353 results about "Multiple language" patented technology

The invention discloses a cross-border e-commerce compliance intelligent auditing platform and a multi-language contract analysis method, and relates to the field of cross-border contract compliance auditing. In the multi-modal data access step, customs codes, laws and regulations and other multi-source data are collected, and 18 kinds of language contract texts are analyzed; in the cross-language semantic alignment step, a knowledge graph is constructed, and multi-language legal concept mapping is achieved; in the compliance risk reasoning step, a rule engine and an agent cooperatively check a contract, and the compliance conclusion confidence is calculated; the dynamic risk assessment step adopts an LSTM network to analyze historical data and predict a risk trend; in the multi-language report generation step, a multi-format bilingual or multilingual report is generated based on a template engine, encrypted and archived. According to the invention, cross-border contracts are audited efficiently and intelligently, dynamic adaptation laws and regulations are analyzed in multiple languages, compliance risks are identified accurately, and a multi-language report is generated quickly; therefore, the checking efficiency is improved, the manual workload is reduced, the compliance risk is reduced, and the enterprise cross-border business competitiveness and the risk response capability are enhanced.

Aspects of the present disclosure relate to detecting hallucinations in language model outputs. Embodiments include receiving a user query. Embodiments further include prompting a language processing machine learning model to generate responses to the user query in each language of a set of multiple languages. Embodiments further include receiving the responses from the language processing machine learning model in response to the prompting. Embodiments further include creating embedding representations of the responses. Embodiments further include calculating, based on the embedding representations, a degree of semantic similarity between the responses. Embodiments further include determining that a response of the responses contains a model hallucination based on comparing the degree of semantic similarity between the responses to a threshold.

The invention belongs to the technical field of document content extraction, and particularly discloses a document content extraction method and system based on multimodal model collaboration, a terminal and a medium. Comprising the following steps: identifying the type of an input to-be-processed document, and judging the document type; on the basis of the type identification result, calling a multi-modal model to analyze the document content, and outputting space coordinates, visual features and semantic features of document elements; generating a content sequence according with a reading habit through a semantic sequence reconstruction algorithm; paragraph boundary detection, paragraph recombination and semantic association modeling of charts and texts are completed based on the multilayer attention network and the graph neural network; grammar error correction, format optimization and title hierarchy generation are carried out by using a large language model and a hierarchical classification network; and converting the identification result into a structured output file. According to the method, the processing requirements of different types of documents can be considered, and high-precision analysis and efficient output are realized under the scenes of complex layouts, multiple languages and formula tables.

The invention relates to a method for improving the internal accuracy of a response from an artificial intelligence (AI) trained in multiple languages, in particular a generative AI trained in multiple languages. The invention further relates to a data processing system for carrying out the method.

A document information extraction service can utilize an LLM to provide support for extracting information from documents in multiple languages. A trained first machine learning model can map data extracted from a first master document of a first document type in a first language. The mappings can be corrected via user input to obtain ground truth data for the master document. The ground truth data can be translated into a second language and optionally corrected to obtain translated ground truth data. An LLM can generate a training dataset of fake documents of the first document type that contain text in the second language based at least in part on the translated ground truth data. The trained first machine learning model can be trained further with the training dataset and deployed to extract data from documents of the first document type that contain text in the second language.

The invention relates to the technical field of artificial intelligence, provides a multi-language task execution method, device and equipment and a medium, is applied to financial and medical health care service scenes, and can acquire text data and visual action data of multiple languages according to an execution instruction and perform preprocessing to realize standardized processing of multi-modal data; constructing a cross-language word vector semantic space based on adversarial training according to the multi-language features to realize preliminary word vector alignment; performing multi-language semantic alignment on the multi-language features according to a cross-language word vector semantic space, further breaking language barriers, and realizing depth mapping and alignment among different language semantics; a multi-language culture knowledge graph is constructed, so that culture knowledge is introduced, and the culture perception ability is improved; fusion is carried out through a gating fusion mechanism, fusion features with language attributes and cultural attributes can be obtained, and therefore multi-language tasks can be executed more accurately by integrating multi-language information and cultural knowledge.

A system and method for classification. A method includes identifying candidate entities among text data by applying at least one entity identification rule to the text data. Inputs are constructed based on the identified candidate entities, where each input includes a first portion of text indicating a candidate entity and at least one second portion of text and where the at least one second portion of text of each input is adjacent to the first portion of text of the input. Multiple language models are applied to the inputs, where each language model is trained to identify a respective set of entities and where outputs of the language models include at least one portion of entity-indicating text for each input. Based on the outputs of the language models, at least one named entity in the text data is determined.

The invention particularly relates to a code writing, checking and modifying method based on a large model. According to the code writing, checking and modifying method based on the large model, a fine-tuning large language model is adopted to analyze user requirements, and multi-language initial codes are generated; executing deep code review through a static analysis engine, wherein the deep code review comprises syntax tree verification, security vulnerability mode matching and API dependency chain detection; executing the automatically generated test case set in an isolation environment, monitoring operation indexes in real time, and generating a defect thermodynamic diagram; and driving the large model to perform iterative code reconstruction based on a defect analysis result until it is confirmed that a preset quality standard is met. The code writing, checking and modifying method based on the large model is suitable for multiple languages, the defect rate of codes is reduced, the software development period is prolonged, the labor cost is reduced, and the vulnerability detection rate is greatly increased.

The present invention facilitates communication between sign language users and machines by translating sign language and text using AI models, deep learning computer vision, and word embeddings. Users interact via sign language, captured and processed through deep learning and NLP modules. The system converts sign language videos into text, constructs coherent sentences, and generates contextually appropriate responses using a Retrieve and Generate (RAG) model. Responses are translated back into sign language videos, spelling out words not found in the dictionary. If requested, a human agent can respond. Key features include high-accuracy recognition, context-aware response generation, dynamic vocabulary updates, and optional human interaction. The method ensures efficient processing with LLM, embedding techniques, and deep learning, optimizing translation accuracy and user experience. The system adapts to multiple languages and dialects by training on specific sign languages, making it applicable globally.

The invention discloses an uncertainty guided few-sample harmful speech detection method (U-GIFT). According to the method, a pre-training language model is finely adjusted based on a small number of labeled samples, and a semi-supervised self-training and uncertainty guiding strategy is combined. Monte Carlo Dropout is started in the reasoning stage, multiple times of random forward propagation are carried out to obtain sample posterior distribution, prediction entropy and information gain are calculated, pseudo-label samples are sorted and screened, and only high-confidence samples are selected to be added into a training set. And in order to reduce the influence of a pseudo labeling error, designing a stability weighting mechanism, giving a sample weight according to a prediction variance, and constructing a joint loss function, so that the model preferentially learns a stable sample to improve the detection performance. According to the method, the semantic and attention mechanism of the pre-training model is utilized, the detection effect is remarkably improved under the conditions of few samples, imbalance, multiple languages and cross domains, models such as BERT, RoBERTa, XLM-R, LLaMA2 and DeepSeek-R1 are compatible, and the method is suitable for content auditing and risk prevention and control.

The invention relates to a memory retrieval method, a memory generation method and related devices. The method comprises the following steps: acquiring an input search word, and translating the input search word into a search word of a target language; performing memory retrieval by using the input retrieval word and the retrieval word of the target language according to the memory index to obtain a retrieval result; wherein the memory index is established according to at least one of an original and a copy of the memory information; the copy is obtained by translating the original into a target language. The method can match and support multiple languages.

The invention provides an eBPF program automatic generation method, system and device based on retrieval enhancement and thinking chain reasoning, and the method comprises the steps: constructing an eBPF semantic knowledge base, and extracting and structuring a plurality of program examples with semantic annotations; after a natural language task description input by a user is received, the natural language task description is coded into a semantic vector, and accurate retrieval of related contexts is achieved in combination with semantic indexes; based on a retrieval result, constructing a structured prompt, and guiding a language model to gradually generate an eBPF program according to stages; an eBPF verifier is used for carrying out legality check on the generated program, and behavior testing is carried out in combination with task input; when structural errors or semantic deviations are found, the model is guided to be automatically repaired based on error information. The method can be operated on a multi-language model platform, supports the output of two styles of BCC and BPFtrace, and has the advantages of controllable structure, accurate semantics, stable deployment and the like. Experimental results show that the automatic generation efficiency and accuracy of the eBPF program can be remarkably improved.

The invention provides a low-resource multi-language large model training method and system for personalized course learning, and belongs to the technical field of large language models, and the method comprises the steps: S1, collecting training samples of multiple languages; performing de-duplication and de-noising processing on each sample, then unifying data formats, and adding language attributes as language labels; s2, initializing parameters of an adaptive sampling scheduler and a dynamic loss scheduler; and S3, taking the pre-trained large language model as a base model, and adding an adaptive sampling scheduler and a dynamic loss scheduler in the training process. According to the method, the dependence on low-resource language annotation data is reduced, and the training weights of different language samples can be adaptively balanced; and dynamically matching the multi-language task difficulty with the model learning progress.

The present disclosure generally relates to evaluating and enhancing LLM responses. In some implementations, a system includes multiple language models with different specialized roles that work together to improve response reliability and transparency. A responder model can generate initial responses to user queries, providing diverse perspectives on the same input. An evaluator model can assess and combines responses from the responder models into an accurate and reliable output. A reporter model can generate summaries and alerts about response quality and confidence levels, providing transparency to users about the decision-making process. An artificial intelligence (AI) engine can manage the flow of information between the different models, orchestrating their interactions and ensuring proper sequencing of operations. A retrieval system can provide additional context from external knowledge sources, allowing the system to generate accurate and well-informed responses.

The invention discloses an image-text report generation method fusing a multi-mode large language model and an RAG mechanism, and belongs to the technical field of text processing. The method comprises the following steps: firstly, converting a PDF document into an image, identifying and extracting contents such as texts, tables and charts through a multi-modal model, and constructing a searchable knowledge fragment library; then, based on user query, adopting a hybrid retrieval strategy to obtain related evidence, and utilizing a large language model to generate a Markdown report containing an image placeholder; and meanwhile, a text graph module is called to generate an illustrated graph, and finally visual report output of image-text fusion is realized. The method supports multi-modal content understanding, cross-modal retrieval and collaborative generation, has good generalization, accuracy and practicability, and is suitable for multi-field and multi-language complex document processing and report generation.

The embodiment of the invention provides a multi-language switching method and device.The method comprises the steps that under the condition that a multi-language item is compiled, all translated texts in the multi-language item are translated to obtain a multi-language resource file; under the condition of running the multi-language project, receiving a switching instruction carrying a target language, and obtaining a target language resource file corresponding to the target language from the multi-language resource file; and querying from the target language resource file to obtain a target translation text corresponding to each translated text, and replacing the translated text with the target translation text. According to the method, all the translated texts are automatically translated in the project compiling stage, the multi-language resource files can be generated after compiling is completed, language resource switching can be automatically completed during project running, existing service codes do not need to be manually modified, and therefore the labor cost of multi-language / internationalized project development is saved.

The invention provides a multi-agent collaborative cross-language system translation method and system based on loAs, and relates to the technical field of intelligent translation.The method comprises the steps that to-be-translated text information and conference theme information are obtained, and first fusion semantic information is obtained; obtaining first target semantic information through the first translation agent; obtaining second target semantic information through a second translation agent; obtaining second fusion semantic information through the first target semantic information and the conference theme information, and determining a second target language translation text; and acquiring third fused semantic information through the second target semantic information and the conference theme information, and determining the first target language translation text. According to the method and the device, the translation accuracy of key texts can be improved by referring to conference theme information, and multi-language semantic information can be mutually corrected through a cross attention mechanism to improve the semantic consistency of multi-language translation, so that the overall translation accuracy is improved, and the communication cost is reduced.

The invention discloses a retrieval method and system based on semantic keyword classification and multi-language intelligent icons, and belongs to the technical field of icon retrieval. The method comprises the following steps: constructing a CLIP model supporting multiple languages; collecting a multi-source icon data set, and taking the multi-source icon data set as a training set to train the CLIP model to obtain a trained CLIP model; processing the icons in the icon library and the associated text tags by using the trained CLIP model to generate a corresponding picture feature vector library and a text feature vector library; extracting a to-be-matched text feature vector of the input text by using the trained CLIP model, and matching the to-be-matched text feature vector with a text feature vector in a text feature vector library to obtain a preliminary matching result; and calculating the similarity between the picture feature vector in the icon set corresponding to the preliminary matching result and the to-be-matched text feature vector in the semantic space to obtain a final matching result. According to the method, the limitation of a single language is broken through, and the retrieval accuracy is improved.

Provided are an information processing device, a method for controlling the information processing device, and a program, whereby the convenience of a user can be improved by using a plurality of language models to improve the efficiency of work for obtaining a response having a comprehensive viewpoint.　An information processing device 1 comprises: a response result acquisition unit 33 that transmits, to a plurality of language models 3 (for example, language models 3-1 to 3-3), a request prompt including content of a user request , and acquires a response result for the request content from each of the plurality of language models 3; and an integration processing unit 35 that generates an integration instruction prompt for instructing the language models 3 to integrate the response results of each of the plurality of language models 3. The response result acquisition unit 33 transmits the integration instruction prompt to the language models 3 to thereby acquire an integrated response result in which the response results of the plurality of language models 3 are reflected.

A system is provided.SOLUTION: A system comprising: means for a user to upload an article; means for a server to parse the uploaded article; means for the server to automatically translate the parsed article into multiple languages; means for the server to extract and embed SEO keywords for the translated article; means for the server to generate and link relevant images and video to the article; and means for the server to distribute the processed article to platforms in each country.SELECTED DRAWING: Figure 1

To provide a recommendation system or the like for suitably proposing an LLM capable of appropriately answering a user's request.SOLUTION: In a recommendation system, a task execution unit receives a request including text data from a user and causes a first language model, which is one language model designated by the user from a plurality of language models, to generate a first answer to the request. The evaluation information acquisition unit acquires evaluation information on each of the plurality of language models from a predetermined database. The recommendation information generation unit causes a second language model having an execution environment different from that of the first language model to select a recommended language model for generating an answer to the request from the plurality of language models on the basis of the request, the first answer, and the evaluation information. Further, the recommendation information generation unit causes the second language model to generate recommendation information including a message for presenting the recommended language model to the user.SELECTED DRAWING: Figure 2

The invention discloses a knowledge graph query method and device based on multi-language model result fusion. The method comprises the following steps: multi-query result generation: obtaining a query problem; generating query statements under the plurality of model parameters by utilizing the plurality of large language models, and performing query in the knowledge graph by respectively using each query statement; multiple query results are synthesized, wherein reply integration is conducted on the execution conditions of the multiple query statements generated by the same large language model and subjected to parameter adjustment, and model-level query reply is obtained; and for a plurality of model-level query answers, eliminating the model-level query answers with relatively low semantic similarity, and merging the model-level query answers which are not eliminated to obtain a result fusion query answer. By the adoption of the technical scheme, the multiple query statements under the multiple model parameters of the multiple language models are beneficial to comprehensively obtaining information from the knowledge graph, then the query replies are integrated and merged, and noise in the query replies can be effectively reduced and avoided.

The present disclosure provides a multilingual keyboard comprising a first set of keys with Organic Light-Emitting Diode (OLED) displays and matte-finished key caps, produced through a blending process for clear visibility and aesthetic appeal. The multilingual keyboard includes at least two sets of keys: the first set with OLED displays for dynamic character updates based on the selected language, and a second set with fixed functions for a familiar typing experience. The multilingual keyboard incorporates a language switching system that enables seamless transitions between multiple languages with real-time updates on the OLED displays of the first set of keys. Designed to replace multiple keyboards and complex software mappings, the multilingual keyboard offers a versatile typing solution.

The invention discloses an entry translation method and device, a medium and equipment. The method comprises the steps that firstly, an Android source entry file is obtained and analyzed, all key value pairs in the Android source entry file are extracted, and the key value pairs are basic data of entry translation; then, the extracted key value pairs are translated through a preset external translation interface, and high efficiency and accuracy of translation are ensured. And after translation is completed, a target language resource file conforming to an Android internationalized directory structure is generated according to a translation result, and the structure enables the application to support multiple languages. Finally, format verification and grammar error correction processing are carried out on the generated target language resource file, the format verification comprises checking whether the initials of the English entries are capitalized or not, and the initials are converted into capitalized forms to conform to English grammar specifications; the grammar error correction relates to escape processing of placeholders and special characters so as to ensure the grammar correctness and compiling stability of resource files. And finally obtaining a high-quality source entry file translation result.

The system according to this embodiment aims to understand the context of a conversation and suggest appropriate responses in multiple languages. [Solution] The system according to the embodiment comprises a collection unit, an analysis unit, a suggestion unit, and a translation unit. The collection unit collects the other party's spoken and nonverbal information. The analysis unit analyzes the information collected by the collection unit and understands the context of the conversation. The suggestion unit suggests an appropriate response based on the context understood by the analysis unit. The translation unit translates the response suggested by the suggestion unit into multiple languages.

This application discloses a training method and related apparatus, equipment, and medium for a speech feature extraction model. The training method includes: acquiring sample speech data; converting the sample recognition text of the first sample speech into a first phonetic symbol sequence based on the pronunciation rules of the language to which the first sample speech belongs, and aligning the first feature sequence of the first sample speech with the first phonetic symbol sequence to determine the first sample phonetic symbol of the first speech frame in the first sample speech; training a phonetic symbol prediction model based on the first sample phonetic symbol of the first speech frame; predicting the second feature sequence of the second sample speech based on the phonetic symbol prediction model to obtain the second phonetic symbol sequence of the second sample speech; and training a speech feature extraction model based on the first speech frame and its first sample phonetic symbol, and the second speech frame and its second sample phonetic symbol. This approach can improve the model's general representation capability for multiple languages ​​while reducing model training costs.

The invention discloses a dynamic translation file management method and system for a multi-language system. The method comprises the following steps: acquiring the change of a configuration file based on an instruction input by a user; matching a to-be-downloaded translated file according to the change of the configuration file, and downloading a corresponding translated file from the public storage server; verifying the integrity and the modification state of the downloaded translated file; and dynamically loading the translated file passing the verification so as to update the language content of the multi-language system. According to the invention, the unified configuration file is adopted to correspond to the translated files of multiple languages at the same time, and the public storage server is utilized to store each translated file, so that the translated files are updated in real time and the language content of the multi-language system is dynamically updated by monitoring the change of the configuration file.

The invention provides a multilingual semantic analysis and decision support system and method for an international climate negotiation scene. The system comprises a data processing module which is used for utilizing a natural language processing model to automatically generate structured data which is composed of multiple languages and corresponds to corpus data related to a target theme; the database construction module is used for carrying out vector database index construction of hierarchical indexing on the structured data according to an issue, a language and time, and generating a corpus vector database; the content generation module is used for responding to the received content generation instruction, converting the content generation instruction into a query vector, performing retrieval from the corpus vector database according to the query vector to obtain target corpus data corresponding to the content generation instruction, and generating the target corpus data according to a pre-trained language generation model; and performing language generation processing on the content generation indication and the target corpus data to obtain a content generation result associated with the content generation indication.

This invention discloses a speech model compression method, electronic device, and storage medium. It is specifically designed for large-scale sequence-to-sequence speech recognition models with an encoder-decoder architecture. This method avoids cumbersome backpropagation computation by sequentially pruning the decoder and encoder. It can reduce the parameters of a Whisper-large model by approximately 60% without backpropagation or retraining, with almost no impact on the model's performance on various datasets. Furthermore, this method is applicable to multilingual datasets, and the pruned model maintains good robustness and generalization capabilities across multiple languages. This innovation significantly lowers the barrier to large-scale model deployment, making it easier to apply in resource-constrained environments.

The embodiment of the invention discloses a multi-language entry management method, electronic equipment and a storage medium. The method comprises the steps that a structured form file is constructed to serve as a unified data source of multi-language entries, wherein the structured form file comprises a key column used for storing a unique key name and a plurality of language columns used for storing translation texts of all languages; reading the structured table file, constructing a mapping relationship between a key name and a translated text for each language type, and generating a corresponding language entry file; and reading the plurality of language entry files, extracting and combining mapping relationships between key names and translated texts in the files, and generating the structured form file. By means of the mode, centralized management and bidirectional reversible conversion of the multi-language entries can be achieved, development and maintenance cost is reduced, multi-language consistency and quality are improved, and proofreading and auditing of the entries are facilitated.