42 results about "Machine translated" patented technology

The invention discloses a text translation method and device, electronic equipment and a storage medium, and the method comprises the steps: translating an original text through employing a neural machine translation algorithm, and obtaining an initial translated text; inputting the initial translation text into a pre-constructed culture knowledge base, and outputting a content adjustment strategy corresponding to the initial translation text through the culture knowledge base; processing the initial translated text according to the content adjustment strategy to obtain an intermediate translated text, and then calling a pre-constructed policy database to verify the intermediate translated text; and generating a target translated text corresponding to the original text according to a verification result of the intermediate translated text. According to the technical scheme provided by the embodiment of the invention, the problems of cultural semantic deviation, content violation risk and the like in cross-regional translated text can be solved, ambiguity in the translated text is eliminated, and the accuracy and flexibility of a text translation result are improved.

A speech-to-speech translation method comprises transcribing speech spoken in a source language into transcribed text data in the source language using an on-premises speech recognition model. The transcribed text data is translated into translated text data in a target language using a first on-premises machine translation model. The translated text data is reverse translated into retranslated text data in the source language using a second, different on-premises machine translation model. The transcribed text and the retranslated text are displayed on a screen. The method also involves synthesizing, using an on-premises speech synthesis model, translated speech data in the target language based on the translated text data and play back, in response to a user confirmation, translated speech in the target language based on the translated speech data in the target language.

The application is suitable for the technical field of machine translation, and provides a machine translation method and device based on imitation learning and terminal equipment, the method comprising: inputting a to-be-translated original text into a non-autoregressive model to obtain a first initial translation; when the first initial translation passes through a deletion classification decoder, deleting an error word in the first initial translation, and outputting a second initial translation; when the second initial translation passes through an insertion classification decoder, using a placeholder to fill a missing word position in the second initial translation, and outputting a third initial translation; and when the third initial translation passes through a word classification decoder, replacing the placeholder with a correct word, and outputting a final translation. The application combines the advantages of autoregressive models and non-autoregressive models, and can realize parallel acceleration while ensuring the quality of the translation.

This disclosure solves various technological problems described above by using large language models to improve machine translations for proper usage of terminology, gender, or other suitable predetermined linguistics. Such improvements may be manifested by resolution of certain grammar, morphology, or other linguistic issues in machine-translation outputs incorporating specific terminology, such as brand names and gendered terms, as disclosed herein.

The present application relates to a large model Lao translation method and system based on agent cooperation and cross-language semantic alignment, belonging to the field of natural language processing and machine translation. In the independent multilingual translation generation stage, the target Lao translation corresponding to the source sentence, the reference language translation and the linguistically similar language translation are generated respectively; cross-language semantic alignment and correction are carried out, the English translation is taken as the semantic alignment reference, the Thai translation provides morphological and syntactic guidance, and through the iterative information interaction between agents, the Lao translation is continuously evaluated and corrected; in the semantic clarification and strengthening stage, the key semantic fragments of the source sentence are extracted by analyzing the agent, the overall structure is simplified, and the Lao agent is assisted to generate the final high-fidelity translation result. The present application significantly improves the semantic accuracy and structural rationality of the Lao translation result; effectively solves the semantic illusion problem easily produced by the large language model in the Lao machine translation scene due to data scarcity.

The embodiments of this disclosure relate to post-editing methods and apparatuses, electronic devices, and storage media, and pertain to the field of machine translation technology. The post-editing method includes: acquiring a target source text and a target machine-translated text, wherein the target machine-translated text is the machine-translated version of the target source text; inputting the target source text and the target machine-translated text into a pre-trained post-editing model, and correcting the target machine-translated text using the post-editing model; the correction methods of the post-editing model include deleting tokens, inserting placeholders, and replacing placeholders with tokens; and outputting the target post-edited text obtained by the post-editing model correcting the target machine-translated text.

An electronic apparatus includes a memory configured to store instructions, and a processor configured to execute the instructions to receive a text of a first language, generate, based on the text of the first language, style information indicating a translation style to be applied to the text of the first language and machine-translate the text of the first language into text of a second language based on the generated style information.

The invention discloses a multilingual off-line simultaneous interpretation method based on a small bidirectional translation model. According to the method, voice front-end enhancement, automatic voice recognition, neural machine translation and text-to-voice conversion are sequentially executed locally on a terminal, and a unified sub-word list and a shared model are adopted to automatically judge translation between at least two languages and generate a target text / voice in a stream mode. In order to adapt to the mobile terminal, the translation model is quantized by an integer and deployed in a lightweight format, and the time delay is reduced in combination with beam width search, early stop and cache mechanisms. The voice front end adopts a causal U-shaped network to jointly suppress reverberation and noise, room impulse response and multi-noise are introduced in the training stage, a signal-to-noise ratio strategy is configured, and time sequence weighted filtering and mean square error gain are combined in the reasoning stage to improve intelligibility. The device is composed of a mobile terminal and a Bluetooth earphone, supports a collaborative chain of earphone side front enhancement and mobile phone side decoding, and realizes stable simultaneous transmission output in a weak network / no network scene.

This invention discloses a machine translation hallucination detection method based on a large language model. The method comprises the following steps: selecting a pre-defined large language model prompt template according to the translation direction; establishing word alignment relationships between the machine translation and the source language sentence; searching for semantically matched, semantically conflicting, and semantically missing word pairs between the machine translation and the source language sentence; using feedback information from the error response sample set to perform multi-dimensional risk scoring for machine translation hallucination detection and automatically optimizing the large language model prompt template; performing a seven-dimensional machine translation hallucination risk probability score, namely, semantic addition risk probability, syntactic logic addition risk probability, format noise risk probability, over-translation risk probability, factual error risk probability, implicit fiction risk probability, and fabrication information risk probability; and summarizing the seven-dimensional machine translation hallucination risk probability scores to predict whether machine translation hallucination exists in the machine translation.

Provided are computer-implemented systems and methods for providing improved rule-based machine translation, including a memory having a plurality of text translations, each text translation comprising an input text element and an output text element; at a processor: generating a text translation cluster comprising a plurality of matching text translations in the plurality of text translations, the text translation cluster associating the plurality of matching text translations; receiving the plurality of matching text translations associated with the text translation cluster; generating a first translation rule corresponding to the input text elements of the text translation cluster and a second translation rule corresponding to the output text elements; generating a machine translation rule based on the first translation rule and the second translation rule, the machine translation rule comprising a template and at least one converter; receiving an input text element for translation; and determining an output translated text element.

The invention provides a traditional medicine AI translation engine, which converts literatures such as traditional Chinese medicine and Ayurbark from a concept space to a modern medicine semantic space. A neural machine translation and semantic mapping combined double-stage method is adopted, ancient Chinese, Sanskrit and the like are translated firstly, then a traditional concept is automatically annotated as a modern term (such as liver depression and qi stagnation)-liver dysfunction caused by emotion depression), and a knowledge graph associated entity is constructed. And multi-language and multi-format output is supported, and semantic annotation teaching materials or computable knowledge data are generated, so that non-traditional Chinese medicine background researchers can understand medical cases and take the medical cases into evidence-based research. International propagation and academic mining of traditional medicine are promoted, traditional Chinese medicine digitization is assisted, and great strategic significance is achieved.

To provide a translation support device and a program for improving the efficiency of translation using machine translation software.SOLUTION: A translation support device divides a source text to be translated that is written in a first language into predetermined segments, generates a translated text in a second language that is different from the first language for each of the segments, associates the source text for each of the segments with the corresponding translated text and outputs the associated source text and translated text as parallel translation information, outputs, to a large scale language model, the parallel translation information and revision instruction information that instructs revision of the parallel translation information, acquires an output from the large scale language model, and generates revised parallel translation information.SELECTED DRAWING: Figure 1

The invention relates to the technical field of API (Application Program Interface) data processing and machine translation, and discloses a precise translation and asynchronous blocking processing method for API response body content, which comprises the following steps that: a tree path encoder identifies a field to be translated by utilizing a reflection mechanism; and generating task metadata containing a path vector of the to-be-translated field in the JSON nested structure according to a path coding formula. The identifier generator and the serializer generate a global request serial number and task snapshot data and distribute asynchronous tasks. A regular cleaner generates a cleaned text, and a semantic compensation extraction algorithm executes logic truncation and splices a semantic compensation window to generate a to-be-translated input sequence. And the thread pool concurrently schedules a third-party engine to obtain translation blocks, and the structure reconstruction engine executes in-situ backfilling to generate a translated response body after eliminating redundant translation contents. And the exponential backoff retry device processes the exception, and the callback polling module adopts a double-track delivery strategy to deliver the result. The problem of context deficiency is solved, and translation continuity is guaranteed.

The application relates to a kind of based on latent variable's paragraph-level neural machine translation method, belong to natural language processing application technical field.First in the encoder-decoder neural machine translation model, respectively, the source language sentence and its last sentence are encoded, and the latent variable is generated by the attention mechanism fusion coding information.The distribution of latent variable is considered as a kind of Gaussian distribution, which contains the context information required by paragraph-level translation.The latent variable is mapped to the shallow paragraph relationship distribution by the conversion matrix through the discriminator pre-trained on the paragraph relationship analysis task.Finally, the discriminator is applied to the training process of paragraph-level neural machine translation model, the generation of latent variable is guided, and the sentence of target language is finally generated.The application utilizes latent variable to represent the information required in paragraph-level neural machine translation, and guides the generation process of latent variable by introducing the discriminator, so as to improve the effect and quality of paragraph-level neural machine translation.

The embodiment of the invention provides a text translation method and device based on double models, electronic equipment and a storage medium, and belongs to the technical field of text translation. The method comprises the steps of obtaining a target text; performing neural machine translation on the target text to obtain an initial translation text; performing text splicing on the target text and the initial translation text to obtain a double-source contrast text; generating a prompt text based on the double-source contrast text to obtain a translation adaptation optimization indication text; based on a preset translation optimization model, performing text analysis on the translation adaptation optimization indication text to obtain a translation optimization task; and on the basis of the translation optimization model, the double-source contrast text and the translation optimization task, translated text retouching is carried out, and a target translated text is obtained. According to the embodiment of the invention, the text translation accuracy can be improved.

The invention relates to a large model multi-language translation method and device based on MoE and representation decoupling, and belongs to the technical field of natural language processing. A Chinese-English-Lao-Thai multi-language translation framework can effectively share similar language knowledge through model parameter sharing to improve translation performance in different translation directions, but parameter sharing is introduced to cause the problem that parameters of different languages interfere with one another. Aiming at the problem of parameter interference among different languages in multi-language translation, the invention provides the method, and the method comprises three parts of data preprocessing, training of a large-model multi-language translation model based on MoE and characterization decoupling, and translation by using the trained large-model multi-language translation model based on MoE and characterization decoupling. The invention also relates to a device formed by modules integrated by the three parts. According to the method, the problem of multi-language parameter interference is relieved in combination with a sparse MoE framework and representation decoupling, and the Chinese-English-Lao-Thai multi-language machine translation quality is effectively improved.

The present application relates to the technical field of natural language processing, and provides a translation method and device, electronic equipment and a storage medium, wherein the method comprises: determining to-be-translated data and a scene label corresponding to the to-be-translated data; performing dependency syntax analysis on the to-be-translated data to obtain syntax tree data representing the grammatical dependency relationship between words in the to-be-translated data; and translating the to-be-translated data based on the scene label and the syntax tree data to obtain target translation data corresponding to the to-be-translated data, which not only effectively solves the problems of part-of-speech misjudgment of polysemous words and structural confusion in translation from an agglutinative language to a fusional language, but also takes into account the grammatical accuracy of the language structure and the business professionalism of the specific field expression, greatly improving the translation accuracy and the final text quality of machine translation in the vertical field and complex conversational scenarios.

The application relates to a kind of Chinese and Vietnamese neural machine translation methods based on fusion bilingual alignment entity, belong to natural language processing technical field. Entity translation has been a difficulty in machine translation, because of its one-to-one translation characteristics, it is rarely in training corpus, and entity word can be excluded from the translation vocabulary, thereby leading to model unable to predict it in the derivation stage;The application incorporates the constraint position prompt information at the encoding end sentence representation level;In the decoding end, the pointer network mechanism is incorporated, so that the model can copy the output source sentence vocabulary. Experimental results show that the method effectively solves the entity word translation problem in sentence under low resource, compared with the baseline model, improves 2.84 BLEU score.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for transferring style formatting on translated text using neural networks. In particular, in some embodiments, the disclosed systems obtain an input text string, the input text string comprising a style formatting element. In addition, in some embodiments, the disclosed systems generate a modified input text string from the input text string, the modified input text string comprising a coded tag identifying the style formatting element. Moreover, in some embodiments, the disclosed systems generate, utilizing a neural machine translation model, a translated text string from the modified input text string. Furthermore, in some embodiments, the disclosed systems apply the style formatting element to a word of the translated text string based on the coded tag of the modified input text string.

This invention provides a machine translation method based on entropy-driven unsupervised reinforcement learning, belonging to the fields of natural language processing and machine translation. It aims to solve the problems of pattern collapse and consensus bias caused by the lack of reference translations in existing unsupervised machine translation reinforcement learning. The method includes: constructing the source sentence as an input sequence with a fine-tuning instruction format; during the autoregressive generation process, using word-level entropy as uncertainty feedback to dynamically adjust the sampling temperature and generate a diverse set of candidate translations; combining majority voting consensus and the model's intrinsic confidence to construct a confidence-weighted hybrid distribution, from which high-quality pseudo-labels are selected as supervision signals; in the policy optimization stage, entropy is used to identify high-uncertainty positions, and the advantage function is reweighted to focus on optimizing highly difficult semantic decision points. This invention can achieve the co-evolution of reward signals and model capabilities under unsupervised conditions, effectively avoiding the homogenization of generated results and significantly improving the quality of machine translation.

The invention particularly relates to an intelligent Japanese polysemy context adaptive translation optimization method, which relates to the technical field of natural language processing and machine translation, and comprises the following steps: executing deterministic shortest path finding calculation in a temporary subgraph, updating node accumulated cost, and finding a minimum total weight path from a sentence head Start node to a sentence tail End node. According to the method, the dynamic correction of the edge weight is completed by fusing the static co-occurrence weight and the vocabulary effective physical distance, so that the weight quantification better fits the actual context of the to-be-translated text; meanwhile, polysemous word splitting is only limited in a temporary dynamic sub-graph of a memory, global network node and edge number explosion is avoided, differentiated weight design of split virtual nodes is combined, so that the shortest path algorithm can achieve deterministic disambiguation, the optimal path is unique, and a selection basis can be traced through a precursor node array; and the precision of Japanese polysemy translation is improved.

The application relates to a machine translation method and device, equipment, medium and product, and the method comprises the following steps: obtaining a text to be translated, wherein the text is a Chinese language text; inputting the text and a pinyin text with a set false value into a machine translation model in parallel to obtain a first foreign language vector sequence, wherein each foreign language vector represents a probability distribution corresponding to each word in the foreign language vector sequence; when there is no foreign language vector with an abnormal probability distribution in the first foreign language vector sequence, the first foreign language vector sequence is determined as a valid vector sequence; otherwise, inputting the text and a standard pinyin text into the machine translation model in parallel to obtain a second foreign language vector sequence as a valid vector sequence; and determining each foreign language vector corresponding word in the foreign language vector sequence according to the valid vector sequence to construct a foreign language text corresponding to the text. The application can perform fault-tolerant translation on the text, and has a wide application scenario.

The embodiment of the invention relates to text style conversion on translated content using a neural network. The present disclosure relates to systems, non-transitory computer-readable media, and methods for up-converting pattern formatting on translated text using neural networks. Specifically, in some embodiments, the disclosed system retrieves an input text string that includes a style formatting element. Further, in some embodiments, the disclosed system generates a modified input text string from the input text string, the modified input text string including an encoded tag identifying the style formatting element. Further, in some embodiments, the disclosed system generates a translated text string from the modified input text string using a neural machine translation model. Further, in some embodiments, the disclosed system applies a style formatting element to the words of the translated text string based on the modified encoding tag of the input text string.

This invention relates to an efficient fine-tuning method for translation parameters of large model based on hierarchical activation difference perception, belonging to the field of natural language processing and large model fine-tuning technology. To address the problems of pre-training bias dominance and hierarchical functional imbalance in existing large language models (LLMs) for low-resource machine translation tasks, this invention includes: firstly, accurately locating model modules sensitive to the machine translation task using cumulative activation intensity and cross-layer activation variance; then, dividing the model hierarchy into task-adaptive layers and legacy inert layers based on activation difference indices; introducing learnable latent gating parameters to generate hierarchical-specific modulation factors; and during efficient parameter fine-tuning, using the modulation factors to amplify the signal in the task-adaptive layer while suppressing the legacy inert layers. This invention effectively improves translation performance from Chinese to low-resource languages ​​such as Lao and Khmer by structurally differentiating the optimization direction.

Disclosed are various embodiments for automated translations for autonomous chat agents. A build service can send a translation request to a machine translation service, the translation request comprising training data in a first language and the translation request specifying a second language. The build service can then receive translated training data from the machine translation service, the translated training data having been translated from the training data into the second language. Next, the build service can create a translated workflow that comprises a translated machine learning model and a translated intent. Subsequently, the build service can add the translated training data to the translated workflow and train the translated machine learning model using the translated training data.

Techniques for generating translated audio output based on media content are disclosed. Text is accessed corresponding to media content. One or more untranslated mouth shape indicia are determined based on the text. The text is parsed into one or more text chunks when one or more dubbing parameters are met. The parsed text is translated from a first spoken language to a second spoken language. One or more translated mouth shape indicia are determined. The one or more translated mouth shape indicia and the one or more untranslated mouth shape indicia are compared based on a predetermined tolerance threshold. A translated audio output is generated based on the translated text.

The invention belongs to the technical field of document processing, and particularly relates to a synonymous sentence recognition method based on semantic feature analysis. Comprising the following steps: segmenting and splicing an input text, and establishing a statement set; performing normalization processing on the statements, eliminating invalid data and performing word segmentation processing; clue words of the candidate synonymous sentences are extracted according to the word segmentation result; screening and filtering, and establishing a statement set index; and generating a translated text by using a machine translation technology, and respectively performing embedded representation on the original statement and the translated text to obtain a polymorphic text representation result. On the basis of semantic representation of the generated words and sentences and translations thereof, word and sentence semantic representation results in various forms can be obtained, synonym features in original words and sentences and translated words are fully utilized to further enhance the accuracy and recognition approach of synonym recognition, judgment on whether the words and sentences are synonymous or not is enhanced, the synonym recognition effect is improved, and the recognition efficiency of the synonym sentences is improved. Efficient data processing of words and sentences and translations thereof is achieved, and a technical basis is provided for practical application of the synonymous sentence recognition method.

Disclosed embodiments may provide techniques for training a machine-learning model using machine translation and multilingual embeddings. A computer-implemented method can include receiving a source document that includes text segments associated with a source language. In some instances, one or more of the text segments are associated with a target label. The computer-implemented method can also include translating the text of the source document to generate a set of translated documents that include text associated with a target language. The computer-implemented method can also include generating a set of labeled multilingual documents by mapping the target label of the source document to corresponding text segments of the set of translated documents. The computer-implemented method can also include encoding the text of the set of labeled multilingual documents into a plurality of multilingual embeddings. The computer-implemented method can also include training a machine-learning model using the plurality of multilingual embeddings.

Aiming at the defects of an existing test method in the field of word sense disambiguation test of machine translation in the aspects of triggering effective word sense conversion, maintaining original sense and maintaining semantic consistency, the invention provides a machine translation polysemous word translation evaluation method based on semantic item trigger word replacement. The method comprises the following steps: firstly, constructing a polysemy semantic item library, and screening source language sentences containing polysemy from a parallel corpus according to the semantic item library; performing natural language processing on the sentence, and identifying a trigger word of a polysemy special definition item in the sentence; replacing the trigger word with a replacement word to generate a variant sentence; performing machine translation and alignment on the original sentences and the variant sentences to obtain expressions of polysemy words in translations; and calculating the similarity of the polysemy translation expression, and judging whether a polysemy disambiguation error exists or not. Through a trigger word replacement strategy, a semantic controlled contrast sentence pair can be constructed, potential errors of a system in polysemous word translation are effectively revealed, and the pertinence and effectiveness of evaluation are improved.

The invention provides a translation model training and translating method and device, equipment and a storage medium, which can be applied to the technical fields of machine translation, model training and the like, and comprises the following steps: obtaining a term translation dictionary which comprises different language names of professional terms in different business fields; based on the term translation dictionary, a training prompt sample of multiple languages is constructed, and the training prompt sample comprises a source language name and a target language name of at least one terminology, a to-be-translated text of the source language and a translated text of the target language; and performing fine tuning on the pre-trained translation model by using the training prompt samples of the multiple languages to obtain a fine-tuned translation model. According to the method, the rich multi-language training prompt samples are used for performing directional optimization and fine tuning on the pre-trained translation model, so that a large amount of time and computing resources are saved, and the translation specialty and accuracy of the translation model are greatly improved by finely adjusting the pre-trained translation model.