JPH11167576A - Document proofreading device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect an error out of a text by collating extracted fact data with each record contained in a fact data base and correcting the detected mismatched data and the expression included in the corresponding text. SOLUTION: A data extraction part 1 analyzes the descriptions concerning the fact data included in a text and extracts the description in a form that can be registered in a fact data base 4 to send it to a consistency verification part 2. The part 2 retrieves the data concerning the fact equal to the extracted fact data out of the data base 4 to check the presence or absence of the points conflicting with each other between the retrieved fact data and the fact data extracted from the text. When the data conflicting with each other are detected at the part 2, an error processing part 3 corrects the extracted fact data based on the data stored in the data base 4 to secure matching between both fact data. The data base 4 stores the fact data which show the facts.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document proofreading apparatus, and more particularly, to a method of detecting or correcting an error of a text document by checking whether the description of the fact in the text document is accurate or not with fact data in an existing database. About what to do.

[0002]

2. Description of the Related Art As a conventional technique for pointing out an error in a text of a newspaper article or the like, for example, a character string constituting a text is cut out into a group of words and collated with a dictionary. To be precise, "Hussein"
If you write an unregistered word "Fusain" where you should write, it is necessary to extract this erroneous part and point out that it is presented as it is. Some types of errors, such as fluctuations in katakana notation that occur when writing foreign words in katakana, are limited to some extent, and there are some that handle the correction up to the correction.

[0003]

These devices are basically non-word errors of the type that fail in morphological analysis, that is, they are not recognized as words as a result of erroneous morphological analysis, It was not possible to point out that the description in the text was inconsistent with the fact.

Most of the work actually performed by the review section of a newspaper company is to judge the consistency of data, such as numbers and names, with common sense and various kinds of knowledge. Could not yet handle the detection of errors as facts that account for a large portion of the errors.

Accordingly, it is an object of the present invention to provide a document proofreading apparatus capable of detecting such a fact error in a text.

[0006]

FIG. 1 shows the configuration of the present invention. In FIG. 1, 1 is a data extraction unit, 2 is a consistency verification unit, 3 is an error processing unit, and 4 is a fact data database.

The above object of the present invention can be achieved by the following inventions described in the claims. (1) In the document proofreading apparatus according to claim 1, a fact database storing data relating to a specific matter, a data extracting unit for extracting fact data from input text, and a fact database for extracting the extracted fact data The system is characterized by comprising a consistency verification unit that checks each record in the data and detects the inconsistency, and an error processing unit that corrects the inconsistent data and the corresponding expression in the text.

(2) In the document proofreading device according to the second aspect, the consistency verification unit according to the first aspect evaluates the possibility of an error for each field in the data in the fact database, The cost of changing field values based on the likelihood of error when the data extracted from the text does not exactly match the data in the fact database and there is more than one possible data in the fact database Is evaluated, the change corresponding to the data in the fact database at the lowest cost is selected, and the content of the error is determined.

(3) In the document proofreading device according to the third aspect, the data extraction unit extracts data relating to a change in fact, and the consistency verification unit checks consistency with the state before the change. When the corresponding data is searched and the data is verified for consistency, the document proofreading device that corrects the corresponding data to the state after the change, when handling the fact data including the due date, the data related to the fact change Is extracted from the text, the consistency verification unit verifies whether the state before the change can exist at the time of writing the extracted data, and the error processing unit further sets the end date of the old fact, It is characterized in that the date of occurrence is set.

(4) In the document proofreading device according to the fourth aspect, each fact data extracted from the text is checked for consistency with existing data in the fact database, and for data having no problem, In a document proofreading device that checks the consistency between fact data described in texts by sequentially registering them in a fact database, when targeting a text corpus, the texts are classified once, and the text in each classification is used as the source. In addition, a unique fact database is constructed for each classification, and the consistency check is performed by the consistency verification unit in the fact database.

(5) In the document proofreading apparatus according to the fifth aspect, each fact data extracted from the text is checked for consistency with existing data in the fact database, and if there is no problem, data is sequentially checked. In a document proofreading device that checks the consistency between fact data described in a text by registering it in a fact database, a consistency verification unit does not process the entire text at once,
It is characterized in that a part corresponding to a specific document structure is extracted with reference to the document structure of the text, and the consistency is determined therein.

The present invention has the following effects. (1) Since fact data is extracted from the input text, the extracted fact data is referred to each record in the fact database, the inconsistency is detected, and the expression of the inconsistent data in the text is corrected. Incorrect and inconsistent data different from the facts in the original text can be accurately displayed and corrected.

(2) When inconsistency between the extracted data extracted from the text and the data in the fact database is detected, the reliability of the two is evaluated, and the one with the lower reliability is determined as an error. Calibration can be performed.

(3) When the fact data is registered in the fact database, the frequency of occurrence of the fact data and the reliability of the information source are checked, and the inconsistency of the data is checked based on the reliability. Since the erroneous data is determined by determining the characteristics, accurate calibration can be performed.

(4) For the data in the fact database, the possibility of error is evaluated for each field, and if the data in the text does not match the data in the fact database,
Furthermore, when there is a plurality of data that may correspond in the fact database, the cost of changing the field value is evaluated based on the possibility of the error, and the change corresponding to the data in the fact database at the lowest cost is selected. Since the content of the error is determined, accurate calibration can be performed.

(5) When the text data relating to the fact change is extracted, the consistency is checked in the state before the change, and if the consistency is obtained, the corresponding data in the fact database is changed to the state after the change. So we can update the facts database to be accurate,
Accurate calibration can be performed.

(6) Since the occurrence date, end date, etc. of each fact are described in the fact database, when the fact is extracted from the text, it is possible to accurately and easily judge the correctness of the date and time of the fact data. it can.

(7) When there is no data in the fact database whose key coincides with the fact data extracted from the text, including the time, and a key other than the time of the extracted fact data exists in the fact database, Since the fact data in the text is corrected so as to represent the temporal context, accurate proofreading can be performed.

(8) For fact data that regularly changes depending on the due date, such as age, the values on the due date of the data are calculated and matched based on the data in the fact database, so such data is accurately calibrated. can do.

(9) Even when the names do not match exactly, such as abbreviations or nicknames, rules for determining the consistency of names are determined, so that the names can be calibrated as matching. (10) For example, if the same record is extracted from the text in the order of Secretary A, Secretary A → Secretary A, Secretary A, in a newspaper or the like, the first record is written without omitting it, and it is omitted sequentially. The second expression of Secretary A can be proofread with Secretary A of State.

(11) Even if a key is specified, a value is not uniquely determined. If a plurality of values exist, it is checked whether or not each value matches. If no matching value exists, it is determined that there is no match. Therefore, accurate calibration can be performed.

(12) Since the fact database is constructed on the basis of each fact data extracted from the reference text group in advance, the fact database can be made accurate and accurate proofreading can be performed. .

(13) Each fact data extracted from the text is checked for consistency with the existing data in the fact database, and data having no problem is sequentially registered, so that the fact data can be stored in the symmetric text set. In this way, it is possible to detect inconsistencies.

(14) Since a plurality of fact databases are provided for each type, and information on the type of the fact database can be extracted from the text to select a fact database to be referred to, accurate calibration can be performed.

(15) Each field has error type information that is likely to occur, and the cost associated with the change is evaluated depending on whether the content of the change in the field value corresponds to the type of error described. Information can be extracted for accurate calibration.

(16) Since the tendency of an error that has actually occurred is analyzed and the type of error that is likely to occur is evaluated, an error can be detected optimally for each text, and accurate proofreading can be performed. Can be.

(17) having a table of factual data changes that occur dependent on an event, to check the consistency of other data in the facts database when a particular event occurs; Accurate calibration can be performed.

(18) For data including a date, when a change in fact is extracted from the text, it is checked whether the state before the change can exist at the time of describing the extracted data,
Since the end date of the old fact and the date of the occurrence of the new fact are set in the fact database, accurate calibration can be performed.

(19) When there is fact data whose end date or date of occurrence is unknown, the reliability of the fact data on the designated date is determined based on the frequency of data fluctuation, the difference between the designated date and the occurrence or end date. As a result of the evaluation, the validity of the data can be accurately determined.

(20) Texts are once classified by a predetermined method, a unique fact database is constructed for each classification based on the text in each classification, the consistency is checked, and data without problems are sequentially registered. Therefore, it is possible to construct a fact data database for each classification, to perform a detailed check, and to perform accurate calibration.

(21) With reference to the tag indicating the classification added to the document structure of the text, a portion corresponding to a specific document structure, for example, a social aspect or a sports aspect, is extracted to determine the consistency, Since no problematic data is sequentially registered, the fact data database of a specific classification can be quickly constructed, the contents thereof can be enhanced and detailed checks can be performed, and accurate calibration can be performed.

[0032]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (1) First Embodiment of the Present Invention A first embodiment of the present invention will be described with reference to FIGS. In FIG. 1, 1 is a data extraction unit, 2 is a consistency verification unit, 3 is an error processing unit, 4 is a fact data database, and FIG. 2 is an explanatory diagram of its operation.

The data extraction unit 1 analyzes the description of the fact data in the text, extracts the description in a form that can be registered in the fact data database 4, and sends it to the consistency verification unit 2. The consistency verification unit 2 searches the fact data database 4 for data relating to the same fact as the extracted fact data, and determines whether there is any contradiction between the searched fact data and the extracted data extracted from the text. Is to check.

When the inconsistency data is detected by the consistency verification unit 2, the error processing unit 3 corrects the extracted data extracted from the text based on the data of the fact data database 4 in order to match the two. Things. The fact data database 4 stores fact data indicating many facts.

The operation of the embodiment shown in FIG. 1 will be described with reference to FIG. When the original text “King Hussein of Iraq starts an attack on Kurdish guerrillas” is input to the data extraction unit 1 shown in FIG. 1 as shown in FIG. As shown in the figure, "Iraq" is extracted as the organization name, "King" as the title, and "Hussein" as the individual name.
Refer to the fact data database 4. Since the fact data database 4 stores "Iraqi President Hussein" and "U.S. President Clinton" as the organization name, post and personal name, respectively,
As a result of the reference in the consistency verification unit 2, "King Hussein of Iraq" and "President Hussein of Iraq" are detected as inconsistent data as the organization name, post and personal name, and the extracted data "King Hussein of Iraq" is incorrect It is displayed on the error processing unit 3.

FIG. 3 shows a detailed view of FIG. 3, the same reference numerals as in FIG. 1 denote the same parts, and the error processing unit 3 includes an error part display unit 3-1 and a processing unit 3-2 as shown in FIG. The extracted data extracted from the text is compared with the fact data database, and the extracted data contradictory to the fact database is determined as an error based on the fact database, and the extracted data is determined by the error part display unit 3-1 to correspond to the expressions in the text. Fact data database 4
Is displayed and presented to the user. If the user confirms this and inputs a correction confirmation operation signal from, for example, a keyboard or the like, the processing unit 3-2 corrects the erroneous portion of the text with the fact data.

However, when the data corresponding to the extracted data from the text is not held in the fact data database 4, only the extracted data is displayed on the error portion display section 3-1. For example, when an update confirmation operation signal is input from a keyboard or the like, the data updating unit 5 newly registers the extracted data held in the consistency verification unit 2 in the fact data database 4.

(2) Second Embodiment of the Present Invention A second embodiment of the present invention will be described with reference to FIGS. FIG. 4 is a diagram showing a second embodiment of the present invention, and FIG. In the figures, the same symbols as those in the other figures indicate the same parts,
Reference numeral 6 denotes a reliability evaluation unit, 7 denotes an error part determination unit, 8 denotes a data update unit, and 9 denotes an error part display unit.

The fact data database 4 stores, in addition to the fact data, the reliability of the fact, for example, 0.9 for an M newspaper, 0.5 for an N newspaper, or the reliability of the number of occurrences of the same fact. The data indicating the gender is entered.

When the consistency verification unit 2 finds mutually inconsistent data from the extracted data extracted from the text and the fact data database 4, the reliability evaluation unit 6 determines the extracted data extracted from the text and the corresponding data. The reliability is calculated based on the data indicating the reliability from the fact data database 4 referred to as described above, a quantitative evaluation result is added, and the result is transmitted to the error part determination unit 7.

The error part judging section 7 compares the evaluation result added to the extracted data with the evaluation result added to the data obtained by referring to the fact data database 4 based on the extracted data to determine the error part. Either is certified.

The data updating unit 8 receives the extracted data transmitted when the erroneous part determination unit 7 determines that the erroneous part exists in the data from the fact data database 4, and based on the received extracted data, Modify the data.

Conversely, if it is determined that the extracted data extracted from the text is incorrect, the extracted data is sent to the erroneous portion display section 9 and presented to the user. At this time, the data referred from the fact data database 4 can be simultaneously displayed and corrected.

For example, as shown in FIG. 5, in the extracted data extracted from the text, "A securities"
When “President” is entered as the post, “C” as the personal name, “Y newspaper” as the information source, and “1” as the number of occurrences, the consistency verification unit 2 shown in FIG. Reference is made to the database 4. Then, as shown in FIG. 5, as the reference data, the organization name “A securities”, the position “President”, the personal name “B”, the information source “X newspaper”, the number of occurrences so far “2”, and the reliability “0” .7 "and the reliability" 0.1 "of the Y newspaper. Here, the reliability “0.7” is registered in advance in the fact data database 4 as the reliability of the article of the X newspaper, and the reliability “0.1” is registered as the reliability of the article of the Y newspaper. Is registered.

As a result, the data shown in FIG. 5 is output from the consistency verification unit 2 to the reliability evaluation unit 6 as the one with the added reliability and the like, and the reliability and the like from the fact data database 4 as shown in FIG. Are also output as the reliability evaluation unit 6 as reference data to which is added. The reliability evaluation unit 6 evaluates the reliability of both by multiplying the number of occurrences by the reliability, and sends it to the error part determination unit 7. The error part determination unit 7 determines the error part by referring to the result of the multiplication, for example, to determine the one having a larger numerical value as having higher reliability.

As a result of the quantitative evaluation of the reliability in the form of the reliability and the probability of occurrence, if the data in the fact data database 4 is determined to be erroneous, the erroneous portion determination unit 7 converts the extracted data into data. Send to update section 8. The data updating unit 8 corrects the data in the fact data database 4 based on this. Conversely, when the extracted data is determined to be incorrect,
The erroneous part determination unit 7 sends the extracted data to the erroneous part display unit 9 and displays this to the user. At this time, the correct data referred from the fact data database 4 is also displayed, and calibration is performed based on this.

In the example shown in FIG. 5, "President of A Securities
Since the reliability of “B” is 2 × 0.7 and the reliability of “A Securities President C” is 1 × 0.1, the error data display unit 9 displays the fact data database 4 with the extracted data. Is displayed to the user, and calibration is performed based on the reference data.

Therefore, in this embodiment, when the fact data is registered in the fact data database 4, background data such as the frequency of occurrence of the fact data, the information source, and the reliability of the information source are registered. First, for example, the reliability of each newspaper is registered, and the fact data is sequentially registered in the fact data database 4 while referring to this.

Another operation explanatory diagram (part 2) of the second embodiment of the present invention will be described with reference to FIG. The possibility of an error is evaluated for each field in the fact data database 4 and a numerical value corresponding to this is added. When the data extracted from the text does not completely match the reference data in the fact data database 4 and there is a plurality of data that may correspond, the cost of changing the field value is evaluated based on the possibility of error. Then, at the lowest cost, a change corresponding to the data in the fact database is selected and determined as an error content.

For example, as shown in FIG. 6, when the field of country name and the field of position are compared, the error probability of a position that is easily erroneous is set to, for example, “2”, and the probability of a country name that is not easily erroneous is set to “3”, which is larger. Keep it. That is, those that are easy to make mistakes are entered in low numerical values.

As shown in FIG. 6, when there is an original text “Jordan President Hussein visits Egypt”, the data extraction unit (omitted in FIG. 4), as shown in FIG. "President" as post
Is extracted as the personal name, and the consistency verification unit 2 shown in FIG. 4 refers to the fact data database 4 based on this.

Then, as possible matches in the fact data database 4, the country name, title and personal name are extracted as "King Hussein of Jordan" and "President Hussein of Iraq", respectively.

At this time, as shown in FIG. 6, since the country name is “3” and the position is “2” as the error probability, the consistency verification unit 2 changes the data corresponding to the data in the fact data database 4 at the lowest cost. (Ie change of title is "2", change of country is "3"), select change of title,
By selecting "Jordan's President Hussein" and "Jordan's King Hussein" as shown in the figure below, the exact data is selected.

(3) Third Embodiment of the Present Invention A third embodiment of the present invention will be described with reference to FIGS. FIG. 7 is for extracting data relating to a change in fact such as, for example, resignation of the Prime Minister, checking consistency with the state before the change, and correcting the corresponding data to the state after the change.

In FIG. 7, 2 is a consistency verification unit, 4 is a fact data database, 8 is a data update unit, 9 is an error part display unit, and 10 is a fact change data extraction unit. The fact change data extracting unit 10 extracts, from the input text, data on a fact change, such as death or resignation, from the input text.

Now, as shown in FIG. 8, for example, when the text "Resign of the Prime Minister of the United Kingdom" is input to the fact change data extraction unit 10, the fact change data extraction unit 10 extracts the fact data and The fact change data is extracted from the fact data using the “resignation” as a key.

By the way, as shown in FIG.
This means that as a prerequisite, the target person must be in the job, and the record of the job must be deleted with “resignation”.

For this reason, as shown in FIG. 8, it is necessary to change the data in the fact data database. In other words, as shown in the fact data database, the country name, title, and personal name are "UK Prime Minister Major", and as shown in the table below, "Major" is deleted.
The data is corrected.

For this reason, the fact change data extraction unit 1
For example, when data relating to a change in fact, such as “Resignation of the British Prime Minister” is extracted, the fact that the data is changed is sent to the consistency verification unit 2.

The consistency verifying unit 2 checks the consistency with the data in the fact data database 4 by referring to the fact data database 4 based on the data before the fact change, “The Prime Minister of Great Britain”. Then, the changed data of the fact that passed the consistency check is transmitted to the data updating unit 8, the personal name of the data “French Primer Major” in the fact data database 4 is deleted, and corrected to the data as shown in FIG. You.

However, rejected data, which was not consistent in the above-described consistency check based on the data before the fact was changed, is sent to the error portion display unit 9 as error data, and is displayed to the user and corrected. Becomes

(4) Fourth Embodiment of the Present Invention A fourth embodiment of the present invention will be described with reference to FIGS. In FIG. 9, the description of the fact data database 4 includes three columns of the date of occurrence, the date of termination, and the date of entry. Based on this, if there is data indicating that Clinton will become President of the United States in 1994, for example, 1
This is to correct the accuracy of the description of Clinton in 992.

The consistency verification unit 12 includes a date consistency discrimination unit 12
-1 is provided to check whether or not the date-extracted extracted data extracted from the original text is accurate. For example, when fact data “US, President, Clinton, 1994, unknown” is described in the fact data database 4 shown in FIG. 9 as the country name, title, name, date of birth, end date as shown in FIG. As shown in FIG. 10, when the original text “President Clinton visits Berlin in 1992” is input to the data extraction unit 1, the data extraction unit 1, as shown in FIG. "US", title "President", name "Clinton", date of birth "199
2), "Unknown" is extracted as the end date.

The fact data is transmitted to the consistency verification unit 12, and the consistency verification unit 12 refers to the fact data database 4 by the country name “USA”, title “President”, and name “Clinton”. Then, fact data as shown in FIG. 10 is obtained.

At this time, the date matching discriminating unit 12-1 sets "1992" as the date of occurrence based on the data extracted from the text.
Is stored, and it is determined that this is a smaller number, that is, an older one as compared with the date of occurrence “1994” referred to from the fact data database 4. Therefore, since it is known that Clinton is incorrect in being the United States President in 1992, as shown in FIG.
"1992" is notified to the error processing unit 3 as an error, and this is displayed to the user together with "US President Clinton 1994", and correct calibration is performed.

(5) Fifth Embodiment of the Present Invention A fifth embodiment of the present invention will be described with reference to FIGS. In FIG. 11, when there is no data having the same key including the extracted fact data and time in the fact data database, but there is data having the key other than time coincident with the fact data in the text, the temporal data is compared with the fact data in the text. It is modified so as to express a proper context. For example, there is a text “A Securities Chairman B 1997.8”, and “A Securities Chairman B
When the data of “End date 19977.5” exists, the text is corrected to “A Securities former chairman B”.

In FIG. 11, the fact data database 4 is provided with items of the occurrence date and the end date. For example, as shown in FIG. 12, in the fact data database 4, the name of the organization, the title, the name, the date of occurrence, and the date of termination are "A
Securities Chairman B Termination Date 19977.5 "is entered.

Now, as shown in FIG. 12, when the text “A president B of A Securities has received information from the prosecutor's office” as shown in FIG. 12, is input to the data extraction unit 1 of FIG. 1 extracts, as fact data, A securities as an organization name, a chairman as a title, B as a name, and 1997.8 as a description date, as shown in FIG.

As a result, the consistency verification unit 12 checks the fact data database 4, and as shown in FIG. 12, the organization name is "A", the title is "Chairman", the name is "B", and the end date is "1997". .5 ".

The date matching discriminating unit 12-1
And recognize that there is already data on the end date for President A Securities B. In addition, A securities B
Although there is extracted data related to the president, the date described in the extracted data from the text does not match the above-mentioned end date, and the date described is later than the end date.

As described above, when there is no other data that can match the extracted data, the consistency verification unit 12 ignores the date specification such as the description date and searches the fact data database 4 for matching data. . Then, as shown in FIG. 12, the matching data is referred to as "A Securities Chairman B".

At this time, the date matching discrimination unit 12-1
Description date 199 with end date 19977.5 in FIG.
7.8 is checked to recognize that 19977.8 is later, and this is notified to the error processing unit 3 together with the coincidence data. As a result, "A Chairman of the Securities B" is first displayed on the error portion display section 3-1. Thereafter, the "Chairman" is corrected by the error processing section 3 to be "Former Chairman", and as shown in FIG. The original text description is modified to indicate that it is not current.

(6) Sixth Embodiment of the Present Invention A sixth embodiment of the present invention will be described with reference to FIG. 13 and FIG. In the sixth embodiment, for fact data that changes regularly, such as an age, the value on the date of description of the extracted data is calculated based on the data in the fact data database to determine consistency. It is.

In FIG. 13, the consistency verification unit 13 is provided with a date calculation discriminating unit 13-1 for calculating a value at a specific date with respect to data that changes with the date, such as age. The algorithm is described.

In FIG. 13, as shown in FIG. 14, in the fact data database 4, the name of the organization, title, age, age, and date of description are “ABC Chairman EF 50 years old 1
994 "is entered.

Now, as shown in FIG. 14, the data extraction unit 1 shown in FIG. 13 reads “1997.6.1 ABC EF Chairman (5.
When the text “0 years old” is input, the data extraction unit 1 sets the name of the organization as AB as shown in FIG.
C, President as title, EF as name, 5 as age
0, 1997 is extracted as fact data as the description date.

Based on this fact data, the consistency verification unit 13 refers to the fact data database 4, and shows ABC as the organization name, chairman as the title, EF as the name, EF as the age, and 50 as the date shown in FIG. 1994 is read. This is sent to the date calculation discriminating unit 13-1.

The date calculation discriminator 13-1 receives the description date 1997 separately transmitted from the data extractor 1, and calculates 50+ (1997-1994) based on the above 1994, 50 and 1997. The age 53 is calculated. This 53 is the processing unit 3-2 of the error processing unit 3.
Is transmitted to

At this time, in the error processing unit 3, a part of the text including this age, "ABC EF chairman (50 years old)" is displayed in the error part display unit 3-1. Amended by Part 3-2, "EF Chairman of ABC (53
Years old) will be proofread.

If the person extracted from the text has, for example, data on the age and the fact data database also describes the age of the same person, the data in the fact data database Calculate the age on the date described in the extracted data and check the consistency. In the example of FIG. 14, the value does not match the calculated value, and is corrected to a value calculated based on the value in the fact data database.

(7) Seventh Embodiment of the Present Invention A seventh embodiment of the present invention will be described with reference to FIGS. In the seventh embodiment, for example, “Ryutaro Hashimoto” may be represented by an abbreviation of “Ryutaro Hashi”, but even in such a case, “Ryutaro Hashimoto” in the text is determined to be correct. .

In FIG. 15, reference numeral 14 denotes an abbreviated name table in which abbreviated names and standard names of personal names are registered in advance. In the example of FIG. 15, “Ryuhashi ← → Ryutaro Hashimoto”, “Clinton ← → Bill Clinton” and the like are registered. When registering in the fact data database 4, it is once converted into a standard expression and registered. For example, register “Japanese Prime Minister Hashimoto Ryu” as “Japanese Prime Minister Ryutaro Hashimoto”.

The consistency verification unit 15 has an abbreviation check unit 15
When the consistency verification unit 15 searches the extracted data for each field value of the fact data database 4, if the names do not match, the abbreviation table 14 is accessed using this name. For example, in the above search, if the terms "Japan" and "Prime Minister" match, but the terms do not match, the abbreviation table 14 is accessed using the name "Hashiryu" of the extracted data and "Hashiryu" By obtaining "Ryutaro Hashimoto" as the standard name for all, the agreement is obtained.

Now, the data extraction unit 1 shown in FIG.
When the original text “Japanese Prime Minister Hashi Ryu” as shown in FIG. 6 is input, the data extraction unit 1 extracts Japan as the country name, Prime Minister as the title, and Ryu Hashi as the personal name as fact data as shown in FIG. I do.

Based on the fact data, the consistency verification unit 15 refers to the fact data database 4 and obtains reference data of Japan as the country name, Prime Minister as the title, and Ryutaro Hashimoto as the personal name, as shown in FIG. However, this reference data matches in the country and title fields but not in the name.

In this case, the abbreviation checking unit 15-1 accesses the abbreviation table 14 using the mismatched name “Hashiryu” and reads out the standard name “Hashimoto Ryutaro” corresponding to “Hashiryu”. Then, based on this, the matching is checked by the consistency verification unit 13 again comparing with the reference data, so that the accuracy of the original text is recognized.

In this way, the rules for determining the matching of names can be determined, and conditions that can be matched even if they do not match exactly, such as abbreviations, can be determined. (8) Eighth Embodiment of the Present Invention An eighth embodiment of the present invention will be described with reference to FIGS. For example, in the description of a newspaper article, in the case of an explanation of the same fact, the entirety is first written without omitting the title, etc., and then abbreviated gradually in some cases.
The embodiment of the present invention corresponds to such a case.

In the eighth embodiment, as shown in FIG. 17, the same facts are expressed in the order of appearance.
6 is provided in the consistency verification unit 17.
6 and an appearance order list creation check unit 1 for searching the appearance order list 16 and checking its omission state.
7-1 is provided.

Now, as shown in FIG. 18, when the original text "Secretary of the United States Regent ... Secretary of the Regent ... Secretary of the Regents ..." is input to the data extraction unit 1, the data extraction unit 1 sequentially extracts "Regan Secretary of State,""Regan," and "Regan Secretary of State," and sequentially sends them to the appearance order list creation check section 17-1.

The appearance order list creation check section 17-1
Based on the extracted fact data, the same facts regarding Regan are listed in the order of appearance 1 as shown in FIG.
Create 6.

After the appearance order list 16 has been created, the appearance order list creation check unit 17-1 checks whether or not the description state conforms to the abbreviation. First, as shown in FIG.
From No. 1 and No. 1. Compare the two expressions. Thereby, No. No. 1 “Regan US Secretary of State” No. 1
Since it is determined that the expression of “Secretary of Reagan” is omitted in No. 2, No. The expression of 1 is accepted.

Next, from the appearance order list 16, 2 and No. 2. Compare expressions 3 Thereby, No. No. 2 from "Regan Secretary" For example, it is determined that the “Secretary of State for Reagan” of No. 3 is long and not omitted, and as shown in FIG. The expression of 2 is judged as reject.

Of course, no. 2, No. If the expression of 3 is the same as the previous expression, it is also accepted. As described above, in general, the title is omitted after the title is specified in detail, so that a case where the back expression is omitted from the front expression or the same case is judged as pass. This makes it possible to check the consistency of the elements in the list based on the rules for describing the order of appearance of the same data and restrictions on elements such as adjacent titles.

(9) Ninth Embodiment of the Present Invention A ninth embodiment of the present invention will be described with reference to FIGS. For example, in the case where a plurality of persons exist in the same title as in the company's office, even if the company name and the title are specified, there are a plurality of persons, so that a single value cannot be determined for the key item. For this reason, since a plurality of records exist in the database, it is necessary to obtain consistency with all of them.

For example, in the fact data database 4, for a record in which the name of the organization is “Company A” and the title is “Management”, the names of all the members are entered in FIG. Register as shown in 4-0.

Further, the consistency verification unit 18 is provided with the same key item all check unit 18-1. Even if a key is designated as a specific title such as a company name and a managing director, a unique value is not determined, When there are a plurality of values, the same key item all check unit 18-1 checks the consistency of all records registered for the same key, and determines that there is no matching when there is no matching value.

For example, when the names “AB” and “CD” are registered in the fact data database 4 shown in FIG. 19 as the title “Management” of the organization name “Company A”, as shown in FIG. When the original text “XY managing director of Company A” as shown in FIG. 20 is input to the data extracting unit 1 shown in FIG. 19, the data extracting unit 1 obtains “Company A” as the organization name and the title as shown in FIG. And "XY" as the name are extracted as fact data.

As a result, the same key item all check section 18
-1 refers to the same key item section 4-0 with “Company A” and “Management” as key items, checks all of a plurality of registered records that match these key items, and
Check whether there is a match with “company”, “management”, and “XY”.

As a result, there is no problem if there is at least one match, but in the case shown in FIG. 20, since there is no match, it is determined that the check result is not consistent as shown in FIG. The result is displayed on the error processing unit 3.

(10) Tenth Embodiment of the Present Invention A tenth embodiment of the present invention will be described with reference to FIG. The tenth embodiment relates to a method for constructing a fact data database 4 used in the document proofreading apparatus of the present invention.

The data extracting unit 1 sends the extracted fact data to the consistency verifying unit 2 or sends it to the data updating unit 19. Which one is selected is controlled by the update signal. For example, when the update signal is “1”, the fact data extracted from the reference text is sent to the data update unit 19,
When the update signal is “0”, fact data extracted from the check target text is sent to the consistency verification unit 2.

The reference text is for constructing the fact data database 4, and the description content is composed of a text group of accurate content which has been strictly checked in advance.

The data updating unit 19 registers the fact data extracted by the data extracting unit 1 based on the reference text in the fact data database 4. In FIG.
First, a reference text is input to the data extraction unit 1. At this time, the update signal is set to, for example, “1”. As a result, the data extracting unit 1 sends the fact data extracted from the reference text to the data updating unit 19. Then, the data updating unit 19 sequentially updates the fact data database based on the fact data, and constructs the fact data database.

After the fact data database is corrected in this way, the update signal is set to “0” and the text to be checked is input to the data extraction unit 1. Data extraction unit 1
The fact data extracted by the verification unit 2
As a result, an error detection process is performed while referring to the fact data database 4 updated earlier.

As described above, the fact data database 4 is updated at any time with the reference text to make it accurate, so that the text to be checked can be accurately corrected. (11) Eleventh Embodiment of the Present Invention An eleventh embodiment of the present invention will be described with reference to FIG.
In the eleventh embodiment, the consistency with already registered data is checked, and if there is no inconsistency, the data is registered.

For example, texts such as “Company A's Managing AB, Company A's Managing CD, Company A's Managing EF ...” are input, and the data extracting unit 1 changes the organization name, title, and name to “Company A AB ”,“ Company A, Managing C ”
D ”,“ Company A, Managing EF ”... Are extracted and sequentially sent to the consistency verification unit 20.

Thus, the consistency verifying section 20 first refers to the fact data database 4 by the name of the organization, the title, and the name “Company A, Managing AB”. As a result, when the data “President AB of Company A” is referred to from the fact data database 4, the contradiction detecting unit 20-1 checks this, and it is contradictory that AB of Company A has different titles of the managing director and the president. Therefore, it is determined that there is no consistency, and this is displayed on the error part display unit 3-1.

However, when the fact data database 4 is referred to by “Company A managing CD”, “Company A managing X”
Even if the data "Y" is referred to, the inconsistency detection unit 20-1
Checks this and judges that there is no contradiction even if there are a plurality of persons with different names in the management, and sends the "Company A's management CD" to the data update unit 5 and updates the fact data database 4 Update by Similarly, "Company A
The data of "management EF" is also sent to the data update unit 5, and updates the fact data database 4. Therefore, when the fact data "Company A Director EF" is subsequently extracted from the text extraction unit 1 and the data "Company A Managing EF" is referred to from the fact data database 4 by the consistency verification unit 20, the contradiction detection unit 20 -1 is determined to be inconsistent between the extracted fact data and the referenced data, and this is displayed on the error display section 3-1.

In this way, each fact data extracted from the text is checked for consistency with the existing data in the fact data database 4, and if there is no problem, the data is sequentially registered. Can be checked for consistency between fact data described in the above, and it is possible to extract inconsistent portions between fact data in the text.

(12) Twelfth Embodiment of the Present Invention A twelfth embodiment of the present invention will be described with reference to FIG.
In the twelfth embodiment, the fact data database 4 is divided into a plurality of types, for example, databases 4-1 4-2, 4-3 for each of the political field, the economic field, and the sports field.
.. Are divided and a database corresponding to the text type is referred to.

The data detecting section 21 includes a type detecting section 21.
-1 is provided, and the type of the text is definitely determined based on the extracted fact data. For example, when the word "Prime Minister" is detected, the text type is determined to be "politics". When the word "baseball" is detected, the text type is determined to be "sports", and the database is identified accordingly. D
A B identification signal is output.

The reference DB control unit 22 converts the data 4-1 to 4-2, 4-3,... Corresponding to the DB identification signal from the fact data database 4 in response to the DB identification signal.
・ Select one of the following.

In FIG. 23, for example, when a text “Yokozuna Wakanohana has won” is input to the data extraction unit 21, the data extraction unit 21 outputs “Yokozuna Wakanohana Win”.
Is extracted, and the type detection unit 21-1
Thus, the type of the text is identified as “sports” from the word “Yokozuna”, and a sports DB identification signal corresponding to the type is output to the reference DB control unit 22.

Thus, the reference DB control unit 22 performs control so as to select the sports database 4-3. The consistency verification unit 2 executes the sports database 4-3.
The connection is controlled so that can be referred to.

In this state, the consistency verifying unit 2 reads “Yokozuna Wakanohana victory” by referring to the sports database 4-3 based on the fact data “Yokozuna Wakanohana victory”, judges that there is no consistency, and makes an error. This is displayed on the partial display unit 3-1.

In this case, since the fact data database 4 is configured for each type, it is possible to quickly and accurately verify whether or not there is consistency. (13) Thirteenth Embodiment of the Present Invention A thirteenth embodiment of the present invention will be described with reference to FIGS. In the thirteenth embodiment, a portion which is likely to be erroneous is predicted in advance and an accurate check is performed. In lesser-known regions, such as the Middle East, confusion between neighboring countries and names can occur. For this reason, for Jordan in the Middle and Near East, which is not very familiar in Japan, describe nearby Lebanon and Iraq as easy-to-mix country names, and set the error probability in the confusion possibility table, such as numerical value 1, as low as possible. Fill in numerical values.

Similarly, immediately after the title of a person who has been promoted or resigned has been changed, there is a high possibility that an old post will be written for that person because he does not have the latest data. For this reason, it is highly likely that Mr. Blair, the newly appointed British Prime Minister, will be called in the post of Labor Party leader, so that the error probability is set to a low value in the confusion possibility table like 1.5, for example. Please fill in.

As shown in FIG. 24, a cost changing unit 23-1 is provided in the consistency verifying unit 23, and when a plurality of reference data exist, a unit having a low error probability is obtained. As described above, the confusion possibility table 24 predicts a portion that is likely to be erroneous in advance, and displays the error probability with a numerical value that is smaller as the error is more likely. The confusion possibility table 24 has information on the type of error for each field.
FIG. 24 shows small and medium-sized countries that are unfamiliar to the confusion possibility table 24-1 and those that are easily confused with neighboring countries in a less-known region. This indicates that there is a high possibility that the current title and the previous title are easily confused by the promotion or resignation.

For example, as shown in FIG. 25, the original text “President Hussein of Jordan visits Egypt” is input to the data extraction unit 1 shown in FIG. 24, and as shown in FIG. 25, the country name is “Jordan” and the post is “President” and “Hussein” as personal names are extracted as extraction data. Based on this, the consistency verification unit 2 shown in FIG.
3 refers to the fact data database 4.

In the fact data database 4, there is a possibility that the country name, the title, and the personal name may be "King Hussein of Jordan" and "Iraq," respectively.
President Hussein is extracted.

At this time, as shown in FIG. 25, the country name “3” and the post “2” are also extracted as error probabilities. The consistency verification unit 23 recognizes that the confusion possibility table (country name) 24-1 should be referred to from “Jordan” existing in the extracted data, and the cost change unit 23-1 recognizes the confusion possibility table 24-1. From −1, Jordan is judged to be Iraq or Lebanon and the error probability is numerical value “1”, and the lowest numerical value is judged to be easy to error. Then, in this case, as shown in FIG. 25, if the text "Jordan" is corrected to "Iraq", it is recognized that the data matches the data in the fact data database 4.

(14) Fourteenth Embodiment of the Present Invention A fourteenth embodiment of the present invention will be described with reference to FIG.
In the fourteenth embodiment, first, the result of error detection / correction using a training text is recorded, the tendency of an actually occurring error is analyzed, and the type of error that is likely to occur is evaluated. Is changed so as to perform optimal error detection / correction for each text.

In the present invention, first, the training text is input to the data extracting unit 1, the fact data is extracted by the data extracting unit 1, transmitted to the consistency verifying unit 26, the fact data database 4 is referred to, That is, the provisional error portion is detected by the provisional error detection unit 26-1. At this time, provisional errors for the entire training text are detected, and as a result, a set of provisional errors is obtained.

This set of provisional errors is sent to the error tendency analysis unit 26.
With -2, the tendency of the error exists is analyzed. As a result, for example, it is determined whether an error easily occurs in a country name or an error easily occurs in a personal name.

When this tendency is transmitted, the parameter adjusting unit 26-3 reduces the parameter, for example, the numerical value of the error probability shown in FIG. Improve ability.

Thereafter, the error detection unit 26-4 repeats the error detection once again, and checks whether or not the error detection is correctly performed on the error-prone portion as described above. In this way, a highly accurate error detection result is obtained.

The above description is for training texts. For ordinary texts, the fact data extracted by the data extraction unit 1 is transmitted to the error detection unit 26-4.
The fact data database 4 is referred to based on the adjustment result.

(15) Fifteenth Embodiment of the Present Invention A fifteenth embodiment of the present invention will be described with reference to FIGS. For example, when a company goes bankrupt, an officer of the company resigns, so it is necessary to delete the company officer from the fact data database. In the fifteenth embodiment, a table is provided for fact data changes that occur in response to a certain event, and when a particular event occurs, other data in the fact data database 4 is updated accordingly. , To maintain data consistency.

As shown in FIG. 27, the dependent event table 27
Is prepared, and the contents of the change of the fact data that occurs following a certain event are shown. For example, it indicates that officers will be deleted for the event of bankruptcy. As an example of an event, as shown in FIG. 28, an officer loses his / her position with the event of bankruptcy, so it is necessary to delete officer data. FIG. 28
As shown in the above, all positions held by the person in connection with the event of the assassination of the dignitary will be dismissed, and it will be necessary to correct the data of the person again.

In FIG. 27, a text is input to the data extraction unit 1. The data extraction unit 1 performs a process of extracting fact data, and this fact data is transmitted to the consistency verification unit 26. When the consistency verification unit 26 detects in this fact data that bankruptcy, assassination, or a search item in the dependent event table 27 indicates, for example, bankruptcy of Company A, the dependent event search unit 26- 1 is made to search the dependent event table 27 to search for items to be performed depending on the event of bankruptcy. Then, they recognize that the officer has been deleted.

As a result, the consistency verification section 26 sends the changed data indicating that the company A officer has been deleted to the data update section 5. On the basis of this, the data update unit 5 sends the fact data database 4
Delete all the data concerning the officers of Company A. In this manner, the data in the fact data database 4 can be processed in accordance with the event.

(16) Sixteenth Embodiment of the Present Invention A sixteenth embodiment of the present invention will be described with reference to FIGS. 29 and 30. For example, if the US President is newly elected,
Based on this, the end date of the old president and the new president, the date of new appointment, etc. are set. In the sixteenth embodiment, after extracting data relating to the fact change from the text, the state before the change is the extracted data. It verifies whether it can exist at the time of description, and sets the end date of the old fact and the date of occurrence of the new fact.

For this purpose, as shown in FIG. 29, the consistency verification section 28 is provided with a pre-change state detection section 28-1 for verifying whether a state before change can exist at the time of describing the extracted data. An update data creation unit 2 for setting an end date of an old fact and an occurrence date of a new fact in the error processing unit 29.
9-1 is provided.

Now, as shown in FIG. 30, in the fact data database 4, the country name, title, name, date of birth, and date of end are respectively "US, President, Bush, 199".
When the data “0, unknown” is registered, the text “President Clinton is first elected in the 1992 presidential election” is input to the data extraction unit 1, as shown in FIG. And fact data extracted from this,
The fact data “Clinton, President, 1992, presidential election, first election” is sent to the integrity verification unit 28.

By the way, at this time, only the fact that Bush became US President in 1990 is described in the fact data database 4. At this time, from the fact data extracted from the text, it is determined that Clinton was first elected president in 1992 by "Clinton, 1992, presidential election, first election". Before Clinton can become President for the first time in 1992, another person must be President before that. However, the pre-change state detector 28-1 of the integrity verification unit 28 reads the “US President Bush 1990 in FIG. The end date is unknown, ”confirms that Bush is the former president.

On the other hand, Bush's term will end in 1992 due to the inauguration of Clinton's presidency, which is recognized by the error processing unit 29. Error processing unit 2
9 recognizes that new data needs to be added to the fact data database 4 based on Clinton's inauguration.

The update data creating unit 29-of the error processing unit 29
1, the update data as shown in FIG. 30 is created,
The updated data is sent to the data updating unit 5 and registered in the fact data database 4, and the data as shown in FIG. 30 is registered.

(17) Seventeenth Embodiment of the Present Invention A seventeenth embodiment of the present invention will be described with reference to FIGS. FIG. 31 is a seventeenth embodiment of the present invention,
FIG. 32 is an explanatory diagram of the operation. When the consistency verification unit searches the fact data database and there is fact data whose end date or date of occurrence is unknown, the reliability at the data designated date is determined by the frequency of data fluctuation, the designated date and the date of occurrence or end date. The evaluation is based on the difference between the fact data and the consistency check is performed only on fact data having reliability within a certain threshold on a specific date.

For example, if the president changes frequently in a short period of time in a politically unstable country, the threshold value is set to, for example, two years, and those having a difference of two years from the designated date are not checked.

FIG. 32 shows an example of how to evaluate the reliability of data for which only incomplete dates are specified in a country where the political situation is stable. FIG.
In the second example, the fact data database 4 only describes that Clinton is the United States President in 1997, as shown in FIG.

Now, as shown in FIG. 32, the text “President US Bush visits Berlin in 1991” is input to the data extraction unit 1 shown in FIG. , The name and the date of occurrence are respectively extracted as “USA, President, Bush, 1991”. Actually, the extracted fact data includes a visit to Berlin, but this part is not relevant to the date specification and is omitted.

In the fact data of President Clinton in the fact data database 4 shown in FIG. 32, neither the date of birth nor the date of termination is specified, so the possibility that Clinton was President in 1991 cannot be completely denied. As such, data confirming the status of the President is,
In the case of the year, the probability that the same person is in the same position six years ago is quantitatively evaluated.

The term of office of the US President is a maximum of two terms (1 term 4
Year), the possibility is very small. Therefore the threshold value Th ₁ and 6 years, when it exceeds than the threshold value, in this case, even without the checked facts data the extracted, can conflict with each other and data Jisshu is 32 It can be assumed that it is very low as shown in FIG.

In the above case, when the fact data extracted from the data extracting unit 1 by the text shown in FIG. 32 and referred to by the consistency verification unit 30 to the fact data database 4, the existence of the data shown in FIG. Detect. Threshold limit determination unit 30-1 this time, stops to check it recognizes that the data of the text with a difference more than 6 years of threshold Th _1, and outputs the text as having been aligned.

[0145] Further, the threshold value can be appropriately determined by its symmetry, for example, defined matters and the threshold value Th ₀ example 2 years about President in political unstable countries, data with a difference of more than 2 years without checking , Can pass the integrity verification.

(18) Eighteenth Embodiment of the Present Invention The eighteenth embodiment of the present invention will be described with reference to FIG. In the eighteenth embodiment of the present invention, a text is once classified for each specific classification, a unique database is constructed for each classification based on the text in this classification, and the consistency is checked for each classification. It was made.

As shown in FIG. 33, the eighteenth embodiment includes a fact data database 4, a data extraction unit 32, a data update unit 33, a consistency verification unit 34, and the like. The fact data database 4 is composed of classification databases 4-1 4-2,... 4-n classified into, for example, the political field, the economic field, the sports field, and the like. The category-specific database 4-1 stores data classified into the political field, and the category-specific database 4-2 stores data classified into the economic field. The classification-specific database 4-n stores data classified into the sports field.

The text 31 is a text corpus in which texts of a plurality of classification types such as a political field, an economic field, a sports field, etc. exist. A text corpus is an aggregate of a plurality of texts, and need not be composed of a plurality of types.

The data extracting unit 32 extracts fact data from text, and has a partial text extracting unit 32-1 for classifying fact data extracted from text into a plurality of predetermined classifications. .

The data updating section 33 stores the fact data for each category transmitted from the data extracting section 32 in a database for each category of the same category as the category and constructs a unique database for each category.

The consistency verifying unit 34 checks the consistency of the fact data extracted from the text 31 with the data stored in the fact data database 4.
Based on the classification by the partial text extraction unit 32-1, a classification checking unit 34 checks the consistency with the data stored in the classification database of the same classification as the classification.
-1.

For example, a text corpus consisting of one page of a one-month newspaper article is input to the data extraction unit 32 as the text 31. Thereby, the data extraction unit 32
From this, fact data is extracted, and the partial text extracting unit 32-1 determines in advance that the fact data is classified into, for example, the political field or the economic field using the extracted word as a key. Divide into categories. This is sent to the data update unit 33 and the consistency verification unit 34.

In the consistency verification section 34, the classification-based checking section 3
According to 4-1, the consistency is checked by referring to the classification-specific database corresponding to the divided classification of the fact data. At this time, a specific classification database already referred to, for example, a classification database 4 in the political field.
If no inconsistency is found by referring to No. 1 and the same is not stored, the data updating unit 33 is notified to store the same. Thereby, the data updating unit 33 stores the fact data in the classification-specific database of the classification.

In this way, a unique database can be constructed for each classification. Therefore, an accurate database for each classification can be constructed, and correct calibration can be performed.

(19) Nineteenth Embodiment of the Present Invention A nineteenth embodiment of the present invention will be described with reference to FIG. In the nineteenth embodiment of the present invention, instead of processing the entire text at once, a specific type of text is referred to by referring to the document structure of the text, that is, the classification indicated by the tag added to the document title or body. The part corresponding to the document structure is extracted, the consistency is determined in the extracted parts, and those having consistent consistency are sequentially stored in a database of the type, that is, the classification.

As shown in FIG. 34, the nineteenth embodiment includes a fact data database 4, a data update unit 33, a data extraction unit 35, a consistency verification unit 34, and the like. The fact data database 4, data update unit 33, consistency verification unit 34, and the like are the same as those shown in FIG.

The data extracting section 35 is for extracting fact data from the text.
And extracts the part of the fact data of the specific classification designated in advance by referring to the tag of the document structure of the text added to indicate the classification type. For example, when the classification is designated as a political field, only fact data of a text related to politics is extracted.

Now, for the specific classification extracting unit 35-1,
For example, a designation of a political field is designated as a classification designation, and a text is input to the data extraction unit 35. Thereby, the specific classification extraction unit 35-1 extracts fact data from only the text in the political field by referring to the tag of the document structure of the text, adds this classification designation item, and adds it to the data update unit 3
3 and the consistency verification unit 34.

The consistency verification unit 34 checks the consistency by referring to the classification-specific database in the fact data database 4 corresponding to the classification. At this time, there is no inconsistency with the data stored in the classification database,
Moreover, if there is not the same data, the data update unit 33 is notified to store the fact data. As a result, the data updating unit 33 stores the fact data in the classification-specific database of the classification.

As described above, a database unique to a classification specified by a classification can be quickly constructed, and can be enhanced when the content of the database for a specific classification is insufficient.

[0161]

The present invention has the following effects. (1) Since fact data is extracted from the input text, the extracted fact data is referred to each record in the fact database, the inconsistency is detected, and the expression of the inconsistent data in the text is corrected. Incorrect and inconsistent data different from the facts in the original text can be accurately displayed and corrected.

(2) For the data in the fact database, the possibility of error was evaluated for each field, and the data in the text did not match the data in the fact database.
Furthermore, when there is a plurality of data that may correspond in the fact database, the cost of changing the field value is evaluated based on the possibility of the error, and the change corresponding to the data in the fact database at the lowest cost is selected. Since the content of the error is determined, accurate calibration can be performed.

(3) For data including a due date, when a fact change is extracted from the text, it is checked whether or not the state before the change can exist at the time of describing the extracted data.
Since the end date of the old fact and the date of the occurrence of the new fact are set in the fact database, accurate calibration can be performed.

(4) Texts are once classified by a predetermined method, a unique fact database is constructed for each classification based on the text in each classification, the consistency is checked, and data without problems are sequentially registered. Therefore, it is possible to build a fact database for each classification, and it is possible to perform detailed checks,
Accurate calibration can be performed.

(5) With reference to the tag indicating the classification added to the document structure of the text, a portion corresponding to a specific document structure, for example, a social aspect or a sports aspect, is extracted to determine consistency. Since data having no problem is sequentially registered, the fact data database of a specific classification can be quickly constructed, the contents thereof can be enhanced, a detailed check can be performed, and accurate calibration can be performed.

[Brief description of the drawings]

FIG. 1 is an embodiment of the present invention.

FIG. 2 is an operation explanatory diagram of one embodiment of the present invention.

FIG. 3 is a detailed view of the first embodiment of the present invention.

FIG. 4 is a second embodiment of the present invention.

FIG. 5 is an operation explanatory view (1) of the second embodiment of the present invention.

FIG. 6 is an operation explanatory diagram (part 2) of the second embodiment of the present invention.

FIG. 7 is a third embodiment of the present invention.

FIG. 8 is an operation explanatory diagram of the third embodiment of the present invention.

FIG. 9 is a fourth embodiment of the present invention.

FIG. 10 is an operation explanatory diagram of the fourth embodiment of the present invention.

FIG. 11 is a fifth embodiment of the present invention.

FIG. 12 is an explanatory diagram of an operation according to the fifth embodiment of the present invention.

FIG. 13 is a sixth embodiment of the present invention.

FIG. 14 is an explanatory diagram of the operation of the sixth embodiment of the present invention.

FIG. 15 is a seventh embodiment of the present invention.

FIG. 16 is an operation explanatory diagram of the seventh embodiment of the present invention.

FIG. 17 shows an eighth embodiment of the present invention.

FIG. 18 is an operation explanatory view of the eighth embodiment of the present invention.

FIG. 19 shows a ninth embodiment of the present invention.

FIG. 20 is an operation explanatory diagram of the ninth embodiment of the present invention.

FIG. 21 is a tenth embodiment of the present invention.

FIG. 22 shows an eleventh embodiment of the present invention.

FIG. 23 shows a twelfth embodiment of the present invention.

FIG. 24 shows a thirteenth embodiment of the present invention.

FIG. 25 is an operation explanatory view of the thirteenth embodiment of the present invention.

FIG. 26 shows a fourteenth embodiment of the present invention.

FIG. 27 is a fifteenth embodiment of the present invention.

FIG. 28 is an operation explanatory diagram of the fifteenth embodiment of the present invention.

FIG. 29 shows a sixteenth embodiment of the present invention.

FIG. 30 is an operation explanatory diagram of the sixteenth embodiment of the present invention.

FIG. 31 shows a seventeenth embodiment of the present invention.

FIG. 32 is an operation explanatory diagram of the seventeenth embodiment of the present invention.

FIG. 33 shows an eighteenth embodiment of the present invention.

FIG. 34 shows a nineteenth embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Data extraction part 2 Consistency verification part 3 Error processing part 3-1 Error part display part 3-2 Processing part 4 Actual data database 5 Data update part 6 Reliability evaluation part 7 Error part judgment part

Claims (5)

[Claims] 1. A fact database storing data relating to a specific matter, a data extracting unit for extracting fact data from input text, and collating the extracted fact data with each record in the fact database. A document proofreading device comprising: a consistency verification unit that detects a match; and an error processing unit that corrects mismatched data and a corresponding expression in a text. 2. The consistency verification unit evaluates the possibility of an error for each field in the data in the fact database, and makes the data extracted from the text completely different from the data in the fact database. When there are multiple pieces of data that do not match and may correspond in the database, the cost of changing the field value is evaluated based on the possibility of error, and the corresponding change in the data in the database at the lowest cost is determined. 2. The document proofreading apparatus according to claim 1, wherein the content of the error is determined by selection. 3. The data extraction unit extracts data relating to a change in fact, and the consistency verification unit checks the consistency with respect to the state before the change, and the corresponding data is searched for consistency. Regarding the verified data, in the document proofreading device that corrects the corresponding data to the state after the change, when handling the fact data including the due date, after extracting the data related to the fact change from the text, the consistency verification unit Document proofreading characterized by verifying whether the state before the change can exist at the time of describing the extracted data, and further setting the end date of the old fact and the date of occurrence of the new fact in the error processing unit. apparatus. 4. Each fact data extracted from a text is
In a document proofreading device that checks the consistency with existing data in the fact database and checks the consistency between fact data described in the text by sequentially registering data having no problem in the fact database, When targeting a text corpus, it is necessary to classify the text once, build a unique fact database for each classification based on the text in each classification, and check the consistency in the consistency verification unit in it. Characteristic proofreading device. 5. Each fact data extracted from a text is
In a document proofreading device that checks consistency with existing data in the fact database and sequentially registers data with no problem in the fact database to check the consistency between fact data described in the text, The character verification unit does not process the entire text at once, but extracts the part corresponding to a specific document structure by referring to the document structure of the text, and judges consistency in the extracted part. Document proofreading device.