JPH01205263A - Method for extracting referring construction - Google Patents

Abstract

PURPOSE:To efficiently process a document by allocating a diagram and presenting a necessary document without allowing a user to take a trouble separately to designate a referring construction. CONSTITUTION:The title method is provided with a referring point candidate extracting part 8 for analyzing sentence data and extracting a referring point candidate, a referring key candidate extracting part 9 for analyzing reference information such as the diagram, referring document and sentence and extracting a referring key candidate made into the key of a reference, and a referring construction deciding part 7 for checking the referring point candidate and the referring key candidate and deciding the referring point and the referring key to mutually correspond. Consequently, the point to have ambiguity, in which the diagram, document and preceding chapter and paragraph are referred in the original, is extracted, the referring construction to exist between a substance to be referred and it can be correctly extracted, and based on the extracted referring construction, the diagram can be arranged and the necessary reference information can be properly displayed without giving a load to an operator. Thus, document processing efficiency can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a reference structure extraction method for extracting ambiguous reference structures.

(Prior art) This is a system that extracts locations such as "Fig.

There is Japanese Patent Application Laid-Open No. 61-21570.

However, as a reference structure in a text, m
Intra-sentence references that refer to the chapter or section that appears in the text, such as ``as mentioned in Chapter 1,'' and references to bibliography, such as ``r ~ is a literature (in υ''). There are literature references.

Conventional document processing cannot extract these reference structures, so when document numbers or chapter/section numbers shift due to document insertion or text correction, it is possible to A problem arose where the numbers no longer matched correctly.

Also, refer to Figures 1 and 3.
(See Figures 2 and 3), "In Figures 1 to 3"
2, 3 or 1 to 3), the reference structure could not be correctly extracted.

Similarly for document references, r~ is document (1, 2).
(See References (G) and ■), and the same can be said.

(Problem to be Solved by the Invention) Conventional document processing systems cannot extract intra-sentence references or literature references contained in a text, and the user has to correct discrepancies in numbers one by one.

Regarding chart references, ambiguous and complex reference structures could not be extracted correctly, and charts could not be allocated correctly.

The present invention was made against this background.
The purpose is to extract various kinds of reference information contained in text data, and to enable the layout of figures and tables, numbering of documents, etc.

[Structure of the invention]

(Means for Solving the Problems) The present invention provides a reference point extracting means for analyzing text data and extracting reference point candidates, and a reference point extraction means for analyzing text data and extracting reference point candidates, and analyzing reference information such as figures, tables, reference documents, or texts to extract reference point candidates. A reference key extracting means for extracting a reference key candidate serving as a key, and a reference structure extracting means for collating a reference location candidate and a reference key candidate to determine mutually corresponding reference locations and reference keys. It is characterized by the presence of

(Function) In the present invention, based on the reference structure that refers to figures, tables, documents, etc. in the 7-text data, it is possible to allocate figures and tables to the vicinity of reference points, unify presentation numbers of referenced documents, etc. It has a shaping processing section.

Therefore, even if the user does not specify the reference structure one by one,
You can allocate figures and tables, present necessary literature, etc.
One document can be processed efficiently.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a general configuration diagram of a document processing system according to a first embodiment of the present invention.

An input unit 2, which is operated by an operator and is comprised of a keyboard, a mouse, or the like, is used to input reference information such as original text data and charts. Original text data and reference information input from the input section 2 are stored in the document storage section 1.

The document management section 5 sends the original text data and reference information input from the input section 2 or stored in the document storage section 1 to the display control section 4, and displays them on the output section 3 as shown in FIG. , to the reference structure extraction unit 13 to extract reference structures.

Next, the moving object of the reference structure extraction unit 13 will be explained using FIG. 2(a) as an example.

If the data sent from the document management unit 5 is original data, the reference structure extraction unit 13 extracts the reference point candidate extraction unit 8.
If it is reference information, it is sent to the reference key candidate extraction unit 9.

The reference point candidate extraction unit 8 is stored in the reference point extraction rule dictionary 11. For example, by taking the reference point extraction rules described in the format shown in Figure 4 and the reference terms and reference symbols stored in the reference reserved word dictionary IO in the format shown in Figure 3, for example, Reference point candidates are extracted according to the following flow.

For example, the format of Figure 3 is that the word "figure" is a reference term that refers to a figure, and raj2Q
The category J is the area before and after the number, that is, "Figure 1"
It shows that there is a form called "(Figure) 1".

Referring to the same figure, C: The word FigJ is in the rforJ category, indicating that only the forms in front of the number, rFig, IJ, are possible. The rbetJ category for the term "[]" which is a literature reference is between the terms (
This shows that only forms such as r (1, 21J) in which a number can be entered only in between) are possible.

The reference symbol category runiJ indicates that there are no candidates. In other words, only the reference key in the format "3.1" matches the reference location "3.1 Figure".

rsplJ indicates that candidates are created by dividing the number. In other words, for the reference ``1 Figure 1, 3'', two reference point candidates ``Figure 1'' and ``Figure 3'' are extracted.

rconJ shows that it creates candidates that complement the continuity of numbers. In other words, for the reference "Figures 1 to 3", three reference location candidates are extracted: r-Figure IJ 1 Figure 2'' and ``Figure 3''.

The 1st Oth sentence in Figure 2 (a) matches rules ■ and ■, the 11th sentence matches rules ■ and ■, the 12th sentence matches rules ■ and 20, and the 11th sentence matches rules Reference point candidates as shown in FIG. 6 are extracted.

FIG. 6 is composed of column start and end positions where reference point candidates exist, and reference attributes and patterns.

In the 10th sentence, the category of the reference symbol "-" is rconJ
and runiJ, so the first candidates are "Figure 1", "Figure 2", and "
The second candidate is runiJ.
It is extracted as a pattern that refers to "Figure 1-3".

Similarly to the reference point candidate extraction section 8, the reference key candidate extraction section 9 also uses the reference key extraction rules stored in the reference key extraction rule dictionary 12 described in the form as shown in FIG. 7 and reference reserved words, for example. Reference key candidates are extracted based on the reference terms stored in the dictionary IO.

For example, the extraction result for the reference information shown in Figure 8 is that the 8th sentence matches rules A and D, the 23rd and 24th sentences match rules A and G, and so on as shown in Figure 9. become.

Figure 9 shows that there is a reference key) found in the 8th sentence, the reference attribute is ``r diagram'', the pattern is ``2'', and the starting sentence of the figure is 1 and the ending sentence is 8. ing.

Next, the reference structure determination unit 7 determines whether the reference pattern of the reference point candidate (Fig. 6) matches the reference pattern of the reference key candidate (Fig. 9), and determines the reference structure only for the matched reference pattern. and writes it into the reference structure storage unit 6.

Since there is no reference pattern that matches the second candidate of the 10th sentence in FIG. 2(a) and there is only a pattern that matches the first candidate, the second candidate is deleted.

As a result, the reference structure becomes as shown in FIG.

There are three references in the 10th sentence, so three cells are created. Each cell has, for example, a statement number, an attribute for reference to the start and end position, a pattern, a pointer to the entity it is referring to, a pointer to the cell with the next younger pattern which is a sibling relationship between the referenced places, and the same which is a twin relationship. It has a pointer to the cell that has the pattern. For example, in the 10th sentence, ``Figure 1'', there is a pointer as the reference to ``Figure 3'' in the 10th sentence, a pointer to the reference entity, and a pointer to the 12th sentence.
Twin pointers to the sentence 1, Figure 2 are stored.

Once the reference structure is determined in this way, the document management section 5, for example, sends an instruction to the display control section 4 to display the extracted reference section in reverse video based on the contents of the reference structure storage section 6. As a result, the display on the output section 3 becomes as shown in FIG. 2(b).

For example, when the input section 2 instructs the inverted display section of FIG. 2(b) to display reference information, the entity of FIG. 2, for example, is displayed as shown in FIG. 2(c).

In the above embodiment, the dictionary, rules, and reference structures as shown in FIGS. 3, 4, 6, 7, 9, and 10 have been explained, but the present invention is not limited to this. It is not something that will be done.

For example, it is possible to modify the reference location "FIG. 2(a)" to extract two types of candidates, "FIG. 2" and "FIG. 2(a)."

Further, it is also possible to perform control such that extraction of reference point candidates and extraction of reference key candidates are performed in parallel in separate processes.

Alternatively, it is also possible to perform control such that reference key candidates are extracted in advance, and reference point candidates are extracted and the reference structure is determined at the same time.

Furthermore, the determined reference structure can also be changed according to the user's specifications. (Figure 2 (C)) Based on the reference structure, the reference number is changed from “Figure 1” to “Figure 1” or rFi
It is also possible to change it to gure tJ, etc., or change it to a serial number for the entire document or a serial number closed within the hall.

Furthermore, if the reference numbers are shifted due to correction of the original text data, it is possible to unify the reference numbers according to the sibling relationship between reference parts.

Also, when assigning reference information such as charts to the vicinity of a reference point, the first twin relationship (the 10th
It is also possible to assign it to the 10th sentence of the diagram ("Figure 2") or conversely to the end (12th sentence, 1 Figure 2). It is also possible to have the user specify which one to allocate to.

〔Effect of the invention〕

As described above, according to the present invention, ambiguous passages in the original text that refer to figures, tables, documents, and the aforementioned chapters and sections are extracted, and references between the referenced entities are extracted. The structure can be extracted correctly. Based on the extracted reference structure, charts can be arranged and necessary reference information can be displayed as needed without placing a burden on the operator, which is very effective.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of an embodiment of the present invention, FIG. 2 is a diagram showing an example of a screen display, FIG. 3 is a diagram showing an example of a dictionary, and FIG.
The figure shows an example of a reference point extraction rule, Figure 5 is a flowchart of applying the rule, Figure 6 shows an example of reference structure storage by the reference point candidate extractor, and Figure 7 is an example of a reference key extraction rule. Figure 8 is a diagram showing an example of displaying reference information, Figure 9 is a diagram showing an example of displaying reference information.
The figure shows an example of reference structure storage by the reference key candidate extraction unit, and FIG. 10 is a diagram showing the state of cells of the reference structure. 1...Document storage section 2...Input section 3...
・Output section 4...Display control section 5...
Document management section 6...Reference structure storage section 7...
・Reference structure determination section 8... Reference point candidate extraction section 9... Reference key candidate extraction section 10... Reference reserved word dictionary 11... Reference point extraction rule dictionary 12... Reference key extraction rule dictionary 13 ...Reference structure extraction department agent Patent attorney Nori Chika Ken Yudo Mitsuyuki Matsuyama Fig. 1 Subexamination rejection (a) (b) (C) Fig. 2 Fig. 3 Tsuk〉: Ryo F ~ Police officer stationed C3 []: Morning-Island 4th. Ko No. 5N [1: Final daughter No. 9 No. 7 Figure 8 Figure 9 Figure 10

Claims (3)

[Claims] (1) An input means for inputting text data and reference information; a reference point extracting means for extracting reference points quoting reference information from the text data according to the degree of ambiguity; a reference key extracting means for extracting key information serving as a reference key; and determining a reference structure by comparing reference point candidates extracted by the reference point extracting means and reference key candidates extracted by the reference key extracting means. 1. A reference structure extraction method comprising: a reference structure determining means for determining a reference structure; and an output means for appropriately formatting and outputting the text data based on the reference structure determined by the reference structure determining means. (2) The reference structure extraction method according to claim 1, wherein the output means displays the presence of the reference structure in the text data by display attributes such as inversion and high brightness. (3) The reference structure extraction method according to claim 2, wherein the input means specifies, deletes, corrects, etc. the reference structure by specifying the reference structure displayed by the output means.