a read word dictionary. A reading extraction unit that extracts the reading of the word in the error-possible part by referring to the attached word dictionary, and other words that have the same reading as the reading of the word extracted by the reading extraction unit are extracted from the reading word dictionary. And a homophone extraction unit that outputs the extracted word as a correct answer candidate.

According to a seventh aspect of the present invention, there is provided the document proofreading apparatus according to the first, second, third, fourth, or fifth aspect, wherein the correct candidate expanding section corresponds to an error notation. An expansion database that describes a plurality of expansion data having correct candidates and constraints, and an expansion unit that expands the error-possible portion into a correct answer candidate using the expansion data in the expansion database that matches the error-possible portion; A condition checking unit that checks whether the correct answer candidate output from the developing unit satisfies the constraint condition for the correct answer candidate and leaves only the correct answer candidate that matches the constraint condition. .

According to an eighth aspect of the present invention, in the document proofreading apparatus according to any one of the first, second, third, fourth, and fifth aspects, the correct candidate developing unit includes a plurality of Japanese input means. Select a reference development database based on a plurality of development databases corresponding to each of the plurality of development databases for developing possible error portions into correct answers and information for specifying a Japanese input unit when the text is created. With reference to the reference destination control unit and the expanded database of the selected reference destination,
And a development processing unit that develops the error-possible part into a correct answer candidate.

According to a ninth aspect of the present invention, there is provided a document proofreading apparatus which has a plurality of expanded data, first corrects a part of a text to be proofed with each expanded data, and as a result, develops expanded data having the highest evaluation value. It is characterized in that proofreading of the entire text is performed using the proofreading.

According to the document proofreading devices of the first to seventh aspects, it is possible to eliminate the possibility that excessively pointing out correct candidates. Further, according to the document proofreading devices of the eighth and ninth aspects, it is possible to use expanded data suitable for the text to be proofread.

[0020]

FIG. 1 is a diagram showing a configuration example of a document proofreading apparatus according to the present invention. In the figure, 100 is a morphological analysis unit, 200 is an error detection unit, 300 is a correct answer candidate developing unit, 400 is a correct answer candidate verification unit, 410 is an occurrence probability assignment unit, 420 is an error probability calculation unit, and 430 is an error candidate selection unit. Each part is shown.

FIG. 1A is a diagram showing an outline of a document proofreading apparatus according to the present invention. The morphological analysis unit 100 decomposes the input text into word strings, and converts the obtained word strings into error part detection units 20.
Pass to 0. The erroneous part detection unit 200 detects an erroneous part (a part that may have an error) from the received word string, and passes the erroneous part to the correct answer candidate developing unit 300. The correct answer candidate developing unit 300 estimates the type of error and generates a correct word or word string candidate (correct answer candidate) corresponding to the error part. The correct candidate verification unit 400 verifies each correct candidate,
A correct answer candidate with a high degree of correctness is selected. Note that the document proofreading apparatus of the present invention is actually realized by a computer and software.

FIG. 1B is a diagram showing an example of the configuration of the correct answer candidate verifying unit. The correct answer candidate verification unit 400 includes the occurrence probability assignment unit 4
10, error probability calculation section 420, error candidate selection section 430,
It has a word occurrence probability database 440. The occurrence probability giving unit 410 refers to the database 440 (word occurrence probability database) relating to the occurrence probabilities of a single word or a word string, and calculates a word or a word string (correct candidate or Erroneous words, etc.)
Output the probability of occurrence of What is the probability of occurrence of a word or word string?
When a word or word string is arbitrarily selected in a text or corpus (sentence collection), it means the probability that it is a specified word or word string. The word occurrence probability database is a database that stores multiple pairs of words or word strings and occurrence probabilities, such as word occurrence probability safety 0.001 guarantee 0.002 guarantee 0.001 walking 0.0005 arc 0.001 It is.

The error probability calculation unit 420 calculates the error probability of the correct answer candidate based on the occurrence probability of the word or word string output from the occurrence probability giving unit 410. The error probability means a probability that a word or word string of a correct answer candidate is mistakenly changed to a word or word string of an error part. The error candidate selection unit 430 selects a candidate suitable for the correct candidate from the correct candidate group output from the correct candidate expanding unit 300 based on the error probability passed from the error probability calculation unit 420.

FIG. 2 is a diagram for explaining a first example of error probability calculation in the error probability calculation section. In the illustrated example, the original text is "Mr. Saito Matsumoto's efforts bear fruit". It is assumed that “Matsumoto” and “Saito” have been detected as error portions by the error detection unit 200. The correct answer candidate developing unit 300 estimates a homonym error, generates a correct answer candidate “Matsumoto” corresponding to the error part “Matsumoto”, and generates a correct answer candidate “Saito” corresponding to the error part “Saito”. I do.
The occurrence probability assigning unit 410 stores the word occurrence probability database 4
With reference to 40, the probability of occurrence in the same sound group = 0.1 is assigned to the erroneous part “Matsumoto”, and the correct answer candidate “Matsumoto”
And the probability of occurrence in the same sound group = 0.2 for the error part "Saito", and the correct answer candidate "Saito" for the correct part. The probability of occurrence in the same sound group = 0.2 is given.

The error probability calculation unit 420 calculates the error probability of the correct answer candidate by using the following formula: error probability = 0.01 × error destination occurrence probability / error source occurrence probability, for example. When the occurrence probability of the error part “Matsumoto” = 0.1 and the occurrence probability of the correct answer candidate “Matsumoto” = 0.02 are substituted into the equation (1), the error probability of “Matsumoto” becomes 0.5. Similarly, substituting the probability of occurrence of the erroneous part “Saito” = 0.2 and the probability of occurrence of the correct answer candidate “Saito” = 0.2 into the above equation gives the error probability of “Saito” = 0.1.

FIG. 3 is a diagram for explaining a second example of the error probability calculation in the error probability calculator. In the illustrated example, the original text is “topic on security”. It is assumed that “guarantee” is detected by the error detection unit 200 as an error part. Correct answer candidate expansion unit 300
Predicts a homonym error and generates correct answer candidates “guarantee” and “compensation” corresponding to the error part “guarantee”. The occurrence probability assigning unit 410 refers to the word occurrence probability database 440, assigns an occurrence probability of 0.2 in the same sound group to the error part “guarantee”, and assigns the same sound group to the correct answer candidate “guarantee”. Within the same sound group for the correct candidate "compensation" =
0.1 is given. Further, the occurrence probability giving unit 410 generates the occurrence probability = 0.0 for the word string “security” in the context.
02, and the probability of occurrence = 0.0
01, and the probability of occurrence = 0.0 for “safety compensation”
01 is assigned.

The error probability calculation unit 420 calculates the error probability of the correct answer candidate according to the following formula: error probability of the correct answer = occurrence probability in context / single occurrence probability. (2)
In the formula, “guarantee”, “guarantee”, “compensation”, “security”,
When the occurrence probabilities of “security guarantee” and “security compensation” are substituted, the error probability of “guarantee” = 0.02 / 0.1 = 0.2 The error probability of “guarantee” = 0.001 / 0.2 = 0 0.005 Error probability of “compensation” = 0.001 / 0.1 = 0.01 The error candidate selection unit 430 outputs “guarantee” having the largest error probability as a verified correct answer candidate.

FIG. 4 is a diagram for explaining a third example of the error probability calculation in the error probability calculator. In the illustrated example, the original text is “change clothes”. It is assumed that “change” is detected by the error detection unit 200 as an error part. The correct answer candidate developing section 300 estimates the homonymous error and generates correct answer candidates “change” and “buy” corresponding to the error part “change”.

The occurrence probability assigning unit 410 extracts the co-occurrence pattern of the error part “change” and the particles “ni” and “ga” from the word occurrence probability database 440, and outputs the correct answer candidate “change” and the particles “ni” and “ga”. ”
The co-occurrence pattern of the correct answer candidate “buy” and the particles “ni” and “ga” is extracted. In the illustrated example, the co-occurrence pattern is changed to a co-occurrence pattern.

The error probability calculator 420 compares the co-occurrence pattern of the word in the error part with the co-occurrence pattern of the correct candidate word, and calculates the error probability of the correct candidate based on the comparison result. In the illustrated example, the word “change” in the error part
Is the same as the co-occurrence pattern of the correct answer word "replace", the error probability of "replace" is increased. Further, since the co-occurrence pattern of the word “change” in the error part and the co-occurrence pattern of the correct candidate word “buy” are different, the error probability of “buy” is reduced.

FIG. 5 is a diagram showing an example of the configuration of the occurrence probability giving section according to the present invention. In the figure, reference numeral 411 denotes a relative occurrence probability calculation unit, 412 denotes an occurrence probability writing unit, and 441 denotes a word dictionary with priority information in the expanded group.

The word dictionary 441 with priority information in the expanded group is first selected when a conversion key is pressed in a group of the same sound (this is an expanded group) like a kana-kanji dictionary of a word processor. Words are arranged in order from word. For example, in a development group called “hosho”, words “guarantee”, “guarantee”, “compensation”, and “walk” are described. In this example, the probability of occurrence of “guarantee”> “guarantee”
Occurrence probability> “compensation” occurrence probability> “walking” occurrence probability. For example, the nth word and the nth word in the expansion group
Assuming that there is a 0.001 occurrence probability difference from the first word, the relative occurrence probability is known.

The relative occurrence probability calculation unit 411 receives a word (or word string) related to the correct answer candidate and the error probability of the correct answer candidate. The relative occurrence probability calculation unit 411 calculates the relative occurrence probability of the input word or word string, with reference to the word dictionary 441 with the priority information in the expanded group. The occurrence probability writing unit 412 adds a relative occurrence probability to the word or word string input to the relative occurrence probability calculation unit 411.

FIG. 6 is a diagram showing a first example of the configuration of the correct answer candidate developing section according to the present invention. In the figure, reference numeral 311 denotes a reading extraction unit, reference numeral 312 denotes a homophone extraction unit, and reference numeral 313 denotes a word dictionary with reading.

The read word notation dictionary 313 stores a plurality of pairs of words (or word strings) and readings, such as safety guarantees.

An error part is input to the reading extraction unit 311. The reading extraction unit 311 searches the word dictionary with reading 313 using the input notation of the error part as a key, and extracts the reading of the error part. The extracted reading is passed to the homophone extraction unit 312. The homophone extraction unit 312 searches the word dictionary with reading 313 using the passed pronunciation as a key, and extracts homonyms. The extracted homonyms are output as correct answer candidates.

FIG. 7 is a diagram showing a second example of the configuration of the correct candidate expanding section according to the present invention. In the figure, reference numeral 321 denotes a developing unit,
Reference numeral 322 denotes a condition inspection unit, and 323 denotes a development database.

The expansion database has a certain notation,
It is the original correct notation candidate (correct answer candidate) written assuming that it is incorrect. The development database stores development data such as → → → → づ づ → → ず ず 保証 → 保証 → → 保証 →. For example, the left side of the expanded data “Oh → Oh” corresponds to an error portion, and the right side corresponds to a correct answer candidate. The same applies to other expanded data. For example, if there is a word "difficult", a correct answer candidate "difficult" can be generated using expanded data "zu-zu".

The correct answer candidate in the expanded data has a format that can describe its own, preceding and following parts of speech, and constraints on notation. For example, if the expanded data is raw → gender (valid only when it comes to the end of a word string), it is possible to generate a correct answer candidate called “validity” corresponding to the error part “valid raw”. I can do it.

An error part is input to the developing unit 321. The expansion unit 321 generates a correct answer candidate group corresponding to the input error part with reference to the expansion database 323, and outputs the correct answer candidate group as a first correct answer candidate group. The first correct answer candidate group is input to the condition checking unit 322. The condition checking unit 322 checks the constraint added to each of the correct answer candidates belonging to the first correct answer candidate group,
Only a group of correct answer candidates meeting the constraint conditions is output as a second correct answer candidate group.

FIG. 8 is a diagram showing a third example of the configuration of the correct answer expanding section according to the present invention. In the figure, reference numeral 331 denotes an expansion processing unit, 332 denotes a reference destination control unit, and 333 to 335 denote expansion databases.

For example, OAK, ATOK, and MS-IME are known as Japanese input means.
For example, the development database 333 corresponds to OAK, the development database 334 corresponds to ATOK, and the development database 335 corresponds to MS-IME.

The reference destination control section 332 collects setting information relating to the Japanese input means from the operating system of the computer or the additional information of the document, and selects the most appropriate development database. The expansion processing unit 331 refers to the selected expansion database and generates a correct answer candidate corresponding to the input error part.

FIG. 9 is a diagram showing another example of the configuration of the document proofreading apparatus of the present invention. In the figure, reference numerals 501 to 503 denote an error correction unit, 504 denotes a correction performance comparison and evaluation unit, 505 denotes a selection unit, and 506 denotes a correction processing unit for the entire text.

Each of the error correction units 501 to 503
It has a configuration as shown in FIG. However, expanded data and constraint conditions used in each error correction unit are different from each other. A part of the text is input to the first error correction unit 501, the second error correction unit 502, and the third error correction unit 503. The correction performance comparison and evaluation unit 504
Automatically or through interaction with the user, the difference between the correction results by the error correction units is detected, and what is correct is evaluated. The selection unit 505 selects an error correction unit having the best correction performance based on the evaluation result of the correction performance comparison evaluation unit 504. Correction processing is performed on the entire text using the selected error correction unit.

[0046]

As described above, according to the present invention,
When presenting the correct answer candidate to the user or using it as a hypothesis for the next verification, instead of presenting all, only show those with high error probability or sort from high error probability to low error probability For example, it is possible to improve the correction rate and to make the calibration work performed by the user more efficient by means such as presenting in order. In addition, it is possible to always achieve optimal error relevance and reproducibility for variations in the occurrence probability due to input means or user habits by extracting data from the kana-kanji conversion dictionary and adjusting the expansion type. .

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a configuration example of a document proofreading device of the present invention.

FIG. 2 is a diagram illustrating a first example of error probability calculation in an error probability calculation unit.

FIG. 3 is a diagram illustrating a second example of error probability calculation in the error probability calculation unit.

FIG. 4 is a diagram illustrating a third example of error probability calculation in the error probability calculation unit.

FIG. 5 is a diagram illustrating a configuration example of an occurrence probability providing unit according to the present invention.

FIG. 6 is a diagram illustrating a first configuration example of a correct answer candidate developing unit according to the present invention.

FIG. 7 is a diagram illustrating a second configuration example of the correct answer candidate developing unit according to the present invention.

FIG. 8 is a diagram showing a third example of the configuration of the correct answer candidate developing unit according to the present invention.

FIG. 9 is a diagram showing another configuration example of the document composition device of the present invention.

[Explanation of symbols]

 Reference Signs List 100 Morphological analysis unit 200 Error part detection unit 300 Correct answer candidate expansion unit 311 Reading extraction unit 312 Homophone extraction unit 313 Reading word notation dictionary 321 Expansion unit 322 Condition inspection unit 323 Expansion database 331 Expansion processing unit 332 Reference control unit 333 Expansion Database 334 Expansion database 335 Expansion database 400 Correct answer candidate verification unit 410 Occurrence probability assignment unit 420 Error probability calculation unit 430 Error candidate selection unit 440 Word occurrence probability database 411 Relative occurrence probability calculation unit 412 Occurrence probability writing unit 441 Expansion group priority information Attached word dictionary 501 first error correction unit 502 second error correction unit 503 third error correction unit 504 correction performance comparison evaluation unit 505 selection unit 506 correction processing unit for the entire text

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁶ Identification code FIG06F 15/40 370J

Claims (9)

[Claims] A morphological analysis unit that converts an input text into a word string; an error part detection unit that extracts an error-possible part from a word string obtained as a result of morphological analysis; and an error part extraction unit. A correct answer candidate developing unit that generates a correct answer candidate for the extracted error-possible part, and one or more correct answer candidates obtained as a result of the expansion of the correct candidate expanding unit are verified and verified. A document proofreading apparatus comprising: a correct answer candidate verifying unit that narrows down only possible correct answer candidates, wherein the correct candidate verifying unit refers to a database relating to the occurrence probability of a word or a word string; Based on the occurrence probability of the word or word string output from the occurrence probability giving unit, and the occurrence probability giving unit that outputs the occurrence probability of the word or word string required to calculate It makes the likelihood of the assay of the correct candidate,
An error probability calculation unit that assigns an error probability to each of the correct answer candidates; and an error candidate selection unit that selects an answer candidate having a predetermined threshold or more by referring to the error probability assigned to each of the correct answer candidates by the error probability calculation unit. A document proofreading device comprising: 2. The document proofreading apparatus according to claim 1, wherein the error probability calculation unit calculates the error probability based on a ratio between an occurrence probability of an error-probable part existing in the text and an occurrence probability of a correct answer candidate. . 3. An error probability calculation unit calculates an error probability for each correct candidate by referring to a ratio between an occurrence probability of each correct candidate word independently and an occurrence probability as a word string in a context in a text. 2. The document composition device according to claim 1, wherein the calculation is performed. 4. An error probability calculation unit calculates a co-occurrence probability that each correct answer co-occurs with a word group to be tested, and a pattern of the co-occurrence probability obtained as a result of the calculation and a possibility of error in the text. 2. The document proofreading apparatus according to claim 1, wherein an error probability is calculated based on a degree of similarity with a co-occurrence probability pattern in which a part co-occurs with the test target word group. 5. An occurrence probability assigning unit, comprising: a word dictionary with priority information in an expansion group having priority information in the group to be expanded; and a word dictionary in the word dictionary corresponding to an input word or word string. 2. The document proofreading apparatus according to claim 1, further comprising: a relative occurrence probability calculation unit that calculates an occurrence probability for the word or the word string based on the priority information in the above. 6. A correct answer candidate developing unit comprising: a reading word dictionary; a reading extraction unit for referring to the reading word dictionary to extract a reading of a word in an error-prone portion; and a word extracted by the reading extraction unit. 3. A homophone extraction unit for extracting another word having the same reading as the reading of the word from the word dictionary with reading, and outputting the extracted word as a correct answer candidate. Claim 3,
The document proofreading device according to claim 4 or 5. 7. An expansion database in which a correct answer candidate expansion unit describes a plurality of expansion data having an error notation, a correct answer candidate corresponding thereto, and constraint conditions, and expansion data in the expansion database adapted to an error-possible portion. And a developing unit that expands the error-possible part into a correct candidate, and investigates whether the correct candidate output from the developing unit satisfies a constraint condition for the correct candidate, and determines whether the correct candidate matches the constraint condition. And a condition inspection unit that leaves only the condition.
Or the document proofreading device according to claim 5. 8. A correct answer candidate developing unit, comprising: a plurality of expanded databases corresponding to each of the plurality of Japanese input means for expanding an error-possible portion into a correct answer candidate; A reference control unit for selecting a reference development database based on the information specifying the input means; and a development processing unit for referencing the selected reference development database and developing the error-prone part as a correct answer candidate 6. The document proofreading device according to claim 1, wherein the document proofreading device comprises: 9. A text having a plurality of decompressed data, a part of the text to be proofread is first corrected by each decompressed data, and as a result, the decompressed data having the highest evaluation value is used for the entire text. A document proofreading device for performing proofreading.