KR101633556B1 - Apparatus for grammatical error correction and method using the same - Google Patents

Abstract

A grammar error correction apparatus and a grammar error correction method using the same are disclosed. An error candidate generation unit for generating an error candidate word and an error candidate group based on a grammar error corpus for the input text; A router for extracting a context frame set based on the error candidate word and the error candidate group; A score calculating unit for calculating an n-gram score for the context frame set; And a word modifying unit for modifying the input text using the n-gram score. Therefore, it is possible to efficiently correct grammatical errors of foreign language learners by comparing the n-gram scores according to the set of context frames for the grammatical errors in the input text provided by the user, and by correcting the grammatical errors included in the input text, Can be effectively performed.

Description

[0001] APPARATUS FOR GRAMMATICAL ERROR CORRECTION AND METHOD USING THE SAME [0002]

The present invention relates to a grammar error correcting apparatus and a grammar error correcting method using the same, and more particularly, to a grammar error correcting apparatus for correcting a grammar error of an input text and a grammar error correcting method using the same.

As the demand for foreign language skills increases in modern society, which is becoming globalized and internationalized, a foreign language education system capable of efficiently learning foreign languages is actively being studied. In addition, with the development of information and communication technology, foreign language learning using an information processing device such as a smart phone, a tablet PC, a portable multimedia player (PMP), a personal digital assistant (PDA) Especially, as the user 's learning demand for foreign language grammar increases, a system for correcting grammatical errors in foreign language writing inputted from a user using information processing devices and providing correction information for errors is commercialized.

For example, Microsoft's MS Word is a representative program that corrects grammatical errors in foreign language writing. MS Word can provide grammar information to the user by displaying a detected error by performing a grammar check such as spelling or spelling of user-created text. However, there is a problem that it is difficult to correct grammatical errors based on the parts of speech information of words because MS Word corrects simple grammatical errors such as spelling of words contained in text or case difference of sentences. Thus, grammatical errors are detected from foreign language learner's sentences using a database or a model in which the format or grammar rules of a structure in which a foreign language sentence is expressed, as in the structure analysis rule disclosed in Japanese Patent Application Laid-Open No. 10-2009-0096905 , And Japanese Patent Laid-Open Publication No. 10-2013-0059795 discloses a method of analyzing a foreign language sentence based on part-of-speech information, and then analyzing a sentence based on regular grammar rules and contect free grammar rules A method of correcting errors in the grammar of foreign language learners through the classification process has been proposed. However, since there are various types of structures and grammatical rules in which foreign language sentences are expressed, it is difficult to precisely generate grammatical rules used for correcting errors in grammar in the above-described conventional techniques, There is a problem that accuracy and efficiency are inferior.

In order to solve the above problems, it is an object of the present invention to provide a grammar error correcting device for correcting a grammar error included in an input text by comparing n-gram scores according to a set of context frames for a grammar error, And a method for correcting a grammar error using the same.

Another object of the present invention is to provide a grammar error correcting apparatus for correcting a grammatical error contained in an input text when a word of a context frame in which a high n-gram score is calculated matches in all context frames, And to provide a method for correcting a grammar error using the same.

According to an aspect of the present invention, there is provided an apparatus for generating an error candidate, comprising: an error candidate generation unit for generating an error candidate word and an error candidate group based on a grammar error corpus for an input text; A router for extracting a context frame set based on the error candidate word and the error candidate group; A score calculating unit for calculating an n-gram score for the context frame set; And a word correcting unit for correcting the input text using the n-gram score.

Here, the error candidate generation unit may generate, as the error candidate word and the error candidate group, n words having the highest frequency for a specific word based on the grammar corpus constructed in advance.

Meanwhile, the router can extract the context frame set having a high error correction capability for the context frame set by measuring the error correction performance of the error candidate word and the error candidate group.

Meanwhile, the score calculating unit may calculate the n-gram score from the existing native speaker document as a word bili number or a probability value corresponding to each word list of the context frame set.

On the other hand, the score calculator may calculate the count information when the number of words of the context frame set is 5 or less, the forward n-gram score, and the backward n-gram score when the number of words is five or more .

On the other hand, the word correcting unit can modify the input text when the words of the context frame in which the high n-gram score is calculated match for all the context frames.

According to another aspect of the present invention, there is provided a method for generating an error candidate word and an error candidate word based on a grammar error corpus for an input text. Extracting a context frame set based on the error candidate word and the error candidate group; Calculating an n-gram score for the context frame set; And correcting the input text using the n-gram score.

In this case, the generating step may generate the n most frequent words for the specific word as the error candidate word and the error candidate word based on the grammar corpus constructed in advance.

Meanwhile, the extracting step may extract the context frame set having a high error correction performance for the context frame set by measuring the error correction performance for the error candidate word and the error candidate set.

Meanwhile, the calculating step may calculate the n-gram score from the existing native speaker document as a word number or a probability value corresponding to each word list of the context frame set.

On the other hand, the calculating step may calculate the count information when the number of words of the context frame set is 5 or less, the forward n-gram score, and the backward n-gram score when the number of words is five or more have.

On the other hand, the modifying step may correct the input text when the words of the context frame in which the high n-gram score is calculated are consistent for all the context frames.

When the grammar error correcting apparatus and the grammar error correcting method according to the present invention as described above are used, the n-gram score is compared according to the set of context frames for grammatical errors in the input text received from the user, Can be modified efficiently.

In addition, it corrects the grammatical errors included in the input text so that the learner can appropriately deal with grammatical errors and effectively perform foreign language learning.

1 is a block diagram of a grammar error correction apparatus according to an embodiment of the present invention.
2 is a detailed configuration diagram of the router of FIG.
3 is a detailed configuration diagram of the word correcting unit of FIG.
4 is a flowchart illustrating a method for correcting a grammar error using a grammar error correction apparatus according to an embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, A, B, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

An embodiment of the present invention selects an optimal context frame set for each grammar error candidate group using a router. In the sentence below, the word after the article is used as the important calibration information to correct the article word in front of the equipment.

We need (an → Ø) equipment to solve problems.

In the example below, to correct the verb words after One, the one in front is used as the important correction information that determines the verb form.

One (are → is) deemed to die at a later stage.

There are also cases where two forward or backward directions or several context frames need to be considered. However, as the range and number of context frames increase, the precision increases, but you can fix the unwanted errors. That is, the defect compensation (Recall) falls. The grammatical error correcting device having the highest F-score by selecting the optimum context frame direction, range, and word count and a grammatical error correction method using the same are provided.

FIG. 2 is a detailed configuration diagram of a router for extracting a most appropriate context frame set for each error candidate word and an error candidate group, and FIG. 3 is a detailed configuration diagram of a word correcting unit for correcting a case in which a word of the error candidate group in which a high n-gram score is calculated is identical for all context frames, and FIG. 4 is a detailed configuration diagram of a grammar error And a method for correcting a grammar error using a correction device.

Referring to FIG. 1, a grammar error correcting apparatus according to an embodiment of the present invention includes an error candidate generating unit 100, a router 200, a score calculating unit 300, and a word correcting unit 400.

The error candidate generation unit 100 generates an error candidate word and an error candidate group based on the grammar error corpus for the input text. The error candidate generation unit 100 generates n error words having the highest frequency for a specific word as error candidate words and error candidates based on the grammar corpus constructed in advance.

The error candidate generation unit 100 may extract the error candidate word and the error candidate group from the tagged grammar error corpus. For example, if a is tagged as an, and a is tagged as changing to the, then a can be changed to an or the. Therefore, it can be expressed as follows.

a -> {a, an, the}

That is, the error candidate word a can be modified by itself or by an, the. Here, a is the error candidate word, and {a, an, the} is the error candidate.

2 is a detailed configuration diagram of the router of FIG.

The router 200 extracts a context frame set based on the error candidate word and the error candidate group. The router 200 extracts the most appropriate context frame set for the context frame set using the performance calculator 210 that measures the error correction performance for the error candidate word and the error candidate.

The context frame is composed of n previous words and m later words based on the correction target position. This is represented by (n; m). For example, "There is another way to look at it." (1; 1) = way to look, (1; 2) = way to look at in the sentence. If the number of words in the context frame is 5, the n-gram count information of the word is used, but if the number of words is 5 or more, the directionality is considered. For example, count (way; look) information is calculated for (1; 1), but forward and reverse directions are considered for (5; (5: 5 >), the forward n-gram score is calculated in the forward direction and the backward n-gram score is calculated in the reverse direction (5;

The context frame set consists of different context frames. In the number of words constituting the context frame, the minimum number is one and the maximum number is three, centering on the error candidate word. For example, when the set of context frames is M, M = {(1; 1)} {(1; 1), (1; 2)}, ..., 1), (1; 2), (3; 1)}.

The router 200 calculates the most appropriate context frame set from the context frame sets, which are each error candidate word and error candidate group. The router 200 is trained through a router trainer. The input of the router 200 is an error candidate word and an error candidate group, and the output is a context frame set.

The router model can be represented as a single table. The router model is the most appropriate context frame set for each error candidate. The router model is shown in Table 1 below.

Errors Candidate words and error candidates Router output (a → {a, an, the})
(an → {an, a, the}}
(the → {the, a, the}}
(a → {a, Ø)
(an → {an, Ø)
(the → {the, Ø)
(Ø → {a, an, the}} {(1; 2), (2; 1)}
{(1; 2)}
{(1; 2), (2; 1)}
{(1; 2)}
{(1; 1)}
{(1; 2)}
{(1; 1)}

The router 200 completes the table in the router model. Grammar error tagging data and performance calculator 210 are required for router training. The performance calculator 210 calculates a figure of merit indicating how accurate the error candidate is when the error is corrected. In addition, the performance calculator 210 may calculate a figure of merit indicating how much error has been captured. The router 200 generates all possible context frame sets to train the router. All possible context frame sets are {(1; 1)}, {(1; 2), (1; 1)}, {1; 1, 1, 1, 2, 1, 2, 3, 4, 5, 1)}, {(1; 2), (1; 3), (1; , (3; 1), (1; 1)}, ..., ..., .... The router 200 calculates the performance using the performance calculator 210 for all possible context frame aggregates. The router 200 sets the calculated result set of the context frame having high performance as a result value of the router 200 and stores it in the router model. The router 200 performs all error candidate words and error candidates in this manner and completes the router model as shown in Table 1.

The score calculation unit 300 calculates an n-gram score for the set of context frames. The score calculation unit 300 calculates an n-gram score using a word-based number or a probability value corresponding to each word sequence of the context frame set in the existing native-language document. The score calculator 300 calculates the count information when the number of words of the context frame set is 5 or less, the forward n-gram score in the forward direction and the backward n-gram score in the case of five or more words, and the backward n-gram score in the n-gram score do.

The score calculation unit 300 may use the count information of the Google n-gram corpus or use an n-gram model based on a RNN (Recurrent Neural Network). If the number of words in the context frame is 5 or more, the score calculation unit 300 calculates an n-gram score using an RNN-based n-gram model. RNN-based n-grams can be calculated by dividing them into forward and backward words according to forward and backward directions. The score indicates how much of the sequence of consecutive words in the frame appears in the corpus. For example, the input I am a boy is much larger than the actual English native speaker corpus than I is a boy, so the n-gram score has a high score.

3 is a detailed configuration diagram of the word correcting unit of FIG.

The word correcting unit 400 corrects the input text with the degree of matching of the highest word using the n-gram score. The word correcting unit 400 corrects the case where the words of the context frame in which the high n-gram score is calculated coincide with all the context frames.

The word correcting unit 400 decides whether or not to correct the error candidates. The word correcting unit 400 obtains the most appropriate context frame set corresponding to the error candidate word and the error candidate group from the router 200. The word correcting unit 400 calculates a score using the score calculating unit 300 for all the context frames of the context frame set. The word correcting unit 400 calculates an n-gram score using each context frame for each error candidate group, and determines whether the word of the error candidate group, which has the highest n-gram score for each context frame, Calibration is carried out. For example, if the word of an error candidate is {(1; 1), (2; 1), (1; 2)} for the error candidate word a-> {a, an, The government 400 performs the following calculation as shown in Table 2 in order to make a correction to the following sentence a.

I have an apple and ...

(1: 1) (2: 1) (1; 2) A C (Have a apple) C (I have a apple C (a apple and) An C (Have an apple) C (I have an apple C (an apple and) The C (Have the apple) C (I have the apple C (the apple and)

If the n-gram score is highest when an is inserted in all cases for (1; 1), (2; 1), (1; 2), the word correcting unit 400 changes a to an.

4 is a flowchart illustrating a method for correcting a grammar error using a grammar error correction apparatus according to an embodiment of the present invention.

The grammar error correction device generates an error candidate word and an error candidate word based on the grammar error corpus for the input text (510). The grammar error correction device generates n most frequently used words for a certain word as error candidate words and error candidates based on a grammar corpus constructed in advance.

The grammar error correction device extracts a context frame set based on the error candidate word and the error candidates (520). The grammar error correction unit extracts the most appropriate context frame set for the context frame set by using the performance calculator that measures the error correction performance for the candidate word and the candidate word for error. The grammatical error correction device calculates the error correction performance using the performance calculator that measures the correctness of error correction and how much error has been caught, and extracts the most appropriate context frame set with high error correction performance.

The grammar error correction device calculates an n-gram score for the context frame set (530). The grammar error correction device calculates an n-gram score from a word dictionary or a probability value corresponding to each word sequence of the context frame set in an existing native speaker document. The grammar error correction device calculates the count information when the number of words in the context frame set is 5 or less, the forward n-gram score in the forward direction in the case of 5 or more, and the backward n-gram score in the n-gram score.

The grammatical error correction unit corrects the input text using the n-gram score and the degree of match of the highest word (540). The grammatical error correction device corrects the case where the words of the context frame that have a high n-gram score are matched for all the context frames. The grammatical error correction device finds the words of the context frame in which the high n-gram score is calculated, and corrects the input text to the words of the context frame found if the words of the found context frame match for all the context frames.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the present invention as defined by the following claims It can be understood that

100: error candidate generation unit 200: router
300: score calculating unit 400: word counting unit

Claims (12)

An error candidate generation unit for generating an error candidate word and an error candidate group based on a grammar error corpus for the input text;
A router for extracting a context frame set based on the error candidate word and the error candidate group;
A score calculating unit for calculating an n-gram score for the context frame set; And
And a word correcting unit for correcting the input text using the n-gram score,
The router includes:
Wherein a context frame set having a high error correction performance is extracted from the context frame set by measuring an error correction performance of the error candidate word and the error candidate group. The method according to claim 1,
Wherein the error candidate generating unit comprises:
Wherein the grammar error correcting unit generates n words having the highest frequency for a specific word as the error candidate word and the error candidate group based on a grammar corpus constructed in advance. delete The method according to claim 1,
The point calculating unit calculates,
The grammar error correcting device calculates the n-gram score from the existing native speaker document as a word bins or a probability value corresponding to each word list of the context frame set. The method according to claim 1,
The point calculating unit calculates,
Wherein the counting information is counted when the number of words of the context frame set is 5 or less and the forward n-gram score and the number of backward n-grams are calculated when the number of words of the context frame set is 5 or more. The method according to claim 1,
The word correcting unit,
And corrects the input text if the words of the context frame in which the high n-gram score is calculated match for all the context frames. Generating an error candidate word and an error candidate group based on a grammar error corpus for the input text;
Extracting a context frame set based on the error candidate word and the error candidate group;
Calculating an n-gram score for the context frame set; And
Modifying the input text using the n-gram score,
Wherein the extracting of the context frame set comprises:
Wherein a context frame set having a high error correction performance is extracted from the context frame set by measuring an error correction performance of the error candidate word and the error candidate group. 8. The method of claim 7,
Wherein the generating comprises:
A grammar error correction method for generating n words having the highest frequency for a specific word as the error candidate word and the error candidate word based on a grammar corpus constructed in advance. delete 8. The method of claim 7,
Wherein the calculating step comprises:
Grammar error correction method for calculating the n-gram score from a word-of-mouth count or a probability value corresponding to each word list of the context frame set in an existing native-language document. 8. The method of claim 7,
Wherein the calculating step comprises:
Counting information when the number of words of the context frame set is equal to or less than five, and forward n-gram score and backward n-gram score when the number of words of the context frame set is five or more. 8. The method of claim 7,
Wherein the modifying comprises:
A grammar error correction method for correcting the input text when a word of a context frame set having a high n-gram score is matched for all context frame sets.

Images