JP2003196636A - Notation error detection processing method using machine learning method having teacher, its processing device and its processing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform highly accurate detection by using a machine learning method having a teacher, concerning notation error detection processing. <P>SOLUTION: Positive example data D (correct notation) and negative example data E (wrong notation) are stored in a teacher's data storage part 11 as teacher's data. A solution-identity pair extraction part 12 extracts pairs of the solution and the identity from the negative example data E and the positive example data D in the teacher's data storage part 11, and a machine learning part 13 estimates the kind of solution probably acquired in the case of a set of identities by the machine learning method and stores the result in a learning result data storage part 14. An identity extraction part 15 extracts the set of identities from inputted data 2, and an error detection part 16 estimates whether a notation is wrong or not from the set of the identities on reference to the learning result data storage part 14, and outputs an estimation result 3. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to notation error detection processing, and more particularly to a notation error detection processing method using a supervised machine learning method, a processing device for realizing the processing, and a computer for executing the processing. And for the program.

[0002]

2. Description of the Related Art In the case of Japanese, word detection error detection is
It's much more difficult than in English. In the case of English, words are written in words, so basically, by preparing a word dictionary and rules for transforming word ends, it is possible to check spelling of words with high accuracy. On the other hand, in the case of Japanese, since it is not written in words, it is difficult to perform the processing with high accuracy even if the processing is limited to word notation errors.

In addition to typographical errors, typographical errors include grammatical errors such as operational errors of the particles "te", "ni", "wo", and "ha".

The following are the main conventional techniques for detecting Japanese typographical errors.

References 1 to 3 below describe conventional techniques for detecting typographical errors based on word dictionaries, dictionaries in which hiragana sequences are registered, and dictionaries in which concatenation conditions are described. In these conventional methods, if there is something that does not exist in the word dictionary or the dictionary in which the hiragana sequence is registered, it is determined that there is a typographical error, and if there is an unsatisfactory concatenation in the dictionary that describes the concatenation condition, it is a typographical error. To determine. [Reference 1: Kazuhiro Notomi, Japanese document proofreading support tool h
Development of sp, Information Processing Society of Japan, Research Presentation (Digital Document), (1997), pp.9-16] [Reference 2: Kazuma Kawahara et al., Error detection method using dictionary extracted from corpus, IPSJ 54th National Convention, (1997), pp.2-21-2-22] [Reference 3: Shinki Shiraki et al., Making Japanese spell checker by registering a large number of hiragana sequences, Annual Conference of the Language Processing Society of Japan Convention, (1997), pp.445-448] In addition, the occurrence probability of each character string is calculated based on a probability model using ngram for each character, and a portion where a character string with a low occurrence probability appears is determined to be a writing error. Conventional methods and the like are described in References 4 to 6 below. [Reference 4: Tetsuro Araki et al., Error detection and correction method for Japanese sentences by dual Markov model, IPSJ Natural Language Processing Research Group NL97-5, (1997), pp.29-35] [Reference 5: Takaaki Matsuyama et al., Oc by n-gram
Experiment and consideration on estimation of precision and recall for examination of r error detection ability, Annual Conference of the Language Processing Society of Japan (1996), p
p.129-132] [Reference 6: Koichi Takeuchi et al., Construction of OCR error correction system using statistical language model, IPSJ Journal, Vol.40, No.6, (1999)] Among the conventional methods, the method utilizing the ngram probability in Reference 5 is mainly an optical character reading device (Optica
l Character Reader (OCR) is used for notation error detection in an error correction system. OCR
In the case of an error correction system, as a premise, the appearance rate of notation errors is as high as 5 to 10%, which is higher than the probability that a person normally makes an error when writing. Therefore, the recall and precision of detection of typographical errors are likely to be high, which is a relatively easy problem setting.

The method of Takeuchi et al., Which seems to be the best among the above-mentioned conventional methods, that is, the conventional method described in Reference 6 (hereinafter referred to as conventional method A) will be briefly described below. To do.

In the conventional method A, first, the text for which the typographical error is to be detected is shifted by one character from the beginning, three consecutive characters are extracted, and the appearance probability of the extracted part in the corpus (set of correct Japanese sentences) is Tp or less. In this case, -1 is added to each of the three consecutive characters, and a character whose given value is Ts or more is determined to be an error. For example, Tp = 0, Ts
= -2. Here, since Tp = 0, it is not necessary to find the appearance probability, and the corpus
All you have to do is check if a sequence of characters appears. When Tp> 0, it is determined that there is an error even if the extracted part appears in the corpus. However, even if the appearance probability is low, if it appears in the corpus, it may not be considered as an error, so Tp> 0 is not appropriate, and Tp = 0 is good.

As a supplementary explanation of the conventional method A, consider that error detection is performed on the Japanese expression "detection of negative things and zeros". At this time, from the beginning of Japanese expression, "negative things"
Cut out three consecutive characters such as "nothing zero", check whether these are in the corpus, and give -1 to the three characters if there are not three cut out characters. In this case, there was no "nothing zero" and "nothing zero", so tr as shown in FIG.
The score is given by igram, and as a result, "-2"
The "thing" and "zero" parts that become points are determined to be incorrect.
This conventional method A uses characters 3 that appear frequently in the corpus.
It is a method of detecting an error by properly combining -gram.

However, after all, the processing of the conventional method A is to determine whether or not the expression exists in the corpus. That is, the conventional method A is very similar to the other conventional methods described above in which it is considered an error if something that is not in the dictionary appears.

Regarding the machine learning method, the following reference 7
It is known that learning from only positive examples is generally difficult as described in. [Reference 7: Yokomori Kan et al., Learning a formal language-focusing on learning from positive examples-, IPSJ, Vol.32, No.3,
(1991), pp226-235] Furthermore, incorrect notation data used as a teacher signal (negative example)
Are generally considered to be more difficult to obtain than correct notation data (positive example).

[0011]

Conventionally, a highly accurate processing cannot be expected with a processing method using a machine learning method in which only positive examples are used as teacher signals, and the acquisition of negative examples used as teacher signals. Since it is difficult to do so, a processing method using a machine learning method in which both positive examples and negative examples are used as teacher signals has not been realized in the sentence notation error detection process.

An object of the present invention is to realize a highly accurate notation error detection process by using a machine learning method in which positive examples and negative examples are used as teacher signals.

Another object of the present invention is to realize a highly accurate notation error detection process by efficiently and automatically generating a negative example as a teacher signal and using it as a teacher signal of a machine learning method. .

[0014]

In order to solve the above problems, the present invention is a processing method for detecting a notation error, in which positive example data that is correct notation and negative example that is incorrect notation. A process of extracting a pair of a feature and a solution from teacher data including data, performing machine learning using the pair of the feature and the solution as a teacher signal, and storing a learning result in a learning result data storage unit, And extracting a feature from the acquired data and detecting an error in the notation of the data based on the learning result.

Further, the present invention is a processing method for detecting a notation error, and when the input case does not exist in the positive example data which is the correct notation prepared in advance, the general appearance of the case A process of calculating a probability and a probability that the case appears in the positive example data are calculated based on the general appearance probability, and the case in which the probability exceeds a predetermined threshold is negative example data. And a processing step to extract a pair of a feature and a solution from teacher data including positive example data and the negative example data, machine learning is performed using the pair of the feature and the solution as a teacher signal, and a learning result The process of storing the learning result data in the memory and the features extracted from the input data,
And a processing step of detecting a notation error of the data based on the learning result.

Further, the present invention is a processor for detecting a notation error, in which a feature and a solution are identified from teacher data including positive example data which is correct notation and negative example data which is wrong notation. Machine learning processing means for extracting a pair, performing machine learning using the pair of the feature and the solution as a teacher signal, and storing the learning result in the learning result data storage unit, and extracting the feature from the input data, and performing the learning. Error detection processing means for detecting an error in the notation of the data based on the result.

Further, the present invention is a processor for detecting a notation error, and when the input case does not exist in the positive example data which is the correct notation prepared in advance, the general appearance of the case. Appearance probability calculation processing means for calculating a probability, and a probability that the case appears in the positive example data are calculated based on the general appearance probability, and the case where the probability exceeds a predetermined threshold is negative. Of the negative example acquisition processing means as the example data of, and a pair of the feature and the solution is extracted from the teacher data including the positive example data and the negative example data, and the pair of the feature and the solution is used as the teacher signal. Machine learning processing means for performing machine learning and storing the learning result in a learning result data storage unit, and error detection for extracting features from input data and detecting an error in notation of the data based on the learning result Processing means.

Further, the present invention is a program for causing a computer to execute a process of detecting a notation error, and a teacher including positive example data that is correct notation and negative example data that is wrong notation. A process of extracting a feature-solution pair from the data, performing machine learning using the feature-solution pair as a teacher signal, and storing the learning result in the learning result data storage unit, and extracting the feature from the input data. Then, the computer is caused to execute a process of detecting an error in notation of the data based on the learning result.

Further, the present invention is a program for causing a computer to execute a process of detecting a notation error, and when an input case does not exist in positive example data which is a correct notation prepared in advance, A process of calculating a general appearance probability of the case, calculating the probability that the case appears in the positive example data based on the general appearance probability,
A process of setting the case in which the probability exceeds a predetermined threshold as negative example data, and extracting a pair of a feature and a solution from teacher data including positive example data and the negative example data, and extracting the feature. Machine learning is performed by using a pair of a and a solution as a teacher signal, a learning result is stored in a learning result data storage unit, and a feature is extracted from input data, and the learning is stored in the learning result data storage unit. The computer is caused to execute a process of detecting an error in the notation of the data based on the result.

The program for realizing each means or function or element of the present invention by a computer can be stored in an appropriate recording medium such as a portable medium memory, a semiconductor memory, a hard disk, which can be read by a computer. It is provided by being recorded in a recording medium or is provided by transmission and reception using various communication networks via a communication interface.

[0021]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, as a first embodiment of the present invention, a process of detecting a Japanese writing error by concatenation will be described.

FIG. 1 shows a configuration example of a notation error detection device 1 according to the first embodiment of the present invention.

The notation error detection device 1 includes a teacher data storage unit 11, a solution-feature pair extraction unit 12, a machine learning unit 13, and
It has a learning result data storage unit 14, a feature extraction unit 15, and an error detection unit 16.

The teacher data storage unit 11 is means for storing data (teacher data) which is a teacher signal when the machine learning method is carried out. The teacher data storage unit 11 stores examples of correct notation (positive example) and examples of incorrect notation (negative example) as teacher data. As a positive example, for example, a corpus which is a set of correct sentences may be used. In the negative example, since it is an erroneous notation and there is no general data, the one generated by hand beforehand is used. Alternatively, a negative example prediction processing method as described later may be used to generate the positive example.

The solution-feature pair extraction unit 12 is means for extracting a set of a solution of a case and a set of features for each case of the teacher data stored in the teacher data storage unit 11.

The machine learning unit 13 includes a solution-feature pair extraction unit 12
It is a means for learning, by a machine learning method, what kind of features and what kind of solutions are likely to occur from the set of the solution and the set of features extracted by. The learning result is stored in the learning result data storage unit 14.

The feature extraction unit 15 is a means for extracting a set of features from the data 2 which is a notation error detection target and passing the extracted feature set to the error detection unit 16.

The error detection unit 16 refers to the learning result data in the learning result data storage unit 14 to find out what kind of solution is likely to occur in the case of the feature set passed from the feature extraction unit 15, that is, notation. It is means for estimating whether or not there is an error and outputting the estimation result 3.

FIG. 2 shows an example of the data structure of the teacher data storage unit 11. The teacher data storage unit 11 stores teacher data that is a set of problems and solutions. For example, the teacher data in which the gap (represented by <|>) of each character of the sentence is used as a problem and the solution (correct answer, error) of the connection of the gap is associated is stored. Of the teacher data in FIG. 2, “problem-solution: <|> can be used in the explained method-error” is an example of negative example data E, and “problem-solution: the explained method <| > Can be used-positive ”is an example of positive example data D.

FIG. 3 shows a processing flowchart of the notation error detection processing. Before the notation error detection process, the positive example data D
And the negative example data E is stored in the teacher data storage unit 11.

First, the solution-feature pair extraction unit 12 extracts a set of a solution and a feature set from the teacher data storage unit 11 for each case (step S1). The feature means a small unit of information used for analysis. As features, the following is extracted for each character gap that is the target of connection determination.

A character string of 1-5 gram for each of the front and back connections, and a character string of 1-5 gram including the target (gap) (however, the target gap (<|>) is treated as one character. ) Prefix and postfix words (word extraction uses existing processing means (not shown in FIG. 1) for performing morphological analysis), etc. ・ Part of speech of prefix and postfix words, for example, " If the “problem-solution” is “<|> can be used in the described method-error”, the features as shown in FIG. 4 are extracted. That is, the following features are extracted.

Feature: Prefix "By the way", Prefix "By the method", Prefix "By the method", Prefix "By the law", Prefix "By", Postfix "By" Subsequent “use”, postfix “use”, postfix “use”, postfix “o”,
＞ ”,“ Method with <|> ”,“ Using method with <|> ”,“ With <|> ”,“ Using <|> ”, Prefix“ de ”, Postfix“ o ” , Prefix "particle", Postfix "particle" Next, the machine learning unit 13 uses the set of the extracted solution and the set of features to identify what kind of feature is likely to be and what kind of solution is likely to occur. After learning, the learning result is stored in the learning result data storage unit 14 (step S2).

As a machine learning method, for example, a decision list method, a maximum entropy method, a support vector machine method, or the like is used.

The decision list method uses a set of features and classification destinations as rules, stores them in a list in a predetermined priority order, and when an input to be detected is given,
This is a method in which the input data and rule features are compared from the highest priority in the list, and the classification destination of the rules whose features match is used as the input classification destination.

In the maximum entropy method, when the set of features f _j (1 ≦ j ≦ k) set in advance is F, and the expression that means entropy is maximized while satisfying a predetermined conditional expression, The probability distribution p (a, b) of
Of the probabilities of each classification obtained according to the probability distribution,
In this method, the classification having the largest probability value is used.

The support vector machine method is a method of classifying data consisting of two classes by dividing a space into hyperplanes.

For the decision list method and the maximum entropy method, the following reference 8 is given. For the support vector machine method, the following references 9 and 10 are given.
Explained. [Reference 8: Maki Murata, Masao Uchiyama, Kiyotaka Uchimoto, Masei,
Hitoshi Isahara, Ambiguity Resolving Experiments Using Various Machine Learning Methods, IEICE Language Understanding and Communication Research Group,
NCL2001-2, (2001)] [Reference 9: Nello Cristianini and John Shawe-Tay
lor, An Introduction to Support Vector Machines an
d other kernel-based learning methods, (Cambridge U
niversity Press, 2000)] [Reference 10: Taku Kudoh, Tinysvm: Support Vector
machines, (http://cl.aist-nara.ac.jp/taku-ku//soft
ware / Tiny SVM / index.html, 2000)] Note that the machine learning unit 13 is not limited to the above method, and any method can be used as long as it is a machine learning method with a teacher.

After that, the feature extraction unit 15 inputs the data 2 for which a solution is to be obtained (step S3), extracts a set of features from the data 2 in the same manner as the processing in the solution-feature pair extraction unit 12, and extracts them. To the error detector 16 (step S
4).

The error detection unit 16 specifies what kind of solution is likely to occur in the case of the passed feature set based on the learning result data of the learning result data storage unit 14, and specifies the specified solution, that is, the notation error. An estimation result 3 indicating whether or not it is output (step S5).

For example, when the problem to be analyzed is the connection of the gap <|>, the data 2 is "<│> by the method described.
Can be used ”, the estimation result 3 of“ error ”is output.

Next, a second embodiment of the present invention will be described.

The positive example data D in the teacher data storage unit 11 can be acquired relatively easily because a corpus or the like can be used. However, since the negative example data E cannot be easily acquired, it is generated manually, but the burden of such generation work is large.

Further, since the processing accuracy increases as the amount of teacher data increases, it is desirable to prepare as much teacher data as possible.

Therefore, a method of predicting negative example data from a large amount of positive example data will be considered.

As a simple method of predicting a negative example from a positive example, a method of making all the negative examples that do not appear in the data of the known positive examples can be considered. However, in reality, it is possible that there are positive examples that have not yet appeared, so using such a simple method leads to the determination that many positive examples that have not yet appeared are negative examples. However, the negative example generated by such a method cannot be applied to high-precision processing.

For example, assuming that all large-scale existing corpus (set of Japanese sentences) is correct, the existing corpus is considered as a correct sentence (positive example), and a notation error is made using this positive example. The method of predicting (negative example) can automatically generate a negative example.

As a result, a large number of negative examples of teacher data are provided, the burden of generation work is reduced, and high-precision notation error detection processing using a machine learning method with teacher data can be realized. .

The notation error detection device 1 in the present embodiment first calculates a general appearance probability p (x) of an unknown case x to be determined as a positive example or a negative example. Next, this appearance probability p
If it is unnatural that it does not appear in the known positive example data D in (x), that is, it is known in spite of the state that it has a high general appearance probability and naturally appears in the positive example data D. When it does not appear in the positive example data D of, the degree of the negative example of the case x is estimated to be high, and the case x having the degree of the negative example higher than a predetermined value is set as the negative example data E. And
The notation error detection process is performed by a machine learning method using the negative example data E and the positive example data D as teacher signals.

FIG. 5 shows a configuration example of the notation error detection device 1 according to the second embodiment of the present invention.

The notation error detection device 1 includes a teacher data storage unit 11, a solution-feature pair extraction unit 12, a machine learning unit 13, and
Feature extraction unit 15, error detection unit 16, and presence determination unit 21
, Appearance probability estimation unit 22, and negative example degree calculation unit 23
And a negative example acquisition unit 24 and a positive example data storage unit 25.

The teacher data storage unit 11, the solution-feature pair extraction unit 12, the machine learning unit 13, the feature extraction unit 15, and the error detection unit 16 detect the notation error described in the first embodiment. Since it is the same as each means of the apparatus 1, the description thereof will be omitted (see FIG. 1).

The existence determining unit 21 determines that the case x of the corpus 20 which is a set of Japanese sentences to which positive or negative information is not added is the positive example data D stored in the positive example data storage unit 25. It is a means of determining whether or not it exists.

The appearance probability estimation unit 22 is means for calculating a general appearance probability (frequency) p (x) of the case x when the case x does not exist in the positive example data storage unit 25.

The negative example degree calculation unit 23 determines the appearance probability p.
This is means for calculating the negative example degree Q (x) of the case x based on (x).

When the negative example degree Q (x) of the case x received from the negative example degree calculating unit 23 exceeds a predetermined value, the negative example obtaining unit 24 sets the case x to the negative example data E. The case x is stored in the teacher data storage unit 11 as the teacher data (negative example data E) of the problem-solution concept.

FIG. 6 shows a processing flow chart of the processing for acquiring the negative example data which is the learning data in the second embodiment.

The existence determination unit 21 of the notation error detection device 1 is
Enter a sentence whose positive or negative example is unknown from the corpus 20, shift the gaps of the characters one by one from the beginning of the sentence, and each gap is the target of the connection check, and the gap is preceded by 1 〜5gram character string a, followed by 1〜
A character string b of 5 gram is taken out, and a case x = (a, b) which is this arbitrary pair is generated (step S11).
Here, 25 cases (pairs) will be generated.

Then, it is checked whether or not the 25 concatenated abs of case x are present in the positive example data storage unit 25 (step S
12) If the connection ab does not exist in the positive example data storage unit 25, the case x is passed to the appearance probability estimation unit 22 (step S13).

The appearance probability estimation unit 22 estimates a general appearance probability p (x) of the case x (step S14).

For example, the positive example data D in the positive example data storage unit 25 has a binary relation (a, b), and the binary terms a and b
Assuming that and are independent of each other, the binary relation (a,
The probability p (x) of occurrence of b) is the product p (a) × p of the probability p of the occurrence of a and b in the positive example data storage unit 25 as p (a) and p (b). (B). That is, by assuming each case x to be a binary relation (a, b) and assuming that the terms a and b are independent, a general appearance probability p of each case x is obtained.
(X) is calculated from the probabilities of the terms a and b.

Then, the negative example degree calculation unit 23 uses the appearance probability p (x) of the case x to obtain the probability Q (x) that the case x appears in the positive example data storage unit 25 (step S15). ).

At this time, assuming that the number of positive example data D in the positive example data storage unit 25 is n and each is independent, the probability that the case x does not appear after one trial is 1-
Since it is p (x) and this occurs n times in succession, the probability that the case x does not appear in the positive example data D of the positive example data storage unit 25 is (1-p (x)) ⁿ , Probability Q (x) = 1 that the case x appears in the positive case data D as well
-(1-p (x)) ⁿ .

By the way, "the probability Q (x) is small" means that the probability that the case x appears in the positive example data D of the positive example data storage unit 25 is low.
Positive example data (corpus) means that it is guaranteed not to appear stochastically because it is small,
It means that the case x can be a positive example.

On the contrary, "the probability Q (x) is large" means that the case x has a high probability of appearing in the positive example data D, and stochastically appears in the same corpus. It should be done, and there is a contradiction that it did not actually appear. This contradiction negates the general probability of occurrence p (x) or various independent assumptions.

Here, "if case x is a positive example,
The general probability of occurrence p (x) and various independent assumptions are correct. ”, This inconsistency results in“ case x
Cannot be a positive example. Will be derived. That is, "the probability Q that the case x appears in the positive example data D
(X) is the probability Q that case x cannot be a positive example.
(X) ”is meant. In that sense, Q
(X) means the degree of a negative example. Therefore, this Q (x) is defined as the “negative example degree”, and the Q of case x is
It is assumed that the larger the value of (x), the greater the degree of the negative example of the case x.

Then, the negative example acquisition unit 24 determines that the maximum Q
When the value of (x) is high, its value is Q _max , and x is x
and _max, the more the gap value is larger the Q (x _max), and that there is a high possibility of articulation not valid, Q (x _max) value of, is larger than a predetermined value, an example data that gap negative It is saved as E in the teacher data storage unit 11 (step S16). The negative example data E and the degree Q (x _max ) of the negative example may be stored in the teacher data storage unit 11.

By performing the above steps S11 to S15 for all the gaps in the sentence, the negative example data E is obtained by using the frequency information of the positive example data D in the positive example data storage unit 25. The positive example data D and the negative example data E can be prepared as teacher data in the teacher data storage unit 11.

Since the subsequent processing is the same as the error detection processing described in the first embodiment, the description will be omitted.

The present invention has been described above with reference to the embodiments, but the present invention can be variously modified within the scope of the gist thereof.

For example, the appearance probability estimation unit 22 of the notation error detection device 1 may calculate the general appearance probability p (x) of the case x by any method, and the method described in the embodiment of the present invention. It is not limited to.

Further, as the positive example data D in the teacher data storage unit 11, the positive example data D stored in the positive example data storage unit 25 can be used. Data can also be used.

[0073]

As described above, the present invention performs notation error detection processing using the machine learning method in which positive examples and negative examples are used as teacher signals. The present invention, which also uses the information of the negative example, can obtain a processing result with significantly higher accuracy than the processing method using only the positive example.

Further, according to the present invention, the frequency information of the positive example is used to perform the process of extracting the negative example from the positive example, and the extracted negative example is used as the teacher signal of the machine learning method. The present invention, which uses the information of the negative example automatically extracted from the positive example, generates the negative example in the problem that the positive example exists but it is difficult to obtain the negative example such as the notation error detection. It is possible to reduce the processing load.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a configuration example of a writing error detection device according to a first exemplary embodiment.

FIG. 2 is a diagram illustrating a configuration example of a teacher data storage unit.

FIG. 3 is a process flow chart of a notation error detection process.

FIG. 4 is a diagram showing an example of features.

FIG. 5 is a diagram showing a configuration example of a writing error detection device according to a second exemplary embodiment.

FIG. 6 is a process flowchart of a negative example data acquisition process.

FIG. 7 is a diagram for supplementarily explaining a conventional method.

[Explanation of symbols]

1 Notation error detector 2 data 3 estimation results 11 Teacher data storage 12 Solution-feature pair extraction unit 13 Machine learning department 14 Learning result data storage 15 Feature extraction unit 16 Error detector 20 corpus 21 presence determination unit 22 Appearance probability estimation unit 23 Negative example degree calculation unit 24 Negative example acquisition part 25 Positive example data storage

Claims (6)

[Claims] 1. A processing method for detecting a notation error, wherein a feature-solution pair is extracted from teacher data including positive example data that is correct notation and negative example data that is incorrect notation. , A process of performing machine learning by using the extracted feature-solution pair as a teacher signal and storing the learning result in the learning result data storage unit, and extracting the feature from the input data and storing it in the learning result data storage unit. A notation error detection processing method using a supervised machine learning method, comprising: a processing step of detecting a notation error based on the stored learning result. 2. A processing method for detecting a notation error, wherein a general appearance probability of the case is calculated when an inputted case does not exist in positive example data which is a correct notation prepared in advance. A processing step, and a processing step in which the probability that the case appears in the positive example data is calculated based on the general appearance probability, and the case in which the probability exceeds a predetermined threshold is negative example data. And extracting a pair of a feature and a solution from teacher data including positive example data and the negative example data, machine learning is performed by using the pair of the feature and the solution as a teacher signal, and the learning result is a learning result data. A supervised machine learning method characterized by comprising a processing step of storing in a storage unit and a processing step of extracting features from input data and detecting an error in notation of the data based on the learning result. Error detection process using Law. 3. A processor for detecting a notation error, wherein a feature-solution pair is extracted from teacher data containing positive example data that is correct notation and negative example data that is incorrect notation. , Machine learning processing means for performing machine learning using a pair of the feature and the solution as a teacher signal, and storing the learning result in a learning result data storage unit, and extracting the feature from the input data to obtain the learning result. And an error detection processing means for detecting an error in notation of the data, a notation error detection processing apparatus using a supervised machine learning method. 4. A processing device for detecting a notation error, wherein a general appearance probability of said case is calculated when an inputted case does not exist in positive example data which is a correct notation prepared in advance. Appearance probability calculation processing means, the probability that the case appears in the positive example data is calculated based on the general appearance probability, and the case in which the probability exceeds a predetermined threshold is negative example data. A negative example acquisition processing means for extracting a pair of feature and solution from the teacher data including the positive example data and the negative example data, and performs machine learning using the extracted feature and solution pair as a teacher signal. Machine learning processing means for storing the learning result in the learning result data storage unit, extracting features from the input data, and making a notation error based on the learning result stored in the learning result data storage unit. Error detection processing means for detecting Notation error detection processing apparatus using supervised machine learning methods, characterized in that it comprises. 5. A program for causing a computer to execute a process of detecting a notation error, wherein the teacher data includes positive example data that is correct notation and negative example data that is wrong notation, and A process of extracting a pair with a solution, performing machine learning by using the pair of the feature and the solution as a teacher signal, and storing the learning result in a learning result data storage unit, extracting the feature from the input data, and performing the learning A notation error detection processing program using a supervised machine learning method, which causes a computer to execute a process for detecting an error in notation of the data based on the result. 6. A program for causing a computer to execute a process of detecting a notation error, wherein when an input case does not exist in positive example data which is a correct notation prepared in advance, the general case of the case is described. And a process of calculating a typical appearance probability, the probability that the case appears in the positive example data is calculated based on the general appearance probability, and the case where the probability exceeds a predetermined threshold is negative. A process of processing as example data, a pair of a feature and a solution is extracted from teacher data including positive example data and the negative example data, and machine learning is performed by using the pair of the feature and the solution as a teacher signal. A method of causing a computer to execute a process of storing a result in a learning result data storage unit and a process of extracting a feature from input data and detecting an error in notation of the data based on the learning result To Teacher There notation error detection processing program using a machine learning method.