An artificial intelligence-based language expression iterative training system and a method of using the same

Abstract

An AI-based iterative language expression training system and its usage method include the following steps: Constructing a language expression training task library, where each task unit contains first descriptive information (such as topic, keywords) and second descriptive information (such as standard examples, structural analysis); arranging the task units according to difficulty or logical order to form a training sequence; the user learns according to an iterative forward training method, starting from the first task unit and attempting to generate or express corresponding language content based on the first descriptive information; the system calls an AI language model (such as BERT, GPT) to evaluate the user-generated content in multiple dimensions (including logical coherence, semantic accuracy, and subjective emotional consistency); if the evaluation fails, the system pushes second descriptive information for the user to learn and reverts to the first task unit to start again; if the evaluation passes, the user proceeds to the next task unit until all forward training is completed; then, the training sequence is logically reversed, and iterative reverse training is performed, starting from the original last task unit, requiring the user to reverse-engineer or parse the first descriptive information based on the second descriptive information, also employing an iterative reinforcement mechanism of evaluation-learning-reverting. This invention significantly improves users' abilities and proficiency in language organization, logical expression, and creative writing through a closed loop of forward and reverse iterative training and AI intelligent evaluation.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence education applications and language learning technology, specifically relating to an iterative training system for language expression based on artificial intelligence and its usage method. Background Technology

[0002] In Chinese language education, foreign language learning, and creative writing instruction, effectively improving learners' language organization, logical expression, and writing skills has always been a core challenge. Traditional training methods, such as teacher correction, memorizing model essays, and random exercises, suffer from problems such as long feedback cycles, lack of systematic approach, and insufficient personalization. In recent years, while digital learning tools have provided some convenience, most still remain at the level of simple multiple-choice questions, fill-in-the-blank exercises, or vocabulary flashcard memorization, failing to effectively assess and guide the logic, coherence, and creativity of complex language units such as sentences and paragraphs.

[0003] Meanwhile, existing methods mostly employ unidirectional or random practice patterns, lacking a systematic closed-loop mechanism that allows for repeated reinforcement and forced regression after errors. Learners are prone to problems such as "positional memory" (remembering the order of practice rather than the content) or "one-way association" (only able to write down answers based on prompts, unable to reverse-analyze). Although powerful generative AIs such as GPT have emerged, how to systematically and methodically integrate their capabilities into the language expression training process to form an efficient and reliable training method with a closed loop of "teaching, learning, testing, and evaluation" remains an urgent problem to be solved. Summary of the Invention

[0004] To overcome the shortcomings of the existing technologies, this invention provides an iterative training system for language expression based on artificial intelligence and its usage method. This system deeply integrates AI's semantic understanding and generation capabilities, and creatively designs an iterative reinforcement process with a two-stage forward-reverse sequence and error-based rollback mechanism, aiming to systematically improve users' deep language expression abilities.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An iterative training system for language expression based on artificial intelligence and its usage method include the following steps: Step 1: Construct a language expression training task library. This library contains i task units (i≥1), each task unit storing interrelated first descriptive information Ai1 and second descriptive information Ai2. Ai1 is a prompt or triggering information, such as a writing topic, a few keywords, or an incomplete opening; Ai2 is a reference example or analysis, such as a standard short essay, a logical connection of keywords, or a reasonable continuation of an opening.

[0006] Step 2: Arrange the task units according to the increasing language difficulty or logical deduction order to form a fixed training sequence from the first unit to the i-th unit.

[0007] Step 3: Iterative forward training. The goal of this stage is to train the ability to generate data "from prompts".

[0008] Starting with Unit 1, users attempt to write a paragraph based on the prompt (A11). If unable to write, the system displays a standard example (A12) for learning. If successful, the system uses an AI model (such as GPT) to perform a multi-dimensional intelligent comparison (comparing logic, semantics, sentiment, etc.) between the user's writing and the standard example. If the AI ​​determines "failure," the system again displays the standard example, and the user must retrain from the beginning (Unit 1). Only by continuously and correctly completing all units can the user pass this stage.

[0009] Step 4: Logical Reversal. After completing the forward training, the system automatically reverses the order of the task units to prepare for reverse training.

[0010] Step 5: Iterative reverse training. The goal of this stage is to train the ability to reverse analyze "from result to cause / from example to core".

[0011] The user starts with the first unit after inversion (i.e., the original last unit). The system presents only one standard example (e.g., A(i2)) and asks the user to deduce what hint (Ai1) this example was based on. The user's answer is compared with the Ai1 stored in the system using AI. Similarly, any failure (inability to write or incorrect writing) will require the user to start over from the first task unit of the inversion training. Training only ends when all inversion parsing is completed correctly and consecutively.

[0012] The beneficial effects of this invention are: it breaks through the limitations of traditional language learning tools that can only score objective questions, enabling real-time, multi-dimensional evaluation (logic, semantics, emotion) of the subjective expression quality of sentences and paragraphs, providing accurate feedback. It not only trains positive "conception-expression" abilities but also strengthens "analysis-induction" abilities through reverse training, achieving true bidirectional association and deep understanding. The error-reversal mechanism creates "spaced repetition" and "necessary difficulty," greatly enhancing the solidity of memory and skills. It integrates fragmented practice into a rigorous closed-loop teaching methodology of "teaching (examples), learning (user attempts), testing (AI evaluation), and assessment (feedback and reset)," ensuring the controllability and efficiency of training effects. Attached Figure Description

[0013] Figure 1 This is a flowchart of the iterative forward + reverse training method of the present invention.

[0014] Figure 2 This is a schematic diagram of a forward sequence embodiment of the present invention applied to creative writing (theme-model essay) training.

[0015] Figure 3 This is a schematic diagram illustrating the logical reversal and inversion of the present invention applied to creative writing training.

[0016] Figure 4 This is a schematic diagram of an orthogonal embodiment of the present invention applied to logical deduction (keyword-paragraph) training.

[0017] Figure 5 This is a schematic diagram of an embodiment of the present invention applied to logic inversion and reverse order training. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Example 1: Creative Writing Training like Figure 2 As shown, a training set containing three task units is constructed. Unit 1: Ai1 = "A stranger knocks on the door on a rainy night", Ai2 = a 300-word suspenseful micro-fiction sample; Unit 2: Ai1 = "Discovering an ancient key", Ai2 = a 300-word adventure-style sample; Unit 3: Ai1 = "Receiving an unsigned letter", Ai2 = a 300-word emotional-style sample.

[0020] The user begins sequential training. The system presents the topic "A stranger knocks on the door on a rainy night," and the user attempts to write. Upon completion, the AI ​​model evaluates the logical coherence and suspenseful atmosphere of the work. If the score is below a threshold, the system displays a sample essay, which the user must study before starting to write again from the first topic (rainy night). Sequential training only ends after the user has successfully completed writing on three consecutive high-quality topics.

[0021] Subsequently, the system logic is reversed. Reverse training then begins, as follows: Figure 3 As shown, the system first displays a sample letter from Unit 3 (a letter with a specific emotional style). The user needs to deduce the theme of this sample letter. The user answers, "Received an unsigned letter." The AI ​​compares the semantic similarity; if correct, it proceeds to the next item (displaying an adventure sample letter from Unit 2). Any errors require starting from the first item in reverse order (the sample letter from Unit 3) and working backwards.

[0022] Example 2: Logical Deduction Training like Figure 4As shown, a training set containing three task units is constructed. Unit 1: Ai1 = "Spring, kite, children", Ai2 = "A spring breeze caresses the face, children in the square laugh and fly their kites, colorful kites vie for beauty in the blue sky"; Unit 2: Ai1 = "Dusk, old tree, returning birds", Ai2 = "As the sun sets, several birds returning to their nests chirp among the branches of an old locust tree, as if calling to their distant companions"; Unit 3: Ai1 = "Late night, lamplight, desk", Ai2 = "In the still of the night, only the desk lamp is still lit, its warm glow illuminating the open pages of the book."

[0023] The training method is the same as in Example 1. Forward training requires users to write fluent and elegant sentences based on keyword combinations, and the AI ​​evaluates their semantic relevance and grammatical correctness. Reverse training requires users to read an elegant sentence (…). Figure 5 This method allows for the precise extraction of core keyword combinations. It effectively trains users' ability to extract core ideas and create vivid descriptions.

Claims

1. An iterative training system for language expression based on artificial intelligence and its usage method, characterized in that, Includes the following steps: Step 1: Construct a language expression training task library containing i task units, i≥1. Each task unit stores interrelated first description information Ai1 and second description information Ai2. Ai1 is the prompt information used to trigger language generation, and Ai2 is the standard language expression example or structural analysis information corresponding to Ai1. Step 2: Arrange the task units according to language difficulty or logical order to form a training sequence with the order from the first unit to the i-th unit; Step 3: Perform iterative forward training according to the following iterative method: 3.1 Starting from the first task unit, display the user's first description information A11 and attempt to receive the expression content U11 generated by the user based on A11; 3.1.1 If the user is unable to generate a description, the system will automatically push and display the corresponding second description information A12 for the user to learn. After learning, return to step 3.

1. 3.1.2 If the user generates expression content U11, the pre-trained artificial intelligence language model is invoked to evaluate U11 and A12. The evaluation dimensions include semantic similarity, logical coherence and / or subjective sentiment conformity. 3.1.3 If the evaluation result is lower than the preset threshold, it is judged as a failure. The system displays A12 for the user to compare and learn, and returns to step 3.1; 3.1.4 If the evaluation result reaches or exceeds the preset threshold, it is determined to be passed, and then proceed to the next step 3.2; 3.2 Enter the second task unit, display its first description information A21, and receive the user's expression content U21; 3.2.1 If the user is unable to generate or fails the evaluation, the system displays A22 for learning and forces a return to step 3.1 (i.e., restarting from the first task unit). 3.2.2 If the evaluation passes, proceed to the next task unit; and so on, until the i-th task unit; 3.3 In the i-th task unit, repeat the above evaluation-learning-backoff mechanism until the user correctly completes the i-th task unit, and end the forward training; Step 4: Logically reverse the training sequence to generate an inverted training sequence, with the order being i, i-1, ..., 1; Step 5: Perform iterative reverse training according to the following iterative method: 5.1 Starting from the first task unit of the reverse sequence (i.e. the original i-th unit), show the user its second description information Ai2, and attempt to receive the content Vi2 associated with Ai1, which is deduced or parsed by the user based on Ai2; 5.1.1 If the user is unable to generate the information, the system will push and display the corresponding first description information Ai1 for learning. After learning is complete, return to step 5.

1. 5.1.2 If the user generates content Vi2, then the AI ​​model is invoked to evaluate Vi2 and Ai1; 5.1.3 If the evaluation fails, the system displays Ai1 for learning and returns to step 5.1; 5.1.4 If the assessment is successful, proceed to the next step, 5.

2. 5.2 Proceed to the next task unit in the reverse sequence (i.e., the original i-1th unit), display its second description information A(i-1)2, and receive the user's reverse-engineered content V(i-1)2; 5.2.1 If the user is unable to generate or fails the evaluation, the system displays A(i-1)1 for learning and forces a return to step 5.1 (i.e., restarting from the first task unit in reverse order). 5.2.2 If the evaluation passes, proceed to the next task unit; and so on, until the last task unit of the reverse sequence (i.e. the original first unit). 5.3 Repeat the above mechanism in the last task unit of the reverse order until the user completes it correctly, and end the reverse order training.

2. The system and its method of use according to claim 1, characterized in that, The first descriptive information Ai1 and the second descriptive information Ai2 include, but are not limited to, the following pairing forms: writing topic and standard model, keyword combination and paragraph example, logical starting point and deduction path, oral scenario and appropriate expression, emotional instruction and corresponding text.

3. The system and its method of use according to claim 1, characterized in that, The pre-trained AI language models include, but are not limited to, BERT models, GPT models, or variants thereof based on the Transformer architecture; when evaluating semantic similarity, cosine similarity is used to calculate the embedding vector distance between user-generated content and standard examples; when evaluating logical coherence, the perplexity of the model regarding the connection relationships between sentences is used; when evaluating subjective sentiment conformity, a sentiment classifier is used to match the sentiment of user text with that of the instruction.

Images