Method of Analyzing Sentences for English Learning

Abstract

A method for analyzing sentences for English learning is disclosed. The most significant feature of the method of the present invention is that, in sentence parsing, it utilizes tags based on traditional educational grammar, but processes even ungrammatical sentences through bad tags, thereby accurately identifying what constitutes an error. Furthermore, through universality calculations, sentences with frequently used structures can be analyzed more accurately, making it particularly useful for English learning.

Description

Technology Field

[0001] The present invention relates to a sentence analysis method for English learning, and more specifically, to a sentence analysis method that utilizes traditional educational grammar in the development of a sentence parser but processes even ungrammatical error sentences through bad tags to accurately identify what constitutes an error, and furthermore, through universality calculation, can analyze sentences with frequently used structures more accurately. Background Technology

[0002] Online English learning should be explained by dividing it into two categories: learning that uses the internet merely as a medium, and learning that proceeds by accessing an English learning system online. Most online English learning commonly seen around us involves uploading videos, Flash content, or web pages to a server for learners to access and view; this falls under the category of learning that uses the internet as a medium. In contrast, an English learning system refers to a system that automatically evaluates learners, designs learning plans tailored to learning objectives based on the evaluation results, and manages the progress of that learning. However, such English learning systems remain at the research stage and are not yet widely adopted.

[0003] While there may be various reasons why English learning systems have not been widely adopted, the most certain reason is that computer systems are not yet smart enough to teach anyone. Here, "not smart" refers not to limitations in computing technology or processing speed, but rather to limitations in the computer's natural language processing capabilities. A learning system can only function properly if the computer can understand human speech; if it cannot, no matter how excellent the learning system is, it is of no use. The engineering technology that enables computers to understand human speech—that is, natural language—is called "Natural Language Processing," and the representative software that performs this function is the "Sentence Parser." Sentence parsers have a history spanning several decades, and widely known examples include Stanford University's Stanford Parser, Carnegie Mellon University's Link Grammar Parser, and the University of Berkeley's Berkeley Parser; there have also been numerous attempts to implement online English learning using these parsers. However, the performance of these sentence analyzers is not yet excellent enough to be applied to actual services, and since they are typically developed as part of computer engineering, they use machine grammar specialized for engineering rather than the traditional grammar inherent to language. Consequently, there are inevitably many difficulties in implementing English learning systems based on these machine grammars. In particular, because these sentence analyzers cannot process sentences with ungrammatical errors that frequently occur among English learners, it is not easy to create a proper English learning system.

[0004] The prior art related to sentence analyzers for English learning includes the following. In Korean Patent Publication No. 10-2014-0021838 (Title of Invention: Method for Detecting Grammatical Errors and Error Detection Device Therefor), errors in a sentence are detected by simultaneously applying a method that separates a written input sentence into morpheme units to tag parts of speech, detects simple grammatical writing errors for each tagged column of parts of speech, and simultaneously detects syntactic analysis errors by parsing based on rules defined by the connection relationships between adjacent parts of speech, and a method that generates a sequence of morphemes grouped by an arbitrary window size from the separated morphemes, searches each sequence of morphemes in an example-based database, and detects errors for each morpheme based on examples by analyzing the frequency at which each morpheme and its corresponding sequence of morphemes are aligned together. Additionally, in Korean Patent Registration No. 10-0892004 (Title of Invention: Device and Method for Providing Verb-Centered Automatic Grammar Error Detection and Correction Information in an English Writing Learning System), input A technology is disclosed that automatically detects verb-centered grammatical errors in writing and provides correction information based on the detected grammatical errors. Additionally, Korean Patent Registration No. 10-15225220000 (Title of Invention: Example-based Error Detection System and Method for Automatic Writing Evaluation) discloses a technology that collects example sentences containing various writing styles to build an example-based database, separates the input sentence into morpheme units when a written input sentence is input, generates a sequence of morphemes grouped by a certain window size from the separated morphemes, and then analyzes each sequence of morphemes using the frequency of the sequence of morphemes appearing in the example-based database to detect errors for each morpheme and suggest corrections for the errors.

[0005] Table 1 below is a diagram showing the set of tags used in a sentence analyzer at Stanford University as an example of prior art. Referring to Table 1, it can be seen that there are a total of 62 tags, but tags for sentence errors are not included. Most sentence analyzers currently under development use this number of tags.

[0006] The aforementioned prior art technologies, which detect grammatical errors using analysis results obtained from general sentence analyzers, employ a method where learners' erroneous sentences are not processed at all (system error), or even if processed, they may be handled in a structure completely different from the learner's intent (analysis error), making it difficult to utilize the results for learning. Therefore, to create a proper English learning system, a learning sentence analysis method is required that utilizes traditional educational tags while also being capable of processing learners' erroneous sentences. The problem to be solved

[0007] Therefore, the present invention is proposed to satisfy the aforementioned necessity, and the problem that the present invention aims to solve is to provide a sentence analysis method for English learning that utilizes traditional educational grammar rather than mechanical grammar specialized for engineering, processes even the learner's erroneous sentences to determine what constitutes an error, and can analyze sentences with structures frequently used for English educational purposes more accurately. means of solving the problem

[0008] To achieve the above objective, the present invention provides a method for constructing an optimal tag set that can extract all syntactic and morphological learning elements constituting any English sentence without causing performance degradation. The tag set includes bad tags capable of tagging errors frequently made by learners, and provides a method for the bad tags to function effectively during the sentence analysis process. Furthermore, an English learning sentence analyzer according to one aspect of the present invention is characterized by having a post-process that corrects the analysis results to suit educational purposes through universality calculation. Effects of the invention

[0009] According to the present invention as described above, since the result of sentence analysis consists of learning tags that follow traditional educational grammar, various learning elements constituting the sentence can be extracted along with sentence structure information, making them immediately usable in an English learning system. In particular, it is possible to process even the learner's ungrammatical error sentences and identify what the problem is through error tags (Bad Tags), thereby enabling the provision of accurate feedback to the learner. Furthermore, the accuracy of the analysis can be improved for sentences with frequently used general structures, making it possible to perform sentence analysis optimized for English learning. Brief explanation of the drawing

[0010] FIG. 1 is a flowchart illustrating the process of implementing a sentence analysis method for English learning, which includes a post-processing step according to an additional embodiment of the present invention. Specific details for implementing the invention

[0011] Table 2 is a set of tags with newly added error tags to apply the sentence analysis method for English learning according to the present invention, which is not present in Table 1; a total of 182 tags are used in addition to the tags in Table 1. Since sentence analyzers based on a statistical approach usually produce results based on probability calculations using MLE (Maximum Likely Estimation), as the number of tags increases, the computational capacity increases and the probability inevitably decreases. Therefore, it is important to configure the tags of a learning sentence analyzer as a minimum set of tags within a range that allows for the explanation of all grammatical components constituting a sentence.

[0012] Unlike native speakers, when learners encounter complex sentences, it becomes difficult to distinguish where the subject begins and ends and where the verb begins and ends.

[0024] If this distinction is made accurately, interpreting the sentence becomes much easier. General sentence analyzers do not use tags for the subject part (SUBJECTIVE) and the predicate part (PREDicative). In the case of a very simple sentence consisting of a single subject pronoun (PRP), such as 'I am a boy.', the subject and verb can be quickly distinguished from the analysis results of a general sentence analyzer, as shown in Fig. 1. However, in the case of a complex sentence such as 'Going on holiday always makes me feel nervous.', an English learning system using a general sentence analyzer would have to create and insert a separate complex routine just to distinguish the subject and predicate parts from the analysis results of the sentence analyzer, and it would not be easy to obtain accurate results with such a routine alone.

[0013] General sentence analyzers typically use only one SBAR tag for subordinate clauses. However, subordinate clauses exist in various forms, such as subordinate interrogative clauses, nominal usages of relative clauses, and adverbial usages, and the interpretation of a sentence can vary significantly depending on the usage. However, according to the sentence analysis method of the present invention, these usages are identified directly through various subordinate clause tags.

[0014] When a verb is used with a particle such as up, down, or out to express a completely different meaning, it is called a phrasal verb; when the original meaning of the preposition remains, it is called a prepositional verb; and when the two are combined, it is called a phrasal-prepositional verb. Conventional sentence analyzers usually use only a single participle tag (PRT) and do not perform such detailed classification. However, native speakers frequently use these phrasal verbs and prepositional verbs, and their meanings can vary significantly depending on the situation. According to the sentence analysis method of the present invention, which utilizes a tag set including error tags and postprocessing, it is possible to distinguish between them, thereby providing significant assistance in sentence interpretation. In other words, the sentence analysis tags used in the sentence analysis method of the present invention include bad tags that can be used to tag error structures frequently caused by learners. These error tags target errors frequently made by learners in sentence substructures, such as noun phrases and adjective phrases. This approach is based on the observation that even low-level learners do not construct nonsensical sentences like "Cat mouse the chased," but rather make errors through specific patterns within phrase-based structures.

[0015] The sentence analysis method for English learning is intended for learners of English, not for native English speakers. Therefore, naturally, there is no need to process special sentences that are typically found in literature or news. Rather, it would be more educationally effective to process sentences commonly used by as many people as possible more accurately. Accordingly, the present invention may include a post-processing step in which a universality check is performed once more according to an option after the sentence analysis is primarily completed. Explanation of the symbols

[0016] -

Claims

Claim 1 A method for analyzing sentences for English learning implemented by a computational processing unit including a computer, comprising: a first step of pre-configuring and preparing a sentence analysis tag set by adding a set of error tags (bad tags), which are tags applied to error structures at the phrase level, to a set of error-free tags that are subdivided to enable grammatical analysis of all components of a sentence and are used for analyzing sentences that are grammatically error-free; a second step of applying a sentence provided by an English learner to the sentence analysis tag set of the first step to obtain an analysis result in which sentence data suitable for English learning is extracted by separating the tagging data of error structures and the tagging data of error-free structures to prevent interference; and a third step of providing sentence error information based on the analysis result to an English learner; wherein the set of error tags is a learning tag further subdivided by word-by-word tagging.

Images