JPS6225324A - Document understanding system - Google Patents

Abstract

PURPOSE:To extract a bibliography item automatically by making divisions according to the structure of a fixed form document. CONSTITUTION:Grammar which expresses the structure of a document is used and the syntax of a description is analyzed by the grammar to grasp the structure of an unknown input document. This grammar expressed a document which is inputted as a character sequence as a set of rectangular areas and contains variables of quantities indicating the absolute or relative sized of the rectangular areas or the absolute or relative relation among the rectangular areas. Further, a searching method for the rectangular areas can be specified. Further, a rectangular area is expressed as a set of searched areas and the format of the document is expressed finely by the hierachical expression. The formats of various documents expressed according to the grammar are stored previously in a memory. When a character sequence is inputted as an unknown document, a syntax analysis part searches for rectangular areas by a searching method specified by a document format to analyze the structure of the document.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to a document understanding method, and in particular, a document understanding method suitable as an input section of an electronic document file device [Background of the Invention] A conventional electronic document file device The content of each page of a document is simply stored as a character sequence, and bibliographic information (document name, publication date, document number, etc.) required as secondary information for searching is entered using the code input method on the keyboard. It had to be given from outside. However, in order to save labor in file creation work, it is desirable to automatically understand the title, author, etc. written in a document and automatically generate bibliographic data. In order to further advance the search, it is necessary to automatically detect captions of figures and tables, chapter/section titles, or automatically extract keywords by understanding the text itself. In addition, it is common to include graphical commands in the character series in a document, but dividing the character series of the target document into parts such as title, author, summary, main text, and graphical commands is useful for various searches. It was requested for the purpose of

In the prior art, methods have been proposed to facilitate document searches by inputting headings separately in word processors, etc. For example, "Hitachi Japanese Word Processor Wor
An example of such a method is disclosed in the dPal 25 Operating Instructions (Ji), page J3s (Hitachi, Ltd., first edition published on July 1, 1980). but,
This method does not specify bibliographic items themselves, but rather inputs sentences that the user remembers, and in order to use bibliographic items for full-fledged searches, it is necessary for Inchfusa to provide bibliographic items separately. Ta.

[Purpose of the invention]

The purpose of the present invention is to target stylized general documents,
The object of the present invention is to provide a document understanding method that enables automatic extraction of bibliographic items by dividing the document according to its structure.

[Summary of the invention]

In order to achieve this purpose, the present invention uses a grammar that expresses the structure of a document, and parses the description expressed by this grammar.

6 In the above grammar, a document given as a character sequence is expressed as a set of rectangular areas, and the absolute or relative size of the rectangular areas and the absolute size between the rectangular areas are Contains quantities that represent physical or relative relationships as variables. Additionally, the search method for the rectangular area can be specified. Furthermore, it is expressed as a set of rectangular search areas, and by such a hierarchical representation.

Document formats can be expressed down to the smallest detail.

For each type of document, a document format expressed according to the following grammar is stored in memory in advance. When the syntax analysis unit receives a character sequence representing an unknown document, it searches a rectangular area according to the search method specified in the document format.
Information on whether the search was successful or not and numerical values representing parameters determined during the search (absolute or relative size of rectangular areas and absolute or relative relationship between rectangular areas) are extracted. The syntax analysis unit assigns the numerical values of the above parameters to variables in the document format and performs the following analysis.
Sequentially proceed with structural analysis of documents.

Before giving a detailed explanation of the present invention, a detailed explanation of the present invention will be given.

FIG. 1 shows an example of a part of a technical paper having a certain format. This page is generated by a sequence of character codes and formatting information indicating where they should be placed directly above and in what format they should be printed. In the following explanation, a technical paper will be taken as an example, but the grammar format is only slightly different even for other documents, and the present invention can be applied by changing a part of the grammar. It is not limited to one example of a paper.

Next, an example of a grammar expressing the structure of a document (hereinafter abbreviated as F-grammar) will be shown.

(deforn + F (Sat orm Fl (10901040)) (
form F'2- (form F3- ))) (defo
rm Fl (form FRII (01001050)S t
r i n 'paper)) (form Fl2 ('?Xmjn';'Xmax ?Ymin
7Yvaax)shrink)) (defmac LINE-1(%1); 1st line (
paint 7Y1 (mode LN Y LES
S)) (paint?Y2 (mode OUT Y
LESS))(form%1 (0'?Ill?
YL? Y2)) (defmac LINE-2(%1
); 2nd line (point ?Y2 (n+ode OU
T Y LESS)) (paint 'Nia Y3
(mode ding N Y LESS)) (are
a (0'i'W ?Y2 ?H))) (paint
'i'Y4 (mode OUT Y LES
S)) (area (0'?lj 7Y37t())
)(form %1 (0'7W 7V3 ?Y4
)) The above grammar will be explained with reference to the example of FIG.

The first deform F... indicates that the format F is constructed with formats F2 and F3 arranged horizontally at the bottom of II2F1, as shown in FIG. In FIG. 1, F, Fl, F2. The portion F3 is shown surrounded by a broken line. 4 enclosed in parentheses following format name F1
The numerical value 10 90 10 4Q is the relative Fl when the total area F is 100X100
When the value of a parameter that indicates the size of is known, it is sufficient to directly write the value.

The following deform Fl... indicates that the format F1 is further composed of formats Fil and Fl2 arranged vertically. 5triB' paper in Fil indicates that character data "paper" exists in Fil. This string specification may or may not be present, and if it is not specified, the presence of character data is not checked. F
In l2? The letters marked with ``represent variables,'' which vary depending on the target.

4 variables ('?Xm1n "7Xmax ?Ymin ?yma
x) is I? 12, and the values found through the search are substituted into these variables as will be described later.

5hrink Indicates that the rectangular area corresponding to the format is reduced until the character component is circumscribed. 5hrink specification may or may not be specified.

The part after the next defmac LINE-1 (%1) is a simplified part of the definition of the -th line LINE-] by the macro definition. The format %1 is (form%l(0?Iil '7Y1 ?Y2))
Defined as a macro. ? H is the horizontal size of the format, ? H represents the vertical size of the format. ;The following is a comment.

Pa1nt indicates searching for a point that satisfies a certain condition and assigning it to a variable. Search conditions are specified by mode. IN/OUT indicates whether the search point is a change point from a blank to a character or a change from a character to a blank, Y indicates a search @ (X or Y), and LESS indicates a search direction. area indicates the area of the search range.

In understanding a document, the user refers to expressions written in accordance with the grammar and sequentially checks whether the rectangular area described therein exists in the document. When a rectangular area described including a variable is searched, the numerical value of that variable is obtained, and from now on, the numerical value is substituted into the variable and used.

Next, calculations between rectangular areas will be explained. In actual documents, areas with shapes other than rectangles also appear. Figure 3 (A
) and (B) are examples of regions having shapes other than rectangles. Further, (C) shows an example in which one rectangular area is separated into two rectangular areas. FIGS. 3A and 3B can be considered as the sum or difference of two rectangular areas, respectively, as indicated by broken lines.

In addition, (C) can be expressed simply if it is considered that two rectangular areas are connected and virtually combined into one rectangular area. To enable operations between rectangular areas like this,
Define virtual transfer of area as follows.

(lIlap&form F (spaca 'squeeze?H) (position (?XO?YO) ('?Xm1n
'? Xmax 7Ymj-n 7ymax))) 4゜5apc to explain the meaning of this virtual transfer using Figure 4
Is e the width as a new format F? W, height? A rectangular area of H is set to indicate that the transfer is to be performed within this area. p
position represents the upper left coordinates of the rectangular area of the transfer destination. 4 values (?Xm1n 7XIIIax':'Ymin ?y
The transfer source rectangular area indicated by max))) is copied to the above transfer destination.

By combining the virtual transfers described above, a region with a complex shape as shown in FIG. 3 can be expressed by calculations between two or more rectangular regions. For example, the third
Figure (A) can be expressed as two rectangular areas of different sizes transferred adjacently.

As can be seen from the above explanation, the document grammar proposed by the present invention understands the structure of a document as a combination of rectangular areas,
Expressing the relationship between rectangular areas using grammar increases the expressiveness of the document, and it is difficult to handle in cases where the number of lines in an area is undefined or whether a rectangular area appears is undefined. You can also describe the objects that existed. Therefore, a wide variety of documents can be analyzed.

[Embodiments of the invention]

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

FIG. 4 is a block diagram showing the configuration of an apparatus employing a document processing method according to an embodiment of the present invention.

Each part of the device is connected to bus 1, and the overall operation is controlled by control unit 2.
controlled by Information (character sequence) on the document 3 is input as a character code sequence from the keyboard 4 and stored in the memory 51 via the bus 1.

Memory 5] forms part of the memory 5 together with 52 to 55, which will be described later. Instead of obtaining the character code series 51 from the keyboard 4, it may be read from a file device such as a magnetic disk. Further, this character code series may be obtained by reading a character pattern on a sheet of paper using a character recognition device. Further, the character code series may include format control data.

By subjecting the input character sequence to a known transcription program process by the control unit 2, the transcription character sequence corresponding to the image of one page is stored in the memory 52 in the form of matrix data.

It is assumed that the format data of the target document written in accordance with the above-mentioned grammar is stored in the memory 53 in advance.

The control unit 2 uses this format data to perform the above-described document understanding process of the neat character series.

Here, the document understanding process refers to dividing the neat character sequence of the above-mentioned page image into a plurality of rectangular areas and classifying each of the rectangular areas. Of each area obtained as a result of document understanding, for an area predetermined as a search target area, the character sequence of that part is registered in the search data memory area 54 as search information. Data may be processed and edited upon registration. The input document search information obtained as described above is output to file 6, and the character code series of the document is output to file 7. When outputting the character code series of the document to the file 7, it may be output separately in units of a plurality of rectangular areas that are decomposed into one.

Further, file 6 and file 7 may be the same.

The details of the document understanding process will be described below. 6 and 7 are diagrams illustrating the flow of document understanding processing. The process flow is PAD (Program Analyzes
isDiagram) format. In step 101, the digit coordinates X(i) Y(i,) of each character code on the page image are extracted. Reference numerals 103, 104, and 105 indicate whether the initialization 2 main body of the parsing process is complete.

At 103, the format data stored in the memories 5 and 3 is copied to the working memory 55, and various tables and program internal variables are initialized.

The main body 104 of the syntax analysis process is composed of ], 06 to 120. Reference numeral 106 controls the processes 107 to 119 to be repeatedly performed until the end determination is made in 120.

At step 107, the statement in the format data is extracted.

An unprocessed statement refers to a line in which there are variables whose values have not been determined or whose corresponding document area has not yet been determined. 108 is 109 if there are no unprocessed statements remaining
~'', this is a determination to skip the process of step 19.

In this case, a termination determination will be made.

If the statement retrieved in 107 is an unprocessed statement, the processing in 109 to 117 cannot be performed (11
The processing of 0 to 117 is shown in FIG. ) 109 is a part that determines the type of statement and branches, and the processing in parts 110 to 117 changes depending on the type of statement. In Figures 6 and 7 and in the following description, the form statement, i.e. (form FO (?Xm1n ?Xmax ?Ymin ?ym
Although only the case of ax)5hrink) will be described, processing specific to that statement is similarly performed for other statements as well.

1 ], O to 119 are parts that process the predicate form. In step 110, it is checked whether the format name FO has been registered, and if not, in step 111, FO is registered in the format table. 111, the variable name ('i'Xm1n ?Xmax '7Ymin 7Ym
Check whether the character string written in position ax) is a variable or a numerical value, and if it is a variable, check whether it has been registered or not, and if it is not registered, register it in the variable table. If the variable has been registered, it is checked whether its value has been determined, and if it has not been determined, the forI11 process ends (in this case, this statement is not completed). If confirmed, rewrite the variable name in the statement with the above value.

As a specific example, ? Xm1n=O,? Xmax=90°? Ymin, 7Ymax: When unregistered, the bending statement is rewritten as (form FO (090?Ymin?Ymax)shrink) and the variable? Ymin and 7Ymax are registered in the variable table and their values are undetermined.

At step 112, a branch is made depending on whether or not all of the variable names in statements 1 to 1 have been rewritten to numerical values, and if all have been rewritten to numerical values, the form execution process at step 113 is performed. Details of the form execution process are represented by 114 to 118. ]-14 indicates that the following process is repeated for the character i extracted in step 102. 115, the X of the letter i
Are coordinates and Y coordinates X(i) Y(i) variables in the statement? Xm1n 7Xmax 'i'Ymin 'i
'Compared to the value corresponding to Ymax? Xm1n<X(i)<? Xmax?
Ymin<Y(i)<? It is determined whether the character Ymax is satisfied. In step 116, - when the above condition is met, the character i is registered in the component table of FO. At 117. , if a character that satisfies the above conditions does not exist, a parsing failure flag is set.

As explained above, by the processes 106 to 117, it is possible to detect that a structure corresponding to the statement form in the format data exists in the input character series. When string is specified in the form statement, whether or not the specified character data exists in the input character series can be determined by word matching. When matching words,
A synonym dictionary may be referred to, and if synonyms exist, they may be treated as the same. Statements other than form can also be analyzed in the same way. When variables are included in a statement, the parameters determined during analysis are assigned to the variables in the statement, and the results are used in other statements.

In step 118, the analysis failure flag is checked, and if the analysis fails, the process backs up and tries again. In this case, control is provided to return to the parsed statement, rewrite the variables to which the parameters were assigned to their previous state, and search for other possibilities.

In step 119, it is detected whether the analysis failure flag is not set or whether there is an analysis failure flag after going back and retrying.
Make a termination judgment.

105 is a part that receives and passes data obtained as a result of analysis to the outside. The data to be passed to the outside includes the coordinates on the document of the rectangular area detected corresponding to the format name.

When the analysis fails at a stage 1 in which an analysis failure flag is specified, the document is unintelligible, and in this case, the extract processing is performed again. For example, the final result or intermediate result of document understanding may be displayed on the display 8 and the keyboard 4 may be used for correction in a man-machine manner.

〔Effect of the invention〕

As explained above, according to the present invention, it is possible to automatically analyze target documents to be stored, and the need to input bibliographic information from a keyboard is eliminated or greatly reduced, reducing human labor. is extremely simplified. Another advantage is that even if the format of the target document changes, it can be handled immediately by changing the format data.

[Brief explanation of the drawing]

Fig. 1 is a reference diagram showing an example of a document, Figs. 2, 3 and 4 are explanatory diagrams for explaining the present invention in detail, and Fig. 5 shows the configuration of a device implementing the document processing method of the present invention. Block Diagram,
FIG. 6 is a flowchart for explaining the processing in the control section 2 in FIG. 5. ]...Bus, 2...Control unit, 3...Document, 4...
Keyboard, 5...Memory, 6,7...File,
8...Display.

Claims (1)

[Scope of Claims] 1. A memory that stores an expression written in accordance with a grammar that describes a document as a set of rectangular areas, a means for obtaining a character code series representing the document, and a method for obtaining a character code series representing the document; and a syntactic analysis unit that analyzes a document by searching for a rectangular area specified by an expression written according to the above grammar. A document understanding method characterized by generating data representing the meaning of a rectangular area. 2. In the document understanding method described in claim 1, the grammar includes a quantity representing the absolute or relative size of the rectangular areas and the absolute or relative relationship between the rectangular areas as variables. It also includes a description of a method for searching the rectangular area, and searches for a rectangular area specified by an expression written according to the grammar from the character code series,
Assign the value determined from the search result to the variable in the above expression,
A document understanding method characterized by having a syntax analysis unit that performs analysis until there are no unknown variables. 3. In the document understanding method described in claim 1, the document is characterized in that the grammar includes one virtual rectangular area generated by performing calculations from a plurality of spatially separated rectangular areas. understanding method.