KR20240081201A - Language Learing Method Using Exploratory language-learning system based on ML, NLP and a pattern-based reference library - Google Patents

Abstract

The present invention provides a language-learning method by means of an exploratory language-learning system based on machine learning, natural language processing, and a pattern-based reference library using the exploratory language-learning system based on machine learning, natural language processing, and a pattern-based reference library, comprising: a first step of determining whether a logged-in user has selected a learning sentence; a second step of after the first step, segmenting the learning sentence if the user has selected the learning sentence; a third step of displaying a lesson matched with a pattern within the sentence on a screen; a fourth step of proceeding with the lesson by the user and, after learning, displaying the results of the learning lesson; and a fifth step of after the fourth step, generating a quiz by a quiz generation bot by inquiring about the user's interest in sentence learning and lesson learning from a database if the user requests the quiz, wherein after the fifth step, the user receives the quiz, performs a quiz learning step, and then a learning history is stored in the database.

Description

Language learning method using machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system {Language Learning Method Using Exploratory language-learning system based on ML, NLP and a pattern-based reference library}

The present invention relates to a language learning method using machine learning, natural language processing, and a pattern-based reference library-based exploratory language learning system. More specifically, it relates to a computer designed as a learning aid for individuals learning a foreign language or students using their native language. As a base system, it enables navigation and a self-learning mode in which instances generated for the target language are analyzed by artificial intelligence (AI) and natural language processing components and then automatically mapped to appropriate items in the associated reference library. , a language learning system based on machine learning, natural language processing, and pattern-based reference libraries that allows users to explore parts of speech, word meanings, syntactic structures, grammar, and idiom patterns used in the language instance. It is about a language learning method using a new form of machine learning, natural language processing, and a pattern-based reference library-based exploratory language learning system that allows learning based sentences.

With the development of Internet services, the number of people enjoying video content from other countries using personal terminals or electronic devices is increasing.

In particular, due to the influence of K-POP, the number of people in English-speaking countries watching not only Korean pop videos but also various video contents such as movies is increasing.

However, for example, when you want to learn Korean through subtitles or look up the English version of the Korean language while watching a Korean video using a video viewing program such as Viki, which uses English, there is no immediate corresponding service. This is the situation.

[Prior art literature]

Republic of Korea Patent Registration No. 10-1578991 (registered on December 14, 2015) (Name of invention:

Memorizing web browsing operations using ＤＯＭ isomorphism)

The present invention was proposed in consideration of the above-described conventional situation, and the purpose of the present invention is to provide a computer-based system designed as a learning aid for individuals learning a foreign language or students using their native language, enabling navigation and learning the target language. A self-learning mode in which instances generated for a language are analyzed by artificial intelligence (AI) and natural language processing components and then automatically mapped to appropriate items in a linked reference library, allowing the user to identify parts of speech and words used in the language instance. A new form that allows users to learn sentences based on their desired language using a search language learning system based on machine learning, natural language processing, and pattern-based reference libraries that enable exploration of meaning, syntactic structure, grammar, and idiom patterns. It provides an exploratory language learning system based on machine learning, natural language processing, and pattern-based reference libraries.

In the language learning method using machine learning, natural language processing, and pattern-based reference library-based search language learning system using system learning, natural language processing, and pattern-based reference library-based search language learning system according to an embodiment of the present invention,

The first step is to determine whether the logged in user has selected a learning sentence,

After the first step, a second step of sentence decomposition when the user selects a learning sentence,

The third step of displaying lessons matching the pattern in the sentence on the screen,

In the third step, the user proceeds with the lesson and, after learning, the fourth step in which the results of the learning lesson are displayed,

After the fourth step, if the user requests a quiz, a fifth step in which the quiz creation bot creates a quiz by searching the database for the user's sentence learning, lesson learning interests,

A language learning method is provided through machine learning, natural language processing, and pattern-based reference library-based navigation language learning system, where the user receives the quiz, performs the quiz learning step, and then the learning history is stored in a database. do.

In addition, it is characterized by decomposing sentences through four databases (pattern DB, morpheme DB, grammar GLR (Generalized LR Parser), and reference DB), and displaying the decomposed sentences in the form of a message so that they can be learned.

In addition, after the user proceeds with the lesson, the learned learning lesson results are sent as a message, and the learning lesson history is stored in the learning lesson history database for ML.

In addition, the lesson learning is presented in consideration of the level according to the user's initial input information, and the lesson learning material is stored and updated by level in the database.

Additionally, the quiz creation bot is characterized in that it searches the Internet and creates a quiz.

In addition, the exploratory language learning system based on machine learning, natural language processing, and pattern-based reference libraries is

An input means for collecting or inputting instances occurring in a language;

A morphological analyzer that generates linguistic units, words, stems, affixes, and pronunciation symbols from sentences input through the above input means, as well as morphological reductions and parts of speech;

a reference library indexed by word morphology and syntactic structure patterns;

A lexical pattern matcher that processes the content processed by the word form analyzer with reference to a reference library,

A phrase-structure parser that recognizes larger phrases and clause structures by processing the word structure processed by the vocabulary pattern matcher into a syntactic structure;

A library pattern-matcher that processes the syntax structure processed through the syntax structure parser by referring to the reference library, and

A language learning method is provided through a machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system that includes a reference explorer that allows users to view and explore related reference materials.

As described above, according to the machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system and method according to the present invention, it is a computer-based system designed as a learning aid for individuals learning a foreign language or students using their native language. , a self-learning mode that enables navigation and where instances generated for the target language are analyzed by artificial intelligence (AI) and natural language processing components and then automatically mapped to appropriate items in the associated reference library, allowing the user to It has the effect of allowing you to explore the parts of speech, word meanings, syntactic structures, grammar, and idiom patterns used in that language instance.

Figure 1 is a functional diagram schematically explaining the learning process in the machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system according to the present invention.
Figures 2 and 3 are flowcharts showing an example of a language learning method through the machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system of Figure 1.

Hereinafter, a language learning method using machine learning, natural language processing, and a pattern-based reference library-based exploratory language learning system according to the present invention will be described in detail with reference to the accompanying drawings.

The applicant of the present invention (Mirinae Technology) helps foreigners learn Korean through machine translation, etc. through self-developed sentence analysis tools, Korean lessons by level, and Q&A when foreigners study Korean using machine translation.

Recently, many foreigners who have been exposed to Korean culture through K-pop and K-drama have started studying Korean using Google and Papago Translate. Helps me learn.

The translation function in current conventional technology is the result of combining machine learning and natural language processing.

This method has no problems with translation between Latin languages, but it is not compatible with Hangul, which has different grammar and rules.

Therefore, the present invention uses machine learning, natural language processing, and the Jeongum language engine, a proprietary Korean compiler, to not only translate sentences but also decompose them by up to 95%.

The Korean compiler is a tool used to convert the rules that make up Korean grammar into other languages. As the structure becomes more sophisticated, translation is done based on rules.

Existing machine learning and artificial neural networks analyze based on translated sentences, but compilers analyze based on unique rules.

Existing Korean translation technology only provides possible meanings from nearby words and cannot even explain the implied meaning.

It is insufficient to understand the syntactic and grammatical structures of sentences necessary for learners.

Mirinae's technology combines over 600 sentence patterns used in the Korean language to analyze the syntax of sentences in an in-depth and ecological way.

“Currently, this analysis method is the only one in the industry,” he said, introducing the Hangul static sound engine of the applicant (Mirinae Technology) of the present invention.

In particular, it has the ability to create a computer language and is created from the perspective of a person learning Korean, so it is bound to be unrivaled.

For example, when analyzing ‘I can speak Korean,’ I divide it into ‘I can speak Korean’, ‘I can speak Korean’, ‘I can speak Korean’, ‘I can speak Korean’, and then the subject sentence, object, and verb of each sentence are analyzed separately, and even the grammar rules are classified one by one. It also interprets how each meaning changes and what meaning it contains.

Thanks to this, you can learn by understanding the hidden meaning of K-pop lyrics or movie lines, and you can also take in-depth learning through separate lessons and libraries.

In addition, existing education methods take several years or more to reflect new words, but the learning method of the applicant of the present invention (Mirienae Technologies) can quickly respond to the latest trends, such as immediately reflecting new words.

A strategy to increase accessibility is to provide services not only based on web browsers but also by introducing browser extensions.

Figure 1 is a functional diagram schematically explaining the learning process in the machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system according to the present invention.

Referring to Figure 1, the present invention is a computer-based system designed as a learning aid for individuals learning a foreign language or students using their native language. According to the invention, a self-learning mode that enables navigation and where instances generated for the target language are analyzed by artificial intelligence (AI) and natural language processing components and then automatically mapped to appropriate items in the associated reference library. It allows users to explore the parts of speech, word meanings, syntactic structures, grammar, and idiom patterns used in that language instance.

A reference library can contain extensive material on each element of the target language, including usage notes, alternative example uses, links to external reference sites, and book citations. It is indexed by morphological and syntactic patterns found by Machine Learning (ML) and Natural Language Processing (NLP) components, enabling a form of bottom-up learning, where organically generated text drives the learning and discovery process, creating structured It is structured as an extension of the typical top-down approach of courseware and class curriculum.

The invention presented herein can be applied to any combination of target and learner languages by covering multiple learner languages in a particular target language simply by replicating appropriately trained analyzers, parsers and library items, and reference material in each learner language. It can be applied. Versions of the reference material in the target language itself can be used by native speakers of the target language in schools and other educational contexts.

How it Works

One embodiment of the invention includes the following components; Other embodiments are possible, as described in the Examples section below.

1. An input system for collecting or inputting instances occurring in a language.

2. A morphological analyzer that reduces the form of input characters such as linguistic units, words, stems, affixes, and phonetic symbols and generates parts of speech.

3. A phrase-structure parser that recognizes larger phrase and clause structures from the above word reductions.

4. Reference library indexed by morphological and syntactic patterns.

5. Pattern-matching system that finds reference library items in which patterns exist through analysis and parsing of searched instances.

6. A presentation system that allows users to view and navigate related reference material.

In Figure 1, a machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system includes an input means for collecting or inputting instances generated in a language, and sentences input through the input means into linguistic units, words, and stems. , a morphological analyzer (10) that reduces the morphological form of input characters with affixes and phonetic symbols and generates parts of speech;

a reference library indexed by word morphology and syntactic structure patterns (50);

A lexical pattern matcher (20) that processes the content processed by the lexical analyzer with reference to a reference library,

A phrase-structure parser (30) that recognizes larger phrases and clause structures by processing the word structure processed by the word pattern matcher (20) into a syntactic structure;

A library pattern-matcher (40) that processes the syntax structure processed through the syntax structure parser (30) by referring to the reference library, and

Includes a reference explorer 60 that allows users to view and navigate related reference materials.

The input means is a means that allows a specific language to be displayed programmatically or hardware-wise within a specific video (streaming or not limited thereto) (device), or a video input device that provides a service for outputting movies, plays, and various videos. It may be an input device that can be operated in connection with the output of the text screen. However, it is not limited to this, and may be an input device that is connected to a separate video output device as an external input device and operates by overlaying text on the video or displaying it at a specific location on the screen. This input device can be operated in conjunction with the language learning system according to the present invention.

These components operate as shown in Figure 1.

Step ①, the user enters the text they want to search for. In the example above, the student is learning Korean and types the sentence “I can ride a bicycle” in Korean.

② Step, the morpheme analyzer processes the input sentence and generates a breakdown of the sentence. Step ③, separate and label various types of affixes, separate verb stems, and assign appropriate parts of speech to all morphemes. Current morphological analyzers are typically deep neural networks, and the present invention allows the use of existing analyzers or new analyzers specifically trained for this purpose. This is covered in more detail in the word form analyzer section below.

Step ④, an initial “lexical” pass of the pattern-matcher is formed over the morpheme-based patterns of the library. Matched patterns can annotate and transform morpheme structures, creating potential complex morpheme structures, such as the “ㄹ can” (= can do) auxiliary verb pattern in ⑤. These annotations and translations are passed on to the reference explorer, but can also be of great help in the next syntactic parser step.

In step ⑥, syntactic structure parsing (parsing) is applied to the annotated modified word structures, and larger and possibly overlapping phrase and clause structures are identified, as shown in step ⑦. This parsing step is performed using standard NLP or computer language parsing methods such as chunking grammars or recursive-descent parsers. The present invention allows the use of such parsing techniques. In some embodiments of the present invention, step ④ or step ⑥ can be omitted without loss of generality, but both cannot be omitted.

Step ⑧, another pass of the pattern-matcher is applied, which includes patterns referencing syntactic structures and morphological structures. It attaches a different set of annotations to morphemes and parsing structures, and links that set of annotations back to every referenced item in the library, ready for display and navigation in the reference explorer.

⑨ Steps, all analysis and pattern matching results can be reviewed by the user. In one embodiment of the invention, results are presented graphically and in a progressively interactive format, allowing the user to drill-down to the parts of the analysis that are most interesting (such as related text or icons to find more information). Click to search as if you were penetrating. The presentation system can save the analysis and navigation status of some or all of the input text so that the user can continue learning at a later time.

In other embodiments, some elements of the analysis may incur long latencies (translation and word lookup from external services), in which case the entire analysis may be passed in parts to the ⑨ stage display system for immediate display. For this purpose, low-latency elements are provided, and longer-latency elements are provided asynchronously so that they can be displayed when they are available.

Example

Various elements of the invention may exist in multiple embodiments. Some of them are presented in more detail below.

The morphological analyzer

Form analyzers perform a well-known task in the field of natural language processing, and there are several implementations of them. Currently, the most popular approach is to use deep neural-net models trained on a corpus of existing word-form parsing. There are existing models available in a variety of languages, or new models can be trained.

Most word-form analyzers are not perfect. The state-of-the-art of CNN (Convolutional Neural Network)-based analyzers ranges from 97% to 98%, and generally requires the use of error-correction or accommodation schemes. Since the word-form analysis errors that occur in the present invention can be captured as transformation patterns in the pattern library, this type of error correction becomes part of the word-type structure transformation covered in step ④ of the above operation overview.

Rather than adding to existing word-analyzer construction techniques, the present invention uses state-of-the-art analyzers in a new approach to language learning.

Phrase-structure parser

In a similar way to the word shaper component, syntactic parsing is a well-known element of natural language processing and computer language implementations, as well as chunking grammars, declarative parser generators or ad-hoc recursive descent. hoc recursive-descent) can be used as a general approach to implementing such components, such as parsers.

Ad-hoc recursive-descent parsers have the advantage of accommodating context-sensitive parsing, which may be necessary for some grammars, in the context of natural language parsing.

Additionally, given encoded learning data, a neural network-based analyzer can be trained to recognize and output both word-form destructuring and syntactic structuring.

Extensive and complete syntactic parsing is not essential for this application; parsing has primarily an educational purpose, and often simpler, coarse-grained structures are easier for students to understand. Another purpose is to help recognize grammar or idiom patterns, so only syntactic structures essential for that purpose are needed so that the parser can recognize them.

Pattern-based reference library

The pattern-based reference library is a core component of the present invention and has many aspects and possible embodiments that are novel and worthy of development within the present invention.

Pattern schemes

The structure of the text being studied is linked to the transformation or explanation items in the library by distinguishing patterns in the word morphology and syntactic structure of the text. This pattern definition and matching is the core mechanism of the present invention, and it accommodates all schemas that perform this reliably and efficiently.

One embodiment encodes the morphological structure of some text into a single string, and the patterns of library items are expressed as regular-expressions for tokens of the string. For example, the morpheme structure of the sample Korean sentence in the diagram above (=“I can ride a bicycle”) can be expressed as follows:

;that: NP;is: J

A sequence consisting of a pair of morpheme + part of speech separated by a semicolon. Part of speech is a common tag code used in NLP, where NP = proper noun, NNG = common noun, VV = verb, etc.

Then, a generalized pattern that recognizes the ‘ㄹ/may’ form can be a regular expression:

( ( [ ^: ] +) : V[ A- Z] +) (ㄹ|eul) : ETM;Number: NNB;Be: VV

It matches all verbs preceded by ‘ㄹ’ or ‘eul’ particles (particles) in front of the dependent noun ‘number’ and all conjugations of the auxiliary verb ‘to be’.

As the number of pattern libraries increases, sequential testing of all patterns may be computationally rejected, so search optimization is necessary. A simple example of optimization could be trie-coding of all patterns for common prefixes (tree refers to an ordered tree data structure).

Guided- and sample-trained patterns

The patterns that distinguish each construct of pedagogical value in the target language can be constructed in several ways. In one embodiment, the patterns are hand-coded by a grammar and education expert in the target language. In other cases, an expert uses examples from text that contain instances of a particular pattern and guides the learning of that pattern. In another embodiment, many examples of text containing instances of a pattern are used to train a neural network-based regular expression generator for the pattern.

Direct neural-net structure recognition

The pedagogical structure of the original sentence may also be discovered directly by a specially trained neural network in another embodiment of the reference library indexing scheme. In this case, the training data is a corpus of text component sentences labeled with direct links to reference library items. A two-step approach is also possible, in which a guided set of patterns or a manual set of patterns are exploited to generate a large corpus labeled with an index of library items, resulting in a trained neural network that performs significantly better than bulk regular expression searches. This method is possible.

Multiple learner languages

The morphological analyzer, syntactic parser, and pattern set are specific to each target language (the language being studied) and are developed once for that language. An example of a reference library where all descriptions, labeling, and translations are organized as vectors, split by learner language (student's native language or fluent language) so that the same library and target language component can easily support multiple learner languages. .

Idiom and sentence patterns as well as formal grammar

In some embodiments of the invention, the reference library will contain most or all of the useful standard vocabulary, syntax, and grammatical structures of the target language. In other embodiments, the library could be expanded to include patterns and descriptions for idioms or slang phrases, since the presence of these forms in instances created from a language is a common source of difficulty for language learners. The same mechanisms used to define, recognize, and describe grammatical patterns can be used for idiomatic forms.

Additionally, there are sentence-level patterns that are generally well-known equivalents of common expressions between the two languages. For example, the general form “I hope These sentence-level patterns can also be defined and recognized by the mechanism of the present invention, and thus will also be added to the reference library in some embodiments.

NLP-discovered example uses

A particularly useful pedagogical element of a learning tool, as described here, is to provide many examples of the use of the constructs (words) being studied. One embodiment of this system uses pattern definitions to automatically discover such examples in existing corpora of the target language.

External references

Another useful element in networked embodiments of the present invention is to provide links within reference library items to external sources of additional training or background information. This includes links to relevant pages on traditional learning sites, YouTube videos, online reference books, or cultural or historical sites. Embodiments of the invention that support external references may include these types of links and other citations that may be deemed useful.

Crowd-sourcing

The construction of a reference library containing the standard vocabulary, syntax, and grammatical structures of a language is a relatively limited exercise, similar to that required to create a grammar textbook for a language. Extending the library to include other material such as idioms, slang, sentence patterns, extensive examples and external reference links becomes more of a task that can be implemented incrementally over time.

Possible embodiments of the invention include a crowdsourcing system that invites contributions to the library from all users, with typical crowdsourcing control over quality and content found in other crowdsourcing content services such as Wikipedia or Wiktionary. An intermediate embodiment of this idea involves a generic open-text-feedback system reviewed and curated by the service's language-teaching experts to extract and enter the library. do

Crowdsourcing techniques may also be used in embodiments of the present invention to source translations of learner language reference material as mentioned in the Multiple Learner Languages section above.

Figures 2 and 3 are flowcharts showing an example of a language learning method through the machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system of Figure 1.

Referring to Figures 2 and 3,

First, when a user signs up (S2), the system stores it in the user DB.

Users can select interests in advance (S4), and this information is collected from the operator (e.g., the first screen of the website operated by “Miri Nae Technologies”, https://mirinae.io/#/) through the user login (S6) process. , if you enter the learning sentence you want to learn (S8),

The system receives training sentences and decomposes the sentences. At this time, the sentence is decomposed through four databases (pattern DB, morpheme DB, grammar GLR (Generalized LR Parser), reference DB), and the decomposed sentence is displayed in the form of a message so that it can be learned (S10) (see drawing for reference) This indicates that all results displayed in the shape of a letter, starting with a small original letter and a dotted line, are displayed in the form of a message).

Afterwards, the user learns the sentence (S12), and the learned sentence is sent to the system and stored as a learning history. In other words, it is stored in the learning history DB for ML.

After step S12, a lesson matching the pattern in the sentence is displayed on the screen (S14). Afterwards, when the user wants to proceed with the lesson (S16), the lesson learning screen is displayed (S18). After the lesson learning (S16) step, when the learned lesson results are delivered as a message, the learning lesson history is stored in the learning lesson history database for ML.

(On the other hand, it is of course possible that the lesson learning can be presented considering the level according to the user's initial input information. The lesson learning material is stored and updated by level in the database.)

After step S18, it is determined whether there is a quiz request (S20). If there is a quiz request, a quiz request message is sent to the system, and the user's interest in sentence learning and lesson learning is searched through the database and the quiz creation bot searches the Internet. Create a quiz.

When the user receives the quiz (S24), the user performs the quiz learning step (S26). Afterwards, the learning history is stored in the database (S28), and then it is determined whether to continue learning (S30), and the lesson learning (S26) is performed. S18) can continue.

Although the technical idea of the present invention has been described in detail according to the above preferred embodiments, it should be noted that the above-described embodiments are for illustrative purposes only and are not intended for limitation. Additionally, an expert in the technical field of the present invention will understand that various embodiments are possible within the scope of the technical idea of the present invention.

10: Morphological analyzer
20: Lexical pattern match
30: Phrase-structure parser
40: Library Pattern match
50: reference library
60: reference explorer

Claims (6)

Machine learning, natural language processing, and pattern-based reference library-based search using a language learning system. In the language learning method using machine learning, natural language processing, and pattern-based reference library-based search language learning system,
The first step is to determine whether the logged in user has selected a learning sentence,
After the first step, a second step of sentence decomposition when the user selects a learning sentence,
The third step of displaying lessons matching the pattern in the sentence on the screen,
In the third step, the user proceeds with the lesson and, after learning, the fourth step in which the results of the learning lesson are displayed,
After the fourth step, if the user requests a quiz, a fifth step in which the quiz creation bot creates a quiz by searching the database for the user's sentence learning, lesson learning interests,
A language learning method through machine learning, natural language processing, and pattern-based reference library-based exploration language learning system, wherein the user receives a quiz, performs a quiz learning step, and then the learning history is saved in a database. According to claim 1,
Machine learning and natural language processing that decomposes sentences through four databases (pattern DB, morpheme DB, grammar GLR (Generalized LR Parser), reference DB) and displays the decomposed sentences in the form of a message so that they can be learned. and a language learning method using a pattern-based reference library-based exploratory language learning system. According to claim 1,
Machine learning, natural language processing, and pattern-based reference library-based search, characterized in that after the user proceeds with the lesson, the learned learning lesson results are sent as a message, and the learning lesson history is stored in the learning lesson history database for ML. Language learning method through language learning system. According to claim 1,
The lesson learning is presented in consideration of the level according to the user's initial input information, and the lesson learning material is stored and updated by level in the database. A machine learning, natural language processing, and pattern-based reference library-based search language learning system. Language learning method through. According to claim 1,
A language learning method through a machine learning, natural language processing, and pattern-based reference library-based exploratory language learning system, wherein the quiz creation bot searches the Internet and generates a quiz. The method of claim 1, wherein the exploratory language learning system based on machine learning, natural language processing, and pattern-based reference libraries is
An input means for collecting or inputting instances occurring in a language;
A morphological analyzer that generates linguistic units, words, stems, affixes, and pronunciation symbols from sentences input through the above input means, as well as morphological reductions and parts of speech;
a reference library indexed by word morphology and syntactic structure patterns;
A lexical pattern matcher that processes the content processed by the word form analyzer with reference to a reference library,
A phrase-structure parser that recognizes larger phrases and clause structures by processing the word structure processed by the vocabulary pattern matcher into a syntactic structure;
A library pattern-matcher that processes the syntax structure processed through the syntax structure parser by referring to the reference library, and
A language learning method through a machine learning, natural language processing, and pattern-based reference library-based navigation language learning system that includes a reference explorer that allows users to view and explore related reference materials.

Images