WO2022116841A1 - Text translation method, apparatus and device, and storage medium - Google Patents

Abstract

A text translation method. The text translation method comprises: S110, using a set segmentation label to segment at least two sentences of text to be translated; S120, inputting said text, after segmentation, into an encoder of a machine translation model to obtain a word sequence and a mark sequence of said text; and S130, inputting the word sequence and the mark sequence into a decoder of the machine translation model to obtain a target translation result.

Description

Text translation method, device, device and storage medium

This application claims the priority of the Chinese Patent Application No. 202011408602.2 filed with the China Patent Office on December 04, 2020, the entire contents of which are incorporated herein by reference.

technical field

The present disclosure relates to the technical field of machine translation, for example, to a text translation method, apparatus, device, and storage medium.

Background technique

Machine Translation ((Machine Translation, MT) is an important field in the field of Natural Language Processing (NLP), which aims to use machines to translate one language into another language. After years of development, MT has evolved from a Rule-based methods to statistical-based methods, to neural network-based neural machine translation (NMT). Generally speaking, like many other mainstream NLP tasks, NMT also adopts a sequence-to-sequence structure (Sequence to sequence, seq2seq), which consists of an encoder (encoder) and a decoder (decoder). The encoder encodes the sentence at the source into a vector representation, and then the decoder generates the corresponding translation word by word according to the vector representation.

SUMMARY OF THE INVENTION

The present disclosure provides a text translation method, device, device and storage medium, so as to realize the translation of dialogue text and improve the accuracy of dialogue text translation.

The present disclosure provides a text translation method, including:

At least two sentences of the text to be translated are segmented by setting segmentation labels;

Inputting the segmented text to be translated into the encoder of the machine translation model to obtain the word sequence and tag sequence of the text to be translated;

The word sequence and the token sequence are input into the decoder of the machine translation model to obtain a target translation result.

The present disclosure also provides a text translation device, comprising:

A sentence segmentation module, which is set to use a set segmentation label to segment at least two sentences of the text to be translated;

A sequence acquisition module, configured to input the segmented text to be translated into an encoder of a machine translation model, to obtain a word sequence and a marker sequence of the text to be translated;

The target translation result obtaining module is configured to input the word sequence and the label sequence into the decoder of the machine translation model to obtain the target translation result.

The present disclosure also provides an electronic device, the electronic device comprising:

one or more processing devices;

storage means configured to store one or more instructions;

The one or more instructions, when executed by the one or more processing devices, cause the one or more processing devices to implement the above-described text translation method.

The present disclosure also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by the processing device, implements the above-mentioned text translation method.

Description of drawings

1 is a flowchart of a text translation method provided by an embodiment of the present disclosure;

FIG. 2a is an example of a training machine translation model provided by an embodiment of the present disclosure;

Figure 2b is an effect diagram of a text translation provided by an embodiment of the present disclosure;

3 is a schematic structural diagram of a text translation apparatus provided by an embodiment of the present disclosure;

FIG. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure.

Detailed ways

Embodiments of the present disclosure will be described below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, the present disclosure may, however, be embodied in various forms and should not be construed as limited to the embodiments set forth herein, which are provided for a more thorough and complete understanding this disclosure. The figures and examples of the present disclosure are for illustrative purposes only.

The multiple steps described in the method embodiments of the present disclosure may be performed in different orders and/or in parallel. Furthermore, method embodiments may include additional steps and/or omit performing the illustrated steps. The scope of the present disclosure is not limited in this regard.

As used herein, the term "including" and variations thereof are open-ended inclusions, ie, "including but not limited to". The term "based on" is "based at least in part on." The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Relevant definitions of other terms will be given in the description below.

Concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of functions performed by these devices, modules or units relation.

Modifications of "a" and "a plurality" mentioned in the present disclosure are illustrative rather than limiting, and those skilled in the art should understand that unless the context indicates otherwise, they should be construed as "one or more".

The names of messages or information exchanged between multiple devices in the embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of these messages or information.

Translation systems used in dialogue are often at the single-sentence level. Problems such as personal pronoun omission, punctuation omission, and typos often occur in dialogue. It is difficult for a single-sentence-level translation system to solve these problems, resulting in low accuracy of translation results.

Table 1 is an example table of the translation of dialogue fragments by a translation model.

Table 1

Chinese (ZH)	What happened to Nancy? Did [she] drop cry.
MT	What happened to Nancy? Did you cry?
Reference (Reference, REF)	What happened to Nancy? Did she cry?
EN	What happened to Nancy[? ] drop is not crying ah.
MT	Did Nancy cry?
REF	What happened to Nancy? Did she cry?
EN	How does Nancy [music] typo ?
MT	How happy is Nancy?
REF	What happened to Nancy?

It can be seen that the first example asks "what's wrong with Nancy?", and then the second sentence asks again "are you crying?", omitting "she" at this time. Then the translation system will make up the translation as "you/you", and get the wrong translation. In order to be concise and compact in dialogue, this omission often occurs, especially in Chinese, Japanese, Korean, Vietnamese, etc.

The second example is punctuation omission, which is common in everyday chat scenarios, where spaces are used to represent intervals. However, this can have a huge impact on the translation system. The omitted "?" in the example results in a loss of semantics in the translation result.

The third example is a typo, "le" was mistakenly typed into "happy", resulting in "happy" appearing in the translation result, with completely wrong semantics.

In order to solve the above problem, FIG. 1 is a flowchart of a text translation method provided by an embodiment of the present disclosure. This embodiment can be applied to the case of translating dialogue text, and the method can be executed by a text translation device, which can be It is composed of hardware and/or software, and can generally be integrated in a device with text translation function, which can be an electronic device such as a server or a server cluster. As shown in Figure 1, the method includes the following steps:

Step 110, at least two sentences of the text to be translated are segmented by using a set segmentation label.

The text to be translated may be a text formed by a dialogue between two or more people, and contains at least two sentences, that is, at least two sentences. The preset segmentation tag can be a preset tag used to segment sentences in the text, for example: <sep>. In this embodiment, the text to be translated may be offline text or online text. Among them, offline text can be understood as non-real-time text that has been generated, such as subtitles in film and television dramas; online text can be understood as dialogue text generated in real time.

After obtaining the text to be translated, add a set segmentation tag <sep> between two adjacent sentences to separate the sentences in the text to be translated.

In this embodiment, when the text to be translated is online text, the sentence generated in real time is translated in real time during the actual translation process, and when the dialogue ends, the translation of the dialogue is completed. In order to improve the accuracy of translation, historical dialogue information can be referred to when translating the currently generated dialogue.

When the text to be translated is online text, the method of dividing at least two sentences of the text to be translated by setting segmentation labels may be: obtaining the current sentence and a set number of forward sentences to form the text to be translated; A set number of forward sentences are split using a set split label.

The set number can be set by the developer, for example: set to 5. The forward sentence can be understood as the above information of the current sentence. In this embodiment, the text to be translated is composed of a current sentence and a set number of forward sentences. The advantage of this is that when translating the current sentence, you can refer to the above information, thereby improving the accuracy of the translation.

Step 120: Input the segmented text to be translated into the encoder of the machine translation model to obtain a word sequence and a token sequence of the text to be translated.

The role of the encoder is to compile the text to be translated into a vector, a sequence. The word sequence can be understood as a sequence consisting of the values corresponding to the words contained in the text to be translated; the tag sequence can be understood as a sequence formed by adding tags to the words contained in the text to be translated, and the tag sequence is used to indicate that the words in the text to be translated are omitted. Pronouns or omitted punctuation or typos or normal words. The role of the token sequence is to assist the decoder in translating the word sequence.

Step 130: Input the word sequence and the token sequence into the decoder of the machine translation model to obtain the target translation result.

The decoder is used to parse or translate the vector generated by the encoder to obtain the target translation result.

In this embodiment, when the text to be translated is online text, the following two methods can be used for translation, one is to translate the current sentence with reference to the above information, and the other is to translate the current sentence with reference to the historical translation result.

The process of using the first method of translation can be as follows: after the text to be translated composed of the current sentence and a set number of forward sentences is divided by the set segmentation label, input into the machine translation model to obtain the target translation result, and then translate from the target The translation result corresponding to the current sentence is cut out from the result. That is, the above information is re-translated to prevent the wrong results of historical translation from being propagated downward.

The process of using the second method of translation may be: after dividing the text to be translated consisting of the current sentence and a set number of forward sentences with a set segmentation label, obtain the historical translation results corresponding to the set number of forward sentences, and The historical translation results and the text to be translated after segmentation are input into the machine translation model to obtain the target translation result, and then the translation result corresponding to the current sentence is cut out from the target translation result. In this embodiment, since the historical translation result corresponding to the forward sentence is input, the machine translation model does not need to translate the forward sentence during the translation process, but only translates the current sentence with reference to the historical translation result, thereby saving computation. quantity, and maintains the coherence of translation.

In this embodiment, the machine translation model consists of an encoder-decoder (encoder-decoder). The training process of the machine translation model can be as follows: obtaining the original text and the original translation result of the original text; segmenting at least two sentences of the original text using a set segmentation label; preprocessing the segmented original text according to the set rules, Obtain the training text; add tags to the training text according to the set rules to obtain the original tag sequence; train the machine translation model based on the training text, the original translation result and the original tag sequence.

When training the machine translation model, due to the scarcity of dialogue data, we introduce a large number of text and sentence-level data to simulate the use of context. In order to increase the expected diversity, the original text needs to be preprocessed with set rules. The setting rules may include at least one of the following: discarding pronouns, discarding punctuation marks, and replacing words with typos.

The training text is tagged according to the set rules, and the original tag sequence can be obtained in the following way: if the original text is preprocessed to discard the pronoun, the tag added to the word after the pronoun in the training text is the first set value ; If the original text is preprocessed to discard punctuation, the mark added to the punctuation position in the training text is the second set value; if the original text is preprocessed to replace words with typos, the The mark added to the misspelled word in the training text is the third set value; the mark added to the words that have not been preprocessed in the training text is the fourth set value.

The first setting value may be 2, the second setting value may be 3, the third setting value may be 1, and the fourth setting value may be 0. Exemplarily, Table 2 is an example table for generating training samples provided by an embodiment of the present disclosure.

Table 2

As shown in Table 2, x ⁽¹⁾ and x ⁽²⁾ are two consecutive Chinese sentences, and y ⁽¹⁾ and y ⁽²⁾ are their corresponding translations, respectively. In order to use the context information, use the set segmentation tag <sep> to connect two sentences (in the dialogue scene, it is to connect multiple sentences above and below), and get x _d , which corresponds to y composed of y ⁽¹⁾ and y ⁽²⁾ _d . Then randomly discard some pronouns and punctuation from this sentence according to the set rules, replace some words as typos, and get a new sentence x' _d , for each position of the new sentence, mark it according to the set rules, for discarded The last word of the subject of , is marked as 2, punctuation is marked as 3, typos are marked as 1, and unprocessed words are marked as 0, so that the sequence of tokens l' _x of the same length as the sequence of _x'd is obtained.

In this embodiment, the process of training the machine translation model based on the training text, the original translation result and the original label sequence may be as follows: input the training text into the decoder of the machine translation model to obtain the training label sequence and the training word sequence; The sequence and the training tag sequence are input into the decoder of the machine translation model to obtain the training translation result; the first loss function is calculated according to the training tag sequence and the original tag sequence; the second loss function is calculated according to the training translation result and the original translation result; based on the first loss function and the second loss function train the encoder, and train the decoder according to the second loss function.

The encoder has the function of compiling text into sequences of words and tokens. The decoder has the function of decoding the word sequence and the token sequence and outputting the translation result. Exemplarily, FIG. 2a is an example of training a machine translation model provided by an embodiment of the present disclosure. As shown in Figure 2a, x' _d is the training text. After inputting x' _d into the encoder, the training label sequence L _SL is obtained. The output of the decoder is the training translation result L _MT , where l' _x is the original label sequence, and y _d is Original translation result. The first loss function is obtained according to L _SL and l' _x , and the second loss function is obtained according to L _MT and y _d . The parameters in the encoder are trained according to the first loss function and the second loss function, and the parameters in the decoder are trained according to the second loss function, until the machine translation model reaches the translation accuracy.

Exemplarily, in order to verify the translation effect of the translation method in the embodiment of the present disclosure, the translation model in this embodiment is compared with the translation results of other translation models. Table 3 is a comparison table of translation results. Among them, BASE is the original translation model, DIALREPAIR and DIALROBUST are translation models improved from the original translation model, and DIALMTL is the translation model of this embodiment.

table 3

As can be seen from Table 3, the overall translation effect of the translation model in the embodiment of the present disclosure, the translation accuracy rate of lost main sentences, the translation accuracy rate of lost punctuation sentences, and the translation accuracy rate of sentences containing typos all reach the best relative to other translation models. .

At the same time, the embodiment of the present disclosure also verifies the overall translation effect (BLEU) and the translation accuracy (Accuracy) of the lost subject sentence when the method in this embodiment is used to translate offline text and online text. As shown in Figure 2b, context_length refers to the maximum number of sentences the model can use for the online text (online) each time; in the case of oneline-cut, for each new sentence obtained, it is combined with the previous dialogue label. , translate it, and then only intercept the last sentence; online-fd refers to the use of historical translation information for translation; offline refers to the translation scene of offline text. As can be seen from Figure 2b, because offline uses the most context, it can achieve the best effect on the task of subject completion, which requires a lot of context, and the overall BLEU quality is also high. In the online mode, online-fd uses historical translation information to continue translation, maintaining the consistency of translation before and after, and the BLEU result is the best, but due to possible error propagation, the accuracy of the subject is slightly lower than the online-cut method.

In the technical solution of the embodiment of the present disclosure, at least two sentences of the text to be translated are firstly segmented using a set segmentation label, and then the segmented text to be translated is input into the encoder of the machine translation model to obtain word sequences and tags of the text to be translated sequence, and finally input the word sequence and tag sequence into the decoder of the machine translation model to obtain the target translation result. In the text translation method provided by the embodiment of the present disclosure, the encoder of the machine translation model compiles the text to be translated into a word sequence and a marker sequence, and the decoder simultaneously decodes the word sequence and the marker sequence to obtain the final translation result, which can realize the dialogue text to improve the accuracy of translation of dialogue texts.

FIG. 3 is a schematic structural diagram of a text translation apparatus provided by an embodiment of the present disclosure. As shown in Figure 3, the device includes:

The sentence segmentation module 210 is configured to divide at least two sentences of the text to be translated by using a set segmentation label; the sequence acquisition module 220 is configured to input the segmented text to be translated into the encoder of the machine translation model, and obtain the text to be translated. The word sequence and the marker sequence; the target translation result acquisition module 230 is configured to input the word sequence and the marker sequence into the decoder of the machine translation model to obtain the target translation result.

Optionally, the text to be translated includes online text; the sentence segmentation module 210 is set to:

Obtain the current sentence and the set number of forward sentences to form the text to be translated; use the set segmentation label to segment the current sentence and the set number of forward sentences.

Optionally, after obtaining the target translation result, it also includes: a translation result interception module, which is set to:

Cut out the translation result corresponding to the current sentence from the target translation result.

Optionally, it also includes: a module for obtaining historical translation results, set to:

Get the historical translation results corresponding to a set number of forward sentences.

Optionally, the sequence obtaining module 220 is further set to:

Input the historical translation results and the segmented text to be translated into the encoder of the machine translation model to obtain the word sequence and tag sequence of the text to be translated.

Optionally, it also includes: a machine translation model training module, set to:

Obtain the original text and the original translation result of the original text; use the set segmentation labels to segment at least two sentences of the original text; preprocess the segmented original text according to the set rules to obtain the training text; Add tags to the training text to obtain the original tag sequence; train the machine translation model based on the training text, the original translation result and the original tag sequence.

Optionally, the machine translation model training module is also set to:

Input the training text into the decoder of the machine translation model to obtain the training tag sequence and the training word sequence; input the training word sequence and the training tag sequence into the decoder of the machine translation model to obtain the training translation result; calculate according to the training tag sequence and the original tag sequence The first loss function; the second loss function is calculated according to the training translation result and the original translation result; the encoder is trained according to the first loss function and the second loss function, and the decoder is trained according to the second loss function.

Optionally, the setting rule includes at least one of the following: discarding pronouns, discarding punctuation marks, and replacing words with typos.

Optionally, the machine translation model training module is also set to:

If the original text is preprocessed to discard pronouns, the mark added to the words after the pronoun in the training text is the first set value; if the original text is preprocessed to discard punctuation, the training text will be marked The mark added at the point position is the second set value; if the original text is preprocessed to replace words with typos, the mark added to the typo in the training text is the third set value; The preprocessed word-added token is the fourth set value.

The foregoing apparatus can execute the methods provided by all the foregoing embodiments of the present disclosure, and has functional modules and effects corresponding to executing the foregoing methods. For technical details not described in detail in this embodiment, reference may be made to the methods provided by all the foregoing embodiments of the present disclosure.

Referring next to FIG. 4 , it shows a schematic structural diagram of an electronic device 300 suitable for implementing an embodiment of the present disclosure. The electronic devices in the embodiments of the present disclosure may include, but are not limited to, such as mobile phones, notebook computers, digital broadcast receivers, personal digital assistants (Personal Digital Assistants, PDAs), tablet computers (PADs), portable multimedia players (Portable Media Players) , PMP), mobile terminals such as in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital (Television, TV), desktop computers, etc., or various forms of servers, such as independent servers or server clusters. The electronic device shown in FIG. 4 is only an example, and should not impose any limitation on the function and scope of use of the embodiments of the present disclosure.

As shown in FIG. 4 , the electronic device 300 may include a processing device (eg, a central processing unit, a graphics processor, etc.) 301, which may be stored in accordance with a program stored in a read-only storage device (Read-Only Memory, ROM) 302 or from a storage Device 308 loads a program into Random Access Memory (RAM) 303 to perform various appropriate actions and processes. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. An Input/Output (I/O) interface 305 is also connected to the bus 304 .

Typically, the following devices can be connected to the I/O interface 305: input devices 306 including, for example, a touch screen, touch pad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, a Liquid Crystal Display (LCD) Output device 307 , speaker, vibrator, etc.; storage device 308 including, eg, magnetic tape, hard disk, etc.; and communication device 309 . Communication means 309 may allow electronic device 300 to communicate wirelessly or by wire with other devices to exchange data. Although FIG. 4 shows the electronic device 300 having various means, it is not required to implement or have all of the illustrated means. More or fewer devices may alternatively be implemented or provided.

According to embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing a recommended method of a word. In such an embodiment, the computer program may be downloaded and installed from the network via the communication device 309, or from the storage device 308, or from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are executed.

The computer-readable medium described above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or a combination of any of the above. Examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer disks, hard disks, RAM, ROM, Erasable Programmable Read-Only Memory (EPROM) or flash memory), optical fiber, portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In this 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 present disclosure, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave with computer-readable program code embodied thereon. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device . The program code embodied on the computer-readable medium may be transmitted by any suitable medium, including but not limited to: electric wire, optical fiber cable, radio frequency (RF), etc., or any suitable combination of the above.

In some embodiments, clients and servers can communicate using any currently known or future developed network protocol, such as HyperText Transfer Protocol (HTTP), and can communicate with digital data in any form or medium. Communication (eg, a communication network) interconnects. Examples of communication networks include Local Area Networks (LANs), Wide Area Networks (WANs), the Internet (eg, the Internet), and peer-to-peer networks (eg, ad hoc peer-to-peer networks), as well as any currently Known or future developed networks.

The above-mentioned computer-readable medium may be included in the above-mentioned electronic device; or may exist alone without being assembled 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 equipment, the electronic equipment: at least two sentences of the text to be translated are divided by setting segmentation tags; The latter text to be translated is input into the encoder of the machine translation model to obtain the word sequence and tag sequence of the text to be translated; the word sequence and the tag sequence are input into the decoder of the machine translation model to obtain the target translation result.

Computer program code for performing operations of the present disclosure may be written in one or more programming languages, including but not limited to object-oriented programming languages—such as Java, Smalltalk, C++, and This includes 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. Where a remote computer is involved, the remote computer may be connected to the user's computer through any kind of network, including a LAN or WAN, or may be connected to an external computer (eg, using an Internet service provider to connect through 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 the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more logical functions for implementing the specified functions executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the blocks 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 the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also 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 in dedicated hardware-based systems that perform the specified functions or operations , or can be implemented in a combination of dedicated hardware and computer instructions.

The units involved in the embodiments of the present disclosure may be implemented in a software manner, and may also be implemented in a hardware manner. Among them, the name of the unit does not constitute a limitation of the unit itself in one case.

The functions described herein above may be performed, at least in part, by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (Application Specific Standard Products) Standard Parts, ASSP), system on chip (System on Chip, SOC), complex programmable logic device (Complex Programmable Logic Device, CPLD) and so on.

In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in connection with the instruction execution system, apparatus or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. Examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, RAM, ROM, EPROM or flash memory, optical fibers, CD-ROMs, optical storage devices, magnetic storage devices, or Any suitable combination of the above.

According to one or more of the embodiments of the present disclosure, the embodiments of the present disclosure disclose a text translation method, including:

At least two sentences of the text to be translated are segmented by setting segmentation labels; the segmented text to be translated is input into the encoder of the machine translation model to obtain the word sequence and tag sequence of the text to be translated; The sequence and the marker sequence are input into the decoder of the machine translation model to obtain the target translation result.

The text to be translated includes online text; the at least two sentences of the text to be translated are segmented by setting segmentation labels, including:

Obtain the current sentence and a set number of forward sentences to form the text to be translated; use the set segmentation label to segment the current sentence and the set number of forward sentences.

After the obtaining the target translation result, the method further includes:

The translation result corresponding to the current sentence is cut out from the target translation result.

After the current sentence and the set number of forward sentences are segmented using the set segmentation label, the method further includes:

Obtain the historical translation results corresponding to the set number of forward sentences;

Inputting the segmented text to be translated into the encoder of the machine translation model to obtain the word sequence and tag sequence of the text to be translated, including:

Input the historical translation result and the segmented text to be translated into the encoder of the machine translation model to obtain a word sequence and a token sequence of the text to be translated.

The training process of the machine translation model is as follows:

Obtain the original text and the original translation result of the original text; use the set segmentation label to segment at least two sentences of the original text; perform preprocessing on the segmented original text according to the set rules to obtain training text; adding tags to the training text according to the set rule to obtain an original tag sequence; training the machine translation model based on the training text, the original translation result and the original tag sequence.

The machine translation model is trained based on the training text, the original translation result and the original tag sequence, including:

Input the training text into the decoder of the machine translation model to obtain a training label sequence and a training word sequence; input the training word sequence and the training label sequence into the decoder of the machine translation model to obtain a training translation result ; Calculate the first loss function according to the training label sequence and the original label sequence; Calculate the second loss function according to the training translation result and the original translation result; Calculate the second loss function according to the first loss function and the second loss function trains the encoder and trains the decoder according to the second loss function.

The setting rules include at least one of the following: discarding pronouns, discarding punctuation marks, and replacing words with typos; adding marks to the training text according to the setting rules to obtain an original mark sequence, including:

If the preprocessing of discarding pronouns is performed on the original text, the mark added to the words after the pronouns in the training text is the first set value; if the original text is processed by discarding punctuation preprocessing, the mark added to the punctuation position in the training text is the second set value; if the original text is preprocessed by replacing words with typos, then the The mark added to the misspelled word is the third set value; the mark added to the words that have not been preprocessed in the training text is the fourth set value.

Claims (10)

1. A text translation method comprising:

   At least two sentences of the text to be translated are segmented by setting segmentation labels;

   Inputting the segmented text to be translated into the encoder of the machine translation model to obtain the word sequence and tag sequence of the text to be translated;

   The word sequence and the token sequence are input into the decoder of the machine translation model to obtain a target translation result.
2. The method according to claim 1, wherein the to-be-translated text comprises online text; the at least two sentences of the to-be-translated text are segmented by setting segmentation labels, comprising:

   Obtain the current sentence and a set number of forward sentences to form the text to be translated;

   The current sentence and the set number of forward sentences are segmented using the set segmentation label.
3. The method according to claim 2, after the obtaining the target translation result, further comprising:

   The translation result corresponding to the current sentence is cut out from the target translation result.
4. The method according to claim 2, after the current sentence and the set number of forward sentences are segmented using the set segmentation label, further comprising:

   Obtain the historical translation results corresponding to the set number of forward sentences;

   Inputting the segmented text to be translated into the encoder of the machine translation model to obtain the word sequence and tag sequence of the text to be translated, including:

   Input the historical translation result and the segmented text to be translated into the encoder of the machine translation model to obtain a word sequence and a token sequence of the text to be translated.
5. The method according to claim 1, wherein the training process of the machine translation model is:

   obtaining the original text and the original translation result of the original text;

   Segmenting at least two sentences of the original text using the set segmentation label;

   Preprocessing the segmented original text according to the set rules to obtain training text;

   Add a mark to the training text according to the set rule to obtain an original mark sequence;

   The machine translation model is trained based on the training text, the original translation result and the original token sequence.
6. The method according to claim 5, wherein training the machine translation model based on the training text, the original translation result and the original tag sequence comprises:

   Inputting the training text into the decoder of the machine translation model to obtain a training label sequence and a training word sequence;

   Inputting the training word sequence and the training label sequence into the decoder of the machine translation model to obtain a training translation result;

   Calculate a first loss function according to the training label sequence and the original label sequence;

   Calculate a second loss function according to the training translation result and the original translation result;

   The encoder is trained according to the first loss function and the second loss function, and the decoder is trained according to the second loss function.
7. The method according to claim 5 or 6, wherein the setting rules include at least one of the following: discarding pronouns, discarding punctuation marks, and replacing words with typos;

   The adding a mark to the training text according to the set rule to obtain the original mark sequence, including:

   In the case that the preprocessing of discarding pronouns is performed on the original text, the mark added to the words after the pronouns in the training text is a first set value;

   In the case that the preprocessing of discarding punctuation is performed on the original text, the mark added to the punctuation position in the training text is a second set value;

   In the case where the original text is preprocessed by replacing words with typos, the mark added to the typos in the training text is a third set value;

   The tags added to the words that have not been preprocessed in the training text are the fourth set value.
8. A text translation device, comprising:

   A sentence segmentation module, which is set to use a set segmentation label to segment at least two sentences of the text to be translated;

   A sequence acquisition module, configured to input the segmented text to be translated into an encoder of a machine translation model to obtain a word sequence and a tag sequence of the text to be translated;

   The target translation result obtaining module is configured to input the word sequence and the label sequence into the decoder of the machine translation model to obtain the target translation result.
9. An electronic device comprising:

   at least one processing device;

   a storage device configured to store at least one instruction;

   The at least one instruction, when executed by the at least one processing device, causes the at least one processing device to implement the text translation method of any one of claims 1-7.
10. A computer-readable storage medium storing a computer program, wherein when the program is executed by a processing device, the text translation method according to any one of claims 1-7 is implemented.

Images