JPH06203020A - Method an device for recognizing and generating text format - Google Patents

Abstract

PURPOSE:To generate a conversion result document having a text format being equivalent to an iput by eliminating text format information and executing various conversion processing of a machine traslation, etc., with respect to a document subjected to text format without a troublesome manual work. CONSTITUTION:With regard to the combination of each row of an input file, the degree of coincidence is calculated and page length is inferred, A range in which the degree of coincidence is a prescribed value or above at the same line interval as the number of pages and which is continued from a page boundary is recognized as a page header or a page footer. A range in which a column whose null character rate is a prescribed value or above is continued is decided to be the boundary of the step set area, the step set and the chart of each page are recognized, and a text is extracted. By the ratio of the number of specific characters, the recognition of a chart area, and the recognition of a processing unnecessary line are executed. After the conversion processing is executed, a conversion result document having a text format being equivalent to an input is generated. In such a manner, the manhour of the eliminating work of text format information, and a text format work after the translation can be reduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text format recognition generation method and apparatus for separating format information and text information from document data formatted as text and resynthesizing the text information. It is suitable for automatic translation targeting a character code data string of a document distributed by a medium and a character code data string generated by an optical character recognition device from a document printed on paper or the like, and a database construction, The present invention relates to a text format recognition generation method and apparatus suitable for use as preprocessing for performing keyword extraction and other text processing.

[0002]

2. Description of the Related Art In a document created by a so-called word processor, format (document format) information such as double space, left edge setting, and enlarged character is a text file (a file containing only character code information that does not include format information). ), It is represented as a blank line or blank. The need to analyze the converted text file and remove the blanks and blank lines due to the format information is due to the "machine translation e-mail system" (Fumito Nishino, Naoto Nakamura, IPSJ Natural Language Processing Research Group 7
5-5 (1990.1)). However, there is no report on how to implement the method, and eventually it will be removed manually.

In addition, when there are figures and formulas in the text, the figures and formulas often do not require translation,
Machine-translation processing on these often results in meaninglessness. For such untranslated parts,
By a manual process called pre-editing, a symbol indicating that no translation is required is inserted in the text.

When recognizing text of a document using an optical character recognizing device, a column editor, a chart, etc. are displayed by a dedicated editor immediately after reading an image in page units or after recognizing characters. It is necessary to manually specify the areas and the order of connection of each area.

Regarding the automatic embedding of a clean copy command for designating a title line or a paragraph in accordance with the document structure of the original sentence after the machine translation,
In the above-mentioned document "Machine translation electronic mail system",
The realization has been reported. However, the page length, the page header, the page footer, the column, and the like have to be manually specified and the document format has to be reconstructed.

[0006]

In the above conventional technique, in order to perform machine translation on text-formatted text data such as page headers, page footers, columns, and chart allocation, these page headers are used. It is necessary to manually cut out areas of pages, page footers, columns, and charts and specify the order of these areas. Therefore, there is a problem that a lot of man-hours are required to process a large amount of data.

Further, since it is also necessary for humans to manually perform pre-editing for inserting a symbol indicating a translation-unnecessary portion, a lot of man-hours are required to process a large amount of data. There is a problem.

Furthermore, a text format program for embedding a clean copy command similar to the original text in the translation result cannot always give a format equivalent to the text-formatted document of the original text to the translation result. If the equivalent format cannot be added, it is necessary to manually correct the document resulting from the processing of the text format program. Therefore, there is a problem that many man-hours are required to process a large amount of data.

In view of the problems in the above conventional example, an object of the present invention is to provide a text format recognition generating method and apparatus capable of automating each of the above-mentioned troublesome operations manually performed by humans. is there.

[0010]

To achieve the above object, a text format recognition generation method according to the present invention is provided with a page header, a page footer, a column,
Input a text-formatted text file such as chart layout, estimate the page length of the input text file, recognize the page footer and / or page header, and specify areas such as columns and charts. A step of recognizing, a step of recognizing a type indicating whether the recognized area is a text area or a figure area, a step of determining a connection order of the recognized areas, and a text spanning a plurality of areas according to the connection order. A step of extracting a chart, a step of performing a predetermined conversion on the extracted text and the chart, and a step of giving a format equivalent to the input text file to the text and the chart of the conversion result Is characterized by.

Further, the text format recognition generation device according to the present invention has means for inputting a text-formatted text file such as page header attachment, page footer attachment, column setting, chart allocation, etc., and page length of the input text file. Means for recognizing, a means for recognizing a page footer and / or a page header, a means for recognizing an area such as a column or a chart, and a means for recognizing a type indicating whether the recognized area is a text area or a chart area. A means for determining a connection order of the recognized areas, a means for extracting a text and a chart that span a plurality of areas in accordance with the connection order, a means for applying a predetermined conversion to the extracted text and a chart, Add the same format as the input text file to the converted text and figures. Characterized by comprising a means.

[0012]

The text format recognition generation method according to the present invention is applied to, for example, a machine translation system including an input device, a display device, a file storage device, a translation device, and a system device. First, a text-formatted document to be processed (here, a translation target) is input via an input device and stored in a file storage device as an input file.

The system device processes according to a page length estimation method, a page header recognition method, a page footer recognition method, a region recognition method, a region type recognition method, a region connection order determination method, a text extraction method, a translation method, and a text format generation method. And the input device, display device, file storage device, and translation device.

The page length estimation method reads some or all of the lines from the beginning of the input file, assigns a line number from the beginning to each read line, and puts a line and the end of the file from the end. Calculates the degree of coincidence with a row. The difference between the line numbers of the two lines for which the degree of coincidence is calculated is used as the row offset, and the operation of accumulating the pair of the degree of coincidence and the row offset is performed for each row for which the degree of coincidence is a certain value or more.
The accumulated result is used as the row coincidence degree calculation result. In addition, counting the frequency of row offset in the row matching degree calculation result,
When the row offset is assumed to be the page length, the page offset is estimated to be the row offset having the largest ratio of the frequency of the row offset to the number of pages calculated from the number of read rows.

In the page header recognition method, a line having a line offset equal to the page length among the line offsets in the line matching degree calculation result, and a line continuous from the start line of the page toward the end of the file is referred to as the page header. To do. Also, when there is a number in a column that does not match for each page in the line recognized as the page header, that column is recognized as a page number field.

In the page footer recognition method, a line having a line offset equal to the page length among the line offsets in the line coincidence calculation result, and a line continuous from the end line of the page toward the beginning of the file is called the page footer. To do. Also, when there is a number in a column that does not match for each page in the line recognized as the page footer, the column is recognized as a page number field.

The operation of the area recognition method is as follows.

When part or all of the input file is read from the beginning and the line length does not reach a certain column, it is assumed that there is a space character in that column of the line, and the line feed character that indicates the end of the line is a space character. Assuming that the character is a character, the number of blank characters in each column is counted, and the range where columns in which the ratio of the number of blank characters to the number of read lines is a certain value or more is assumed to be a column boundary.

Next, in the area recognition method, data forming one page is read according to the estimated page length, and the following processing is performed for each page.

If a column from one column to another column is assumed to be a column, all the characters in the range of a line are blank characters, or the length of the line up to the range. When is not reached, the region of the column is divided immediately before and after the line.

When a column between one column and another column is assumed to be one column, and a line has a character other than a space character in a column which is assumed to be a column boundary, If there is a region of columns adjacent to each other across the boundary, those regions end immediately before that row, and it is assumed that there is a region of columns with the width including the adjacent columns from that line. Continue to recognize.

The area recognition method operates as described above.

The area type recognition method is as follows.
If the ratio of the number of characters such as' + ','-',' | ', and'‖' that are frequently used to draw charts and the total number of characters excluding blank characters and line feed characters in the area is a certain value or more. It is recognized that a certain area is a figure area and the other areas are text areas.

The operation of the area connection order determination method is as follows.

If there are areas that are vertically adjacent to an area in a page in the input file, those areas are connected from top to bottom, and if there are areas that are horizontally adjacent, those areas are connected. Area from the bottommost area on the left side to the topmost area on the right side, but when determining these connections, the areas that have already been established are not reconnected and the connection is determined. Assuming that the already-existing areas are skipped and connected, one row of connection order relationships is determined for each area in the page.

Next, in the area connection order determination method, it is assumed that an area in a page in an input file is connected to an area at the end of the next page to the uppermost area on the leftmost side of the next page.

Next, immediately after the text area, for all places where the chart area is connected, the text area is skipped and connected to the next area, and for the skipped chart area, the original order is changed. The connection order of those chart areas is determined, and as a result, the connection order of the two columns is determined.

Next, when another text area is connected immediately after a certain text area, the difference in the used word frequency distribution between the two text areas is large, and the connection of the words at the boundary of the areas is not appropriate. And if there is a text area similar to the text area in the adjacent column behind the other text area, and the word distribution of the word at the boundary of the area is appropriate, that text area is Change the connection order of the areas so that they are connected immediately after the certain text area.

The operation of the area connection order determination method is as described above.

In the text extraction method, when extracting text from a series of areas, the text is extracted only from the area recognized as the text area, the texts are connected according to the connection order of the areas, and the text file is extracted. To generate. Text is extracted from the chart area for each area to generate a text file.

The translation unnecessary line recognition method is as follows.
'+', '-', 'X', '÷', '^', '±', equal sign, inequality sign, 'Σ', etc. A line with a ratio of the number of characters frequently used to draw a chart such as'‖ 'and the total number of characters excluding blank characters and newline characters in the line is a certain value or more is determined to be an untranslated line. , Add a translation unnecessary designation to that line. Further, the threshold value for this determination is changed according to the type of the area to which the row belongs.

The translation device and the translation method translate the text file generated by the extracted text by a known method. The text format recognition generation method of the present invention can be applied not only to machine translation but also to the case where various conversions are performed on extracted texts and charts and output in the same format as the original document.

The text format generation method works as follows.

First, when the text areas adjacent in the connection order have the same column, the text areas are integrated.

If the result of translating the text in the diagram area exceeds the number of lines and columns occupied by the original area, increase the number of lines and columns in the area where the translated result is embedded so that the translation result can be accommodated. And, with the increase,
Move or shrink another area adjacent to that area to fit on the page.

The text format generation method also includes
If the result of translating the text does not fit in the original formatted text area, create a new page with a format that conforms to the recognized page header, page footer, and column, and then create the page area. To
Put the part where the translation result extends beyond the area of the original formatted text.

If a blank space is generated in the area after the result of translating the text is put in the area corresponding to the original area of the text, a blank character or a line feed character is inserted in the blank area. Also, if there are only blank characters or newline characters in all areas of a page, that page is deleted.

After that, the system device stores the processing result in the file storage device or displays it via the display device.

[0039]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 15 shows the configuration of an apparatus to which the text format recognition generation method according to an embodiment of the present invention is applied. This device is an input device 15001, a file storage device 15002, a system device 15003, a translation device 150.
04 and a display device 15005.

When translation is performed, the original text is input by the input device 15001 and stored as an input file in the file storage device 15002 via the system device 15003. The original text is a formatted document converted to a text file. System unit 150
03 reads this input file and recognizes its text format. After that, translation processing is performed by the translation device 15004. Further, the system device 1500
3 generates a translation result document having a format equivalent to the previously recognized text format of the original sentence.

The operation procedure in this embodiment will be described in detail below. First, prior to the recognition and translation processing of the original text format, the file storage device 150 is input via the input device 15001 and the system device 15003.
The input file (original text) is stored in 02.

FIG. 1 is a flow chart showing a text format recognition method and a translation method. The text format recognition method and the translation method operate in the system unit 15003 of FIG.

FIG. 16 shows an example of text data of an input file stored in the file storage device 15002. For easy understanding of the explanation, it is assumed that the data shown in this figure is subjected to a text format recognition process and a translation process.

The text of FIG. 16 is line 16001-line 1
As indicated by 6046, there are 46 rows of data. Number 1
The line feed character is indicated by "▽" as indicated by "▽" in 6047. This '▽' is the right end of each line. The left end of each row is the position of “1” at the left end of row 16002.

Row 16001, row 16003, row 1601
5, row 16021, row 16023, row 16024, row 1
6026, line 16044, and line 16046 have a line feed character at the left end of the line, and these lines do not have normal character data. Line 16004 to line 16014, line 16016
~ Line 16020, line 16022, line 16027 to line 16
043, and line 16045 are filled with blank characters from the left edge of the line to just before the character visible in this figure.

In FIG. 16, the characters representing the text are shown by appropriate numbers and symbols ':', but this is for the purpose of simplifying the explanation, and it is assumed that various characters are actually used. .

Next, the procedures of the text format recognition method and the translation method will be specifically described with reference to FIG.

First, in page length estimation step 1001, a part or all of the input file is read,
Estimate the page length in text format of the input file. Details of this step will be described later with reference to FIGS. 2, 3, and 4. Next, in step 1002, the page header is recognized, and in step 1003, the page footer is recognized. Details of the page header recognition step 1002 will be described later with reference to FIG. Page footer recognition step 1
Details of 003 will be described later with reference to FIG.

Next, in step 1004, an area such as a column is recognized, in step 1005 the area type is recognized, and in step 1006 the connection order of areas is determined. Details of the area recognition step 1004 will be described later with reference to FIGS. 7 and 8. Details of the area type recognition step 1005 will be described later with reference to FIG. 9. Area connection order determination step 1006
Will be described later in detail with reference to FIGS. 10 and 11.

Next, in step 1007, the texts and figures are extracted. Text and chart extraction step 1007
Will be described later in detail with reference to FIG. Next, in step 1008, the translation unnecessary portion is recognized, and in step 1009, machine translation processing is performed. Then, in step 1010, a text format is generated, the format is added to the result of the machine translation, and the process ends. Details of the translation unnecessary portion recognition processing step 1008 will be described later with reference to FIG. Details of the text format generation process are shown in FIG.
Will be described later with reference to.

Next, referring to FIGS. 2, 3 and 4,
The page length estimation step of step 1001 of FIG. 1 will be described.

First, in step 2001 of FIG.
Empty the input line group buffer. The already-input-line-group buffer is a buffer that reads an input file from the file storage device 15002 and stores it in the system device 15003. For the sake of simplicity, the word "buffer" is omitted in the figure and is simply described as "already input row group". The same applies to other buffers and counters.

Next, in step 2002, the value of the current input line number counter is set to 0. The current input line number counter is a counter for numbering the lines of the data accumulated in the already input line group. Next, in step 2003, the row matching score calculation result buffer is emptied. The row matching score calculation result buffer is a buffer for accumulating the matching score calculation results for various pairs of rows in order to estimate the page length.

Subsequently, in steps 3001 to 3006 of FIG. 3, the degree of coincidence is calculated for the set of rows while reading the rows of the input file.

First, in decision step 3001, it is checked whether the first 300 lines of the input file being processed have already been processed or the end of the file has been reached. The reason why the number of lines is 300 is that it is possible to estimate the page in most cases by reading about 300 lines. At the moment, still 1
Since no line has been read, this condition is not satisfied and the process proceeds to step 3002.

In step 3002, one line is read from the input file and the data is stored in the current input line buffer. At this point, the data in row 16001 of FIG. 16 is the current input row. In the following step 3003,
Add 1 to the value of the current input line number counter. At the moment,
Since the value of the current input line number counter is 0, its value becomes 1 in this step 3003.

At the next step 3004, the degree of coincidence between each row stored in the already-input row group buffer and the current input row stored in the current input row buffer is calculated. Since the input line group buffer is empty at this time, the degree of coincidence is not calculated in this step. Since the degree of coincidence is not calculated, the degree of coincidence is not added to the line coincidence degree calculation result buffer.

At the next step 3005, the set data of the current input line number of the current input line number counter and the current input line data of the current input line buffer (current input line number, current input line) is converted into the already input line group buffer. Add to. Then, the process proceeds to step 3001.

At present, the process proceeds to step 3002 through step 3001.

In step 3002, the next line from the input file is read into the current input line buffer and the data is set as the current input line. At this point, line 160 of FIG.
The data of 02 becomes the current input line. Continued Step 3003
At 1, add 1 to the value of the current input line number counter. Since the value of the current input line number is 1 at this point, the value is 2 in this step 3003.

At the next step 3004, the degree of coincidence between the row stored in the already-input row group buffer and the current input row is calculated. At the moment, the already input line group buffer has 160 lines.
Since 01 is stored, in this step, the degree of coincidence between the current input row, that is, row 16002 and row 16001 is calculated, and that value is set in the row coincidence degree buffer.

The degree of coincidence between lines is defined as the result of dividing the number of columns in which the same character is the same in two lines by the average of the lengths of the two lines. Lines 16001 and 16
In 002, since there is no column in which the same column has the same character, the matching degree is 0, and this value is set in the row matching degree buffer. Since the matching degree value is 0, no data is added to the row matching degree calculation result buffer. In this embodiment, when the value of the matching score is 0.75 or more, data is added to the row matching score calculation result buffer in this step.

At the next step 3005, the set data of (current input line number, current input line) is added to the already input line group buffer. As a result, in the already input row group buffer, the row 16
001, row 16002, and row numbers of these rows are stored. Then, the process proceeds to step 3001.

Here, the process again proceeds to step 3002 through step 3001.

In step 3002, the next line from the input file is read into the current input line buffer and the data is set as the current input line. At this point, line 160 of FIG.
The data of 03 becomes the current input line. Continued Step 3003
At 1, add 1 to the value of the current input line number counter. Since the value of the current input line number is 2 at this point, the value is 3 in this step 3003.

At the next step 3004, the degree of coincidence between the row stored in the already-input row group buffer and the current input row is calculated. At the moment, the already input line group buffer has 160 lines.
Since 01 and row 16002 are stored, in this step, the degree of coincidence between these rows and the current input row, that is, row 16003 is calculated, and that value is set in the row coincidence degree buffer.

In the comparison between line 16001 and line 16003, since the leftmost column has the same line feed character, the matching degree is 1, and this value is set in the matching degree buffer. In addition,
The line feed character shall be counted as one character. Row 160
01 and the row 16003 have a constant degree of coincidence of 0.75.
As described above, the row offset is 2 which is the difference between the row number 1 of the row 16001 and the row number 3 of the row 16003. Therefore, three sets of data (already input row number, row offset, matching degree), that is, (1, 2, 1) are added to the row matching degree calculation result buffer.

FIG. 17 shows an example of data stored in the row coincidence degree calculation result buffer. 17007 shows the group data (1,2,1) just added. In FIG. 17, the input line number 170
02, row offset 17003, and degree of coincidence 1700
In addition to 6, serial number 17001, number of matching columns 1700
4 and line length (average of line lengths of two lines) 17005 are shown at the same time.

In the same step 3004, the degree of coincidence between the current input row 16003 and all the rows in the already-input row group buffer is calculated, so rows 16002 and 1600 are calculated.
The degree of agreement of 3 is then calculated. Since there is no column with the same character in these two lines, the degree of coincidence is 0.
Therefore, regarding these two lines, nothing is added to the line coincidence degree calculation result of FIG.

In the next step 3005, the set data of (current input line number, current input line) is added to the already input line group buffer. As a result, in the already input row group buffer, the row 16
001, row 16002, row 16003, and row numbers of these rows are stored. Then, the process proceeds to step 3001.

Similarly, while reading the lines, the degree of coincidence between each line of the already input line group buffer and the current input line is calculated,
Data having a matching degree of 0.75 or more is added to the row matching degree calculation result of FIG. 17 is the same as FIG.
The result of calculating the degree of line matching as described above from the input file is shown below.

When the line 160046 in FIG. 16 is read, and the process reaches the decision step 3001 again after the steps 3004 and 3005, the end of the file has been reached, so the process advances to step 3006. Step 3
At 006, the number of lines actually read is set in the read line number counter. At present, its value is 46.

Next, the process proceeds to step 4001 in FIG. In step 4001, the page length estimation calculation result buffer is emptied. The page length estimation calculation result buffer is a buffer that stores the result of statistically processing the line matching degree calculation result.

In step 4002, the following processing is performed on the row coincidence degree calculation result (FIG. 17). First, the number of sets of data having the same row offset is taken as the row offset frequency. The expected number of pages is the result of dividing the number of read rows by the row offset frequency, and the result is the result of dividing the row offset frequency by the expected number of pages. For each row offset value,
Creates a set of data (row offset, row offset frequency, expected number of pages, reliability) and adds it to the page length calculation result buffer. FIG. 18 shows the results obtained by arranging the results of this step in descending order of reliability at the present time.

In step 4003, the page length estimation result is determined. That is, regarding the page length estimation calculation result of FIG. 18, the value of the row offset of the group data having the reliability of 1 or more and the maximum is set in the page length buffer. When the maximum reliability is less than 1, the number of read rows is set in the page length buffer. Currently 1
Since the maximum value of reliability is 3 as shown in 8001,
The row offset value 23 of the group data is set as the page length.

In the next step 4004, the line having the longest line length is detected from among the lines in the already-input line group buffer, and the line length is set in the maximum line length buffer. At this time,
The line length includes the line feed character at the end of the line. At present, as can be seen from FIG. 16, the maximum line length is 53.

As described above, the page length estimation step 100 of FIG.
The operation of 1 ends.

Next, the page header recognition step 100.
Go to 2. The details of this step will be described with reference to FIG. First, in step 5001, each line is viewed downward from the beginning of the already-input-line group buffer, and the line number of those lines is equal to the already-input line number in the line coincidence calculation result (FIG. 17). There is group data whose value is equal to the page length, and a range in which such lines continue from the head of the already-input line group buffer is detected. Recognize the detected line as a page header and copy it to the page header storage area. Hereinafter, this copied and stored data is simply referred to as a page header.

Further, each line recognized as the page header is compared with the lines below by the number of lines equal to the page length,
When there is a number in a column where characters do not match, that column is recognized as the number of pages, and the character in that column of the page header is set to '$'.

At this point, if the line offset value in the line coincidence calculation result is equal to the page length value 23 in the range where the lines are continuous from the beginning of the already-input line group buffer when viewed downward, 17 17008, 17009, 17010
There is. Since the line number of 17008 is 1, this is shown in FIG.
6 row 16001 is shown. This indicates row 16002 because the row number of 17009 is 2. Since the row number of 17010 is 3, this indicates the row 16003. Therefore, these three lines are recognized as a page header and copied to the page header storage area.

Next, in this same step 5001, each line of these page headers is compared with the line below by the number of lines equal to the page length value 23. Ie line 16001 and line 1
6024 is compared, row 16002 and row 16025 are compared, and row 16003 and row 1606 are compared. As a result,
Since the 41st column is different between line 16002 and line 16025 and the character in that column is a number, it is recognized as the number of pages and the character in this column in the page header is changed to '$'. To do.

In the following step 5002, the line number of the line next to the line recognized as the page header in the already input line group is set as the text upper limit number. At this point, the lines recognized as the page header are the line 16001, the line 16002, and the line 16003.
The line number 4 of 004 is set as the text upper limit number.

This is the end of the operation of the page header recognition step 1002 shown in FIG.

Next, the page footer recognition step 100.
Go to 3. The details of this step will be described with reference to FIG. The operation in this page footer recognition step is almost the same as the operation procedure in the page header recognition step 1002 in FIG.

First, in step 6001, the line having a line number equal to the page length in the already input line group is set as the page end line. At present, the line 16023 is the last line of the page. Next, look at each line upwards from the last line of the page,
In the row matching degree calculation result (FIG. 17), there is group data in which the row numbers of those rows are equal to the already input row numbers and the row offset values thereof are equal to the page length, and from the last row of the page, Detect a range of continuous lines. Recognize the detected line as a page footer and copy it to the page footer storage area. Hereinafter, this copied and stored data is simply referred to as a page footer.

At the present time, if the line offset value in the line coincidence degree calculation result is equal to the page length value 23 in the range where the lines are continuous from the end line of the page as viewed upward,
7 17011, 17012, 17013. 17
Since the line number of 011 is 21, this indicates the line 16021. The line number of 17012 is 22, so this is line 1
6022 is shown. Since the line number of 17013 is 23, this indicates the line 16023. Therefore, these three lines are recognized as a page footer and copied to the page footer storage area.

Next, each line recognized as a page footer in the same step 6001 is compared with the line below by the number of lines equal to the page length. If there is a number in a column in which the characters do not match, the column is paged. Recognize it as a number and set the character in that column of the page footer to '$'.

At this time, each line of the page footer is compared with the line below by the number of lines equal to the page length value 23. That is, row 16021 and row 16044 are compared and row 160
22 and row 16045 are compared, and row 16023 and row 160
Compare 46. As a result, row 16022 and row 1604
Since the 22nd column is different between 5 and the character in that column is a number, it is recognized as the number of pages, and the character in this column in the page footer is changed to '$'.

In the following step 6002, the line number of the line immediately before the line recognized as the page footer in the already input line group is set as the text lower limit number. Currently, the lines recognized as page footers are line 16021, line 1
6022, line 16023, so the line 160 immediately before it
Set line number 20 of 20 to the text lower limit number.

The page footer recognition step 100 is completed.
The operation of 3 ends.

FIG. 19 shows the results of estimation and recognition of the page length, page header, and page footer described above.

Next, the process proceeds to area recognition step 1004.
The details of this step will be described with reference to FIGS. 7 and 8.

First, in step 7001, for each line of the already-input line group buffer, if the line length of a certain line does not reach a certain column, it is considered that there is a blank character in that column, and each column (that is, FIG. Count the blank characters (16 vertically). The line feed character is regarded as a space character. The tab character is assumed to have the tab expanded to the required number of white space characters. Next, in step 7002, for each row of the already input row group, the result of dividing the number of blank characters for each column by the number of read rows is defined as the blank character rate.

FIG. 20 shows the result of arranging the column-by-column blank character ratio calculation results, which are the processing results of the current step 7001 and step 7002, in descending order of the blank character ratio.

Next, in step 7003, the range where columns with a blank character ratio of a certain value or more continue is recognized as the boundary of the column, and the start column and end column of each column are determined. The result is stored as the basic column recognition result.

In the present embodiment, the constant value of the blank character ratio is 0.75 or more. At present, the range of continuous columns in this range is the 53rd column shown in 20001 and the 1st column shown in 20002 to 20003 of FIG.
It can be recognized that the second column and the 25th to 26th columns shown in 20004 to 20005 are boundaries of the region of the column. Therefore, there are two columns, the third column to the 24th column and the 27th column to the 52nd column. Figure 21
Shows the basic column area recognition result which is the result of recognition in this way.

Next, in step 8001 of FIG.
Reopen the input file so that it can be read from the beginning of the input file again. And step 80
In 02, the value of the page number counter is set to 0.

After judgment step 8003, step 8
At 004, 1 is added to the value of the page number counter. At present, the value of the page number counter is 1.

Next, at step 8005, the number of lines indicated by the page length is read from the input file. In this example, rows 16001 to 16023 are read. In the following step 8006, this read 1
Recognize the area within the page by analyzing the data for the page.

In step 8006, based on the contents of the page header, the page footer, the text upper limit line number, the text lower limit line number, and the basic column area recognition result, the existing range of the text and the figure is assumed, and The recognition result of the area in the page is stored in the individual page area recognition result assuming the range of the column at the boundary of the column. Here, the recognition of the break of the area is performed as in (i) and (ii) below. The line feed character is treated as a space character.

(I) Characters within the range of one column assuming that there is a column in a certain line (that is, a range in which a character is considered to exist if the same column as the immediately preceding line exists in that line) A line is considered to be a break in the area if it is all whitespace characters. That is, if there is a region immediately before the line, the region ends immediately before the line, and another region starts immediately after the line.

(Ii) If a column assumed to be a column boundary in a line is not a blank character, the line constitutes an area different from the basic column area recognition result. In other words, if there is a region that is adjacent through that boundary,
It is assumed that the regions end just before the line, and that the line starts a region of the combined width of the adjacent regions.

At the present time, that is, the area recognition result of the first page is 22009 of the individual page area recognition result of FIG.
22010 and 22011. A graphical description of this is shown at 23001 in FIG. The start line 22001 and the end line 2200 of the individual page area recognition result in FIG.
2, start column 22004, end column 22005,
17 is a diagram showing a range of existence of an area in the input file shown in FIG. 16. The page 22003 in each line is the current page number value. Area type 22006, next area 2
The data 2007 and the next connection type 22008 are empty or "-" indicating the end because data is input in the subsequent processing.

After the step 8006 is completed, the process proceeds to the step 8004 through the judgment step 8003 again.

At step 8004, 1 is added to the value of the page number counter. At the moment, its value is 2.

Next, at step 8005, the number of lines indicated by the page length, that is, 23 lines of data is read from the input file. In this example, lines 16024 to 16
046 will be read. In the next step 8006, the data of the second page is analyzed to recognize the area. The result of this step is 22012, 2 in FIG.
2013, 22014, 22015. A graphical description of this is shown at 23002 in FIG.

The step 8006 is ended and the procedure goes to the decision step 8003 again. This time, since the end of the input file has already been reached, the operation of the flowchart of FIG. 8, that is, the area recognition step 1004 is ended.

Next, the processing proceeds to area type recognition step 1005. The details of this step will be described with reference to FIG.

First, in step 9001, FIG.
For each region recognized as 8, “+”, “−”,
Check the ratio between the number of characters that make up the chart, such as' | 'and'‖', and the number of characters other than blank characters and newline characters. If the result is a certain value or more, select the area type. Set as the chart area. The rest is a text area.

In the present embodiment, the threshold for the area type in this step is set to 0.6. In the current input file, the area indicated by 22011 in FIG.
That is, in the region from the third column to the 52nd column of lines 16016 to 16020 in FIG. 16, the number of characters making up the above chart is 134 characters, and the number of characters excluding blank characters and line feed characters is 155 characters. Therefore, the ratio is 0.86, and the type of this area is the chart area.
Since the other areas do not include the characters that make up the above chart, the types of these areas are considered to be text areas.

FIG. 24 shows the result of recognizing the type of area. In the field of area type 24001, only the area indicated by 24002 is shown as "chart", which indicates that it is a chart area. This area is indicated by reference numeral 22011 in FIG. In the area type 24001 of the other area, there is "text", which indicates that the area is a text area.

Thus, the operation of the area type recognition step 1005 is completed.

Next, the area connection order determination step 1006.
Proceed to. The details of this step will be described with reference to FIGS. 10 and 11.

First, in step 10001, for each page in the input file, if there are areas vertically adjacent to a certain area in the page, those areas are connected from top to bottom. If there are areas that are adjacent to each other on the left and right, the areas that connect to the bottom on the left side of these areas are connected to the areas on the right side. The connection of this area is determined from top to bottom and from left to right.It is assumed that the area already connected is skipped and the area is already connected, instead of connecting again. The connection order of one row is determined for each area.

In the next step 10002, the connection of the areas between the pages in the input file is such that the connection order of the areas in a page is from the last area to the leftmost and uppermost area of the next page. To do.

FIG. 25 shows the current input file (FIG. 16).
For this step 10001 and step 10002
The recognition result of is shown. The number of the field of the next area 25002 is represented by the value of the area number of the field 25001. As a result, the connection order of the areas in the input file is determined in a line. The'- 'in 25003 at this point
Indicates the end of the connection order.

FIG. 26 shows a diagrammatic representation of the connection sequence of FIG.

Next, the process proceeds to step 11001 in FIG. In this step 11001, the connection order of one row of regions is traced, and the connection order of different columns is created for each type of area. That is, a connection order of two columns (a column in the text area and a column in the chart area) is created. Each area included in the series of columns related to the text area is called the text main area,
The head of the connection order is set as the head area of the text main area. Each area included in a series of columns related to a chart area is called a chart subarea, and the head of the connection order is set as the head area of the chart subarea. The order of each column shall be such that the order in the original one column connection order is preserved and ordered.

In the next area 28001 column in FIGS. 27 and 28, the step 11001 for the current input file is displayed.
The result is shown. FIG. 29 shows a graphical description of this result.

In the next step 11002, when another text area is connected immediately after one text area,
The difference in the frequency distribution of the used words between the two areas is large, the words at the boundaries of the areas are not properly connected, and the adjacent text area is behind the adjacent text area. It is checked whether there is a text area with similar usage word frequency distribution and proper area boundary word connection. And if such a region exists,
Change the connection order so that the region is connected immediately after the certain region.

In the current input file, the words in each area are not considered, so a concrete example of the operation of this step will be omitted. If the effect of this step is on the current input file, it is, for example, the fourth region 29004 to the sixth region 290 of FIG.
The area connected to 06 is changed from the fourth area 29004 to the fifth area 29005.

In the next step 11003, when areas are connected vertically in the same page according to the basic column area recognition result, the next connection type of the upper area is set to the same column.
Other than that, the next connection type is a separate set.

The next connection type 28002 column in FIG. 28 shows the result of step 11003 for the current input file. Connection from the fourth area to the sixth area 28003,
Connections 28004 from the fifth region to the seventh region are in the same column.

The area connection order determination step 1006 is completed as described above.
Ends the operation.

Next, the text / figure extraction step 100.
Proceed to 7. The details of this step will be described with reference to FIG. First, in step 12001, a list of areas indicated by the text main area is traced, data is extracted from each area and connected to create a text file. However, if the next connection type is the same column, they are not linked and are set as different text files.

In the current input file, as shown in FIGS.
8 and FIG. 29, first, the first region 29
001, the second area 29002, and the fourth area 29
Since the next connection type 004 is connected in a different column, data is extracted from these areas and connected to create a text file. The result is shown in FIG. First area 2
9001 corresponds to 30001, and the second area 29002
Corresponds to 30002, and the fourth region 29004 is 300
Corresponds to 03.

The fourth area 29004 to the sixth area 29
Since the connection to 006 is the same column as shown by 28003 in FIG. 28, it is stored in a different file from the sixth area. In addition, the sixth area 29006 and the fifth area 2900
Since the next connection type is connected to the column 5 by another column, data is extracted from these two areas and connected to create a text file. The result is shown in FIG. The sixth area 29006 corresponds to 31001, and the fifth area 2900
5 corresponds to 31002.

Since the connection from the fifth area to the seventh area is the same column, it is stored in a different file from the seventh area 29007. Since the end is after the seventh area, data is extracted from only the seventh area and stored in the file. The result is shown in FIG.

In the next step 12002, the list of chart sub-areas is traced, data is extracted from each area, and a chart file is created. The data extracted from each area of the list of chart subareas should be in a separate chart file. In the current input file, the chart sub-area indicates the third area 29003, and the next is the end, so data is extracted from this area and stored in the chart file. The result is shown in FIG.

The text / figure extraction step 100 is completed.
The operation of 7 is ended.

Next, a translation unnecessary portion recognition step 1008.
Proceed to. The details of this step will be described with reference to FIG.

First, in step 13001, for each line of each text file extracted from the area whose area type is text area, "+", "-", "|",
Characters that make up a chart such as'‖ ', equal signs, inequality signs,
Number of characters that make up the mathematical expression, such as '/', '*', 'Σ', '±', '÷', 'x', and numbers, and the number of characters excluding whitespace and newline characters in that line A line whose ratio is above a certain value is recognized as a translation-unnecessary line, and the translation-unnecessary designation of that line is inserted.

In the current input file, the text files extracted from the text area are shown in FIG. 30, FIG. 31, and FIG.
It is assumed that only the text file of FIG. 30 changes as a result of the processing of this step.
The processing result of step 13001 is shown in FIG. 340
The lines indicated by 01, 34002, 34003, and 34004 are the lines recognized as the translation unnecessary lines, and the translation unnecessary designations "<<" and ">>" are inserted.

In the next step 13002, for each line of each text file extracted from the area whose area type is the chart area, the characters that make up the chart such as "+", "-", "|", "|", Equal sign, inequality sign, '/',
'*', 'Σ', '±', '÷', 'x', numbers, etc.
Recognize a line containing characters that make up a mathematical expression as a translation-unnecessary line, and insert a translation-unnecessary specification into that line. In the current input file, the chart file extracted from the chart area is shown in Figure 3.
3 is only the chart file, and the result is shown in FIG.

As described above, the translation unnecessary portion recognition step 1008
Ends the operation.

Next, the process proceeds to translation step 1009. In this step, the text files shown in FIGS. 34, 31, and 32 and the chart file shown in FIG. 35 are machine translated, and the results are stored in separate files. Since the translation process in this step operates according to a known translation method, detailed description thereof will be omitted.

Next, the process proceeds to the text format generation step 1010. The details of this step will be described with reference to FIG.

First, in step 14001, regarding the individual page area recognition result (FIG. 28), the list of text areas indicated by the text main area is traced, and areas having the same connection type as the next connection type are integrated into one area. The result is used as the target document area generation result. By this step, FIG. 28 is modified as shown in FIG. End column 3 of the fourth area 36001 and the fifth area 36002
The value in the column of 6003 is changed, and the integrated sixth area 28005 and seventh area 28006 are deleted.
37 graphically describes the contents of FIG.

Next, in step 14002, the translation result of each chart file is embedded in the area from which data is extracted in each chart area indicated by the chart sub-area. At this time,
When the translation result does not fit in the embedding destination area, the embedding destination area is expanded and other areas in the same page are reduced accordingly. For the current input file,
The chart file of FIG. 35, which is the extraction result of the third area, is embedded. In the processing on the input file at the present time, it is not necessary to expand the area for this embedding.

Next, in step 14003, the translation results of each text file are concatenated by putting blank lines at the file boundaries and embedded in the area indicated by the text main area in order. When the translation results of all text files are embedded and there is a space, fill the newline character or the space character. If the text area is insufficient, a new page and column area are generated according to the results of page length estimation, page header recognition, page footer recognition, and the basic column area recognition result. If there is a page number designation '$' in the page header and page footer, replace the page number designation of the generated page with an appropriate page number.

38 and 39 show the results of these steps for the current input file (FIG. 16).
This is an example of the case where a new page is generated as shown in the third page 39001.

Next, in step 14004, the page in which all the areas are filled with only the line feed character or the blank character is deleted. In the present example, as shown in FIGS. 38 and 39, since there is no corresponding page, there is no change in this step.

As described above, the operation of the text format generation step 1010 is completed, and thus the operation of the text format recognition generation method (and machine translation process) of FIG. 1 is completed.

This is the end of the description of the embodiment of the present invention.

Next, a modification of the above embodiment will be described. In the above-described embodiment, the condition of the break of the region when recognizing the region is that the range in which it is assumed that there is a region is a blank character. However, in addition to this, when the left side of a line, which is a range assumed to have a region, has a certain number or more of blank characters, a condition for making a region break immediately before the line can be added. This means that if there is an indentation at the beginning of a paragraph, the area immediately before that can be used to divide the area.

Next, in the above embodiment, when recognizing a column, a part or all of the input file is read, blank characters in each column are counted, and the blank character ratio is calculated. It is assumed that the continuous range is recognized as the boundary of the column area. This is done by calculating the frequency of each character in each column for consecutive lines in a specific range, and defining columns with a high rate of having the same characters even if they are not blank characters in the consecutive lines as the boundaries of the column area. May be recognized.

As a result, for example, a quoted portion may be displayed by adding '>' to the left end of a line in an electronic mail, but this quoted portion can be recognized as a region. Further, if the area is recognized as a character that can form the boundary of the column area, especially if the characters forming the line of the chart are recognized, the line 1601 of the input file (FIG. 16) of the above embodiment is recognized.
The interior of a table cell, such as 6-row 16020, can be recognized as a region, and thus the structure of the table can be preserved and its contents translated.

Next, in the above embodiment, when there is a character other than a blank character or a line feed character in a column, which is assumed to be the boundary of the column area in a certain line, two adjacent areas are separated via the boundary. Suppose the width region begins. The boundary of the column area may be recognized by calculating the blank character ratio for each column in a specific range from that line. As a result, it is possible to recognize a text having a more complicated area structure.

Further, in the above embodiment, the text format of the input file is recognized, and after translation, the translation result text in the same text format is generated. After recognizing the text format, the text format recognition result may be inserted as data with a tag (clean copy command) in the result of concatenating the extracted text files. This facilitates transformation and modification of the text format.

[0151]

As described above, according to the present invention,
When machine translation of page-formatted document data is performed, the text format is recognized from the input document, and the original page format is added to the translated document to have a text format equivalent to the input document. A translated document can be generated. Therefore, there is an effect that it is possible to automate the troublesome work that has been conventionally done manually. Further, it is effective not only when the input file is data created by a word processor and a text formatter, but also when the result of recognizing a printed matter by an optical character recognition device is used as the input file.

[Brief description of drawings]

FIG. 1 is a processing flowchart of a text format recognition generation method of the present invention.

FIG. 2 is a part of a processing flowchart of a page length estimation method which is a part of the text format recognition generation method of the present invention.

FIG. 3 is a part of a processing flowchart of a page length estimation method which is a part of the text format recognition generation method of the present invention.

FIG. 4 is a part of a process flowchart of a page length estimation method which is a part of the text format recognition generation method of the present invention.

FIG. 5 is a processing flowchart of a page header recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 6 is a processing flowchart of a page footer recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 7 is a part of a processing flowchart of a region recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 8 is a part of a process flowchart of a region recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 9 is a processing flowchart of a region type recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 10 is a part of a processing flowchart of a region connection order recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 11 is a part of a processing flowchart of a region connection order recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 12 is a processing flowchart of a text / figure extracting method which is a part of the text format recognition generating method of the present invention.

FIG. 13 is a processing flowchart of a translation unnecessary portion recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 14 is a processing flowchart of a text format generation method which is a part of the text format recognition generation method of the present invention.

FIG. 15 is a diagram showing an example of an apparatus on which the text format recognition generation method of the present invention operates.

FIG. 16 is a diagram showing the contents of an input file for explaining the operation of the text format recognition generation method of the present invention.

FIG. 17 is a diagram for explaining the operation of the page length estimation method which is a part of the text format recognition generation method of the present invention.

FIG. 18 is a diagram for explaining the operation of the page length estimation method which is a part of the text format recognition generation method of the present invention.

FIG. 19 is a diagram showing operation results of a page length estimating method, a page header recognizing method, and a page footer recognizing method, which are part of the text format recognition generating method of the present invention.

FIG. 20 is a diagram for explaining the operation of the area recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 21 is a diagram for explaining the operation of the area recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 22 is a diagram showing an operation result of the area recognition method which is a part of the text format recognition generation method of the present invention.

23 is a diagram graphically describing the contents of FIG. 22. FIG.

FIG. 24 is a diagram showing the result of the operation of the area type recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 25 is a diagram for explaining the operation of the area connection order recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 26 is a diagram graphically describing the contents of FIG. 25.

FIG. 27 is a part of a diagram showing an operation result of the area connection order recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 28 is a part of a diagram showing an operation result of the area connection order recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 29 is a diagram graphically describing the contents of FIG. 28.

FIG. 30 is a diagram showing a processing result of a text / figure extracting method which is a part of the text format recognition generating method of the present invention.

FIG. 31 is a diagram showing a processing result of a text / figure extracting method which is a part of the text format recognition generating method of the present invention.

FIG. 32 is a diagram showing a processing result of a text / figure extracting method which is a part of the text format recognition generating method of the present invention.

FIG. 33 is a diagram showing a processing result of a text / figure extracting method which is a part of the text format recognition generating method of the present invention.

FIG. 34 is a diagram showing a processing result of a translation unnecessary portion recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 35 is a diagram showing a processing result of a translation unnecessary portion recognition method which is a part of the text format recognition generation method of the present invention.

FIG. 36 is a diagram showing a processing result of a text format generation method which is a part of the text format recognition generation method of the present invention.

FIG. 37 is a diagram graphically describing the contents of FIG. 36.

FIG. 38 is a diagram showing a processing result of the text format recognition generation method of the present invention.

FIG. 39 is a diagram showing a processing result of the text format recognition generation method of the present invention.

[Explanation of symbols]

15001 ... Input device, 15002 ... File storage device, 15003 ... System device, 15004 ... Translation device, 15005 ... Display device.

Claims (30)

[Claims] 1. A step of inputting a text-formatted text file such as page header attachment, page footing attachment, column arrangement, chart allocation, etc., a step of estimating the page length of the input text file, and page footer and / or Recognizing page headers, recognizing areas such as columns and charts, recognizing the type indicating whether the recognized areas are text areas or chart areas, and determining the connection order of recognized areas According to the above connection order, a step of extracting texts and figures that span a plurality of areas, a step of applying a predetermined conversion to the extracted texts and figures, and an input of the texts and figures of the conversion results. The step of giving the same format as the text file of Text formatting recognition generating method characterized by the. 2. The text format recognition generation method according to claim 1, wherein the predetermined conversion is a translation process. 3. The text format recognition generation method according to claim 2, further comprising recognizing an extracted text and a portion of the chart that does not require translation and adding a translation unnecessary designation to the portion for translation. Text format recognition generation method. 4. The text format recognition generation method according to claim 1, wherein some or all lines are read from the beginning of the input text file, line numbers are given from the beginning to each line, and a certain line is added. And the line on the end side of the file than that line is calculated, and the difference between the line numbers of the two lines that calculated the match is used as the line offset, and the set of two lines with a certain degree of match is equal to or greater than The operation of accumulating a set of the degree of coincidence and the row offset is performed for each row, the accumulation result is used as the row coincidence calculation result, the frequency of the row offset in the row coincidence calculation result is counted, and the row offset is assumed to be the page length. In this case, the ratio of the frequency of row offset to the number of pages calculated from the number of read rows is calculated, and the row offset with the highest frequency ratio is estimated to be the page length. , Text formatting recognition generating method characterized by. 5. The text format recognition generation method according to claim 4, wherein from the line matching degree calculation result, lines having a line offset equal to the page length are consecutive from the start line of the page toward the end of the file. A text format recognition generation method characterized by detecting whether there is such a line and recognizing such a line as a page header. 6. The text format recognition generation method according to claim 5, wherein for each line recognized as a page header, the corresponding line of each page is compared, and if there is a number in a column where characters do not match, that column is selected. A text format recognition and generation method characterized in that it is recognized as a page number field. 7. The text format recognition generation method according to claim 4, wherein from the line matching degree calculation result, lines having a line offset equal to the page length are consecutive from the end line of the page toward the beginning of the file. A text format recognition generation method characterized by detecting whether there is such a line and recognizing such a line as a page footer. 8. The text format recognition generation method according to claim 7, wherein for each line recognized as a page footer, the corresponding lines of each page are compared, and when there is a number in a column where characters do not match, that column is selected. A text format recognition and generation method characterized in that it is recognized as a page number field. 9. The text format recognition generation method according to claim 1, wherein part or all of the lines from the beginning of the input text file are read, and when the line length does not reach a certain column, that line's Assuming that there is a space character in the column and that the line feed character that indicates the end of the line is a space character, the number of space characters in each line is counted for each column, and the ratio of the number of space characters to the number of read lines is a certain value or more. A method for generating and recognizing a text format, characterized in that a range where columns are continuous is assumed to be a column boundary. 10. The text format recognition generation method according to claim 9, wherein if one column from one column to another column is assumed to be one column, all the characters in that range are in that range. White space, or
A text format recognition generation method, characterized in that when the line length does not reach the range, the region of the column is divided immediately before and after the line. 11. The text format recognition generation method according to claim 9, wherein when it is assumed that a column from one column to another column is one column, the characters in the range of a certain line are: A text format recognition generation method, characterized in that, when it starts with a certain number or more of blank characters, the column area is divided into that line and the immediately preceding line. 12. The text format recognition generation method according to claim 9, wherein if one column is assumed to extend from a certain column to another certain column, then at a certain row, at the boundary of the column. When there is a character other than a space character in a column that is assumed to exist, if there is an area of columns that are adjacent through that boundary, those areas will end immediately before that line, and that adjacent column will be combined from that line. A method for generating and recognizing a text format, which is characterized in that recognition of a region is continued on the assumption that a region of a column having a predetermined width exists. 13. The text format recognition generation method according to claim 1, wherein '+', '-',
An area where the ratio of the number of characters such as' | 'and'‖' that are frequently used to draw a chart to the total number of characters excluding blank characters and line feed characters in that area is a certain value or more And a region other than that is recognized as a text region. 14. The text format recognition generation method according to claim 1, wherein when there is an area vertically adjacent to an area in a page in the input text file, those areas are above. To the bottom, and if there are adjacent areas on the left and right, connect from the bottom-most area on the left side of those areas to the top-most area on the right, but when determining these connections. It is assumed that the area for which the connection has already been established will not be reconnected, but the area for which the connection has been determined will be skipped, and as a result, a one-line connection order relationship will be determined for each area in the page. Text format recognition and generation method. 15. The text format recognition generation method according to claim 14, wherein an area in a page in an input text file is connected from an end area to an upper left area on the leftmost side of the next page. A method for generating and recognizing a text format, characterized by: 16. The text format recognition generation method according to claim 14 or 15, wherein after the connection order of the areas is determined, the text area is connected to the chart area immediately after the text area. It is assumed that the areas are skipped and connected to the next area, and the skipped chart areas retain their original order to determine the connection order of those chart areas, and as a result, the text area column and the chart area column are A method for generating and recognizing a text format, which determines a connection order of columns. 17. The text format recognition generation method according to claim 14, wherein when another text area is connected immediately after a certain text area after determining the connection order of the areas, the two text areas are connected. There is a large difference in the used word frequency distribution between the two, the words at the boundary of the area are not properly connected, and the text area and the used word frequency distribution in the adjacent column behind the other text area are Text that is similar and that changes the connection order of areas so that the text area is connected immediately after a certain text area when there is an appropriate text area with the word connection at the boundary of the area. Format recognition generation method. 18. The text format recognition generation method according to claim 1, wherein when extracting the text from a series of areas, the text is extracted only from the area recognized as the text area, A text format recognition generation method, characterized in that the texts are connected according to a connection order. 19. The text format recognition generation method according to claim 3, wherein '+', '-', 'x',
Characters often used to describe mathematical expressions such as '÷', '^', '±', equal sign, inequality sign, 'Σ', and '|', '‖'
For example, a line that has a ratio of the number of characters that are frequently used to draw a chart, etc., and the total number of characters excluding white space characters and line feed characters in the line is a certain value or more is determined to be a translation unnecessary line. Characteristic text format recognition generation method. 20. The text format recognition generation method according to claim 19, wherein the threshold value that is the constant value is changed according to the type of the area to which the line belongs. 21. The text format recognition generation method according to claim 1, wherein when the result of converting the text in the figure area exceeds the number of lines and columns occupied by the original area, an area for embedding the conversion result The number of rows and columns is increased so that the conversion result can be accommodated, and with the increase, another area adjacent to that area can be moved or reduced to fit on the page. Text format recognition generation method. 22. In the text format recognition generation method according to any one of claims 1 to 3, when the result of converting the text does not fit in the original formatted text area, the recognized page header, page footer, and A text format recognition generation method characterized in that a page having a format conforming to a column is newly generated, and a conversion result is set in an area of the generated page. 23. In the text format recognition generation method according to claim 1, when a space is generated in the area after the result of converting the text is put in an area corresponding to the original area of the text. A text format recognition generation method characterized by including a space character or a line feed character in the margin. 24. In the text format recognition generation method according to claim 1, when a format equivalent to that of the input text file is added to the conversion result, all areas of a page are only blank characters or line feed characters. When the page is deleted, the text format recognition generation method characterized by the above. 25. The text format recognition generation method according to claim 1, wherein a page length, a page footer, a page header, a column number, and a column are included in the text and the chart data extracted from the input text file. A text format recognition and generation method characterized in that the recognition result of a text format such as a set width and the recognition result of a line not requiring translation are embedded as tagged data. 26. The text format recognition generation method according to claim 8, wherein instead of calculating the ratio of the number of blank characters in each column, the appearance rate for each character type appearing in each column is calculated, and the appearance rate is constant. It is assumed that the continuous range of the column where the character that is more than the value exists is the boundary of the area.
A text format recognition and generation method characterized by the above. 27. The text format recognition generation method according to claim 8, 9 or 11, wherein, in addition to a blank character and a line feed character, a character frequently used for drawing a chart is regarded as a boundary of the area. A method for generating and recognizing a text format, characterized by: 28. A means for inputting a text-formatted text file such as page header, page footer, column setting, chart layout, etc., means for estimating the page length of the input text file, and page footer and / or A means for recognizing page headers, a means for recognizing areas such as columns and charts, a means for recognizing the type indicating whether the recognized areas are text areas or chart areas, and means for deciding the connection order of the recognized areas. According to the above connection order, a means for extracting texts and figures that span multiple areas, a means for performing a predetermined conversion on the extracted texts and figures, and an input for the texts and figures resulting from the conversions. Text file with a format equivalent to that of the text file of Tsu door recognition generating device. 29. The text format recognition generation device according to claim 28, wherein the predetermined conversion is a translation process. 30. The text format recognition generation device according to claim 29, further comprising: recognizing an extracted text and a translation unnecessary portion of the chart, and adding a translation unnecessary designation to the portion for translation. Text format recognition generator.