CN109635269B - Method and device for post-translation editing of machine-translated text - Google Patents

Abstract

The invention discloses a method for post-translation editing of machine-translated texts, comprising: obtaining source texts and machine-translated texts; Process to obtain the first vector representing the source text; extract the second text feature of the machine translation text through the self-attention mechanism, optimize the second text feature by using the attention mechanism on the first vector; use the feed-forward neural network to optimize the The second text feature is processed to obtain a second vector representing the machine-translated text; and a post-translation edited text of the machine-translated text is generated word by word from left to right according to the first vector and the second vector. The method can improve the processing efficiency and accuracy of the post-translation editing, so that the accuracy of the processed post-editing text is better. A post-translation editing device, equipment, and readable storage medium for machine-translated texts disclosed in the present invention also have the above-mentioned technical effects.

Description

Method and device for post-translation editing of machine-translated text

technical field

The present invention relates to the technical field of automatic text translation, and more specifically, to a post-translation editing method, device, equipment and readable storage medium for machine-translated text.

Background technique

Machine translation, also known as automatic translation, is the process of using a computer to convert a natural source language into another natural target language, generally referring to the translation of sentences and full texts between natural languages. Correspondingly, a machine-translated text refers to a text in another language obtained by translating a text in one language using a computer. Post-editing refers to the process of improving the translated text generated by the machine, so that the machine-translated text is more in line with the human language style.

In the prior art, the automatic processing of post-translation editing is generally realized based on a recurrent neural network. It should be noted that the features of the language text extracted by the cyclic neural network are not fine enough, and the logarithmic linear combination is used to process the source text and the machine translation text, and the features between the source text and the machine translation text cannot be correlated, resulting in the Insufficient representation capabilities of text and machine-translated texts reduce the accuracy of post-editing and reduce the accuracy of post-edited texts obtained by post-editing. Wherein, the post-edited text is the text obtained after post-editing the machine-translated text.

Therefore, how to improve the accuracy of post-translation editing is a problem to be solved by those skilled in the art.

Contents of the invention

The purpose of the present invention is to provide a post-translation editing method, device, equipment and readable storage medium for machine-translated texts, so as to improve the accuracy of post-translation editing.

In order to achieve the above object, the embodiment of the present invention provides the following technical solutions:

A method for post-translation editing of a machine-translated text, comprising:

obtaining source texts and machine translated texts of said source texts;

extracting the first text features of the source text through a self-attention mechanism, and processing the first text features using a feed-forward neural network to obtain a first vector representing the source text;

Extract the second text feature of the machine translation text through the self-attention mechanism, optimize the second text feature by using the attention mechanism on the first vector; utilize the feed-forward neural network to optimize the second text features are processed to obtain a second vector representing the machine-translated text;

A post-translation edited text of the machine-translated text is generated word by word from left to right according to the first vector and the second vector.

Wherein, the first text feature of the source text is extracted through a self-attention mechanism, and the first text feature is processed by a feed-forward neural network to obtain a first vector representing the source text, including:

processing the source text through a residual neural network to obtain the first vector;

Wherein, each network layer in the residual neural network is composed of a self-attention mechanism sublayer and a feedforward neural network sublayer.

Wherein, the optimization of the second text feature by using the attention mechanism on the first vector includes:

Optimize the second text feature according to the attention mechanism processing formula, and the attention mechanism processing formula is:

Wherein, Q represents a query item in the second text feature; K, V represent a pair of key values.

Wherein, said generating the post-translation edited text of the machine-translated text from left to right word by word according to the first vector and the second vector includes:

Generate the post-translation editing text according to the text generation formula, and the text generation formula is:

Wherein, x represents the first vector, m represents the second vector, y represents the post-edited text, P(y|m, x) represents the conditional probability of generating the post-translated edited text; the translated The conditional probability generated by any word in the edited text is: P(y _t |y _<t ,m,x)=Softmax(W _o z _t +b _o ), y _t represents the word generated at time t, W _o and b _o are generation parameters, and Z _t represents the output result after passing through the network layer.

Wherein, after generating the post-translation edited text of the machine-translated text word by word according to the first vector and the second vector, it further includes:

calculating the cross-entropy loss function value of the post-edited text and the standard translation text of the source text;

Judging whether the cross-entropy loss function value is less than a preset threshold;

If not, update the generation parameters according to the value of the cross-entropy loss function, carry the updated generation parameters and perform the step of generating the machine translation text from left to right word by word according to the first vector and the second vector Steps to post-edit text.

Wherein, the calculation of the cross-entropy loss function value of the post-translation editing text and the standard translation text of the source text includes:

Obtain the standard translation text, and extract the third text feature of the standard translation text through a masked self-attention mechanism;

optimizing the third text feature by using an attention mechanism on the first vector, and optimizing the third text feature a second time by using an attention mechanism on the second vector;

Using a feed-forward neural network to process the third text features after the second optimization to obtain a third vector representing the standard translation text;

vectorizing the post-translated edited text into a fourth vector, and calculating a cross-entropy loss function value between the fourth vector and the third vector.

A post-translation editing device for machine-translated text, comprising:

an acquisition module, configured to acquire a source text and a machine-translated text of said source text;

A first processing module, configured to extract first text features of the source text through a self-attention mechanism, and process the first text features using a feedforward neural network to obtain a first vector representing the source text;

The second processing module is used to extract the second text feature of the machine translation text through the self-attention mechanism, optimize the second text feature by using the attention mechanism on the first vector; utilize the feed-forward neural network to optimize After processing the second text feature, obtain the second vector representing the machine translation text;

A generation module, configured to generate a post-translation edited text of the machine-translated text word by word from left to right according to the first vector and the second vector.

Among them, also include:

Calculation module, for calculating the cross-entropy loss function value of the standard translation text of the post-translation edited text and the source text;

A judging module, configured to judge whether the cross-entropy loss function value is less than a preset threshold;

An execution module, configured to update generation parameters according to the value of the cross-entropy loss function when the value of the cross-entropy loss function is not less than a preset threshold, and carry out the steps in the generation module with the updated generation parameters.

A post-translation editing device for machine-translated text, comprising:

memory for storing computer programs;

A processor, configured to implement the steps of the method for post-translation editing of a machine-translated text described in any one of the above when executing the computer program.

A readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for post-translation editing of a machine-translated text described in any one of the above items are realized.

From the above solutions, it can be known that a post-translation editing method of a machine-translated text provided by an embodiment of the present invention includes: obtaining the source text and the machine-translated text of the source text; extracting the first text of the source text through a self-attention mechanism Text features, and use the feedforward neural network to process the first text features to obtain the first vector representing the source text; extract the second text features of the machine translation text through the self-attention mechanism, and pass the The first vector uses the attention mechanism to optimize the second text feature; utilizes the feedforward neural network to process the optimized second text feature to obtain the second vector representing the machine translation text; according to the first A post-edited text of the machine-translated text is generated word by word from left to right by the first vector and the second vector.

It can be seen that the method extracts the text features of the source text and the machine translation text through the self-attention mechanism, and can capture the internal structure of the source text and the machine translation text, so that the extracted text features are more specific and refined, thereby improving the accuracy of machine translation. At the same time, by using the attention mechanism on the first vector of the source text to optimize the second text features of the machine-translated text, thereby associating the features between the source text and the machine-translated text, the translation can be improved. The generalization ability of post-editing; the feed-forward neural network can combine the representation information of different positions to further improve the information representation ability of sentences. Therefore, this method can improve the processing efficiency and accuracy of the post-translation editing, so that the accuracy of the post-translation editing text obtained through processing is better.

Correspondingly, the post-translation editing device, equipment and readable storage medium for machine-translated texts provided by the embodiments of the present invention also have the above-mentioned technical effects.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is a flow chart of a method for post-translation editing of a machine-translated text disclosed in an embodiment of the present invention;

FIG. 2 is a flow chart of another method for post-translation editing of a machine-translated text disclosed in an embodiment of the present invention;

Fig. 3 is a schematic diagram of a post-translation editing device for a machine-translated text disclosed in an embodiment of the present invention;

FIG. 4 is a schematic diagram of a post-translation editing device for a machine-translated text disclosed in an embodiment of the present invention;

Fig. 5 is a schematic diagram of a post-translation editing network model framework disclosed by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The embodiment of the invention discloses a method, device, equipment and readable storage medium for post-translation editing of machine-translated texts, so as to improve the accuracy rate of post-translation editing.

Referring to Fig. 1, a kind of post-translation editing method of machine translation text provided by the embodiment of the present invention comprises:

S101. Obtain the source text and the machine translation text of the source text;

Specifically, the machine-translated text of the source text is the text obtained after machine-translating the source text.

S102. Extract the first text feature of the source text through a self-attention mechanism, and process the first text feature by using a feed-forward neural network to obtain a first vector representing the source text;

S103. Extract the second text feature of the machine-translated text through the self-attention mechanism, optimize the second text feature by using the attention mechanism on the first vector; use the feedforward neural network to process the optimized second text feature, and obtain the representation the second vector of machine translated text;

S104. Generate a post-translation edited text of the machine-translated text word by word from left to right according to the first vector and the second vector.

It should be noted that the attention mechanism imitates the internal process of biological observation behavior, that is, a mechanism that aligns internal experience with external perception to increase the fineness of observation in some areas. Attention mechanism can quickly extract important features of sparse data, so it is widely used in natural language processing tasks. The self-attention mechanism is able to learn the dependencies between different positions within the sentence itself.

The attention mechanism is generally used to process machine translation tasks, but in this application, the attention mechanism is used to process post-translation editing tasks, and combined with the self-attention mechanism to capture the text features of the source text and the machine translation text, not only can Extract specific and fine text features, and can improve the processing efficiency of post-translation editing.

It can be seen that this embodiment provides a method for post-translation editing of machine-translated texts. The method extracts the text features of the source text and the machine-translated text through a self-attention mechanism, and can capture the internal structure of the source text and the machine-translated text, so that The extracted text features are more specific and refined, which can improve the accuracy of post-editing of machine translation texts; at the same time, by using the attention mechanism on the first vector of the source text to optimize the second text features of the machine translation text, thereby By associating the features between the source text and the machine-translated text, the generalization ability of post-translation editing can be improved; the feed-forward neural network can combine the representation information of different positions to further improve the information representation ability of sentences. Therefore, this method can improve the processing efficiency and accuracy of the post-translation editing, so that the accuracy of the post-translation editing text obtained through processing is better.

The embodiment of the present invention discloses another method for post-translation editing of machine-translated texts. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution.

Referring to FIG. 2 , another post-translation editing method for machine-translated text provided by an embodiment of the present invention includes:

S201. Obtain the source text and the machine translation text of the source text;

S202. Extract the first text features of the source text through a self-attention mechanism, and process the first text features by using a feed-forward neural network to obtain a first vector representing the source text;

S203. Extract the second text feature of the machine-translated text through the self-attention mechanism, optimize the second text feature by using the attention mechanism on the first vector; use the feedforward neural network to process the optimized second text feature, and obtain the representation the second vector of machine translated text;

S204. Generate a post-translation edited text of the machine-translated text word by word from left to right according to the first vector and the second vector;

205. Calculating the cross-entropy loss function value of the post-translation edited text and the standard translation text of the source text;

Specifically, the standard translation text of the source text is: after post-editing the machine translation text obtained by machine translation of the source text, the final text conforming to the human language style is obtained. Calculating the cross-entropy loss function value of the post-edited text and the standard translation text can be understood as: judging the similarity between the post-edited text and the standard translation text.

When the cross-entropy loss function value of the post-edited text and the standard translated text is larger, it indicates that the similarity between the post-edited text and the standard translated text is smaller, and it can be considered that the two are not the same, and the post-edited text needs further optimization and optimization. Processing; when the cross-entropy loss function value of the post-edited text and the standard translation text is smaller, it indicates that the similarity between the post-edited text and the standard translation text is greater, and the two can be considered to be the same to a certain extent.

In this embodiment, the sentence-level loss function is taken into consideration, which can provide a better optimization basis for the generation of post-translation edited text.

206. Determine whether the value of the cross-entropy loss function is smaller than a preset threshold; if yes, execute S208; if not, execute S207;

S207. Update the generation parameters according to the value of the cross-entropy loss function, and execute S204 with the updated generation parameters;

Specifically, the loss of the post-edited text can be considered as the difference between the post-edited text and the standard translation, which is generally represented by the edit distance between the two texts. The smaller the edit distance between two texts, the more similar the two texts are.

S208. Determine the generated post-edited text as a standard translation result of the machine translated text.

Wherein, the calculation of the cross-entropy loss function value of the post-translation editing text and the standard translation text of the source text includes:

Obtain the standard translation text, and extract the third text feature of the standard translation text through a masked self-attention mechanism;

optimizing the third text feature by using an attention mechanism on the first vector, and optimizing the third text feature a second time by using an attention mechanism on the second vector;

Using a feed-forward neural network to process the third text features after the second optimization to obtain a third vector representing the standard translation text;

vectorizing the post-translated edited text into a fourth vector, and calculating a cross-entropy loss function value between the fourth vector and the third vector.

It can be seen that this embodiment provides another method for post-translation editing of machine-translated texts. The method extracts the text features of the source text and the machine-translated text through a self-attention mechanism, and can capture the internal structure of the source text and the machine-translated text. The extracted text features are more specific and refined, thereby improving the accuracy of post-editing of machine-translated texts; at the same time, optimizing the second text features of machine-translated texts by using the attention mechanism on the first vector of the source text, Thus, the features between the source text and the machine-translated text are associated, which can improve the generalization ability of post-translation editing; the feed-forward neural network can combine the representation information of different positions to further improve the information representation ability of sentences. Therefore, this method can improve the processing efficiency and accuracy of the post-translation editing, so that the accuracy of the post-translation editing text obtained through processing is better.

Based on any of the above-mentioned embodiments, it should be noted that the first text feature of the source text is extracted through the self-attention mechanism, and the first text feature is processed by a feed-forward neural network to obtain a representation of the source text. The first vector of text, consisting of:

processing the source text through a residual neural network to obtain the first vector;

Wherein, each network layer in the residual neural network is composed of a self-attention mechanism sublayer and a feedforward neural network sublayer.

Based on any of the above embodiments, it should be noted that optimizing the second text feature by using an attention mechanism on the first vector includes:

Optimize the second text feature according to the attention mechanism processing formula, and the attention mechanism processing formula is:

Wherein, Q represents a query item in the second text feature; K, V represent a pair of key values.

Based on any of the above embodiments, it should be noted that the post-translation edited text of the machine-translated text is generated word by word from left to right according to the first vector and the second vector, including:

Generate the post-translation editing text according to the text generation formula, and the text generation formula is:

Wherein, x represents the first vector, m represents the second vector, y represents the post-edited text, P(y|m, x) represents the conditional probability of generating the post-translated edited text; the translated The conditional probability generated by any word in the edited text is: P(y _t |y _<t ,m,x)=Softmax(W _o z _t +b _o ), y _t represents the word generated at time t, W _o and b _o are generation parameters, and Z _t represents the output result after passing through the network layer.

Wherein, if the post-translation editing processing model is constructed according to the post-translation editing method provided by the present invention, the network layer is the last layer of the entire post-translation editing processing model.

A post-translation editing device for machine-translated texts provided by an embodiment of the present invention is introduced below. The post-translation editing device for machine-translated texts described below and the post-translation editing method for machine-translated texts described above can be mutually refer to.

Referring to FIG. 3 , a post-translation editing device for machine-translated text provided by an embodiment of the present invention includes:

An acquisition module 301, configured to acquire a source text and a machine-translated text of the source text;

The first processing module 302 is configured to extract the first text features of the source text through a self-attention mechanism, and process the first text features using a feed-forward neural network to obtain a first vector representing the source text ;

The second processing module 303 is used to extract the second text feature of the machine translation text through the self-attention mechanism, optimize the second text feature by using the attention mechanism on the first vector; utilize the feed-forward neural network to Processing the optimized second text features to obtain a second vector representing the machine-translated text;

A generation module 304, configured to generate a post-translation edited text of the machine-translated text word by word from left to right according to the first vector and the second vector.

Among them, also include:

Calculation module, for calculating the cross-entropy loss function value of the standard translation text of the post-translation edited text and the source text;

A judging module, configured to judge whether the cross-entropy loss function value is less than a preset threshold;

An execution module, configured to update generation parameters according to the value of the cross-entropy loss function when the value of the cross-entropy loss function is not less than a preset threshold, and carry out the steps in the generation module with the updated generation parameters.

Wherein, the calculation module includes:

An acquisition unit, configured to acquire the standard translation text, and extract the third text feature of the standard translation text through a masked self-attention mechanism;

A first optimization unit, configured to optimize the third text feature by using an attention mechanism on the first vector, and optimize the third text feature for a second time by using an attention mechanism on the second vector;

The second optimization unit is configured to use a feed-forward neural network to process the second optimized third text features to obtain a third vector representing the standard translated text;

A calculation unit, configured to vectorize the post-translation edited text into a fourth vector, and calculate a cross-entropy loss function value between the fourth vector and the third vector.

Wherein, the first processing module is specifically used for:

processing the source text through a residual neural network to obtain the first vector;

Wherein, each network layer in the residual neural network is composed of a self-attention mechanism sublayer and a feedforward neural network sublayer.

Wherein, the second processing module is specifically used for:

Optimize the second text feature according to the attention mechanism processing formula, and the attention mechanism processing formula is:

Wherein, Q represents a query item in the second text feature; K, V represent a pair of key values.

Wherein, the generating module is specifically used for:

Generate the post-translation editing text according to the text generation formula, and the text generation formula is:

Wherein, x represents the first vector, m represents the second vector, y represents the post-edited text, P(y|m, x) represents the conditional probability of generating the post-translated edited text; the translated The conditional probability generated by any word in the edited text is: P(y _t |y _<t ,m,x)=Softmax(W _o z _t +b _o ), y _t represents the word generated at time t, W _o and b _o are generation parameters, and Z _t represents the output result after passing through the network layer.

It can be seen that this embodiment provides a post-translation editing device for machine-translated text, including: an acquisition module, a first processing module, a second processing module, and a generation module. First, the source text and the machine translation text of the source text are obtained by the acquisition module; then the first processing module extracts the first text feature of the source text through a self-attention mechanism, and uses a feed-forward neural network to analyze the first text Features are processed to obtain the first vector representing the source text; then the second processing module extracts the second text features of the machine translation text through the self-attention mechanism, and optimizes the first vector by using the attention mechanism on the first vector. The second text feature; use the feed-forward neural network to process the optimized second text feature to obtain the second vector representing the machine translation text; finally the generation module according to the first vector and the second The vector generates a post-edited text of the machine translated text word by word from left to right. In this way, the division of labor and cooperation among the various modules, each performing its own duties, thereby improving the processing efficiency and accuracy of post-translation editing, and making the post-translation editing text obtained after processing more accurate.

A post-translation editing device for machine-translated texts provided by an embodiment of the present invention is introduced below, a post-translation editing device for machine-translated texts described below and a post-translation editing method and device for machine-translated texts described above Can be cross-referenced.

Referring to FIG. 4, a post-translation editing device for machine-translated text provided by an embodiment of the present invention includes:

Memory 401, used to store computer programs;

The processor 402 is configured to implement the steps of the post-translation editing method for machine-translated text described in any of the above embodiments when executing the computer program.

Wherein, the processor may be a central processing unit (CPU) or a graphics processing unit (GPU). GPU has a good advantage when dealing with large-scale data.

The following is an introduction to a readable storage medium provided by an embodiment of the present invention. The readable storage medium described below and the post-translation editing method, device, and equipment for a machine-translated text described above may refer to each other.

A readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for post-translation editing of a machine-translated text as described in any of the above embodiments are realized.

According to the post-translation editing method provided by the present invention, the post-translation editing network model as shown in Figure 5 can be constructed, and the model includes a source text processing network, a machine translation text processing network and a standard translation text processing network, wherein the source text processing network, the machine Both the translated text processing network and the standard translated text processing network are residual networks.

The source text processing network has N layers, and each layer consists of a self-attention mechanism sublayer and a feedforward neural network sublayer. The machine translation text processing network has N layers, and each layer is composed of a self-attention mechanism sublayer, an attention mechanism sublayer and a feedforward neural network sublayer. The attention mechanism sublayer in the machine translation text processing network represents the source text Use attention mechanism. The standard translation text processing network has N layers, and each layer is composed of a self-attention mechanism sublayer with a mask, an attention mechanism sublayer and a feedforward neural network sublayer, and the attention mechanism sublayer in the standard translation text processing network Including: using attention mechanism on source text and using attention mechanism on machine translated text.

进行缩放，最后运用一个softmax函数获得键值K(values)的权重分布。具体可以用如下公式描述：Among them, the attention mechanism calculates the dot product of the query item Q (query) on all key K (key) values through a mapping query (Query) and a set of key-value pairs (key-values), and divides by

Scale, and finally use a softmax function to obtain the weight distribution of the key K (values). Specifically, it can be described by the following formula:

The multi-head attention mechanism allows the model to jointly pay attention to information from different representation subspaces in different positions, which can be expressed by the following formula:

MultiHead(Q,K,V)＝Concat(head ₁ ,...,head _h )W ^o

wherehead _i ＝Attention(QW _i ^Q ,KW _i ^K ,VW _i ^V )

in,

The feedforward neural network contains two linear changes, and the ReLu activation function is used between the linear transformations, which can be expressed by the following formula:

FFN(x)=max(0,xW ₁ +b ₁ )W ₂ +b ₂

Among them, W ₁ , W ₂ , b ₁ , and b ₂ are all trainable parameters.

The Discriminator in Figure 5 is the discriminator in the post-translation editing network model. The discriminator uses a recurrent neural network and chooses to use a bidirectional (Gated Recurrent Unit, GRU) structure to represent sentences. The discriminator reads the post-translation edited text and the standard translation text, and uses a bidirectional GRU to represent the word embedding of the two sentences to obtain the content vector. Given a loss function, the goal of the loss function is to distinguish between the generated text and the reference text. Make the judgment more and more accurate.

The formula for calculating the cross-entropy loss of the discriminator to distinguish the post-edited text from the standard translated text is:

P(y,r)＝sigmoid(W _d ·||H _y -H _r ||+b _d )

The loss function of the discriminator is expressed by the following formula:

L(H _y ,H _r )＝-log(sigmoid(W _d ·||H _y -H _r ||+b _d ))

where ||H _y −H _r || represents the Euclidean distance between the content vectors of the post-edited text and the standard translated text, and both W _d and b _d are trainable parameters.

When the discrimination result output by the discriminator does not meet the preset output conditions, the loss of the post-edited text is calculated, and the loss is fed back to optimize the network parameters of the post-translation editing network model to generate more accurate post-editing text.

Among them, the maximum expected value of the objective function for generating post-edited text is set as:

Sample the generated post-edited text and compute the gradient

生成器的参数更新函数为：/>

so

The parameter update function of the generator is: />

When training the discriminator in a post-editing network model, freeze the generator parameters and minimize the loss function of the discriminator. Specifically, every time the generator is trained for 4 epochs, the discriminator is trained using another epoch, and the training is iterative in turn until the model generator and the discriminator both converge and then stop the training.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (9)

1. A method of post-translation editing of machine-translated text, comprising:

acquiring a source text and a machine translation text of the source text;

extracting first text features of the source text through a self-attention mechanism, and processing the first text features by utilizing a feedforward neural network to obtain a first vector representing the source text;

extracting a second text feature of the machine translated text by a self-attention mechanism, the second text feature being optimized by using an attention mechanism on the first vector; processing the optimized second text feature by using a feedforward neural network to obtain a second vector representing the machine translation text;

generating translated edited text of the machine translated text word by word from left to right according to the first vector and the second vector;

wherein the generating the translated compiled text of the machine translated text word by word from left to right based on the first vector and the second vector comprises:

generating the translated editing text according to a text generation formula, wherein the text generation formula is as follows:

wherein x represents the first vector, m represents the second vector, y represents the translated edit text, and P (y|m, x) represents a conditional probability of generating the translated edit text; the conditional probability of any word generation in the translated editing text is as follows: p (y) _t |y _＜t ,m,x)＝Softmax(W _o ·z _t +b _o )，y _t Representing words generated at time t, W _o And b _o Generating parameters, Z _t Indicating the output result after passing through the network layer.

2. The method of post-translation editing of machine-translated text according to claim 1 wherein the extracting the first text feature of the source text by a self-attention mechanism and processing the first text feature using a feed-forward neural network to obtain a first vector representing the source text comprises:

processing the source text through a residual neural network to obtain the first vector;

wherein each network layer in the residual neural network is comprised of a self-attention machine sublayer and a feed-forward neural network sublayer.

3. The method of post-translation editing of machine-translated text according to claim 2, wherein said optimizing the second text feature by using an attention mechanism on the first vector comprises:

optimizing the second text feature according to an attention mechanism processing formula, wherein the attention mechanism processing formula is as follows:

wherein Q represents a query term in the second text feature; k, V represents a pair of key values.

4. A method of post-translational editing of machine-translated text according to any one of claims 1-3, wherein after said generating of post-translational edited text of said machine-translated text word by word from left to right based on said first vector and said second vector, further comprises:

calculating a cross entropy loss function value of the translated editing text and a standard translation text of the source text;

judging whether the cross entropy loss function value is smaller than a preset threshold value or not;

if not, updating the generation parameters according to the cross entropy loss function value, carrying the updated generation parameters, and executing the step of generating the translated edited text of the machine translation text word by word from left to right according to the first vector and the second vector.

5. The method of post-translational editing of machine-translated text of claim 4, wherein said calculating a cross-entropy loss function value for the post-translational edited text and standard translated text of the source text comprises:

acquiring the standard translation text, and extracting a third text feature of the standard translation text through a masked self-attention mechanism;

optimizing the third text feature by using an attention mechanism on the first vector and optimizing the third text feature a second time by using an attention mechanism on the second vector;

processing the third text feature after the second optimization by using a feedforward neural network to obtain a third vector representing the standard translation text;

and vectorizing the translated edited text into a fourth vector, and calculating a cross entropy loss function value of the fourth vector and the third vector.

6. A post-translation editing apparatus for machine-translating text, comprising:

the acquisition module is used for acquiring a source text and a machine translation text of the source text;

the first processing module is used for extracting first text features of the source text through a self-attention mechanism, and processing the first text features by utilizing a feedforward neural network to obtain a first vector representing the source text;

a second processing module for extracting a second text feature of the machine translated text by a self-attention mechanism, the second text feature being optimized by using an attention mechanism on the first vector; processing the optimized second text feature by using a feedforward neural network to obtain a second vector representing the machine translation text;

the generation module is used for generating translated editing text of the machine translation text word by word from left to right according to the first vector and the second vector;

the generating module is specifically configured to:

generating the translated editing text according to a text generation formula, wherein the text generation formula is as follows:

wherein x represents the first vector, m represents the second vector, y represents the translated edit text, and P (y|m, x) represents a conditional probability of generating the translated edit text; the conditional probability of any word generation in the translated editing text is as follows: p (y) _t |y _＜t ,m,x)＝Softmax(W _o ·z _t +b _o )，y _t Representing words generated at time t, W _o And b _o Generating parameters, Z _t Indicating the output result after passing through the network layer.

7. The machine-translated text post-editing apparatus of claim 6, further comprising:

the calculation module is used for calculating a cross entropy loss function value of the translated editing text and the standard translation text of the source text;

the judging module is used for judging whether the cross entropy loss function value is smaller than a preset threshold value or not;

and the execution module is used for updating the generation parameters according to the cross entropy loss function value when the cross entropy loss function value is not smaller than a preset threshold value, and executing the steps in the generation module with the updated generation parameters.

8. A post-translation editing apparatus for machine-translating text, comprising:

a memory for storing a computer program;

a processor for implementing the steps of the method for post-translational editing of machine translated text according to any one of claims 1 to 5 when executing said computer program.

9. A readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of the method for post-translational editing of machine translated text according to any one of claims 1 to 5.

Images