Child language ability testing method based on large language model superposition adapter

Abstract

The present disclosure provides a large language model superposition adapter-based children's language ability testing method, device, medium, system and terminal. The large language model superposition adapter-based children's language ability testing method comprises: obtaining a pre-trained large language base model and at least one children's corpus; preprocessing the at least one children's corpus to determine a children's language training set; inserting a large language model adapter into the large language base model to determine a to-be-trained children's language dialogue model; inputting the children's language training set into the to-be-trained children's language model for model processing to determine the children's language dialogue model; inputting the voice of a to-be-tested child into the trained children's language dialogue model to determine the voice dialogue predicted by the children's language dialogue model; and determining the language ability detection result of the to-be-tested child according to the voice of the to-be-tested child and the predicted voice dialogue. The present disclosure realizes children's language training and children's language ability evaluation.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and more specifically, to a method, apparatus, medium, system, and terminal for testing children's language abilities based on a large language model overlay adapter. Background Technology

[0002] Chatbots can help children expand their vocabulary and improve their language communication skills during language training. At the same time, they can provide contextualized and personalized learning to help children understand and use language, thereby increasing their interest in learning language.

[0003] Traditional chatbots rely on rules, templates, or statistical models. However, rule-based chatbots are suitable for specific domains or tasks, while rule- and template-based chatbots require manual rule or template writing. For new domains or tasks, rules or templates need to be redesigned and rewritten, resulting in poor scalability. Furthermore, their dependence on pre-written rules or templates can lead to unstable dialogue quality. Traditional chatbots also have weak language understanding capabilities, making it difficult to accurately understand user intent and questions. Statistical model-based chatbots require large amounts of training data to improve model accuracy and generalization ability, but their generalization ability is poor. Finally, traditional chatbots lack human-like characteristics, resulting in stiff and unnatural responses that negatively impact user experience. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method, device, medium, system, and terminal for testing children's language abilities based on a large language model overlay adapter.

[0005] According to a first aspect of this disclosure, a method for testing children's language ability based on a large language model superimposed adapter is provided, comprising:

[0006] Obtain a pre-trained large language foundation model and at least one children's corpus;

[0007] The at least one children's corpus is preprocessed to determine the children's language training set;

[0008] Insert a large language model adapter to be trained into the pre-trained large language base model to determine the child language dialogue model to be trained. The large language model adapter is used to improve the vertical depth of the large language base model in the child language domain.

[0009] The children's language training set is input into the children's language model to be trained for model processing to determine the trained children's language dialogue model;

[0010] The child's voice is input into the trained child language dialogue model to determine the voice dialogue predicted by the trained child language dialogue model.

[0011] The language ability test result of the child under test is determined based on the child's speech and the speech dialogue predicted by the trained child language dialogue model.

[0012] Optionally, an adapter for the large language model to be trained is inserted into the pre-trained large language base model to determine the child language dialogue model to be trained, including:

[0013] Determine a copy of the pre-trained large language foundation model;

[0014] According to the preset scenario, the large language model adapter to be trained with preset parameters is inserted on top of the copy of the pre-trained large language base model through a stacking process to determine the child language dialogue model to be trained.

[0015] Optionally, the step of inputting the children's language training set into the children's language model to be trained for model processing, and determining the trained children's language dialogue model, includes:

[0016] Freeze the parameters of the large language foundation model;

[0017] Based on the adaptation prompts in the children's language training set, the parameters of the large language model adapter are trained to determine the parameters of the large language model adapter corresponding to the preset scenario, and the trained children's language dialogue model is determined.

[0018] Optionally, the adaptation prompts include text instructions and image instructions.

[0019] According to a second aspect of this disclosure, a children's language ability testing device based on a large language model superimposed adapter is provided, comprising:

[0020] The acquisition module is used to acquire a pre-trained large language base model and at least one children's corpus.

[0021] The first determining module is used to preprocess the at least one children's corpus to determine the children's language training set;

[0022] The second determining module is used to insert a large language model adapter to be trained into the pre-trained large language base model to determine the children's language dialogue model to be trained. The large language model adapter is used to improve the vertical depth of the large language base model in the children's language domain.

[0023] The third determining module is used to input the children's language training set into the children's language model to be trained for model processing, and to determine the trained children's language dialogue model.

[0024] The fourth determination module inputs the child's voice into the child language dialogue model to determine the voice dialogue predicted by the trained child language dialogue model.

[0025] The fifth determining module is used to determine the language ability test result of the child under test based on the child's speech and the speech dialogue predicted by the trained child language dialogue model.

[0026] Optionally, the second determining module includes:

[0027] The first determining submodule is used to determine a copy of the pre-trained large language base model;

[0028] The second determining submodule is used to determine the child language dialogue model to be trained by inserting the large language model adapter with preset parameters to be trained on top of the copy of the pre-trained large language base model through a stacking process, according to a preset scenario.

[0029] According to a third aspect of this disclosure, a children's language ability testing device based on a large language model superimposed adapter is provided, comprising:

[0030] processor;

[0031] Memory used to store processor-executable instructions;

[0032] The processor is configured to perform the steps of the method for testing children's language ability based on a large language model overlay adapter provided in the first aspect of this disclosure.

[0033] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the steps of the children's language ability testing method based on a large language model overlay adapter provided in the first aspect of this disclosure.

[0034] According to the fifth aspect of this disclosure, a children's language ability testing system based on a large language model superimposed adapter is provided, comprising:

[0035] A speech recognition module is used to acquire the speech of the child to be tested and convert the speech of the child to be tested into text information.

[0036] A child language dialogue model, wherein the child language dialogue model is connected to the speech recognition module, the text information is input into the child language dialogue model, and the child language dialogue model outputs text response information;

[0037] A voice broadcast module is connected to the child language dialogue model. The text response information is input into the voice broadcast module, and the voice broadcast module plays the text response information by voice.

[0038] According to a sixth aspect of this disclosure, a terminal is provided, comprising the children's language ability testing system based on a large language model overlay adapter provided in the fifth aspect of this disclosure.

[0039] Compared with the prior art, the embodiments of the present invention have at least one of the following beneficial effects:

[0040] By preprocessing at least one children's corpus to determine a children's language training set, an adapter for the large language model to be trained is inserted into the pre-trained large language foundation model to determine a children's language dialogue model. The large language foundation model is adjusted to follow instructions using the large language model adapter, and the verticality of the large language foundation model in the children's language domain is improved. The children's language training set is input into the children's language model to be trained for model processing to determine the children's language dialogue model. By inserting the large language model adapter, the amount of parameter training is reduced and the efficiency of model processing is improved. The speech of the child to be tested is input into the children's language dialogue model to determine the speech dialogue predicted by the children's language dialogue model. The children's language dialogue model can predict speech dialogue based on the speech of the child to be tested, improving the flexibility between the children's language dialogue model and the speech dialogue of the child to be tested. Based on the speech of the child to be tested and the speech dialogue predicted by the children's language dialogue model, the language ability detection result of the child to be tested is determined, realizing the diagnostic detection of children's language ability and promoting children's language development. Attached Figure Description

[0041] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0042] Figure 1 This is a flowchart illustrating a method for testing children's language ability based on a large language model overlay adapter, according to an exemplary embodiment.

[0043] Figure 2 This is a flowchart illustrating a method for determining a child's language model to be trained, according to an exemplary embodiment.

[0044] Figure 3This is a flowchart illustrating a method for determining a trained child language dialogue model according to an exemplary embodiment.

[0045] Figure 4 This is a block diagram illustrating a children's language ability testing device based on a large language model overlay adapter, according to an exemplary embodiment.

[0046] Figure 5 This is a block diagram illustrating a children's language ability testing device based on a large language model overlay adapter, according to another exemplary embodiment.

[0047] Figure 6 This is a block diagram illustrating a children's language ability testing system based on a large language model overlay adapter, according to an exemplary embodiment. Detailed Implementation

[0048] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0049] Figure 1 This is a flowchart illustrating a method for testing children's language abilities based on a large language model overlay adapter, according to an exemplary embodiment. Figure 1 As shown, a method for testing children's language ability based on a large language model superimposed adapter includes S11 to S16.

[0050] S11, obtain a pre-trained large language foundation model and at least one children's corpus.

[0051] In one possible implementation, a large foundation model can be obtained from the open-source community as a pre-trained large language foundation model. The large foundation model can handle large-scale natural language data and has good language understanding capabilities.

[0052] As an example, one or more of the ChatGPT large language model, BLOOM large language model, etc., can be used as the pre-trained large language base model.

[0053] This disclosure uses a large language model based on the GPT-3.5 architecture as a pre-trained large language foundation model.

[0054] In another possible embodiment, the pre-trained large language foundation model can also be a model trained using a pre-set Chinese corpus.

[0055] In one possible embodiment, at least one children's corpus may be a Chinese children's corpus and / or a Chinese children's corpus.

[0056] Those skilled in the art should understand that a children's corpus can be formed by combining multiple data sources and is not limited to Chinese children's corpora or English children's corpora. Other types of children's corpora can also be used, all of which fall within the protection scope of this disclosure.

[0057] S12, preprocess the at least one children's corpus to determine the children's language training set.

[0058] Preprocessing can employ automated keyword filtering methods or manual review by professional educators.

[0059] The preprocessed children's corpus forms a children's language training set, which serves as a guided training dataset for optimizing the pre-trained large language foundation model and training the adapter model of the large language model to be trained.

[0060] S13, Insert the large language model adapter to be trained into the pre-trained large language base model to determine the children's language dialogue model to be trained. The large language model adapter is used to improve the vertical depth of the large language base model in the children's language domain.

[0061] Among them, according to the preset scenario, a large language model adapter corresponding to the preset scenario can be inserted into the large language base model. The large language model adapter trained by the model can improve the vertical depth of the large language base model in the professional knowledge domain of the preset scenario.

[0062] In this disclosure, a large language model adapter is cascaded with the top neural network layer of a large language base model to form a child language dialogue model to be trained. The large language model adapter can adjust the large language base model adapter into a model that follows instructions.

[0063] For example, the large language model adapter to be trained can be an adapter with 1.2 million parameters.

[0064] S14, input the children's language training set into the children's language model to be trained for model processing, and determine the trained children's language dialogue model.

[0065] In this process, the large language model adapter to be trained is trained using a children's language training set. The large language model adapter learns professional knowledge in the children's language domain. The pre-trained large language basic model is then optimized using the children's language training set to determine the trained children's language dialogue model. For example, the optimization process can use the backpropagation algorithm to improve the model performance of the large language basic model.

[0066] In one possible embodiment, the large language model adapter trained by the model can serve as a scenario-preceding model in the children's language domain.

[0067] S15, input the voice of the child to be tested into the trained child language dialogue model, and determine the voice dialogue predicted by the trained child language dialogue model.

[0068] In this disclosure, the trained children's language dialogue model can receive the speech of the child to be tested and conduct interactive voice dialogue with the child to train the child's language ability.

[0069] Among them, the children's language dialogue model can predict and generate dialogue speech based on the speech of the child being tested.

[0070] S16. Based on the child's speech and the speech dialogue predicted by the trained child language dialogue model, determine the child's language ability test result.

[0071] Following the example above, the children's language dialogue model can also record the content of interactive voice dialogues and evaluate the language ability of the children being tested based on the content of the voice dialogues.

[0072] The above technical solution involves preprocessing at least one children's corpus to determine a children's language training set. A large language model adapter to be trained is inserted into the pre-trained large language foundation model to determine a children's language dialogue model. The large language model adapter is used to adjust the large language foundation model to a model that follows instructions, thereby improving the verticality of the large language foundation model in the children's language domain. The children's language training set is input into the children's language model to be trained for model processing to determine the children's language dialogue model. The inserted large language model adapter reduces the amount of parameter training and improves the efficiency of model processing. The speech of the child to be tested is input into the children's language dialogue model to determine the predicted speech dialogue. The children's language dialogue model can predict speech dialogue based on the speech of the child to be tested, improving the flexibility between the children's language dialogue model and the speech dialogue of the child to be tested. Based on the speech of the child to be tested and the speech dialogue predicted by the children's language dialogue model, the language ability detection result of the child to be tested is determined, realizing the diagnostic detection of children's language ability and promoting children's language development.

[0073] Figure 2 This is a flowchart illustrating a method for determining a child's language model to be trained, according to an exemplary embodiment.

[0074] Figure 2 This is a flowchart illustrating a method for determining a child's language model to be trained, according to an exemplary embodiment. Figure 2As shown, in some possible embodiments, a large language model adapter to be trained is inserted into the pre-trained large language base model to determine the child language dialogue model to be trained, including S21 to S22.

[0075] S21, determine a copy of the pre-trained large language foundation model.

[0076] A copy of the pre-trained large language base model is made, and a large language model adapter is inserted into the copy of the large language base model. In various preset scenarios, a scenario-specific model is formed by inserting the large language model adapter into the copy of the large language base model. When switching scenarios, the large language model adapter corresponding to the scenario is switched, and the copy of the large language base model is reset without affecting the large language base model.

[0077] S22, according to the preset scenario, the large language model adapter to be trained with preset parameters is inserted on top of the copy of the pre-trained large language base model through stacking processing to determine the child language dialogue model to be trained.

[0078] The inserted large language model adapter can be an adapter with 1.2 million parameters.

[0079] In this disclosure, an adapter is inserted in a stacking manner to combine multiple adapters into a large language model adapter. The large language model adapter can be adjusted according to factors such as the age, gender, and language environment of the child being tested in order to provide accurate language training and language assessment.

[0080] In one possible embodiment, a large language model adapter to be trained is inserted into a pre-trained large language base model. The large language model adapter to be trained and a copy of the pre-trained large language base model are stacked to form a dedicated model for children's language domain scenarios, namely, a children's language dialogue model to be trained.

[0081] Figure 3 This is a flowchart illustrating a method for determining a trained child language dialogue model according to an exemplary embodiment. Figure 3 As shown, in some possible embodiments, the step of inputting the children's language training set into the children's language model to be trained for model processing and determining the trained children's language dialogue model includes S31 to S32.

[0082] S31, freeze the parameters of the large language foundation model.

[0083] Following the example above, when optimizing the large language base model in the child language model to be trained, the 1.2 million parameters of the Transformer layer at the top of the large language base model are optimized, and other parameters of the large language base model are frozen to reduce the training workload.

[0084] S32, based on the adaptation prompts in the children's language training set, train the parameters of the large language model adapter, determine the parameters of the large language model adapter corresponding to the preset scenario, and determine the trained children's language dialogue model.

[0085] For example, after freezing the parameters of the large language foundation model, the large language model adapter is trained to transform the learnable domain knowledge in the children's language training set into adaptive prompts, and the adaptive prompts are placed before the input text tags of higher transformation layers. A zero initial attention mechanism with zero gating is introduced to adaptively inject teaching cues into the large language foundation model while retaining the pre-trained knowledge of the large language foundation model.

[0086] In another possible embodiment, the adaptation prompts include text instructions and image instructions.

[0087] Image tags can also be added to adaptive cues, thereby extending the input of children's language dialogue models to image input for multimodal reasoning.

[0088] By freezing the parameters of the large language model and training the large language model adapter, the large language model with 7 billion or 13 billion parameters can be fine-tuned, improving training efficiency. By using a high-quality guided children's language training set, the model performance and generalization ability of the children's language dialogue model can be effectively improved, which can meet the needs of children's language development assessment. This disclosure can also be extended to multimodal input, giving the children's language dialogue model image reasoning ability.

[0089] In some possible implementations, image tags are added to adaptive prompts to extend the child language dialogue model to image input for multimodal reasoning.

[0090] Specifically, images are input into the children's language dialogue model to enable the children to recognize images and generate text; an image generation module is added to the children's language dialogue model to input the children's voices to the model and output images.

[0091] In some possible implementations, the children's language dialogue model can be used as a language testing tool to assess the language development level of the child being tested.

[0092] Among them, the children's language dialogue model tests the children's vocabulary, grammar application ability, and sentence comprehension ability by asking them pre-set questions.

[0093] In some possible embodiments, language exercises of different difficulty levels can be determined based on the child's learning progress, learning interest, and language level, providing targeted programs to expand the child's vocabulary, improve the child's language expression ability, and enhance the child's language comprehension ability.

[0094] The above technical solution trains the language skills of children under test according to the scenario, stimulates their interest in language, assesses their language skills by recording the dialogue content, and provides targeted solutions.

[0095] In one possible embodiment, the experimental platform used in the embodiments of this disclosure is Ubuntu 20.04, Python 3.8, using four NVIDIA GEFORCE 3090 GPUs to train and tune the deep learning network, and the PyTorch 1.12.1 deep learning framework, and using miniconda3 to create an independent virtual environment.

[0096] Based on the same concept, this disclosure also provides a children's language ability testing device based on a large language model superimposed adapter. Figure 4 This is a block diagram illustrating a children's language ability testing device based on a large language model overlay adapter, according to an exemplary embodiment. (Refer to...) Figure 4 The children's language ability testing device 100 based on the large language model superimposed adapter includes an acquisition module 110, a first determination module 120, a second determination module 130, a third determination module 140, a fourth determination module 150, and a fifth determination module 160.

[0097] Module 110 is used to acquire a pre-trained large language base model and at least one children's corpus.

[0098] The first determining module 120 is used to preprocess the at least one children's corpus to determine a children's language training set;

[0099] The second determining module 130 is used to insert a large language model adapter to be trained into the pre-trained large language base model to determine the child language dialogue model to be trained. The large language model adapter is used to improve the vertical depth of the large language base model in the child language domain.

[0100] The third determining module 140 is used to input the children's language training set into the children's language model to be trained for model processing, and to determine the trained children's language dialogue model.

[0101] The fourth determining module 150 inputs the voice of the child to be tested into the child language dialogue model and determines the voice dialogue predicted by the trained child language dialogue model.

[0102] The fifth determining module 160 is used to determine the language ability test result of the child under test based on the child's speech and the speech dialogue predicted by the trained child language dialogue model.

[0103] The above technical solution involves preprocessing at least one children's corpus to determine a children's language training set. A large language model adapter to be trained is inserted into the pre-trained large language foundation model to determine a children's language dialogue model. The large language model adapter is used to adjust the large language foundation model to a model that follows instructions, thereby improving the verticality of the large language foundation model in the children's language domain. The children's language training set is input into the children's language model to be trained for model processing to determine the children's language dialogue model. The inserted large language model adapter reduces the amount of parameter training and improves the efficiency of model processing. The speech of the child to be tested is input into the children's language dialogue model to determine the predicted speech dialogue. The children's language dialogue model can predict speech dialogue based on the speech of the child to be tested, improving the flexibility between the children's language dialogue model and the speech dialogue of the child to be tested. Based on the speech of the child to be tested and the speech dialogue predicted by the children's language dialogue model, the language ability detection result of the child to be tested is determined, realizing the diagnostic detection of children's language ability and promoting children's language development.

[0104] Optionally, the second determining module 130 includes:

[0105] The first determining submodule is used to determine a copy of the pre-trained large language base model;

[0106] The second determining submodule is used to determine the child language dialogue model to be trained by inserting the large language model adapter with preset parameters to be trained on top of the copy of the pre-trained large language base model through a stacking process, according to a preset scenario.

[0107] Optionally, the third determining module 140 includes:

[0108] The freeze submodule is used to freeze the parameters of the large language base model;

[0109] The third determining submodule is used to train the parameters of the large language model adapter based on the adaptation prompts in the children's language training set, determine the parameters of the large language model adapter corresponding to the preset scenario, and determine the trained children's language dialogue model.

[0110] Optionally, the adaptation prompts include text instructions and image instructions.

[0111] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0112] Figure 5 This is a block diagram illustrating a children's language ability testing device based on a large language model overlay adapter, according to another exemplary embodiment. Figure 5 As shown, the children's language ability testing device 500 based on a large language model overlay adapter may include: a processor 501 and a memory 502. The children's language ability testing device 500 based on a large language model overlay adapter may also include one or more of the following: a multimedia component 503, an input / output interface 504, and a communication component 505.

[0113] The processor 501 controls the overall operation of the child language ability testing device 500 based on a large language model overlay adapter to complete all or part of the steps in the child language ability testing method based on a large language model overlay adapter described in the first aspect. The memory 502 stores various types of data to support the operation of the child language ability testing device 500 based on a large language model overlay adapter. This data may include, for example, instructions for any application or method operating on the child language ability testing device 500 based on a large language model overlay adapter, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 502 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 503 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 502 or transmitted via the communication component 505. The audio component also includes at least one speaker for outputting audio signals. Input / output interface 504 provides an interface between processor 501 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 505 is used for wired or wireless communication between the child language ability testing device 500 based on the large language model overlay adapter and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IoT, eMTC, or other 5G technologies, or combinations thereof, and is not limited herein. Therefore, the corresponding communication component 305 may include: a Wi-Fi module, a Bluetooth module, an NFC module, etc.

[0114] In another exemplary embodiment, a non-transitory computer-readable storage medium including program instructions is also provided. When executed by a processor, these program instructions implement the steps of the child language ability testing method based on a large language model overlay adapter described above. For example, the computer-readable storage medium may be the aforementioned memory including program instructions, which can be executed by a processor of the child language ability testing device based on a large language model overlay adapter to complete the aforementioned child language ability testing method based on a large language model overlay adapter.

[0115] In another exemplary embodiment, a computer program product is also provided, the computer program product comprising a computer program executable by a programmable device, the computer program having a code portion for performing the above-described method for testing children's language ability based on a large language model overlay adapter when executed by the programmable device.

[0116] Figure 6 This is a block diagram illustrating a children's language ability testing system based on a large language model overlay adapter, according to an exemplary embodiment. Figure 6 As shown, in another exemplary embodiment, a children's language ability testing system based on a large language model overlay adapter is also provided, including:

[0117] A speech recognition module is used to acquire the speech of the child to be tested and convert the speech of the child to be tested into text information.

[0118] A child language dialogue model, wherein the child language dialogue model is connected to the speech recognition module, the text information is input into the child language dialogue model, and the child language dialogue model outputs text response information;

[0119] A voice broadcast module is connected to the child language dialogue model. The text response information is input into the voice broadcast module, and the voice broadcast module plays the text response information by voice.

[0120] In another exemplary embodiment, a terminal is also provided, including a children's language ability testing system based on a large language model overlay adapter.

[0121] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention. The above preferred features can be used in any combination without conflict.

Claims

1. A method for testing children's language abilities based on a large language model superposition adapter, characterized in that, The method comprises the following steps: obtaining a pre-trained large language base model and at least one child corpus; preprocessing the at least one child corpus to determine a child language training set; inserting a large language model adapter to be trained into the pre-trained large language base model to determine a child language dialogue model to be trained, wherein the large language model adapter is used to improve the vertical depth of the large language base model in the field of child language; inputting the child language training set into the child language model to be trained for model processing to determine a trained child language dialogue model; inputting the voice of a child to be tested into the trained child language dialogue model to determine the voice dialogue predicted by the trained child language dialogue model; determining the language ability detection result of the child to be tested according to the voice of the child to be tested and the voice dialogue predicted by the trained child language dialogue model. The method comprises the following steps: determining a copy of the pre-trained large language base model; inserting a large language model adapter to be trained with preset parameters into the top of the copy of the pre-trained large language base model by stacking processing according to a preset scene to determine a child language dialogue model to be trained.

2. The method of claim 1, wherein, The method comprises the following steps: freezing the parameters of the large language base model; training the parameters of the large language model adapter according to the adaptation prompt in the child language training set to determine the parameters of the large language model adapter corresponding to the preset scene and determine a trained child language dialogue model.

3. The method of claim 2, wherein, The adaptation prompt comprises a text instruction and an image instruction.

4. An apparatus of a child language evaluation system based on a large language model, characterized by, The method comprises the following steps: an obtaining module is configured to obtain a pre-trained large language base model and at least one child corpus; a first determining module is configured to preprocess the at least one child corpus to determine a child language training set; a second determining module is configured to insert a large language model adapter to be trained into the pre-trained large language base model to determine a child language dialogue model to be trained, wherein the large language model adapter is used to improve the vertical depth of the large language base model in the field of child language; a third determining module is configured to input the child language training set into the child language model to be trained for model processing to determine a trained child language dialogue model; a fourth determining module is configured to input the voice of a child to be tested into the child language dialogue model to determine the voice dialogue predicted by the trained child language dialogue model; a fifth determining module is configured to determine the language ability detection result of the child to be tested according to the voice of the child to be tested and the voice dialogue predicted by the trained child language dialogue model. The second determining module comprises the following steps: a first determining submodule is configured to determine a copy of the pre-trained large language base model; A second determining sub-module is configured to insert a to-be-trained large language model adapter with preset parameters into the top of a copy of the pre-trained large language base model through a stacking processing manner according to a preset scene, and determine a to-be-trained child language dialogue model. 5.A device for testing children's language ability based on a large language model superposition adapter, characterized in that, Comprise: A processor; A memory for storing processor-executable instructions; The processor is configured to perform the steps of the child language ability testing method based on the large language model superimposed adapter in any one of claims 1-3.

6. A non-transitory computer-readable storage medium having stored thereon computer program instructions, wherein, The program instructions are executed by the processor to implement the steps of the child language ability testing method based on the large language model superimposed adapter in any one of claims 1-3.

7. A large language model superposition adapter-based children's language ability test system for implementing the large language model superposition adapter-based children's language ability test method of claim 1, characterized in that, Comprise: A speech recognition module is configured to obtain the speech of the child to be tested, and convert the speech of the child to be tested into text information through the speech recognition module; A child language dialogue model is connected with the speech recognition module, and the text information is input into the child language dialogue model, and the child language dialogue model outputs text answer information; A voice broadcast module is connected with the child language dialogue model, and the text answer information is input into the voice broadcast module, and the voice broadcast module plays the text answer information through voice.

8. A terminal, characterized by comprising: The child language ability testing system based on the large language model superimposed adapter of claim 7 is comprised.

Images