CN111126078B - Translation method and device - Google Patents

Abstract

The embodiment of the present disclosure discloses a translation method and device. The method adopts a translation model, and the translation model includes an encoder, a classifier and a decoder. The method includes: inputting the vector matrix determined based on the word segmentation sequence of the first text into the encoder to obtain an intermediate representation output by the encoder; inputting the intermediate representation into the classifier , to obtain the classification label output by the classifier; in response to the classification label indicating that the word segmentation at the end of the word segmentation sequence of the first text is an independent translation unit, the intermediate representation is input into the decoder, and the second text output by the decoder is obtained. This method reduces the time consumption of the system while the translation quality is guaranteed, and is easy to improve on the basis of the existing translation model, and realizes the control of the translation unit by using simple and introducing supervised learning.

Description

Method and device for translation

technical field

The present disclosure relates to the field of computer technology, in particular to the field of translation technology, and in particular to a translation method and device.

Background technique

Simultaneous interpretation has developed rapidly in the past two years, and major Internet companies have launched their own simultaneous interpretation products one after another.

Most of the current simultaneous interpretation systems are based on the pipeline (pipeline). First, the text is generated through speech recognition (ASR), and then the sentence segmentation module is called to segment the text, and then the translation model is called to translate the text after the sentence segmentation, and finally the translation result is displayed. . Specifically, the simultaneous interpretation system increases the frequency of sentence segmentation and reduces the granularity of sentence segmentation. Alternatively, the simultaneous interpretation system adopts the wait-k words model. When the recognition result exceeds K words, the translation starts. Every time an additional word is recognized, an additional word is translated. When the sentence segmentation model judges that the sentence can be segmented, the tail part is translated at one time.

However, if the simultaneous interpretation system increases the frequency of calling translation by frequently adding punctuation, it will increase the probability of wrong sentence segmentation, which will affect the quality of translation, and if there is semantic dependence between clauses, it will further increase translation errors; The translation delay is still relatively large. If the simultaneous interpretation system adopts the wait-k words model, it will greatly help reduce the delay of the simultaneous interpretation system, but it still needs to rely on an additional sentence segmentation module, otherwise the translation result will be delayed more than the recognition result. In the case of Chinese-English translation, under the same meaning, the average English sentence length is about 1.25 times that of Chinese (Huang and Zhao, When to finish? Optimal beam search for neural text generation), if there is no "catch-up" at the moment of sentence breaking, English will Always behind Chinese. In addition, the wait-k model requires that a word must be decoded at each moment. If the recognition result information is insufficient, the decoded word may be wrong, which will affect the translation effect.

Contents of the invention

Embodiments of the present disclosure provide translation methods and devices.

In the first aspect, an embodiment of the present disclosure provides a translation method, including: a translation method, using a translation model, the translation model includes an encoder, a classifier, and a decoder, and the method includes: converting word segmentation based on the first text The vector matrix determined by the sequence is input to the encoder to obtain the intermediate representation output by the encoder; the intermediate representation is input to the classifier to obtain the classification label output by the classifier; in response to the classification label, it indicates that the word segmentation at the end of the word segmentation sequence of the first text is independent The translation unit of , inputs the intermediate representation into the decoder, and obtains the second text output by the decoder.

In some embodiments, in response to the classification label indicating that the word segmentation at the end of the word segmentation sequence of the first text is an independent translation unit, the intermediate representation is input into the decoder, and the second text output by the decoder includes: in response to the classification label indication The word segmentation at the end of the word segmentation sequence of the first text is an independent translation unit, and there are translation results corresponding to the word segmentation before the end of the word segmentation sequence of the first text in the historical translation results, the intermediate representation is input into the decoder, and the history In the translation result, the translation result corresponding to the word segment before the tail in the word segment sequence of the first text is used as a constraint, and the constraint is input into the decoder to obtain the second text output by the decoder.

In some embodiments, the training data samples of the translation model are determined based on the following steps: using a word alignment tool to align the training data of the translation model to obtain alignment information of the training data; using the alignment information of the training data as the training data samples of the translation model.

In some embodiments, the word segmentation sequence of the first text is obtained through the following steps: recognizing the input first voice to obtain the first text; performing word segmentation on the first text to obtain the word segmentation sequence of the first text.

In some embodiments, the method further includes: generating a second voice based on the translated second text; and playing the second voice.

In a second aspect, an embodiment of the present disclosure provides a device for translation, using a translation model, the translation model includes an encoder, a classifier, and a decoder, and the device includes: an encoder input unit configured to convert word segmentation based on the first text The sequence-determined vector matrix is input to the encoder to obtain an intermediate representation output by the encoder; the classifier input unit is configured to input the intermediate representation to the classifier to obtain the classification label output by the classifier; the decoder input unit is configured to respond to The classification label indicates that the word segmentation at the end of the word segmentation sequence of the first text is an independent translation unit, and the intermediate representation is input into the decoder to obtain the second text output by the decoder.

In some embodiments, the decoder input unit is further configured to: respond to the classification label indicating that the word segment located at the end of the word segment sequence of the first text is an independent translation unit, and there is a word segment sequence corresponding to the first text in the historical translation results The translation result of the participle before the end in the middle representation is input into the decoder, and the translation result of the participle before the end in the participle sequence corresponding to the first text in the historical translation result is used as a constraint, and the constraint is input into the decoder to obtain the decoding The second text output by the compiler.

In some embodiments, the training data samples of the translation model used in the device are determined based on the following units: a training data alignment unit configured to use a word alignment tool to align the training data of the translation model to obtain alignment information of the training data; The determining unit is configured to use the alignment information of the training data as a training data sample of the translation model.

In some embodiments, the device further includes: a first speech recognition unit configured to recognize the input first speech to obtain the first text; a first text word segmentation unit configured to perform word segmentation on the first text to obtain the first text The token sequence of the text.

In some embodiments, the device further includes: a second speech generation unit configured to generate a second speech based on the translated second text; a second speech playback unit configured to play the second speech.

In a third aspect, an embodiment of the present disclosure provides an electronic device/terminal/server, including: one or more processors; a storage device for storing one or more programs; when one or more programs are used by one or more executed by one or more processors, so that one or more processors implement the method of translation as described above.

In a fourth aspect, an embodiment of the present disclosure provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned translation method is implemented.

The translation method and device provided by the embodiments of the present disclosure, the translation method adopts a translation model, and the translation model includes an encoder, a classifier, and a decoder. The method includes: first, inputting the vector matrix determined based on the word segmentation sequence of the first text into the encoding device to obtain the intermediate representation output by the encoder; after that, input the intermediate representation into the classifier to obtain the classification label, which indicates whether the word at the end of the word segmentation sequence of the first text is an independent translation unit; after that, in response to the classification label Indicates that the word segment at the end of the word segment sequence of the first text is an independent translation unit, the intermediate representation is input into the decoder, and the text output by the decoder is used as the translated second text. In this process, the sentence segmentation module in the traditional system pipeline process is removed, and the translation unit is controlled by the translation model, so that the time consumption of the system is reduced while the translation quality is guaranteed, and, on the basis of the existing translation model, it is easy to Improvement, the control of the translation unit is realized by using simple and supervised learning, and, in some embodiments, each translation of the tail participle in the translation process takes into account the previous part of the participle sequence of the first text The historical translation results of word segmentation can effectively solve the problem of entity reference disambiguation and domain translation.

Description of drawings

Other features, objects and advantages of the present disclosure will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

FIG. 1 is an exemplary system architecture diagram in which the present disclosure can be applied;

Fig. 2a is a schematic flowchart of an embodiment of a translation method according to an embodiment of the present disclosure;

Figure 2b shows an exemplary structural diagram of the translation model in Figure 2a;

FIG. 3 is an exemplary application scenario of a translation method according to an embodiment of the present disclosure;

Fig. 4 is a schematic flowchart of another embodiment of a translation method according to an embodiment of the present disclosure;

Fig. 5 is an exemplary structural diagram of an embodiment of the translation device of the present disclosure;

FIG. 6 is a schematic structural diagram of a computer system suitable for implementing a server according to an embodiment of the present disclosure.

Detailed ways

The present disclosure will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain related inventions, rather than to limit the invention. It should also be noted that, for the convenience of description, only the parts related to the related invention are shown in the drawings.

It should be noted that, in the case of no conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and embodiments.

FIG. 1 shows an exemplary system architecture 100 of an embodiment of the translating method or translating apparatus to which the present disclosure can be applied.

As shown in FIG. 1 , a system architecture 100 may include terminal devices 101 , 102 , 103 , a network 104 and a server 105 . The network 104 is used as a medium for providing communication links between the terminal devices 101 , 102 , 103 and the server 105 . Network 104 may include various connection types, such as wires, wireless communication links, or fiber optic cables, among others.

Users can use terminal devices 101 , 102 , 103 to interact with server 105 via network 104 to receive or send messages and the like. Various communication client applications can be installed on the terminal devices 101, 102, 103, such as translation applications, browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

The terminal devices 101, 102, and 103 may be hardware or software. When the terminal devices 101, 102, and 103 are hardware, they may be various electronic devices supporting browser applications, including but not limited to tablet computers, laptop computers, and desktop computers. When the terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. It can be implemented, for example, as a plurality of software or software modules for providing distributed services, or as a single software or software module. No specific limitation is made here.

The server 105 may be a server that provides various services, such as a background server that provides support for browser applications running on the terminal devices 101 , 102 , and 103 . The background server can analyze and process the data received, such as requests, and feed back the processing results to the terminal device.

It should be noted that the server may be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server is software, it may be implemented as multiple software or software modules for providing distributed services, or as a single software or software module. No specific limitation is made here.

In practice, the translation method provided by the embodiments of the present disclosure can be executed by the terminal equipment 101, 102, 103 and/or the server 105, 106, and the translation device can also be set on the terminal equipment 101, 102, 103 and/or the server 105, 106 in.

It should be understood that the numbers of terminal devices, networks and servers in Fig. 1 are only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.

Continuing to refer to FIG. 2a, FIG. 2a shows a process 200 of an embodiment of the translation method according to the present disclosure. The translation method includes the following steps:

Step 201, input the vector matrix determined based on the word segmentation sequence of the first text into the encoder to obtain the intermediate representation output by the encoder.

In this embodiment, the executing subject of the translation method (for example, the terminal or the server shown in FIG. 1 ) may use a translation model with a classifier to control the translation unit during translation. The translation model here may be a multi-task neural translation model, specifically, may include: an encoder, a classifier, and a decoder.

Among them, the encoder accepts the word segmentation sequence of the first text as input, and encodes the information in the word segmentation sequence into an intermediate representation. The classifier accepts the intermediate representation as input, and determines whether the word segment at the end of the word segment sequence of the first text is an independent translation unit, and if so, inputs the intermediate representation into the decoder. The decoder decodes the intermediate representation into the target language.

Specifically, when the above execution subject detects that the input first text includes K (K is an integer greater than 1) number of word segments, it starts to input the vector matrix determined based on the word segment sequence of the first text into the encoder to Get an intermediate representation of the encoder output.

The above-mentioned translation model can refer to the structure shown in Figure 2b. The upper part of the above-mentioned translation model is a conventional translation model (it can be a translation model in the prior art or a technology developed in the future, and the present application is not limited to this. For example , based on Vaswani and Shazeer's Transformer (Attention is all you need) or RNN, etc.), the lower part of the above translation model is connected to a classification model after the encoder. The update can be optimized by supervised learning, where the classifier and decoder jointly update the parameters at the encoder. Specifically, a large-scale source language-target language translation sentence pair can be constructed first, and then an end-to-end translation model can be used to train on this sentence pair to optimize model parameters. When testing, the source language text is generally given to directly generate the target language text.

In some optional implementations of this embodiment, the training data samples of the translation model are determined based on the following steps: use a word alignment tool to align the training data of the translation model to obtain the alignment information of the training data; use the alignment information of the training data as the translation A sample of training data for the model.

In this implementation, by using word alignment tools (such as TER-based, GIZA++, METEOR) in existing technologies or technologies developed in the future, the training data of the translation model can be aligned to obtain the alignment information of the training data. A concrete example is as follows:

Source text: I |want|a |excellent| wallet.

Target text: I|want|a|magnificent|wallet.

According to the above alignment information, multiple pieces of training data for the new translation model can be generated, as shown in Table 1 below, there are two training targets, which are classification labels (binary classification, that is, whether it can be used as an independent translation unit) and translation targets. It should be noted that the last "|" in the input is the position that needs to be classified at that time, and there are W words after that (W is the window size, 3 is taken here).

Table 1:

enter training target I | want one Classification Label: 1 Translation: I I | want | want one Classification label: -1 Translation: I I | want | a pole Tags: 1Translation: I want I|want|a|excellent Category label: -1 Translation: I want i | want | a | excellent Category Tags: 1 Translation: I want a I |want|a |excellent| wallet Tags: 1Translation: I want a magnificent

Step 202, input the intermediate representation into the classifier, and obtain the classification label output by the classifier.

In this embodiment, the execution subject inputs the intermediate representation into the classifier, and obtains a classification label output by the classifier. The classification label is used to indicate whether the word segment at the end of the word segment sequence of the first text is an independent translation unit.

Wherein, the classifier may be a machine learning network model that performs binary classification on the input in the existing technology or a technology developed in the future, which is not limited in this application. For example, a binary classification network implemented using support vector machines, decision trees, Logistic regression models, and Logistic regression models.

Step 203: In response to the classification label indicating that the word segment at the end of the word segment sequence of the first text is an independent translation unit, the intermediate representation is input into the decoder to obtain the second text output by the decoder.

In this embodiment, if the classification label indicates that the word segment at the end of the word segment sequence of the first text is an independent translation unit, it indicates that an accurate translation can be obtained based on the word segment sequence of the first text. At this time, the intermediate representation can be input into the decoder. The integrity and accuracy of the second text output by the decoder is improved.

In some optional implementations of this embodiment, in response to the classification label indicating that the word segmentation at the end of the word segmentation sequence of the first text is an independent translation unit, inputting the intermediate representation into the decoder includes: in response to the classification label indicating the first text The word segment at the end of the word segment sequence is an independent translation unit, and there is a translation result corresponding to the word segment before the tail in the word segment sequence of the first text in the historical translation result, the intermediate representation is input into the decoder, and the historical translation result is The translation result of the word segment before the tail in the word segment sequence corresponding to the first text is used as a constraint, and the constraint is input into the decoder to obtain the second text output by the decoder.

In this implementation, the historical translation results corresponding to the word segments before the tail in the word segment sequence of the first text can be used to constrain the translation results of the word segment sequence of the first text, which improves the accuracy and timeliness of translation.

In a specific example, the second text output by the decoder is shown in Table 2 below.

Table II:

In the translation method of the above-mentioned embodiments of the present disclosure, a classifier may be used to determine whether the word segment located at the end of the word segment sequence of the first text is an independent translation unit, and when it is determined to be an independent translation unit, the translation of the first text is performed, thereby improving ensure the completeness and accuracy of the translated second text. Compared with the prior art, the time consumption of the system is reduced due to the removal of the sentence segmentation module in the traditional system pipeline process. In some embodiments, when the word segmentation at the tail is determined to be an independent translation unit, the translation of the word segmentation sequence of the first text takes into account the historical translation results of the word segmentation before the tail, so that the disambiguation and disambiguation of entity reference can be effectively solved. Domain translation problems.

An exemplary application scenario of the translation method of the present disclosure is described below with reference to FIG. 3 .

As shown in FIG. 3 , FIG. 3 shows an exemplary application scenario of the translation method according to the present disclosure.

As shown in FIG. 3 , the translation method 300 is run in an electronic device 320, and a translation model is used for translation. The translation model includes an encoder, a classifier and a decoder. The method 300 includes:

The vector matrix 302 determined based on the word segmentation sequence 301 of the first text is input into the encoder 303 to obtain the intermediate representation 304 output by the encoder 303;

Input the intermediate representation 304 into the classifier 305 to obtain the classification label 306 output by the classifier 305;

In response to the classification label 306 indicating that the last word segment 307 in the word segment sequence 301 of the first text is an independent translation unit 308 , the intermediate representation 304 is input to the decoder 309 to obtain a second text 310 output by the decoder 309 .

It should be understood that the application scenario of the translation method shown in FIG. 3 is only an exemplary description of the translation method, and does not represent a limitation to the method. For example, each step shown in FIG. 3 above may further adopt a more detailed implementation method. It is also possible to further add other translation steps on the basis of the above-mentioned FIG. 3 .

Further referring to FIG. 4 , FIG. 4 shows a schematic flowchart of another embodiment of the translation method according to the present disclosure.

As shown in FIG. 4, the translation method 400 of this embodiment may include the following steps:

Step 401, recognize the inputted first voice to obtain the first text.

In this embodiment, the execution subject of the translation method (for example, the terminal or server shown in FIG. 1 ) can use the speech recognition technology in the existing technology or the technology developed in the future to recognize the first speech input in real time, and obtain the second speech a text.

Step 402, perform word segmentation on the first text to obtain a word segmentation sequence of the first text.

In this embodiment, the execution subject may perform word segmentation on the first text according to a preset rule or by using a word segmentation tool in the existing technology or a technology developed in the future, so as to obtain a word segmentation sequence of the first text.

Step 403: Input the vector matrix determined based on the word segmentation sequence of the first text into the encoder to obtain an intermediate representation output by the encoder.

In this embodiment, the executing subject of the translation method (for example, the terminal or the server shown in FIG. 1 ) may use a translation model with a classifier to control the translation unit during translation. The translation model here may be a multi-task neural translation model, specifically, may include: an encoder, a classifier, and a decoder.

Among them, the encoder accepts the word segmentation sequence of the first text as input, and encodes the information in the word segmentation sequence into an intermediate representation. The classifier accepts the intermediate representation as input, and determines whether the word segment at the end of the word segment sequence of the first text is an independent translation unit, and if so, inputs the intermediate representation into the decoder. The decoder decodes the intermediate representation into the target language.

Step 404, input the intermediate representation into the classifier, and obtain the classification label output by the classifier.

In this embodiment, the execution subject inputs the intermediate representation into the classifier, and obtains a classification label output by the classifier. The classification label is used to indicate whether the word segment at the end of the word segment sequence of the first text is an independent translation unit.

Wherein, the classifier may be a machine learning network model that performs binary classification on the input in the existing technology or a technology developed in the future, which is not limited in this application. For example, a binary classification network implemented using support vector machines, decision trees, Logistic regression models, and Logistic regression models.

Step 405: In response to the classification label indicating that the word segment at the end of the word segment sequence of the first text is an independent translation unit, the intermediate representation is input into the decoder to obtain the second text output by the decoder.

In this embodiment, if the classification label indicates that the word segment at the end of the word segment sequence of the first text is an independent translation unit, it indicates that an accurate translation can be obtained based on the word segment sequence of the first text. At this time, the above-mentioned executive body can input the intermediate representation a decoder, thereby improving the integrity and accuracy of the second text output by the decoder.

Step 406: Generate a second voice based on the translated second text.

In this embodiment, the above-mentioned executor can convert the translated second text into speech by using a text-to-speech tool or an AI speech intelligent synthesis tool based on the translated second text.

Step 407, playing the second voice.

In this embodiment, the execution subject may play the second speech generated in step 406, so as to complete the simultaneous interpretation.

The translation method in the embodiment in FIG. 4 of the present disclosure, on the basis of the translation method shown in FIG. 2, refines the process of obtaining the played second voice based on the input first voice, and plays the second voice Voice, which improves the accuracy and efficiency of simultaneous interpretation.

Further referring to FIG. 5 , as an implementation of the methods shown in the above figures, an embodiment of the present disclosure provides an embodiment of a translation device, which is similar to the method embodiments shown in FIGS. 2-4 Correspondingly, the apparatus can be specifically applied to the above-mentioned terminal device or server.

As shown in FIG. 5 , the translation device 500 of this embodiment may include: using a translation model, the translation model includes an encoder, a classifier, and a decoder, and the device includes: an encoder input unit 510 configured to convert the first text based on The vector matrix determined by the word segmentation sequence is input into the encoder to obtain the intermediate representation output by the encoder; the classifier input unit 520 is configured to input the intermediate representation into the classifier to obtain the classification label output by the classifier; the decoder input unit 530 is obtained by It is configured to input the intermediate representation into the decoder to obtain the second text output by the decoder in response to the classification label indicating that the word segment located at the end of the word segment sequence of the first text is an independent translation unit.

In some embodiments, the decoder input unit is further configured to: respond to the classification label indicating that the word segment located at the end of the word segment sequence of the first text is an independent translation unit, and there is a word segment sequence corresponding to the first text in the historical translation results The translation result of the participle before the end in the middle representation is input into the decoder, and the translation result of the participle before the end in the participle sequence corresponding to the first text in the historical translation result is used as a constraint, and the constraint is input into the decoder to obtain the decoding The second text output by the compiler.

In some embodiments, the training data samples of the translation model used in the device are determined based on the following units (not shown in the figure): the training data alignment unit is configured to align the training data of the translation model with a word alignment tool to obtain the training Alignment information of the data; the training data determining unit is configured to use the alignment information of the training data as a training data sample of the translation model.

In some embodiments, the device further includes (not shown in the figure): a first speech recognition unit configured to recognize the inputted first speech to obtain the first text; a first text word segmentation unit configured to Word segmentation is performed on the text to obtain a word segmentation sequence of the first text.

In some embodiments, the device further includes (not shown in the figure): a second speech generation unit configured to generate a second speech based on the translated second text; a second speech playback unit configured to play the second speech Second voice.

It should be understood that each unit recorded in the apparatus 500 corresponds to each step recorded in the method described with reference to FIGS. 2-4 . Therefore, the operations and features described above for the method are also applicable to the device 500 and each unit contained therein, and will not be repeated here.

Referring now to FIG. 6 , it shows a schematic structural diagram of an electronic device (such as the server or terminal device in FIG. 1 ) 600 suitable for implementing the embodiments of the present disclosure. Terminal devices in the embodiments of the present disclosure may include, but are not limited to, devices such as notebook computers and desktop computers. The terminal device/server shown in FIG. 6 is only an example, and should not limit the functions and scope of use of the embodiments of the present disclosure.

As shown in FIG. 6, an electronic device 600 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 601, which may be randomly accessed according to a program stored in a read-only memory (ROM) 602 or loaded from a storage device 608. Various appropriate actions and processes are executed by programs in the memory (RAM) 603 . In the RAM 603, various programs and data necessary for the operation of the electronic device 600 are also stored. The processing device 601, ROM 602, and RAM 603 are connected to each other through a bus 604. An input/output (I/O) interface 605 is also connected to the bus 604 .

Typically, the following devices can be connected to the I/O interface 605: input devices 606 including, for example, a touch screen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, a liquid crystal display (LCD), speaker, vibration an output device 607 such as a computer; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication means 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. While FIG. 6 shows electronic device 600 having various means, it should be understood that implementing or having all of the means shown is not a requirement. More or fewer means may alternatively be implemented or provided. Each block shown in FIG. 6 may represent one device, or may represent multiple devices as required.

In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, where the computer program includes program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via communication means 609, or from storage means 608, or from ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are executed. It should be noted that the computer-readable medium described in the embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the above two. A computer readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In the embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the embodiments of the present disclosure, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, and the computer-readable signal medium may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device . Program code embodied on a computer readable medium may be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

The above-mentioned computer-readable medium may be included in the above-mentioned electronic device, or may exist independently without being incorporated into the electronic device. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed by the electronic device, the electronic device: inputs the vector matrix determined based on the word segmentation sequence of the first text into the encoder to obtain the encoded The intermediate representation of the device output; the intermediate representation is input into the classifier, and the classification label output by the classifier is obtained; in response to the classification label indicating that the word segmentation at the tail in the word segmentation sequence of the first text is an independent translation unit, the intermediate representation is input into the decoder, Get the second text output by the decoder.

Computer program code for carrying out operations of embodiments of the present disclosure may be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, Also included are conventional procedural programming languages - such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (for example, using an Internet service provider to connected via the Internet).

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or portion of code that contains one or more logical functions for implementing specified executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by a dedicated hardware-based system that performs the specified functions or operations , or may be implemented by a combination of dedicated hardware and computer instructions.

The units involved in the embodiments described in the present disclosure may be implemented by software or by hardware. The described units may also be set in a processor, for example, it may be described as: a processor includes an encoder input unit, a classifier input unit, and a decoder input unit. Among them, the names of these units do not constitute a limitation of the unit itself in some cases. For example, the encoder input unit can also be described as "input the vector matrix determined based on the word segmentation sequence of the first text into the encoder, and obtain The unit of the intermediate representation of the encoder output".

The above description is only a preferred embodiment of the present disclosure and an illustration of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in this disclosure is not limited to the technical solution formed by the specific combination of the above-mentioned technical features, but should also cover the technical solutions formed by the above-mentioned technical features or without departing from the above-mentioned inventive concept. Other technical solutions formed by any combination of equivalent features. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

Claims (10)

1. A method of translation, using a translation model, said translation model comprising encoder, classifier and decoder, said method comprising: Inputting the vector matrix determined based on the word segmentation sequence of the first text into the encoder to obtain an intermediate representation output by the encoder; Inputting the intermediate representation into the classifier to obtain a classification label output by the classifier; In response to the classification label indicating that the word segmentation at the tail in the word segmentation sequence of the first text is an independent translation unit, and there is a translation result corresponding to the word segmentation before the tail in the word segmentation sequence of the first text in the historical translation results, the The intermediate representation is input into the decoder, and the translation result of the word segmentation before the tail in the word segmentation sequence corresponding to the first text in the historical translation result is used as a constraint, and the constraint is input into the decoder to obtain the output of the decoder second text. 2. The method according to claim 1, wherein the training data samples of the translation model are determined based on the following steps: Use the word alignment tool to align the training data of the translation model to obtain the alignment information of the training data; The alignment information of the training data is used as a training data sample of the translation model. 3. The method according to claim 1, wherein the word segmentation sequence of the first text is obtained through the following steps: Recognizing the input first voice to obtain the first text; Word segmentation is performed on the first text to obtain a word segmentation sequence of the first text. 4. The method according to any one of claims 1-3, wherein the method further comprises: generating a second voice based on the translated second text; Play the second voice. 5. A device for translation, using a translation model, the translation model includes an encoder, a classifier and a decoder, and the device includes: An encoder input unit configured to input the vector matrix determined based on the word segmentation sequence of the first text into the encoder to obtain an intermediate representation output by the encoder; a classifier input unit configured to input the intermediate representation into the classifier to obtain a classification label output by the classifier; The decoder input unit is configured to indicate in response to the classification label that the word segment located at the tail in the word segment sequence of the first text is an independent translation unit, and there is a corresponding word segment located before the tail in the word segment sequence of the first text in the historical translation result The translation result of the word segmentation, the intermediate representation is input into the decoder, and the translation result of the word segmentation before the tail in the word segmentation sequence corresponding to the first text in the historical translation result is used as a constraint, and the constraint is input into the decoder , to obtain the second text output by the decoder. 6. The device according to claim 5, wherein the training data samples of the translation model used in the device are determined based on the following units: The training data alignment unit is configured to use a word alignment tool to align the training data of the translation model to obtain alignment information of the training data; The training data determination unit is configured to use the alignment information of the training data as a training data sample of the translation model. 7. The apparatus of claim 5, wherein the apparatus further comprises: The first speech recognition unit is configured to recognize the input first speech to obtain the first text; The first text word segmentation unit is configured to perform word segmentation on the first text to obtain a word segmentation sequence of the first text. 8. The device according to any one of claims 5-7, wherein the device further comprises: a second speech generation unit configured to generate a second speech based on the translated second text; The second voice playing unit is configured to play the second voice. 9. An electronic device comprising: one or more processors; storage means for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method according to any one of claims 1-4. 10. A computer-readable medium, on which a computer program is stored, and when the program is executed by a processor, the method according to any one of claims 1-4 is realized.

Images