JP3307336B2 - Document search method, document search device, and recording medium storing document search program - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document search method, a document search apparatus, and a recording medium on which a document search program is recorded, and more particularly, to a document search method useful for a text including an error such as after character recognition of a document image. In addition, the present invention relates to a document search device and a recording medium storing a document search program.

[0002]

2. Description of the Related Art A document printed on paper is digitized by an optical scanner, and is further subjected to character recognition (OCR: Optical Chassis).
racter Reader), and a document image filing system that saves the image has been developed and is already a product. As a very large-scale system, the development of electronic library systems and the like is also progressing. In such a system for storing and retrieving digitized document images, the text after character recognition is stored in association with the document image and used as a search index. In order to obtain a desired one from the stored document images, the user inputs a search formula in which an arbitrary character string is combined with a logical sum or a logical product operator, and issues a search request. In the search process, the text after character recognition is searched based on the existing character string collation technology, and as a result, a document image that seems to contain a character string that satisfies the search request, or a text after character recognition of the document image is used. Output.

However, in general, the result of OCR includes an error. The cause is not only the problem of the limit of the recognition technology, but also various factors such as the resolution of the image and the inclusion of noise, and cannot be avoided by the current technology. Therefore, if the full-text search is performed on the text after the character recognition based on the exact match of the character string in the input search expression as in the case of the ordinary text, there is a problem that a search omission occurs.

As a conventional technique that can solve the problem of search omission, as a conventional technique, each character of a word included in a search expression (hereinafter, “search expression constituent word”) is confused by character recognition (hereinafter, “similar character”). ) To create a group of character strings (hereinafter “similar character strings”), combine them with a logical sum to generate a new search expression (hereinafter “similar character expansion search expression”), and replace it with the original search expression There are ways to prevent searching. Japanese Patent Application Laid-Open No. 4-158478 discloses "information search method and information storage device", and Japanese Patent Application Laid-Open No. 6-19538.
No. 7, "Document search device", Japanese Patent Application Laid-Open No. 7-152774.
Publication "Document Searching Method and Apparatus", JP-A-8-18
0064, “Document Search Method and Document Filing Device”.

[0005]

However, the number of similar character strings formed by expanding the words constituting the search formula with similar characters increases exponentially with respect to the length of the original word. For a word with 10 characters, if each character has an average of 5 similar characters (including yourself), then 5
, That is, 9865625 similar character strings, and the similar character extended search expression is the original that is connected by OR. Since the processing time and the amount of processing memory in the search are generally proportional to the number of words and characters constituting the search expression, the search processing using the similar character extended search expression greatly wastes the search time and the amount of processing memory. Will be.

On the other hand, Japanese Patent Application Laid-Open No. Hei 8-1
The technique described in “Document Search Method and Document Filing Apparatus” of Japanese Patent No. 80064 describes that control is performed so as not to generate a certain number or more of similar character strings. However, a method that does not generate some similar character strings leaves a risk of missing search.

[0007] In view of such problems of the prior art, an object of the present invention is to suppress the increase in the number of similar character strings even if a long character string is replaced with many similar characters, and as a result, processing time and processing memory capacity are reduced. An object of the present invention is to provide a document search method, a document search device, and a storage medium storing a document search program that are capable of suppressing an explosive increase in the number of documents and that do not significantly reduce search accuracy, and that are comfortable for users. .

[0008]

According to the present invention, in order to attain the above object, before forming a similar character string group from search formula constituent words, the search formula constituent words are divided based on a predetermined rule. , A partial search expression is created by combining the divided character strings with a logical product, and a “character string shift division unit” that replaces the original search expression constituent words with the created partial search expression is provided. .

[0009] Further, in order to enable a balance between search speed and accuracy to be adjusted, a parameter for a splitting process performed by the character string shifting and splitting means is determined.
Was provided. The shift division constant determining means is executed with reference to at least one of a search formula input by the search request input means, a progress of the search processing obtained by the search processing monitoring means, or characteristic information of the search data. .

[0010]

Next, embodiments of the present invention will be described with reference to the drawings.

(First Embodiment) FIG. 1 is a block diagram showing a configuration example of the first embodiment of the present invention.

It is assumed that the document image information to be searched and the text resulting from character recognition of the document image information have already been stored in the search data 1a. For the data structure of the recognition result, the storage format of the image, or the correspondence between the recognition result and the image, the conventional technology in this field can be used as it is. In addition, the image and the result of recognizing the image are associated with each other on a one-to-one basis, and they are unique information (for example, identification number).
It can be specified by. The search result for the search expression returns information (for example, an identification number) uniquely indicating a document image (which may be composed of a plurality of images) including a character string satisfying the search expression. .

The user operates the search request input means 1b to
Enter a search expression as a search request. FIG. 2 is an example of the search formula. 2a is an entire search formula, and 2b is an example of a word constituting the search formula, that is, an example of a search formula constituent word.
In this example, three words "computer", "computer", and "system" are connected by parentheses, "+", and "*". “+” Indicates a logical sum in a search expression,
That is, if either of the expressions on both sides of “+” is satisfied, the search is regarded as successful. “*” Is a logical product in the search expression, that is, if both of the expressions on both sides of “*” are satisfied at the same time. , The test is considered to be successful. Parentheses are
It indicates the priority of the operation in the search expression. This figure 2
In the example, a request is made to search for a document image containing either “computer” or “computer” and containing the word “system”.

The character string shift division unit 1c first divides each search expression constituent word included in the search expression input by the search request input unit 1b according to a predetermined rule. Division may be performed with overlap. Considering the search accuracy, it is preferable that there is an overlap. FIG.
Is a result obtained by dividing each search formula constituent word included in the example of the search formula shown in FIG. 4 into a character string having a division length of 3 and a shift width of 1. Here, the division length is the length of the character string created as a result of division, and the shift width is the number of characters to shift the next division start position with respect to the character string cut out immediately before being divided. means. In this example, for example, for “computer”, “compilation”, “computer”, “pew”, “computer”
Are cut out.

Next, the character string shifting / dividing means 1c combines the character strings formed by the division for each of the search expression constituent words by a logical product, thereby replacing the original search expression constituent words. FIG. 4 is an example in which the division examples of the “computer” shown in FIG. 3 are combined by logical product. A new search expression is created by replacing the original search expression constituent word “computer” with the created partial search expression. The same processing is performed for the other search formula constituent words, “computer” and “system”. FIG. 5 is based on the examples of FIG. 3 and FIG.
An example of an output finally created by the character string shifting and dividing unit 1c is shown.

In the similar character expanding means 1d, each search expression constituent word in the new search expression output from the character string shifting and dividing means 1c is expanded by OCR into a character which is likely to be erroneous to create a similar character extended search expression. I do. FIG. 6 is an example of a similar character table that stores characters that are easily erroneous in OCR. This indicates that when the OCR attempts to recognize the character on the left side, a plurality of characters on the right side often appear as errors. For example,
OCR indicates that the character "ko" is apt to be mistaken for "yu, u, mouth, b". This table is OC
It can be created through R recognition experiments. For example, a confusion matrix (confus
What is necessary is just to list and register the values of the ion matrix) that are not less than a certain value.

The similar character expanding means 1d generates all character strings that can appear due to an OCR error with reference to the similar character table for each search expression constituent word of the search expression. Specifically, each character included in the search expression constituent word is replaced with a similar character registered in a similar character table to create a similar character string. FIG. 7 shows an example of a similar character string generated for each of "compi", "mpu", "pyu", and "tuta". These are examples created with reference to the similar character table of FIG.

The number of similar character strings generated by replacing one search formula constituent word with similar characters depends on the number of similar characters of each character constituting the search formula constituent word (self, for example, “ "A" is counted as a similar character, including "A"). For example, in the similar character table of FIG. 6, five characters including “self”
, And six similar characters are defined for "pi". Therefore, the number of similar character strings generated for “compilation” therefrom is 150 as their product.

In the similar character developing means 1d, after the similar character string group is created, the similar character string groups are combined by logical OR for each search formula constituent word, and are replaced with the original search formula constituent words. FIG.
Is an example of a partial search expression in which similar character strings generated by the similar character expansion means 1d from "compi" are combined by a logical sum, and this replaces the search expression constituent word "compi". Remaining character strings generated by character string shift division for “computer” included in the search request entered first, “mpu”, “pyu”, “uta”
The same processing is performed for That is, a group of similar character strings is created, and a search expression combined by a logical sum is created and replaced with the original search expression constituent word. Further, the same applies to a character string generated from the character string shifting / dividing means for the words "computer" and "system" included in the search request input first. FIG. 9 shows an example of a search formula finally generated by the similar character expanding means 1d. As for "computer", the character string was not divided by the character string shifting / dividing means 1c, nor was any similar character found by the similar character developing means 1d, and the character string was the same as the word of the input search formula.

In the search means 1e, the similar character developing means 1d
The search data 1a is again generated by the newly generated search formula.
Search for the text of the recognition result included in. For this part, the conventional full-text search technique for full-text search can be used as it is. As the search result, information for uniquely identifying the searched document image is obtained. If a plurality of documents meet the search condition, information about all documents is output.

The result processing means 1f extracts necessary information from the search data 1a based on the search result of the search means 1e, and outputs the search result. There are several possible output forms depending on the user's desire, such as outputting an image to a display, outputting text after OCR, or outputting both.

The number of character strings and the number of characters constituting the retrieval formula created by the similar character developing means 1d are determined by the length of the original character string developed using similar characters and the number of similar characters used for replacement. Depends heavily. In particular, the length of a character string increases in the order of a power. Generally, the processing time and the processing memory increase in proportion to the number of words and the number of characters in the search formula, and this is a serious problem.

In the present embodiment, the character string shifting / dividing means 1c sets the division length to 3, so that a character string having a maximum length of 3 is processed by the similar character developing means 1d. Of course, it is not necessary to fix the division length to 3 in this example, and it is also possible to set a longer or shorter value. For example, FIG. 10 shows an example in which the division length is 4 and the shift width is 1.

In general, when the division length is shortened, the number of character strings formed by expansion using similar characters certainly decreases, and the retrieval processing itself becomes faster, but a document which should not be originally detected is detected. A case arises. For example, in an extreme case, when the division is performed with the length being 1 and the margin width being 1, the number of similar character strings formed for one search formula constituent word is determined by the number of all characters in the search formula constituent word. Is the total number of similar characters. Therefore, the explosive increase in the number of similar character strings does not occur, but the partial search formula created thereby implies a search for documents containing any of the characters. In other words, if the division is performed in a unit that is too short, the positional relationship information of the characters included in the search expression constituent words is lost, and extra search results that should not be originally searched are generated.

Conversely, if the division length is set to be long, excessive search is reduced, but search time is increased due to an increase in the number of similar character strings, and memory is wasted. That is, the two are in a trade-off relationship with the length to be divided, and by appropriately selecting the length, the processing amount can be suppressed and high search accuracy can be maintained.

Although the above is a discussion on the division length, the same applies to the shift width. If the shift width is large, the character itself or the positional relationship information of the character included in the original search formula constituent word is lost, so that extra search results increase. However, there is an advantage that the number of words or characters of the search expression included in the similar character extended search expression finally formed is reduced, and the search process is sped up.
If the shift width is increased, the opposite characteristic is obtained, and this is also a trade-off relationship. It is desirable to determine what kind of division is to be performed by considering various characteristics of the search in advance or during the search.

(Second Embodiment) FIG. 11 is a block diagram showing a configuration example of a second embodiment of the present invention. In the configuration example of the first embodiment shown in FIG. 1, a search division monitoring unit 11h is connected to the search unit before the character string shift division unit 11c. . For the operation of the other components, see FIG.
Since it is the same as the first embodiment shown in FIG.

The shift division constant determining means 11g has a role of determining operation parameters related to the processing performed by the character string shift division means 11c based on the contents of the search request, the progress of the search processing, and the characteristics of the search data. is there.

FIG. 12 shows a shift division constant determining means 11g.
It is an example of the processing flow of. Here, it is assumed that the search formula is generated from n search formula constituent words. Then, a division length Wk (1 ≦ k ≦ n) and a shift width Sk (1 ≦ k ≦ n) are determined for each search formula constituent word.

First, Wk and Sk are initialized. FIG.
Is an example of a flowchart of the initialization processing 12b. k is 1 to n
While changing this, Wk is set to 3 and Sk is set to 1. That is, for any search expression constituent words, the shift division by the division length 3 and the shift width 1 is set as the initial state. Of course, another value may be set.

In 12c of FIG. 12, Wk and Sk are partially changed as necessary in consideration of information relating to a search, such as a search request, a status of a search process, and characteristics of search data. The search speed and the search accuracy are appropriately adjusted accordingly.

FIG. 14 is an example of a flowchart of the correction processing of Wk and Sk shown in 12c. In this example, the parameters used for the shift division processing are changed with reference to the search request input by the user. In general, katakana characters are simple figures and have many similar shapes, making character recognition difficult. Therefore, the number of similar characters tends to increase, and there is a risk that the number of similar character strings may explosively increase when Wk is increased. Therefore, when the search expression constituent word is a katakana string, the number of similar character strings is suppressed by setting Wk to be short by one and Sk to be long by one. On the other hand, a kanji character string has characteristics opposite to those of katakana, so that even if Wk is increased, the number of similar character strings is unlikely to increase so much.
Therefore, Wk is lengthened by one and Sk is also shortened by one. However, when Sk is 0, the shift division is not established, so that the value is set to 1 or more. In other cases, Wk and Sk set as initial values are left as they are.

FIG. 15 is also an example of a flowchart of the correction process of Wk and Sk shown in 12c. This is also a process that refers to the search request. When the words of the search formula are connected by a logical sum, either of them may be satisfied, and the search result tends to increase. At this time, if Wk is too short, the possibility that any of the short similar character strings is detected in the document increases, and the possibility that an unnecessary document is detected further increases. Therefore, in the case of a search request combined by a logical sum, an excessive search for unnecessary documents can be prevented by lengthening Wk when dividing search expression constituent words at both ends thereof.
Conversely, if words in the search formula are combined by logical OR, the search results tend not to increase very much. Therefore, by shortening Wk and reducing the number of shifted divided similar character strings to be expanded, the accuracy is improved. The processing speed can be increased without significantly affecting the processing.

FIG. 16 also shows Wk and Sk shown in FIG.
13 is another example of the correction processing of FIG. This refers to the progress of the search process and adjusts the division method. This example is effective when one search system is shared by a plurality of users via a network or the like. In the process, the progress of the search process is received from the search process monitoring means 11h in FIG. 11, and if the search process is currently executing the search process of another user, or is waited for the next search process. If there is a search request, the number of similar character strings generated is suppressed by shortening Wk and lengthening Sk. As a result, it is possible to improve the response (turnaround time) of the search result to the user.

FIG. 17 shows another embodiment 12c. This is characterized by referring to the characteristics of the search data. First, as an example of the characteristics of the search data, the content rate of the character type is acquired. If the search formula constituent word is a katakana string and the search data has a certain katakana content ratio or more, a parameter is set so that Wk is lengthened and Sk is shortened to prevent excessive erroneous search of unnecessary documents. Has changed. A high content rate of katakana in the search target document means that there are many katakana character strings,
If Wk is short, there is a high risk that unnecessary documents will be erroneously searched, and therefore, it is necessary to address the problem.

As described above, Wk and Sk at 12c in FIG.
As examples of correction processing, two examples referring to a search request,
One example referring to the progress of the search process and one example referring to the characteristics of the search data are shown separately, but these processes can be connected in multiple stages and applied to the process of the same search formula. Good.

In order to realize the present invention by a computer, for example, in the above-described first embodiment, the character string shift division unit 1c, the similar character expansion unit 1d, and the search unit 1e are provided inside the computer. Computer program that implements the functions of a computer, etc., and stores the computer program on a CD-ROM.
The effects of the present invention are not lost even in a form provided by being recorded on a recording medium represented by a ROM, a floppy disk, or a semiconductor memory. The same applies to the second embodiment.

[0038]

According to the present invention, by performing similar character expansion after dividing search expression constituent words in a search expression, the number of similar character strings created by similar character expansion means can be greatly reduced.
For example, for a character string of length 10, 5
If there are two similar characters, if similar character expansion is performed as it is, a similar character string having a length of 10 will be 5 to the 10th power, and the number of characters will be 97656250. This causes a problem that a large processing time and a large amount of processing memory are required. On the other hand, when a similar character string having a length of 3 is divided into 3 * 8 * 5 by performing expansion using similar characters after dividing by a division length of 3 and a shift width of 1.
The power is 1000, and the number of characters is 3000. As a result,
The search time is reduced, and the amount of memory required for processing is reduced.

However, in the present invention, since the positional relationship information between the character strings is partially lost due to the division, there is a danger that a text that should not be originally searched may be over-searched. On the other hand, in experiments conducted on Japanese patents, it was found that if a division was made with a division length of 3 or more and a shift width of 1, an excessive search would be suppressed to 1% or less, and under certain conditions this would not be a major problem in practice. Confirmed by experiment. of course,
Some parts depend on the type of document and the performance of OCR, but by taking into account those characteristics and determining rules and constants for division, high-speed search has almost no effect on search accuracy and existing full-text search can be performed. This can be realized using an engine.

Also, in the present invention, the division length and the shift width at the time of performing the shift division are adaptively determined during the search process according to the information related to the search, for example, the search request, the progress of the search process, and the characteristics of the data to be searched. adjust. As a result, the balance between processing time and processing speed is automatically set appropriately,
It is possible to provide a user with a stress-free search environment.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an example of a configuration according to a first embodiment of the present invention.

FIG. 2 is an example of a search formula input via a search request input unit.

FIG. 3 is an example in which a search formula constituent word is divided by a division length of 3 and a shift width of 1;

FIG. 4 is an example of a partial search formula generated from a divided character string.

FIG. 5 is an output example of a character string shifting / dividing unit 1c.

FIG. 6 is an example of a similar character table.

FIG. 7 is an example of a similar character string formed by similar characters.

FIG. 8 is an example of a partial search formula generated from a similar character string.

FIG. 9 is an output example of the similar character developing means 1d.

FIG. 10 is an example in which a search-type national census word is divided by a division length of 4 and a shift width of 1;

FIG. 11 is a block diagram illustrating an example of a configuration according to a second embodiment of the present invention.

FIG. 12 is an example of a flowchart of a shift division constant determining unit 11g.

FIG. 13 is an example of a shift division constant initialization process.

FIG. 14 is an example of a shift division constant correction process.

FIG. 15 is an example of a shift division constant correction process.

FIG. 16 is an example of a shift division constant correction process.

FIG. 17 is an example of a shift division constant correction process.

[Explanation of symbols]

 1a, 11a Search data 1b, 11b Search request input means 1c, 11c Character string shift division means 1d, 11d Similar character expansion means 1e, 11e Search means 1f, 11f Result processing means 11g Shift division constant determination means 11h Search processing monitoring means 2a Search formula 2b Search formula constituent words

Continuation of the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 17/30 JICST file (JOIS)

Claims (12)

(57) [Claims] (1)Search described by logical OR or logical AND of words
A document search method that searches for text using an expression
hand, Each word that constitutes the search expression is referred to by referring to a predetermined rule.
First to divide into overlapping substrings of length 2 or more
Steps and Discuss the character string formed by the first step.
A second step for creating a partial search expression joined by logical operations
And Each constituent word in the search formula created in the second step
For each of the letters that make up that word
Create a group of similar strings replaced with characters, and
A third step for creating a search expression in which the column groups are combined by OR
And Using the search formula created in the third step, the text
A fourth step of searching for a kist, The rule referred to in the first step is set in the search request.
A fifth step of determining based on the content; Including
Characteristic document search method. (2)Search described by logical OR or logical AND of words
A document search method that searches for text using an expression
hand, Each word that constitutes the search expression is referred to by referring to a predetermined rule.
First to divide into overlapping substrings of length 2 or more
Steps and Discuss the character string formed by the first step.
A second step for creating a partial search expression joined by logical operations
And Each constituent word in the search formula created in the second step
For each of the letters that make up that word
Create a group of similar strings replaced with characters, and
A third step for creating a search expression in which the column groups are combined by OR
And Using the search formula created in the third step, the text
A fourth step of searching for a kist, Already require the rule referenced in the first step
Search process Based on the degree of congestion inside the search process
And the fifth step Characterized by containing
Document search method. (3)Search described by logical OR or logical AND of words
A document search method that searches for text using an expression
hand, Each word that constitutes the search expression is referred to by referring to a predetermined rule.
First to divide into overlapping substrings of length 2 or more
Steps and Discuss the character string formed by the first step.
A second step for creating a partial search expression joined by logical operations
And Each constituent word in the search formula created in the second step
For each of the letters that make up that word
Create a group of similar strings replaced with characters, and
A third step for creating a search expression in which the column groups are combined by OR
And Using the search formula created in the third step, the text
A fourth step of searching for a kist, The rule referred to in the first step is a search target.
Fifth Decision Based on Characteristic of Text
When A document search method comprising: 4. When a text containing an error mainly after character recognition is searched for using a search expression described by a logical OR or a logical product of words, each text in the search expression is searched for. Create a similar character string group in which each character constituting the word is replaced with a similar character that is apt to be erroneous in the character recognition, and further create a partial search formula in which the similar character strings are combined by logical OR, and In a document search method in which a search is performed after replacing each word in the original search formula using a simple search formula, each word constituting the search formula is referred to by referring to information related to the search before replacement by similar characters Is determined according to the parameters related to the process of dividing the word, and each word constituting the search formula is divided into partial character strings having a length of 2 or more that allow overlapping, and the resulting character strings are logically ANDed. Partial search joined by Create a document search method characterized by previously replacing each word in the original search expression using the partial search expression. 5. The document search method according to claim 4, wherein the information referred to when determining the parameter is a characteristic of a search formula. 6. The characteristic of the search expression is at least one of the character type or length of each word constituting the search expression, or the type of a logical operator constituting the search expression. The document search method according to claim 5 . 7. The document search method according to claim 4, wherein the information referred to when determining the parameter is a congestion degree in a search process of a search process that has already been requested. 8. The document search method according to claim 4, wherein the information referred to when determining the parameter is a characteristic of data to be searched. 9. The document search method according to claim 8, wherein the characteristic of the data to be searched is a content rate by type of a character included in the data to be searched. 10.Search request input means for inputting a search expression;
Refer to the rules that determine each constituent word included in the search expression in advance
And split it into substrings of length 2 or more that allow overlap,
Search for the original by combining the divided substrings with logical product
Create a second search expression that replaces the constituent words included in the expression
Character string shifting and dividing means; For each constituent word in the second search expression,
Similarity where each character is replaced with a similar error-prone character
Create a group of strings and combine the similar strings with a logical OR
Similar character expansion means for creating a third search expression; A search for performing a character string search according to the third search expression
Means, The rule referred to by the character string shifting / dividing means is
Characteristics, already required Search process inside the search process
Congestion degree, the characteristics of the data to be searched
Also determines the shift division constant using any one of them.
With steps and The character string shift division unit determines the shift division constant.
Each component included in the search expression according to the rules determined by the means
A document search device characterized by dividing words. 11.On the computer, Using a search formula described by the logical OR or logical product of words,
A record of a document search program for searching text
In recording media, The search formula is referred to a computer with reference to a predetermined rule.
Partial characters of length 2 or more that allow each word to be composed to overlap
A first step of splitting into columns; Discuss the character string formed by the first step.
A second step for creating a partial search expression joined by logical operations
And Each constituent word in the search formula created in the second step
For each of the letters that make up that word
Create a group of similar strings replaced with characters, and
A third step for creating a search expression in which the column groups are combined by OR
And Using the search formula created in the third step, the text
A fourth step of searching for a kist, The rule referred to in the first step is set in the search request.
A fifth step of determining based on the content; Run
Recording medium on which a document search program for recording is recorded. 12.On the computer, Each component included in the search expression entered via the input means
Words of length 2 or more that allow overlapping by referring to predetermined rules
Is divided into the above substrings, and the divided
Replaces constituent words in the original search expression by combining with products
A character string shift division function for creating a second search formula, For each constituent word in the second search expression,
Similarity where each character is replaced with a similar error-prone character
Create character strings and OR their similar character strings Combine with
A similar character expansion function for creating a third search expression, A search for performing a character string search according to the third search expression
Features and The rule referred to by the character string shift division function is
Characteristics inside the search process of the search process already requested
Congestion degree, the characteristics of the data to be searched
Is a shift division constant determining machine that uses one of them to determine
Noh and realize The character string shift division function determines the shift division constant.
Each configuration included in the search expression according to the rules determined by the function
Split words into substrings of length 2 or more that allow overlapping
Recording a document search program characterized by the fact that
Medium.