38 results about "Model translation" patented technology

The invention relates to the field of the language processing, and provides a decoding method based on a deep neural network translation model. The problems of high model training complexity, large training difficulty and slow decoding speed in a machine translation model are solved. The specific implementation way of the method comprises the following steps: performing word segmentation processing on the to-be-translated sentence to obtain source language vocabulary; step two, performing word alignment on the linguistic data in a preset translation model glossary by using an automatic alignment tool so as to obtain a target language word aligned to the source language vocabulary; and step three, determining a target-side dynamic glossary of the to-be-translated sentence based on the target language word obtained in the step two, and using the sentence decoded by using a column searching method as the output of the translation model according to the pre-constructed translation model, wherein the translation model is the deep neural network based on the threshold residual mechanism and the parallel attention mechanism. Through the decoding method disclosed by the invention, the model translation quality is improved, and the mode coding speed is improved.

Automatically translating a building architecture file format (Industry Foundation Class) to a simulation file, e.g., for EnergyPlus and/or Radiance, in one aspect, may comprise extracting from a building architecture file, data and metadata used by a target simulation tool; creating interoperability data objects and storing the extracted data in the interoperability data objects; preparing a model translation procedure to identify a mapping from a Model View Definition to a translation and transformation function; transforming the extracted data using the data stored in the interoperability data objects, an input Model View Definition template, and the translation and transformation function to convert the extracted data to correct geometric values needed for a target simulation file format used by the target simulation tool; and generating the simulation file in the target simulation file format.

A method and system for process model translation is disclosed herein. The method includes generating a common process model based upon a first process model capable of being utilized by a first simulation program. A second process model is then generated based upon the common process model, the second process model being capable of being utilized by a second simulation program.

Automatically translating a building architecture file format (Industry Foundation Class) to a simulation file, e.g., for EnergyPlus and/or Radiance, in one aspect, may comprise extracting from a building architecture file, data and metadata used by a target simulation tool; creating interoperability data objects and storing the extracted data in the interoperability data objects; preparing a model translation procedure to identify a mapping from a Model View Definition to a translation and transformation function; transforming the extracted data using the data stored in the interoperability data objects, an input Model View Definition template, and the translation and transformation function to convert the extracted data to correct geometric values needed for a target simulation file format used by the target simulation tool; and generating the simulation file in the target simulation file format.

A machine translation method includes translating a source sentence using a first model, determining a back-translation probability of a translation result of the source sentence being back-translated into the source sentence using a second model, applying the back-translation probability to context information extracted from the source sentence in the first model, and retranslating the source sentence using the first model and the context information to which the back-translation probability is applied

A CAD model customized, which represents a mechanical design of an artifact comprising a set of parts with respective dimensions. A configuration model defines an exhaustive range of ways to customize the CAD model by changing a subset the dimensions. A model translation module is arranged to convert the configuration model into: (i) a CSP representing all possible customizations defined by the configuration model, the CSP is defined by: (a) set of integer variables, wherein each variable may attain a finite number of different values, and (b) a set of constraints restricting which variable values that are simultaneously possible for the variables, (ii) a set of CSP variable-dimension pairs, and (iii) a set of CSP variable-parameter pairs. A configuration engine is arranged to: calculate a solution to the CSP, which solution is optimal with respect to a value assigned to each variable in the CSP relative to a predefined optimizing criterion, and assign a parameter value for each CSP variable-parameter pair, the allocated parameter value corresponding to the value assigned to the CSP variable in the optimal solution. A CAD interface is arranged to assign a dimension in the CAD model for each CSP variable-dimension pair, where the assigned dimension corresponds to a value assigned to the CSP variable in the optimal solution.

The invention relates to a method for processing out-of-set words of Chinese-over-neural machine translation integrated with a classification dictionary, and belongs to the technical field of naturallanguage processing. Out-of-set words are classified. Different types of out-of-set words can be processed by adopting different methods. A classification dictionary is built in a targeted manner; wherein a bilingual dictionary is used for solving a translation problem of rare words outside a word list; an entity dictionary is used for solving the problem of inaccurate entity word translation; anda rule dictionary is used for solving the translation problem of numbers, symbols, time, dates and other words. Then, in the preprocessing stage of the model, out-of-set word recognition is performedby querying a classification dictionary. Label replacement is performed on out-of-set words at the encoding end of the model. A translation result with labels is acquired after model translation andtranslation recovery is carried out on the labels by querying the classification dictionary. The classification dictionary is fused into neural machine translation. The out-of-set words can be more accurately translated and the performance and effect of a neural machine translation system are improved.

The invention discloses a translation model optimization method for dynamically adjusting length punishment and translation length. The method comprises the steps of obtaining standard data in a specified language direction as a standard bilingual data set for various index prediction; performing word segmentation operation on the standard bilingual data set, and performing further training to obtain a new training data set; modifying a neural machine translation model decoder part, and automatically predicting the optimal length punishment value of the current sentence pair; performing lengthstatistics to obtain a target statement sub-length; preparing an independent feedforward neural network model so that a translation finally predicted by the model tends to a translation result with the optimal length; and enabling the Transformer neural machine translation model to dynamically adjust the length penalty and the optimal translation sentence length for different sentences. Accordingto the method, the length punishment and the dynamic adjustment of the translation length in the model translation process are realized, the realization is simple, the method is effective, the practicability is high, and the model translation quality improvement effect is obvious.

The invention discloses a neural machine translation system based on Romatized Uygur language, which provides bilingual data with segmented words and proper format for subsequent word alignment processing, for Chinese-Uygur language before constructing a translation system; according to the preprocessing of the system, different preprocessing processes are carried out on corpora according to different characteristics of the corpora, namely Jieba word segmentation is carried out on Chinese corpora, words are continuously cut by using BPE coding, root + affix form word segmentation is firstly carried out on Uygur language corpora, then Roman processing is carried out on Uygur language after word segmentation, and finally processing is carried out by using BPE coding; a translation model is trained by using a Tansformer translation model to obtain a final translation model; for a translation process, firstly, Uygur language is also preprocessed, namely, word segmentation and Roman processing of'word roots + affixes' are carried out, then the processed corpus is translated by using the trained model, and finally translated Chinese is obtained.

The invention relates to a construction method of a semi-supervised neural machine translation model based on word-to-word translation, and belongs to the technical field of natural language processing. The method comprises the steps that firstly, monolingual corpora of a source language and a target language and parallel corpora of the source language and the target language are acquired; training a cross-language language model by using the monolingual corpus; initializing an encoder-decoder of the translation model by using the trained language model; acquiring bilingual dictionaries of thetwo languages; constructing a dictionary prefix tree in the translation model according to the bilingual dictionary; training an auto-encoder in the translation model; training a translation model byusing the parallel corpora of the source language and the target language; training the translation model and training the translation model at the same time; and training the translation model by using the parallel corpora of the source language and the target language through the auto-encoder, and fusing the translation model to obtain a final translation model. The method is simple and effective, the model can be translated normally, and the translation performance of the model is greatly improved.

The invention provides a data processing method and device, and the method and device are used for a decoder which comprises at least one decoding layer. For each decoding layer, the method comprisesthe following steps: carrying out self-attention calculation according to an input reference vector to be decoded to obtain a first sub-layer vector; performing multi-head attention calculation on thefirst sub-layer vector and the input coding vector to obtain a second sub-layer vector; fusing the reference vector to be decoded and the second sub-layer vector to obtain a fusion vector; carrying out linear transformation on the fusion vector to obtain an output decoding vector, so that vector fusion of the high-order sub-layer and the low-order sub-layer is realized, semantic information contained in the high-order sub-layer and grammatical information contained in the low-order sub-layer are effectively utilized, and therefore, effective improvement of a model translation effect is ensured.

The invention discloses a pose computation method for point cloud position of non-cooperative target. According to the method, the optimal matching path is searched synchronously from different initial parameters. Finally the overall optimal matching result is obtained by final comparison. The precision of the computation result of the overall position pose is high. According to the invention, aiming at the incomplete condition of the cloud point information of the non-cooperative target, model translation is adopted to carry out different spot matching. Even if only a pat of point cloud is scanned, the method can obtain the optimal configuration effect as well. The overall position pose computation result is ensured to be accurate. The robustness is high.

The invention relates to a dependency graph network-based Han-Vietnamese neural machine translation method, and belongs to the technical field of natural language processing. The method comprises thefollowing steps: firstly, crawling Chinese-Vietnamese parallel corpora ar from a website, and performing dependency syntax analysis on the obtained Chinese-Vietnamese bilingual corpora through a dependency syntax analysis tool; secondly, converting the obtained source language dependency syntax parsing tree into a dependency graph by using a graph encoder, encoding the dependency graph to obtain dependency graph structure information, sending the dependency graph structure information to an encoding end to be fused with source language sequence information, and sending the fused information toa decoding end to guide a model to translate. According to the invention, the dependency relationship is converted into the dependency graph by fusing the dependency syntax information, and the graphneural network is utilized to realize global structural coding of the dependency graph, so that richer global dependency information is provided for the translation model, and certain help is provided for improving the translation effect of the Han-Van language pairs with syntax differences.

The invention discloses a text recognition method, device and equipment, relates to the technical field of internet, and aims at abnormal information deformed in a to-be-recognized text, abnormal information recognition can be performed after the abnormal information is translated into an original text in combination with a machine model, and the flexibility of abnormal information recognition is improved while the accuracy of a recognition result is ensured. The method comprises the steps of obtaining a plurality of character elements formed by character-level segmentation of a to-be-recognized text; encoding each character element to form a phonetic code vector of the character element through encoding processing; inputting the phonetic code vectors of the character elements into a pre-constructed recognition model to obtain an original text mapped by the to-be-recognized text, the recognition model having a function of performing semantic translation on deformation information in the phonetic code vectors; and judging whether an original text mapped by the to-be-recognized text contains abnormal information or not by utilizing a pre-constructed sensitive word bank.

The invention discloses a neural machine translation method based on a translation memory bank. The method comprises the following steps: crawling massive bilingual data from a network containing foreign language data; preprocessing the bilingual data to obtain processed bilingual data; storing half of the bilingual data as basic data of a translation memory bank, and adopting the other half of the bilingual data to find sentences most similar to each sentence in the training corpus in the translation memory bank by using a search technology; inputting the training corpus data and the translation memory bank data into the neural network model together to start training the model; importing user-defined data into a translation memory bank; searching a text input by a user in real time to find a sentence most similar to the text in a translation memory bank, inputting the sentence into a machine translation model, and guiding model translation at a decoding end by using a gating mechanism; and storing the sentence input by the user and the revised result into a translation memory bank. The advanced neural machine translation technology is applied, the translation effect is excellent, and the high-quality requirements of different users for translated texts are met.

The invention provides a model translation method and device and a computer readable storage medium. The method comprises the steps of obtaining a to-be-translated model and a training data set; converting the neural network layer of the to-be-translated model into a target neural network layer supported by a target processor to obtain a target model; and training the target model according to the training data set to determine parameters of each target neural network layer in the target model. By means of the method, the model generated under the specific framework can be converted into the target model capable of being operated by the target processor, and therefore the implementation scene of the model is expanded.