JP2001338265A - Device and method for recognizing pattern and computer- readable recording medium with program executed by for realizing the device and the method recorded thereon computer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device and a method for recognizing pattern, by which a pattern can be recognized by highly accurately and automatically selecting an optimal context dictionary out of plural context dictionaries at a high speed. SOLUTION: The method for recognizing pattern an inputted pattern for includes the steps of recognizing candidate character strings are recognized and a pattern evaluation value is calculated, each of recognition candidate character strings, a context evaluation value is calculated on the basis of at least two context dictionaries, and a total evaluation value is calculated on the basis of the pattern evaluation value and the context evaluation value, and a recognition candidate character string having the maximum total evaluation value is selected as a recognition character string. In the method, further, it is decided which of the context dictionary, which is the base of calculating processing of the context evaluation value, is valid and the pattern is recognized by using only the context evaluation value, which is calculated on the basis of the context dictionary decided as valid processing, as a context evaluation value.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern recognition apparatus and method for use in discriminating handwritten characters, voice recognition, and the like.

[0002]

2. Description of the Related Art In pattern recognition technology, after recognizing individual patterns of handwritten characters and voices, context information is obtained by a method of collating with a word dictionary or a method of using N-gram statistical information. The mainstream context processing was to output the most likely character string while using it.

In such a method, by using a context dictionary in a specific field, a large context processing effect can be obtained as long as the input contents belong to the field. Therefore, many methods for effectively performing context processing in various fields by using a plurality of context dictionaries created for each field are disclosed.

For example, Japanese Patent Application Laid-Open No. Hei 7-271921 discloses a method in which a device for selecting a context dictionary is provided, and a user selects a context dictionary to be used or sets priorities and then performs recognition. ing.

[0005] In addition, IEICE Technical Report (PRMU98-238)
In the “Online Character String Recognition Method without Frames that Integrates Segmentation, Recognition, and Language Confidence” disclosed in
A method is disclosed in which context processing is performed with reference to all context dictionaries, and a result having the highest evaluation value after context processing is output as a result.

[0006]

However, even if a plurality of word dictionaries and N-gram statistical dictionaries are prepared as context processing dictionaries, the effect of the context processing depends on the contents of the dictionaries. Therefore, it is indispensable that a context dictionary of a field that matches the input content be selected accurately and at high speed in accordance with the input content.

In addition, instead of specifying a context dictionary for each field, context processing is performed on a plurality of context dictionaries relating to different fields, so that a context processing result with the highest matching degree can be used as a recognition result. For example, in the method disclosed in Japanese Patent Application Laid-Open No. Hei 7-271921, the user can select a context dictionary or specify a priority to increase the effect of context processing. Each time is changed, it has to be re-specified, and there is a problem that the burden on the user is large. In addition, users should be aware of information about all available context dictionaries and, if they do not have sufficient knowledge of when to use each context dictionary, specify appropriate context dictionaries. You can't do it.

[0008] Also, IEICE (PRMU98-238)
However, in the method disclosed in JP-A-2003-264, it is necessary to always obtain a context evaluation value from all context dictionaries, and there is a problem that the recognition time increases in proportion to the number of context dictionaries.
Furthermore, if all pattern recognition results are to be output regardless of the context evaluation value, the context dictionary that gives the best overall evaluation value will vary depending on the pattern evaluation value even for the same input content, resulting in an incorrect context dictionary. There was also a risk of selecting.

Further, all context dictionaries must be created as those giving evaluation values on the same basis, and dictionaries having clear uses such as address context dictionaries are equally evaluated. There is also a problem that the features of each context dictionary cannot be reflected in the recognition result.

The present invention provides a pattern recognition apparatus and method capable of automatically selecting an optimum context dictionary from a plurality of context dictionaries at high speed and with high accuracy and performing pattern recognition in order to solve the above problems. The purpose is to provide.

[0011]

In order to achieve the above object, a pattern recognition apparatus according to the present invention recognizes a pattern input section for inputting a pattern, and a recognition candidate character string for the pattern input from the pattern input section. A pattern recognition unit that calculates a pattern evaluation value together with a context that receives a recognition candidate character string recognized by the pattern recognition unit and calculates a context evaluation value based on at least two or more context dictionaries for each recognition candidate character string A processing unit, a comprehensive determination unit that calculates an overall evaluation value based on the pattern evaluation value and the context evaluation value, and a recognition candidate character string having a maximum overall evaluation value based on the overall evaluation value calculated by the overall determination unit. A pattern recognition apparatus including an output determining unit that selects a recognition character string, wherein the context processing unit calculates a context evaluation value, A context calculated based on the context dictionary determined to be a valid process by the valid context determination unit, further comprising an effective context determination unit that determines whether a context evaluation value calculation process based on the pulse dictionary is valid. Pattern recognition is performed using only the evaluation value as a context evaluation value.

With this configuration, the context evaluation value can be calculated based only on the context dictionary that is valid for the calculation of the context evaluation value. Therefore, unnecessary processing for calculating the context evaluation value can be reduced, and the accuracy can be reduced. It becomes possible to recognize a character pattern well.

Further, the pattern recognition apparatus according to the present invention stores priority context information in which information relating to a processing priority of a process of calculating a context evaluation value using a context dictionary is recorded based on a determination result in an effective context determination unit. It is preferable to further include a priority context information management unit that manages the information. By performing only the process of calculating the context evaluation value based on the high-priority context dictionary, the useless process of calculating the context evaluation value can be omitted, and the recognition time can be reduced.

Further, the pattern recognition apparatus according to the present invention further includes a character string determination unit for performing a determination process by the user on the recognition character string output by the output determination unit. It is preferable to update the content of the priority context information based on the character string that has been subjected to the determination processing in the determination unit. This is because the recognition result expected by the user can always be reflected.

Further, the present invention is characterized by software for executing the functions of the above-described pattern recognition apparatus as processing steps of a computer, and more specifically, a step of inputting a pattern, and a step of inputting the pattern. Calculating a pattern evaluation value while recognizing a recognition candidate character string for the pattern; receiving the recognized recognition candidate character string; and generating a context evaluation value for each recognition candidate character string based on at least two or more context dictionaries. Calculating, a step of calculating an overall evaluation value based on the pattern evaluation value and the context evaluation value, and selecting a recognition candidate character string having the largest overall evaluation value as the recognition character string based on the calculated overall evaluation value A pattern evaluation method that includes a context evaluation value calculation process based on any context dictionary in the context evaluation value calculation process. And a pattern recognition method capable of performing pattern recognition using only a context evaluation value calculated based on a context dictionary determined to be a valid process as a context evaluation value, and a method thereof. It is a computer-readable recording medium that records such steps as a program.

With this configuration, the context evaluation value can be calculated based only on the context dictionary that is valid for the calculation of the context evaluation value. Therefore, unnecessary processing for calculating the context evaluation value can be reduced, and the accuracy can be reduced. It is possible to realize a pattern recognition device that can well recognize a character pattern.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A pattern recognition device according to an embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram of a pattern recognition device according to an embodiment of the present invention. In FIG. 1, reference numeral 11 denotes a pattern input unit having a pen input type handwriting input unit or the like in a part of the display screen, 12 denotes a pattern recognition unit, 13 denotes a pattern recognition dictionary, 14 denotes a context processing unit, Reference numeral 15 denotes a plurality of context dictionaries, 16 denotes an effective context determination unit, 17 denotes an output determination unit, and 18 denotes a recognition result output unit including the above-described handwriting input unit.

In FIG. 1, a pattern to be recognized is input from a pattern input unit 11. In the present embodiment, it is a handwritten character string. Then, the input pattern is transmitted to the pattern recognition unit 12.

In the present embodiment, a case where a handwritten character "Akashi" as shown in FIG. 2 is input as a pattern will be described. Of course, the present invention is not limited to this, and may be a pattern input such as a voice input. In FIG. 2, the input form is such that one hand-written character is input in a frame. However, the present invention is not limited to this, and a hand-written character string is input in an input part having no frame. It may be in a form.

The pattern recognizing unit 12 refers to the contents of the pattern recognizing dictionary 13 to collate the inputted patterns and outputs recognition candidate characters to the context processing unit 14 in order of accuracy. The commonly used OC is used to extract recognition candidate characters.
Any recognition engine such as an R recognition engine or an online recognition engine may be used. Also in the calculation of the certainty factor, various methods are conceivable, such as calculating the distance between the input pattern and the dictionary 13 for pattern recognition or calculating based on the feature points. FIG. 3 illustrates an example of a recognition result when the handwritten character “Akashi” illustrated in FIG. 2 is input as a pattern. In FIGS. 3A and 3B, the numbers displayed beside each recognition candidate character indicate the degree of certainty of each recognition candidate character, and the larger the number, the higher the degree of certainty. Is shown.

Then, depending on the combination of the recognition candidate characters,
For example, a recognition candidate character string with a pattern evaluation value as shown in FIG. 4 is generated. Actually, since the recognition candidate character is two characters, the recognition candidate character string is 2 × 2, and there are four candidates, but here, the direct comparison target of “Akashi” and “Tomo name” Two recognition candidate character strings are shown. In FIG. 4, the pattern evaluation value of the recognition candidate character string is represented by the sum of the degrees of certainty of each recognition candidate character. However, the present invention is not particularly limited to this, and is calculated using an evaluation function or the like. Or the like.

Next, the context processing unit 14 calculates a context evaluation value using a plurality of context dictionaries 15 based on the recognition result obtained by the pattern recognition unit 12, and outputs it to the comprehensive judgment unit 17. .

First, the first context dictionary 15 is read from the external memory to the internal memory when the recognition operation is started, and the matching between the recognition candidate character string and the character string registered in the context dictionary 15 read into the internal memory is performed. , The context evaluation value for the recognized candidate character string is calculated for the collated character string.

Next, the other context dictionaries 15 are sequentially read into the internal memory, and the context evaluation value is similarly calculated. Therefore, different context evaluation values are calculated for the same recognition candidate character string according to the features of the context dictionary.

For example, as shown in FIG. 5, it is assumed that a general context dictionary C1 and an address context dictionary C2 are prepared. Then, a context evaluation value using the context dictionary C1 and a context evaluation value using the context dictionary C2 are obtained for each candidate character string.

In the present embodiment, based on the recognition result obtained by the pattern recognition unit 12, the effective context determination unit 1
At 4, a context dictionary capable of performing effective context processing is selected. For example, if the context evaluation values for all the recognition candidate character strings using a certain context dictionary are equal to or smaller than a predetermined threshold, the context processing using the context dictionary is not determined to be valid, and the obtained The context evaluation value is not used for calculating the overall evaluation value. Conversely, if the value is higher than the predetermined threshold value, the context processing using the context dictionary is determined to be valid, and the obtained context evaluation value is used for calculating the overall evaluation value, so that more recognition is possible. Accuracy can be improved.

For example, the threshold value of the context evaluation value is “10”
In the case of FIG. 5, the general context dictionary C
1 is not determined to be valid, only the context processing using the address context dictionary C2 is determined to be valid, and the determination result is notified to the context processing unit 14 and is determined to be valid. Only the context evaluation value using the context dictionary C2 is notified to the comprehensive judgment unit 17.

The overall judgment section 17 obtains an overall evaluation value based on the calculated context evaluation value and the pattern evaluation value calculated by the pattern recognition section 12. In the present embodiment, the total evaluation value is simply obtained by adding the context evaluation value and the pattern evaluation value, but the present invention is not particularly limited to this.

In this case, as shown in FIG. 6, for example, when the context dictionary C1 is used, the total evaluation value of the recognition candidate character string "Tomoname" is larger, and when the context dictionary C2 is used. The recognition candidate character string “Akashi” has a larger overall evaluation value. Therefore, IEICE (PRMU98-238)
According to the method disclosed in Japanese Patent Application Laid-Open No. H11-270, the recognition candidate character string having the largest overall evaluation value is selected, so that C1 is selected as the context dictionary, and an erroneous recognition result of “Tomo name” is recognized. The result is output to the result output unit 17.

The output determining section 18 selects the recognition candidate character string having the highest overall evaluation value calculated by the comprehensive determining section 17 as the recognized character string.

For example, in FIG. 6, in the comprehensive evaluation value using the context dictionary C1, it can be determined that the recognition candidate character string “Tomo name” has a higher comprehensive evaluation value, but the context processing using the context dictionary C1 is not performed. Since it is determined that the character string is not valid, the recognition character string is selected based on the comprehensive evaluation value using the context dictionary C2. That is, in the comprehensive evaluation value using the context dictionary C2, since the recognition candidate character string "Akashi" can be determined to have a higher comprehensive evaluation value, "Akashi" is selected as the recognized character string.

Various methods are conceivable for the determination method in the effective context determination unit 14, including a method of setting a threshold value. For example, the threshold may be determined not based on the evaluation value itself but on whether or not the difference between the maximum context evaluation value and the minimum context evaluation value in the recognition candidate character string is equal to or larger than a predetermined size. Can be This is based on the fact that if the processing of the context evaluation value is not large, the degree of influence on the overall evaluation value is small, and the degree of contribution to the improvement of the recognition accuracy can be considered as the context processing. . That is, in the case where the context evaluation value obtained when the general context dictionary C1 and the address context dictionary C2 are used is as shown in FIG. 7, the maximum context evaluation value and the minimum If it is determined that the case where the difference from the context evaluation value is “10” or more is valid, the address context dictionary C
2 is determined to be valid, but the context dictionary C
1 is not determined to be valid.

Therefore, as shown in FIG. 8, in the comprehensive evaluation value using the general context dictionary C1, although the recognition candidate character string "Tona" can be judged to have a higher comprehensive evaluation value, the general context dictionary C1 Since it is determined that the context processing using is not effective, the recognition character string is selected based on the comprehensive evaluation value using the address context dictionary C2. That is, in the comprehensive evaluation value using the address context dictionary C2, the recognition candidate character string "Akashi" can be determined to have a higher comprehensive evaluation value, so "Akashi" is selected as the recognized character string.

It is also conceivable to make a similar determination not only with the context evaluation value but also with the pattern evaluation value. That is, the difference between the maximum pattern evaluation value and the minimum pattern evaluation value in the recognition candidate character string is equal to or greater than a predetermined size, and the maximum context evaluation value and the minimum context evaluation in the recognition candidate character string If the difference from the value is smaller than a predetermined value, it is determined that context processing using the context dictionary is not valid. This is based on the fact that, when the difference between the pattern evaluation values is sufficiently large, the recognition result can be considered not to change when the processing using the context dictionary that performs only the context processing with a predetermined variation amount or less is enabled. Things.

For example, in the case of FIG. 7, if the threshold value of the pattern evaluation value is (40 × the length of the recognition candidate character string), the threshold value becomes “80”. In context processing using the dynamic context dictionary C1, the difference is' 8
If the difference between the maximum context evaluation value and the minimum context evaluation value among the recognition candidate character strings is equal to or greater than “0” and is equal to or greater than “10”, the address context dictionary C2 is eventually obtained.
Is determined to be valid, but the context dictionary C1
Is not determined to be valid.

Therefore, as shown in FIG. 8, in the comprehensive evaluation value using the general context dictionary C1, although the recognition candidate character string "Tona" can be judged to have a higher comprehensive evaluation value, the general context dictionary C1 Since it is determined that the context processing using is not effective, the recognition character string is selected based on the comprehensive evaluation value using the address context dictionary C2. That is, in the comprehensive evaluation value using the address context dictionary C2, the recognition candidate character string "Akashi" can be determined to have a higher comprehensive evaluation value, so "Akashi" is selected as the recognized character string.

If the method of calculating the context evaluation value using the context dictionary is uniform for all the context dictionaries,
In some cases, it is impossible to simply compare both evaluation values. For example, as shown in FIG. 9, the general context dictionary C1 performs a normal process, whereas a name that cannot be determined as a “place name” in the address context dictionary is set to lower its comprehensive evaluation value. Cases are conceivable.

In this case, the method of determining the validity of the context dictionary based on the magnitude of the threshold value of the context evaluation value itself has a high possibility of erroneous recognition. That is, for example, if it is determined that the context dictionary in which all the context evaluation values are equal to or less than “10” is not valid, in the example of FIG.
1 and the address context dictionary C2 are not determined to be valid context dictionaries.

Therefore, the method of calculating the context evaluation value using the context dictionary including the threshold value needs to be set for each context dictionary. In the example of FIG. 9, a normal threshold value is set for the general context dictionary C1, and all the context evaluation values are set to "-10" for the address context dictionary C2.
If it is determined that the context dictionary is not valid, the address context dictionary C2 is determined to be a valid context dictionary, and “Akashi” is selected as a recognized character string from the comprehensive evaluation value shown in FIG.

Further, depending on the order in which context processing is performed on a plurality of context dictionaries to determine the validity of the context processing, the recognition processing time may fluctuate. That is, first perform context processing using a valid context dictionary,
This is because, when such context processing is effective, by not performing context processing by another context dictionary, the number of times of context dictionary collation can be reduced.

In this case, in order to minimize the recognition processing time, it is necessary to accurately determine the query dictionary priority in the context dictionary. Various methods are conceivable for the method of determining such priorities. For example, it is conceivable to store the result of the previously performed context processing and determine the collation priority based on the result.

FIG. 11 is a block diagram of a pattern recognition apparatus according to an embodiment of the present invention for realizing the above-described method. The basic configuration is the same as that of FIG.
11 is updated, and the priority context information 112 is updated each time.

That is, as the priority context information 112,
It is necessary to collect information such as which context dictionary is used to determine that the context processing is valid, or concludes which context processing result is used as the final output. Therefore, the priority context information management unit 111
Then, the determination results or the recognition results in the effective context determination unit 16 and the output determination unit 18 are collected, and the priority context information 11 is collected.
2 will be updated.

At the time of pattern recognition, the priority context information management unit 111 notifies the context processing unit 14 of the priority context information 112, and the context dictionary 1 to perform context processing based on the priority context information 112.
5 is assigned a priority. By performing the context processing based on the priority, useless context processing can be avoided, and the recognition processing time can be reduced.

Further, since it is conceivable that the recognition result in the output determination unit 18 is not a valid result, for example, FIG.
By providing the character string determination unit 121 that allows the user to determine the final recognition character string as shown in (1), the priority context information 112 reflecting the final recognition result can be obtained.

The value of the context evaluation value in the present embodiment will be examined using experimental values. If the fields match, the field does not match the context dictionary in which the context processing result should be reflected. It should be set to a number of digits that makes it clear from the context dictionary that it should not be reflected in the case. In other words, the context dictionary in which the context processing result should be reflected when the fields match, should be a context evaluation value that is large enough from the viewpoint of the overall evaluation value. For a context dictionary that should not be reflected, a context evaluation value having a number of digits that affects the overall evaluation value should not be used.

As described above, according to the present embodiment, it is determined whether or not context processing using the context dictionary is valid for each context dictionary, and pattern recognition is performed only for those determined to be valid. By making the result output target,
An optimal context dictionary can be automatically and rapidly selected from a plurality of context dictionaries with high accuracy and high-speed pattern recognition can be performed.

Next, the flow of processing of a program for realizing the pattern recognition apparatus according to the embodiment of the present invention will be described. FIG. 13 shows a flowchart of the processing of a program for realizing the pattern recognition device according to the embodiment of the present invention.

In FIG. 13, when a user inputs a pattern such as a handwritten character or a voice (step S).
131), pattern recognition is performed on the input pattern (step S132), and a pattern evaluation value is calculated for the recognition candidate character string (step S133).

Next, using the first context dictionary, context processing is performed on each recognition candidate character string, and a context evaluation value is calculated (step S134). Based on the context evaluation value calculation result for each recognition candidate character string, it is determined whether or not context processing using the context dictionary is valid (step S13).
5). If it is determined that the context processing using the context dictionary is valid (step S135: Yes), the result of the context processing using the context dictionary is used as a context evaluation value. If it is determined that it is not valid (step S13
5: No), the context processing result using the context dictionary is not used for calculating the overall evaluation value, but the context evaluation value is calculated again using the next context dictionary (step S136), and the process of step S135 is repeated. .

Then, an overall evaluation value is calculated based on the obtained pattern evaluation value and context evaluation value (step S13).
7) The recognition candidate character string having the highest overall evaluation value is determined as the recognition character string (step S13).
8).

A recording medium storing a program for realizing the pattern recognition apparatus according to the embodiment of the present invention is shown in FIG.
As shown in the example of the recording medium shown in FIG.
Not only the portable recording medium 142 such as the 2-1 and the floppy disk 142-2, but also any other storage device 141 provided at the end of the communication line, or the recording medium 144 such as a hard disk or a RAM of the computer 143 may be used. When the program is executed, the program is loaded and executed on the main memory.

A recording medium for recording a context dictionary and the like used in the pattern recognition device according to the embodiment of the present invention is also a CD-R as shown in the example of the recording medium shown in FIG.
OM142-1 or floppy (registered trademark) disk 14
Not only the portable recording medium 142 such as 2-2, but also any other storage device 141 provided at the end of the communication line, or a recording medium 144 such as a hard disk or a RAM of the computer 143 may be used, for example, according to the present invention. It is read by the computer 143 when using the pattern recognition device.

(Supplementary Note 1) The effective context determination unit determines whether or not the process of calculating the context evaluation value is valid by taking into account not only the context evaluation value but also the pattern evaluation value. 2. The pattern recognition device according to 1.

When the difference between the pattern evaluation values is sufficiently large, it can be considered that the recognition result does not fluctuate when the processing using the context dictionary that performs only the context processing with a predetermined variation amount or less is enabled. is there.

(Supplementary note 2) The criterion according to claim 1, wherein the effective context judging unit can individually set a criterion for judging whether the process of calculating the context evaluation value is effective or not for each of the context dictionaries. Pattern recognition device.

If the method of calculating the context evaluation value using the context dictionary is uniform for all the context dictionaries, it may be impossible to simply compare the two evaluation values. .

[0058]

As described above, according to the pattern recognition apparatus and method of the present invention, it is determined for each context dictionary whether or not context processing using the context dictionary is effective.
By outputting pattern recognition results only for those that are determined to be valid, the optimal context dictionary can be automatically and quickly selected from among multiple context dictionaries with high accuracy and high recognition rate. Recognition can be performed.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a pattern recognition device according to an embodiment of the present invention.

FIG. 2 is an exemplary diagram of a handwritten character input pattern

FIG. 3 is an exemplary diagram of recognition candidate characters.

FIG. 4 is an exemplary diagram of calculating a pattern evaluation value of a recognition candidate character string;

FIG. 5 is a diagram illustrating an example of calculating a context evaluation value of a recognition candidate character string;

FIG. 6 is a diagram illustrating an example of calculating a comprehensive evaluation value of a recognition candidate character string;

FIG. 7 is an exemplary diagram of calculating a context evaluation value of a recognition candidate character string;

FIG. 8 is an exemplary diagram of calculating a comprehensive evaluation value of a recognition candidate character string;

FIG. 9 is an exemplary diagram of calculating a context evaluation value of a recognition candidate character string;

FIG. 10 is an exemplary diagram of calculating a comprehensive evaluation value of a recognition candidate character string;

FIG. 11 is a configuration diagram of a pattern recognition device according to an embodiment of the present invention.

FIG. 12 is a configuration diagram of a pattern recognition device according to another embodiment of the present invention.

FIG. 13 is a flowchart of a process in the pattern recognition device according to the embodiment of the present invention;

FIG. 14 is an exemplary diagram of a recording medium.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 11 Pattern input part 12 Pattern recognition part 13 Pattern recognition dictionary 14 Context processing part 15 Context dictionary 16 Effective context determination part 17 Recognition result output part 18 Output determination part 111 Priority context information management part 112 Priority context information 121 Character string determination part 141 Line destination storage device 142 Portable recording medium such as CD-ROM or floppy disk 142-1 CD-ROM 142-2 Floppy disk 143 Computer 144 Recording medium such as RAM / hard disk on computer

Claims (5)

[Claims] A pattern input unit for inputting a pattern; a pattern recognition unit for recognizing a candidate character string for the pattern input from the pattern input unit and calculating a pattern evaluation value; A context processing unit that receives the recognized recognition candidate character string, and calculates a context evaluation value based on at least two or more context dictionaries for each of the recognition candidate character strings; based on the pattern evaluation value and the context evaluation value A comprehensive determination unit that calculates a comprehensive evaluation value by using the total evaluation value calculated by the comprehensive determination unit; and an output determination unit that selects the recognition candidate character string having the maximum overall evaluation value as a recognition character string. A pattern recognition device that includes, based on which of the context dictionaries, the context evaluation unit calculates the context evaluation value. And a context evaluation calculated based on the context dictionary determined to be an effective process by the effective context determination unit. A pattern recognition apparatus characterized in that pattern recognition is performed using only a value as the context evaluation value. 2. Priority context information for managing priority context information that records information on the processing priority of the context evaluation value calculation process using the context dictionary based on the determination result in the effective context determination unit. The pattern recognition device according to claim 1, further comprising a management unit. 3. The apparatus according to claim 1, further comprising: a character string determining unit for performing a determination process by the user on the recognition character string output by the output determining unit; 3. The pattern recognition device according to claim 2, wherein the content of the priority context information is updated based on the character string subjected to the above. 4. A step of inputting a pattern, a step of recognizing a recognition candidate character string for the input pattern and calculating a pattern evaluation value, receiving the recognized recognition candidate character string, and receiving each of the recognition candidates. Calculating a context evaluation value for the character string based on at least two or more context dictionaries; calculating an overall evaluation value based on the pattern evaluation value and the context evaluation value; A pattern recognition method including a step of selecting, as a recognition character string, the recognition candidate character string having the largest overall evaluation value based on a value, wherein in the context evaluation value calculation processing step, Determining whether the processing for calculating the context evaluation value is valid, based on the context dictionary determined to be valid. A pattern recognition method characterized by performing pattern recognition using only the output context evaluation value as the context evaluation value. 5. A step of inputting a pattern, a step of recognizing a recognition candidate character string for the input pattern and calculating a pattern evaluation value, receiving the recognized recognition candidate character string, and receiving each of the recognition candidates. Calculating a context evaluation value for the character string based on at least two or more context dictionaries; calculating an overall evaluation value based on the pattern evaluation value and the context evaluation value; and calculating the overall evaluation. A computer-readable recording medium storing a program for causing a computer to execute a pattern recognition method including a step of selecting, as a recognition character string, the recognition candidate character string having the maximum overall evaluation value based on a value. In the context evaluation value calculation processing step, the context Determining whether the value calculation process is valid; and performing pattern recognition using only the context evaluation value calculated based on the context dictionary determined to be a valid process as the context evaluation value. A computer-readable recording medium on which a program to be executed by a computer is recorded.