126results about How to "Improve translation performance" patented technology

The invention discloses a translation model establishing method and system. The translation model establishing method comprises the following steps: respectively generating a regular alignment table, a word semantic vector table and a phrase table according to alignment information of a double-language parallel corpus, subsequently generating a source language phrase semantic vector table of a source language semantic space and a target language phrase semantic vector table of a target language semantic space by using the word semantic vector table and the phrase table, and finally training by using phrase semantic vector tables of different semantic spaces, thereby generating a translation model integrated with semantic information. The result shows that phrase semantic information can be integrated in statistic machine translation, the research shows that the relevance of words or phrases to context words or phrases can be reflected in the semantic information, and compared with a conventional translation method based on words or phrases, the translation model is relatively high in translation quality after the phrase semantic information is integrated, so that the translation property of the statistic machine translation is further improved as compared with that of the prior art.

The invention discloses a method for determining target information. The method includes the following steps that encoding is conducted on to-be-processed text information to obtain a source-end vector expression sequence; according to the source-end vector expression sequence, a source-end context vector corresponding to the first moment is obtained, wherein the source-end context vector is used for expressing to-be-processed source-end content; according to the source-end vector expression sequence and the source-end context vector, a first translation vector and / or a second translation vector are / is determined, wherein the first translation vector indicates the source-end content which is not translated in the source-end vector expression sequence within the first moment, and the second translation vector indicates the source-end content which is translated in the source-end vector expression sequence within the second moment; decoding is conducted on the first translation vector and / or a second translation vector and the source-end context vector so as to obtain the target information of the first moment. The invention further provides a method of translation and a device for determining the target information. According to the method for determining the target information, the model training difficulty of a decoder can be reduced, and the translation effect of a translation system is improved.

A domain-adapted portable machine translation device for closed captions using dynamic translation resources and a method thereof are provided. The device and method improve translation performance with respect to various specialized domains by automatically recognizing style and domain of a caption sentence, dynamically constructing specialized translation module and knowledge, and performing automatic translation with optimal translation resources suitable for a corresponding sentence.Not only does the domain-adapted portable machine translation device for closed captions exhibit improved translation performance for sentences in various domains, but it can also be freely linked with various media devices through input / output terminals.

The invention provides a translation method and device, a translation model training method and a translation device, wherein, the translation method comprises the following steps of: obtaining a sentence to be translated; inputting the sentence to be translated into a translation model to obtain a translation sentence corresponding to the sentence to be translated, wherein the translation model is trained in the following ways: the first sentence in the target corpus and the translated second sentence are used as training samples and training features respectively; inputting the training sample to the coding layer of the translation model to obtain at least one coding vector; inputting at least one coding vector to a weight layer of the translation model to obtain a weight vector; inputting the training feature and the weight vector to a decoding layer of the translation model, obtaining a decoding vector, and obtaining an error of the decoding vector according to the decoding vector;adjusting the weight layer according to an error of the decoding vector, and continuing to train the translation model until a training stop condition is reached.

The invention discloses a statistical machine translation method with error self-diagnosis and self-correction functions. The method comprises the following steps: firstly, defining the translation error type, training an error classifier, carrying out translation error classifying on a test set, then mapping the translation error from a target language terminal to a source language terminal and constructing a repeated word graph network, optimizing the repeated word graph network of the source language, and finally, carrying out word graph decoding to obtain a self-correction result. Compared with the current statistical machine translation method, according to the statistical machine translation method with error self-diagnosis and self-correction functions, the translation error rate is effectively lowered, and the translation performance is improved.

The invention discloses a neural machine translation method based on word prediction. The method includes the step of adding a word prediction mechanism during the training process of an end-to-end neural machine translation system based on an attention mechanism, wherein the step can be specifically divided into two aspects: in the first aspect, the word prediction mechanism is added into a source end namely an encoding end; in the second aspect, the word prediction mechanism is added into a target end namely a decoding end. Compared with existing neural machine translation systems, the neural machine translation method based on the word prediction can fully utilize the information of words and enhance capabilities of the source end and the target end during training, so that the translation quality of the machine translation system can be improved, and during translation, by using the word prediction mechanism of the encoding end to compress a demanded word list, translation efficiency can be greatly improved. Under a practical experiment circumstance, the translation quality and efficiency can both be significantly improved.

The invention relates to the technical field of neural machine translation, and discloses a neural machine translation method oriented to a small language. The method solves the problem of neural machine translation under the condition of lack of parallel corpora. According to the method, a neural machine translation model is constructed and trained through the following steps: 1, obtaining and preprocessing monolingual corpora; 2, respectively training language models of a source language and a target language by utilizing the monolingual corpus; 3, respectively training mapper used for mapping the encoding result of one language to the space of the other language by utilizing the encoding results of the bilingual parallel corpus in the parallel corpus of the small language in the language models of the source language and the target language; 4, training a discriminator model by utilizing the monolingual corpus; and 5, training a translation model by utilizing the language model, themapper, the discriminator model, the bilingual parallel corpus and the monolingual corpus. The method is suitable for translation between small languages only having small-scale parallel corpora.

The invention discloses an ancient Chinese automatic translation method based on multi-feature fusion. The method comprises the following steps: 1) collecting a text, modern text translation data of the text, a text word list and modern Chinese monolingual corpus data; And 2) cleaning the data and constructing an ancient Chinese parallel corpus by using a sentence alignment method. And 3) carryingout word segmentation on the modern text and the ancient text by using a Chinese word segmentation tool; 4) performing topic modeling on the ancient text corpus to generate topics-Word distribution and word-Subject conditional probability distribution 5) using the modern Chinese monolingual corpus to train to obtain a modern Chinese language model; And obtaining an aligned dictionary by using ancient Chinese parallel corpora. 6) on the basis of the attention-based recurrent neural network translation model, fusing statistical machine translation characteristics such as a language model and analignment dictionary, and using an ancient Chinese parallel sentence pair and a word topic sequence training model, and 7) inputting a to-be-translated text by a user, and obtaining a modern text translation by using the model obtained by training in the step 6).

The invention provides method and system for adapting a machine translation system based on a language database to a new field. The method comprises the following steps of: translating a plurality of source language sentences in the new field by using the machine translation system based on the language database which is trained in one field; selecting the source language sentences the evaluation of the translated result of which is lower than a pre-set first evaluation threshold from the plurality of source language sentences; recognizing a text fragment related to the new field from the source language sentences evaluation of the translated result of which is lower than the first evaluation threshold; and updating the machine translation system by using the plurality of source language sentences and the translated results thereof, as well as the text fragment related to the new field and a correct translated text thereof. In the invention, the machine translation system trained well outside the field trains the machine translation system through using the text fragment which is recognized in the process of repeatedly translating the text in the new field and is related to the new filed so as to continuously improve the translation performance of the new field by using the machine translation system.

The invention discloses a multi-field neural machine translation method based on a self-attention mechanism. The invention discloses a multi-field neural machine translation method based on a self-attention mechanism. The multi-field neural machine translation method comprises the following steps: carrying out two important changes on a Transformer, wherein the first change is a self-attention mechanism based on domain perception, and the domain representation is added to a key and a value vector of the original self-attention mechanism, the weight of the attention mechanism is the degree of correlation of the query and domain aware keys, the second change is to add a domain representation learning module to learn a domain vector. The method has the beneficial effects that a domain-aware NMT model architecture is provided on the basis of a neural network architecture Transformer representing the most advanced level at present. A self-attention mechanism based on domain awareness is provided for multi-domain translation. It is known that this is a first attempt on a multi-domain NMT based on a self-attention mechanism. Meanwhile, experiments and analysis also verify that the model can significantly improve the translation effect of each field and can learn the field information of training data.

The invention discloses a training method and device for a speech translation model. The method comprises the steps that model training data including all pieces of sample speech is obtained, then translating is conducted on the obtained sample speech directly by means of a current speech translation model, and predicted translation text is obtained; meanwhile, by means of a current speech recognition model, the obtained sample speech is recognized, and predicted recognition text is obtained; then, according to the obtained predicted translation text and predicted recognition text, parametersof the speech translation model and the speech recognition model are updated. Due to the fact that the speech translation model and the speech recognition model share part of the model parameters, when the parameters of the speech recognition model are updated, the shared model parameters in the speech translation model can be updated, and then the model parameters of the speech translation modelare more accurate; therefore, when speech translating is conducted by means of the speech translation model, the translation performance of the speech translation model can be improved.

The invention designs a machine translating device based on intelligent matching and a method of the machine translating device. The method comprises the following steps of: A step, carrying out a linguistic data pre-processing step: carrying out word frequency counting and marking on sub-domain linguistic data; B step, carrying out a sentence database establishing step: establishing the sentence database by using sub-domain English-to-Chinese linguistic data; C step, carrying out a word database establishing step: establishing a word database according to the linguistic data pre-processing step and the sentence database establishing step; and D step, carrying out a machine translating step: firstly, manually making a phrase framework consisting of Chinese phrase nodes and a translation rule of the phrase framework according to the sentence database establishing step and the word database establishing step, and identifying a rule of a Chinese sentence sub point; automatically generating a sentence framework template base by using the sentence database, grammar identification and the phrase framework; and finding a framework matching sentence.

According to one aspect, there is provided an apparatus for improving a bilingual corpus including a plurality of sentence pairs of a first language and a second language and word alignment information of each of the sentence pairs, the apparatus comprises: an extracting unit for extracting a split candidate from word alignment information of a given sentence pair; a calculating unit for calculating split confidence of said split candidate; a comparing unit for comparing said split confidence and a pre-set threshold; and a splitting unit for splitting said given sentence pair at said split candidate in a case that said split confidence is larger than said pre-set threshold.

An apparatus and method for automatic translation are disclosed. In the apparatus for automatic translation, a User Interface (UI) generation unit generates UIs necessary for start of translation and a translation process. A translation target input unit receives a translation target to be translated from a user. A translation target translation unit translates the translation target received by the translation target input unit and generates results of translation. A display unit includes a touch panel for outputting the results of translation and the UIs in accordance with the location of the user.

The invention belongs to the field of machine translation, particularly relates to a speech translation method, a system and equipment fusing text semantic features, and aims to solve the problem that the existing speech translation method is difficult to fuse information between different modals and cannot fully utilize data in the fields of speech recognition and machine translation, so that the translation performance is poor. The method comprises the following steps: acquiring to-be-translated source language voice data; extracting a voice feature sequence corresponding to the source language voice data; obtaining acoustic representation corresponding to each voice feature; mapping the implicit vector of the acoustic representation to a source language word list, and obtaining the probability that the speech feature sequence is recognized as a word in the source language word list at each moment through a softmax function; filtering the acoustic features, and obtaining semantic features corresponding to the filtered acoustic features through a second encoder; and based on the semantic representation, obtaining a target language translation text corresponding to the source language voice data through a decoder. According to the method, the speech translation performance is improved.

The invention relates to a neural machine translation method based on Multi-BiRNN encoding. Multi-BiRNN encoding is adopted at an encoder end, that is to say that on the basis of using source languagesentences as input sequences, one or more groups of BiRNN are added to encode other input sequences associated with the input sequences; on the basis of Multi-BiRNN-encoded neural machine translation, in the encoding process of a source end, the source language sentence sequences and a dependency syntax tree thereof are considered at the same time, and serialization results of the syntax tree areobtained by means of two different traversal modes respectively and serve as Multi-BiRNN encoding input with the source language sentence sequences; at the output end of each BiRNN, a word is formedin a vector splicing mode. According to the method, vectors obtained by encoding contain more abundant semantic information, the source language sentence sequences and other sequences associated withthe source language sentence sequences are considered at the same time, and the disambiguation function is achieved in the semantic representation process of the source language sentences.

A method and a device for training a neural machine translation model to ensure high translation performance even in a language pair or a domain having a small amount of parallel corpora and solving the problems of over-translation and under-translation caused by the inaccuracy of word-alignment information of an attention network. To this end, bidirectional neural machine translation models are built, and single language corpora are made available for training on the basis of symmetric relation between the models. Also, incomplete alignment information between attention networks of the bidirectional neural machine translation models is normalized to have orthogonal relation so that accurate alignment information may be learned.

The invention relates to a crowdsourcing translation method and platform, and belongs to the technical field of internet language translation. The crowdsourcing translation method comprises the stepsof firstly, matching attributes of a received translation order with attributes of translators, and determining a set number of the translators with the highest matching degree with the attributes ofthe translation order; secondly, pushing the translation order to the set number of the determined translators, and waiting the translators to grab the order; and finally, selecting a translator according to selection standards from the translators grabbing the order, allocating the translation order to the selected translator, and after the selected translator finishes the translation order, feeding back an excitation value corresponding to the translation order to the selected translator. The matching degrees between the translation order and the translators are fully considered; and the matching degree between the selected translator and the translation order is high, so that the final translation effect is improved.

The invention discloses a multi-modal machine translation method based on variational reasoning and multi-task learning. Before the method has translation capability, multi-task modeling is performedon multi-modal information such as images, texts and the like by using a variational reasoning theory, and then a variational multi-modal machine translation model is obtained by training under the condition of giving a sufficient training set, so that the machine translation capability is obtained. And finally, predicting a plurality of translated texts through bundle search and maximum likelihood. The innovation point of the invention lies in that a model capable of integrating multi-modal information such as images into machine translation, namely variational multi-modal machine translation, is created and used. According to the variational model, a set of feature extraction neural network framework confusing image and text semantics is constructed, a modeling process and a self-learning updating process are derived at the same time, a detailed derivation algorithm is given, and an application method is given in an instructive mode.

The embodiment of the invention discloses a method and device for generating a translation model. A specific embodiment of the method comprises the following steps of: obtaining a target object; the method comprises the following steps of: inputting a training sample into a forward encoder and a forward decoder to obtain a forward translation statement segmentation probability sequence; and inputting the training sample into a reverse encoder and a reverse decoder to obtain a reverse translation statement segmented word probability sequence to determine a total loss value, and adjusting modelparameters of the forward encoder, the forward decoder, the reverse encoder and the reverse decoder when it is determined that the model is not trained until it is determined that the model training is completed. According to the embodiment, the influence of front translation errors on rear translation in the translation process is reduced.

The embodiment of the invention discloses a machine translation method and device, electronic equipment and a medium. The method comprises the steps that a to-be-translated source language sentence isacquired; wherein a pre-trained machine translation model is used for giving a target language translation result of the source language sentence and a category to which the source language sentencebelongs at the same time, the category comprises coarse classification and fine classification under the coarse classification, and the machine translation model is used for giving the target languagetranslation result in combination with a classification task of the source language sentence. According to the embodiment of the invention, the trouble of manually selecting the field by a user is avoided; meanwhile, a classification task and a translation task are executed; coarse classification and fine classification to which the coarse classification belongs are given in the classification task. Compared with the prior art, the method has the advantages that each field is classified in the same dimension instead of in the same dimension, so that the model is more targeted in a category classification stage, categories and fields described by source language sentences can be automatically identified, meanwhile, translation content boundaries responsible for each field are clearer, andthe translation effect is improved.

The invention relates to a neural machine translation model training method and a neural machine translation method and device. The neural machine translation method comprises: identifying a named entity in a to-be-translated source statement; replacing the identified named entity with a label corresponding to the category of the named entity to obtain an intermediate source statement; translatingthe intermediate source statement through a neural machine translation model to obtain an intermediate target statement with a label; searching translations of the named entities from a preset namedentity dictionary and / or named entity library; and replacing the corresponding label in the intermediate target statement with the found translation to obtain a target statement corresponding to the source statement to be translated. The problem that the low-frequency named entity is wrongly translated or missed in the machine translation process is solved.

A machine-translation apparatus using multi-level verbal-phrase patterns includes: a simple sentence generation unit for generating an input simple sentence; a basic verbal-phrase pattern-matching unit for trying a match of a semantic code of each case component of the input simple sentence with basic verbal-phrase patterns; a default verbal-phrase pattern matching unit for trying a match of a size and case prepositions of the input simple sentence with default verbal-phrase patterns having a verb identical to that of the input simple sentence; a default word-order matching unit for trying a match of a word-order of the input simple sentence with default word-order verbal-phrase patterns having a case component structure identical to that of the input simple sentence; and a default preposition matching unit for generating a target sentence of an input sentence with default preposition patterns having a context identical to that of the input simple sentence.

The invention discloses a multi-modal machine translation data enhancement method based on image description generation, which comprises the following steps of: training an attention mechanism-based image description generation model by using pre-trained image coding information and corresponding image description; encoding and decoding pictures in the existing multi-modal training data by using the trained image description generation model to generate a corresponding source language image description text; translating the generated source language image description text into a target language, and constructing pseudo data; and adding the constructed pseudo data into the multi-modal training data, fusing the picture information in the multi-modal training data with the source language description information, sending the fused information into a multi-modal machine translation model, and generating a target language translation assisted by the image context information in an autoregressive manner. The diversity of the pseudo data is enriched, the performance can be improved from knowledge refinement, and compared with a common data enhancement method adopting a random replacementmode and the like, the invention has great advantages.

The invention relates to a depthwise separable convolution-based Chinese-Vietnamese neural machine translation method, and belongs to the technical field of resource scarcity type language neural machine translation. According to language characteristics of Vietnamese, Vietnamese is segmented into four different translation granularity sequences of words, syllables, characters and sub-words; secondly, a neural machine translation model is improved by utilizing depthwise separable convolution; a depthwise separable convolutional neural network is added; convolution operation is carried out on different granularity sequences input by the model and more feature data are extracted. By adding the depthwise separable convolutional neural network prior to an encoder embedding layer of the neuralmachine translation model, The depthwise separable convolution-based Chinese-Vietnamese neural machine translation method is constructed in the field of resource scarcity type language machine translation, so that the problem of sparse data of the resource scarcity type language neural machine translation is effectively relieved, and the translation performance is improved.

The invention provides a translation method based on a multi-modal machine translation model, which comprises the following steps: obtaining a source end sentence and a corresponding translation image, and preprocessing the source end sentence and the translation image to obtain the processed source end sentence, the global feature of the translation image and the local feature of the translationimage; establishing a multi-modal machine translation model, and training the multi-modal machine translation model according to the multi-modal machine translation model, the multi-modal machine translation model comprising an encoder and a decoder, and the decoder comprising a context-guided capsule network; translating the processed to-be-translated source end sentence and the corresponding translation image based on a trained multi-modal machine translation model to generate a target end sentence corresponding to the to-be-translated source end sentence. Therefore, the context is introduced into the decoder of the multi-modal machine translation model to guide the capsule network to translate, and introduction of a large number of parameters can be avoided while rich multi-modal representation is dynamically generated, so that the performance of multi-modal machine translation is effectively improved.