Translated text quality evaluation method and device, computer equipment and storage medium

Abstract

The embodiment of the invention provides a translated text quality evaluation method and device, computer equipment and a storage medium. The method comprises the steps of obtaining a translation source text and a reference translation text of a to-be-evaluated translation text, and performing quality evaluation on the reference translation text based on the translation source text to obtain reference quality of the reference translation text; performing quality scoring on the to-be-evaluated translated text through the translated source text to obtain a first quality score; scoring the quality of the to-be-evaluated translation through the reference translation to obtain a second quality score; according to the reference quality of the reference translation, carrying out weight adjustment on the first quality score to obtain a first target quality score, and carrying out weight adjustment on the second quality score to obtain a second target quality score; wherein the weight of the first mass fraction is reduced along with the increase of the reference mass; and obtaining a quality evaluation result of the to-be-evaluated translation based on the first target quality score and the second target quality score. Therefore, the accuracy of performing quality evaluation on the translated text can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to a translation text quality assessment method, apparatus, computer equipment and storage medium. Background Art

[0002] Machine translation refers to the target language text generated by computer programs. Generally, it can be automatically converted by computer programs based on source language text through algorithms. In order to ensure the accuracy and reliability of translation and provide high-quality translation services in various application scenarios, it is usually necessary to conduct quality assessment of machine translation.

[0003] In the related art, the quality of machine translation can be evaluated by comparing the reference translation with the machine translation. However, the quality of reference translations varies. When the machine faces an incorrect reference translation, it will give an incorrect evaluation of the current machine translation, reducing the accuracy of the quality evaluation of the machine translation. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a translation text quality assessment method, apparatus, computer equipment and storage medium, which can effectively improve the accuracy of quality assessment of translation texts.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for evaluating the quality of a translation text, the method comprising:

[0006] Obtaining a translation source text and a reference translation of the translation to be evaluated, and performing a quality evaluation on the reference translation based on the translation source text to obtain a reference quality of the reference translation;

[0007] Scoring the quality of the translation to be evaluated by using the translation source text to obtain a first quality score;

[0008] Scoring the quality of the translation to be evaluated by using the reference translation to obtain a second quality score;

[0009] According to the reference quality of the reference translation, the first quality score is weighted to obtain a first target quality score, and the second quality score is weighted to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases;

[0010] A quality evaluation result of the translation to be evaluated is obtained based on the first target quality score and the second target quality score.

[0011] Accordingly, a second aspect of the embodiments of the present application provides a translation text quality assessment device, the device comprising:

[0012] An evaluation module, used to obtain a translation source text and a reference translation of a translation to be evaluated, and to perform a quality evaluation on the reference translation based on the translation source text to obtain a reference quality of the reference translation;

[0013] A first scoring module, configured to score the quality of the translation to be evaluated using the translation source text to obtain a first quality score;

[0014] A second scoring module, configured to score the quality of the translation to be evaluated using the reference translation to obtain a second quality score;

[0015] an adjustment module, configured to adjust the weight of the first quality score according to the reference quality of the reference translation to obtain a first target quality score, and to adjust the weight of the second quality score to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases;

[0016] An acquisition module is used to obtain a quality evaluation result of the translation to be evaluated based on the first target quality score and the second target quality score.

[0017] In some implementations, the adjustment module is further used to:

[0018] Determining a target weight of the translated source text according to the reference quality of the reference translated text;

[0019] A first target quality score is obtained according to the product of the target weight and the first quality score.

[0020] In some implementations, the adjustment module is further used to:

[0021] Obtaining the complement of the target weight;

[0022] A second target quality score is obtained according to the product of the complement of the target weight and the second quality score.

[0023] In some implementations, the adjustment module is further used to:

[0024] Obtain a first weight coefficient and a second weight coefficient;

[0025] Obtaining a first intermediate coefficient by multiplying the reference mass by the first weight coefficient;

[0026] A target weight of the translation source text is obtained based on the sum of the first intermediate coefficient and the second weight coefficient.

[0027] In some embodiments, the evaluation module is further used to:

[0028] Acquire multiple translation source texts from a translation source text collection, and acquire multiple reference translation texts from a reference translation text collection;

[0029] Comparing each of the translated source texts with the corresponding reference translations to obtain a consistency score for characterizing the quality of the reference translations;

[0030] The sum of the consistency scores of the reference translations is obtained, and the reference quality of the reference translation is obtained by the ratio of the sum of the consistency scores to the number of the reference translations.

[0031] In some embodiments, the evaluation module is further used to:

[0032] Obtain multiple translation source texts from a translation source text collection, and obtain multiple reference translation texts from a reference translation text collection;

[0033] For each of the reference translations, the reference translation is compared with the corresponding translation source text to obtain a consistency score for characterizing the quality of different reference translations;

[0034] The consistency score is used as the reference quality of the corresponding reference translation.

[0035] In some embodiments, the evaluation module is further used to:

[0036] Extracting a first target feature from the translation source text, wherein the first target feature includes at least one of a grammatical feature, a sentence feature, and a semantic feature;

[0037] Extracting a second target feature from the translation to be evaluated, wherein a feature attribute of the second target feature corresponds to a feature attribute of the first target feature;

[0038] An alignment degree index between the first target feature and the corresponding second target feature is calculated, and a first quality score of the translation to be evaluated is determined based on the alignment degree index.

[0039] In some embodiments, the second scoring module is further used to:

[0040] Performing vector encoding on the reference translation to obtain a first vector encoding;

[0041] Performing vector encoding on the translation to be evaluated to obtain a second vector encoding;

[0042] The semantic similarity between the first vector code and the corresponding second vector code is calculated, and a second quality score of the translation to be evaluated is determined based on the semantic similarity.

[0043] To achieve the above-mentioned purpose, the third aspect of the embodiments of the present application proposes a computer device, which includes a memory and a processor, the memory stores a computer program, and the processor implements the translation text quality assessment method described in any one of the embodiments of the first aspect of the present application when executing the computer program.

[0044] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the translation text quality assessment method described in any one of the embodiments of the first aspect of the present application.

[0045] In the embodiment of the present application, a translation source text and a reference translation of a translation to be evaluated are obtained, and quality evaluation of the reference translation is performed based on the translation source text to obtain a reference quality of the reference translation; quality scoring of the translation to be evaluated is performed based on the translation source text to obtain a first quality score; quality scoring of the translation to be evaluated is performed based on the reference translation to obtain a second quality score; according to the reference quality of the reference translation, the weight of the first quality score is adjusted to obtain a first target quality score, and the weight of the second quality score is adjusted to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases; and based on the first target quality score and the second target quality score, a quality evaluation result of the translation to be evaluated is obtained. In this way, by considering the different qualities of different reference translations, the reference quality of the corresponding reference translation can be obtained, and then the first quality score corresponding to the translated source text and the second quality score corresponding to the reference translation can be adjusted according to the reference quality, so that when the reference quality of the reference translation is high, the weight of the first quality score is lower, and when the reference quality of the reference translation is low, the weight of the first quality score is higher, thereby dynamically adjusting the contribution of the reference quality to the quality evaluation of the translation to be evaluated. The present application can achieve accurate evaluation of the quality of the translated text under reference translations of different qualities. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Schematic diagram of the architecture of the translation text quality assessment system provided in the embodiment of the present application;

[0047] Figure 2 is a flow chart of a translation text quality assessment method provided in an embodiment of the present application;

[0048] Figure 3 is an overall flow chart of the translation text quality assessment method provided in the embodiment of the present application;

[0049] Figure 4 It is a schematic diagram of the functional modules of the translation text quality assessment device provided in an embodiment of the present application;

[0050] Figure 5 It is a schematic diagram of the hardware structure of the computer device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0054] Machine translation refers to the target language text generated by computer programs. Generally, it can be automatically converted by computer programs based on source language text through algorithms. In order to ensure the accuracy and reliability of translation and provide high-quality translation services in various application scenarios, it is usually necessary to conduct quality assessment of machine translation.

[0055] In the related art, the quality of machine translation can be evaluated by comparing the reference translation with the machine translation. However, the quality of reference translations varies. When the machine faces an incorrect reference translation, it will give an incorrect evaluation of the current machine translation, reducing the accuracy of the quality evaluation of the machine translation.

[0056] Based on this, the embodiments of the present application provide a translation text quality assessment method, apparatus, computer equipment and storage medium, which can effectively improve the accuracy of quality assessment of machine translations.

[0057] The translation text quality assessment method, apparatus, computer device and storage medium provided in the embodiments of the present application are specifically described through the following embodiments. First, the translation text quality assessment system in the embodiments of the present application is described.

[0058] Please refer to Figure 1 In some implementations, the translation text quality assessment system provided in the embodiments of the present application includes a terminal 11 and a server 12 .

[0059] Exemplarily, the terminal 11 may be a mobile terminal device or a non-mobile terminal device. Specifically, the terminal 11 may be a mobile terminal device such as a smart phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted terminal device, a wearable device, a personal digital assistant, or a non-mobile terminal device such as a personal computer. The terminal 11 may provide a convenient user interaction interface and a data upload portal. For example, when the terminal 11 is a smart phone or a tablet computer, the application on the smart phone or tablet computer may allow the user to conveniently upload documents or text fragments for instant translation quality assessment.

[0060] Exemplarily, the server end 12 can be a high-performance server cluster, a cloud server, a distributed computing system, etc. The server end 12 can centrally process complex algorithms and computing tasks to ensure the efficient operation of the system. For example, when the server end 12 is a high-performance server cluster, the high-performance server cluster can process a large number of translation quality assessment requests from the terminal 11, and perform complex computing tasks, such as calculating the reference quality of the reference translation, scoring the quality of the translation to be assessed by translating the source text, obtaining a first quality score, scoring the quality of the translation to be assessed by the reference translation, obtaining a second quality score, adjusting the weights of the first quality score and the second quality score, etc.

[0061] The translation text quality assessment method in the embodiments of the present application can be illustrated by the following embodiments.

[0062] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first. Moreover, the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or separate consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or separate consent, the necessary user-related data for enabling the normal operation of the embodiment of the present application will be obtained.

[0063] In the embodiments of the present application, the description will be made from the perspective of a translation text quality assessment device, which can be integrated into a computer device. Figure 2 , Figure 2 This is a flowchart of the steps of the translation text quality assessment method provided in the embodiment of the present application. In the embodiment of the present application, the translation text quality assessment device is specifically integrated in a terminal or a server as an example. When the processor on the terminal or the server executes the program instructions corresponding to the translation text quality assessment method, the specific process is as follows:

[0064] Step 101 : obtaining a translation source text and a reference translation of a translation to be evaluated, and performing a quality evaluation on the reference translation based on the translation source text to obtain a reference quality of the reference translation.

[0065] In some implementations, since a high-quality reference translation can better reflect the true level of the translation to be evaluated, and a low-quality reference translation may lead to evaluation bias, in order to ensure that when the quality of the reference translation is poor, the evaluation of the translation to be evaluated does not rely too much on the reference translation, the quality of the reference translation can be quantified to obtain the reference quality, so that the quality of the translation to be evaluated can be accurately and fairly evaluated later.

[0066] The translation to be evaluated may be a translation text output by a machine translation system, the quality of which needs to be evaluated.

[0067] The source text for translation can be the original untranslated text, which is the input of the machine translation task and one of the bases for comparing with the translation to be evaluated when evaluating the translation quality. The translation source text is input into the machine translation system to obtain the translation to be evaluated.

[0068] The reference translation may be obtained manually from the source text, or obtained through crowdsourcing translation or other methods. The reference translation may be used as one of the benchmarks for evaluating the quality of the translation to be evaluated.

[0069] Among them, the reference quality can be a measure of the accuracy of the reference translation relative to the source text, which is used to characterize the translation quality of the reference translation itself.

[0070] Exemplarily, the translation to be evaluated can be obtained by translating the source text through a statistical machine translation system, or it can be obtained by online translation services, such as Google Translate, Baidu Translate and other online platforms. As long as it is a translation that needs to be quality evaluated, it can be used as the translation to be evaluated. This application does not make specific restrictions on the source of the translation to be evaluated.

[0071] Exemplarily, the quality of the reference translation can be evaluated based on the translation source text to obtain the reference quality of the reference translation. Specifically, the trained SRC-on ly indicator model can be obtained to obtain the translation source text set corresponding to all the translations to be evaluated this time, denoted as S = {s1, s2, ..., sn}, where si represents the i-th translation source text. And obtain the reference translation set corresponding to the translation source text set, denoted as R = {r1, r2, ..., rn}, where ri is the reference translation of si, and each translation to be evaluated has corresponding ri and si.

[0072] Exemplarily, the SRC-on ly indicator model can be a bidirectional encoder based on deep learning, or a statistical model based on feature matching, etc.

[0073] Furthermore, the SRC-on ly indicator model can be used to evaluate each translation source text si and its corresponding reference translation ri, determine the extent to which the reference translation ri retains the information content of the translation source text si, and obtain the consistency score of each reference translation ri.

[0074] Exemplarily, the translation source text and the corresponding reference translation can be first encoded with a high-dimensional vector, and then the semantic similarity between the translation source text and the reference translation can be calculated, and the semantic similarity is equivalent to the reference quality of the reference translation. Alternatively, the degree of semantic alignment between the translation source text and the reference translation across languages ​​can be calculated based on the translation source text, and the degree of semantic alignment represents the reference quality of the reference translation. Alternatively, the reference translation can be directly scored by a professional translator to obtain the reference quality of the reference translation. Specifically, the corresponding method for calculating the reference quality can be selected according to the actual situation, and the embodiment of the present application does not specifically limit this.

[0075] In some implementations, the average of the consistency scores of all reference translations in the overall reference translation set may be calculated as the reference quality of each reference translation, or the consistency score of each reference translation may be calculated one by one as the reference quality of each reference translation, or the mode of the consistency scores of all reference translations in the reference translation set may be used as the reference quality of each reference translation, and this embodiment of the present application does not impose any specific limitation on this.

[0076] Exemplarily, when the reference quality is a number other than 0 to 1, the reference quality can be converted to a reference quality between 0 and 1. For example, if the output is a distance metric, the smaller the distance between the translation source and the reference translation, the higher the matching degree, which can be converted to the reference quality through normalization.

[0077] In some implementations, in order to improve the efficiency of evaluating the translation to be evaluated while taking into account the accuracy of the evaluation, the average quality score of all reference translations can be used as the quality score of all reference translations, thereby reducing the consumption of computing resources and making the entire evaluation process more efficient. For example, in step 101, "based on the translation source text, the quality of the reference translation is evaluated to obtain the reference quality of the reference translation" may include:

[0078] (101.a1) obtaining a plurality of translation source texts from a translation source text set, and obtaining a plurality of reference translation texts from a reference translation text set;

[0079] (101.a2) performing consistency comparison between each translation source text and the corresponding reference translation, and obtaining a consistency score for characterizing the quality of the reference translation;

[0080] (101.a3) Obtain the sum of multiple consistency scores of multiple reference translations, and obtain the reference quality of the reference translation by the ratio of the sum of the multiple consistency scores to the number of reference translations.

[0081] The translation source text set may be a set of translation source texts corresponding to the translation text set to be evaluated, and each translation source text may be used as an input of a machine translation system to obtain a corresponding translation text to be evaluated.

[0082] The reference translation set may correspond to the translation source set, and the reference translation set includes a standard translation result for each translation source, which is used as one of the benchmarks for evaluating the quality of the translation to be evaluated.

[0083] Among them, the consistency score can be a quantitative indicator to measure the degree of match between the reference translation and the translation source text in terms of content, meaning and expression, and is used to evaluate the reference quality of the reference translation. The consistency score can be calculated during the evaluation process and can be obtained based on semantic similarity, cross-language semantic alignment or manual scoring. The consistency score can be used to characterize the semantic match between the translation source text and the reference translation, for example, it can be obtained by calculating semantic similarity.

[0084] For example, in an English-Chinese translation scenario, there is a translation source text set S and a corresponding reference translation set R. The translation source text set S includes 5 English sentences, and the reference translation set R includes the standard translations corresponding to the 5 English sentences. It can be understood that there are also 5 translations to be evaluated corresponding to the 5 English sentences.

[0085] Furthermore, the SRC-on ly indicator model can be used to calculate the consistency score of each pair of translation source texts and the corresponding reference translations, and the consistency score can be used as the basis for evaluating the quality of the translation source texts. For example, for s1 and r1, the consistency score is 0.92; for s2 and r2, the consistency score is 0.88; for s3 and r3, the consistency score is 0.95; for s4 and r4, the consistency score is 0.90; for s5 and r5, the consistency score is 0.85.

[0086] Furthermore, the number of reference translations is 5, and the sum of multiple consistency scores is 0.92+0.88+0.95+0.90+0.85=4.50, so the ratio of the sum of multiple consistency scores to the number of reference translations is 0.9. Therefore, the reference quality of the reference translation is 0.9, and when evaluating each translation to be evaluated, 0.9 can be used as the reference quality of the corresponding reference translation.

[0087] By quantifying the reference quality of the reference translation, we can quantify the degree of match between the reference translation and the source text, providing a more objective and quantitative basis for the quality evaluation of the reference translation. In addition, by determining the reference quality of the reference translation through the ratio of the sum of multiple consistency scores to the number of reference translations, we can focus on the overall translation indicators of the reference translation and improve the efficiency and accuracy of the evaluation of the translation to be evaluated.

[0088] In some embodiments, in order to more accurately reflect the quality differences of individual reference translations and avoid the situation where the reference quality of an excellent translation is lowered by a poor-quality translation, the reference quality of each reference translation can be calculated one by one to adjust the evaluation strategy in a targeted manner, especially when automatically adjusting the target weight of the translated source text, the personalized impact of the reference translation quality can be more accurately reflected. In this way, in addition to obtaining the reference quality of the reference translation by the ratio of the sum of multiple consistency scores to the number of reference translations, the present application also proposes a solution for calculating the reference quality of each reference translation one by one and adjusting the target weight of the corresponding translated source text one by one. For example, the "quality evaluation of the reference translation based on the translated source text to obtain the reference quality of the reference translation" in step 101 can also include:

[0089] (101.b1) obtaining a plurality of translation source texts from a translation source text set, and obtaining a plurality of reference translation texts from a reference translation text set;

[0090] (101.b2) For each reference translation, the reference translation is compared with the corresponding translation source text to obtain a consistency score for characterizing the quality of different reference translations;

[0091] (101.b3) The consistency score is used as the reference quality of the corresponding reference translation.

[0092] For example, in an English-Chinese translation scenario, there is a translation source text set S and a corresponding reference translation set R. The translation source text set S includes three English sentences, and the reference translation set R includes the standard translations corresponding to the three English sentences. It can be understood that there are also three translations to be evaluated corresponding to the three English sentences.

[0093] For example, the consistency score between each pair of translation source texts and the corresponding reference translation texts can be calculated by the SRC-on ly indicator model, and the consistency score can be used as the basis for evaluating the quality of the translation source texts. For example, for s1 and r1, the consistency score is 0.92; for s2 and r2, the consistency score is 0.95; for s3 and r3, the consistency score is 0.90.

[0094] That is to say, 0.92 can be used as the reference mass for r1, 0.95 can be used as the reference mass for r2, and 0.90 can be used as the reference mass for r3.

[0095] The corresponding translation set T to be evaluated includes t1 corresponding to r1 and s1, t2 corresponding to r2 and s2, and t3 corresponding to r3 and s3. Then the target weight of the first quality score corresponding to s1 is determined based on the reference quality of r1, which is 0.92; the target weight of the first quality score corresponding to s2 is determined based on the reference quality of r2, which is 0.95; and the target weight of the first quality score corresponding to s3 is determined based on the reference quality of r3, which is 0.90.

[0096] By determining the reference quality of each reference translation one by one, the quality difference of each reference translation can be captured more finely, thereby improving the accuracy of evaluating the translation to be evaluated.

[0097] Step 102: Score the quality of the translation to be evaluated by translating the source text to obtain a first quality score.

[0098] In some implementations, in order to evaluate whether the translation to be evaluated accurately conveys the meaning of the translated source text, the translation to be evaluated may be scored for quality by translating the source text to determine the fidelity and accuracy of the translation to be evaluated.

[0099] The first quality score may be an SRC-only indicator of the translation to be evaluated, that is, a score evaluated only based on a comparison between the translation source text and the translation to be evaluated during the evaluation of the translation to be evaluated.

[0100] For example, the correspondence between the source text and the translation to be evaluated in terms of grammatical features, sentence features, and semantic features can be compared one by one, and the SRC-on ly indicator model can be used to quantify the fidelity and accuracy of the translation to be evaluated. It is assumed that the SRC-on ly indicator model takes into account factors such as vocabulary matching, grammatical structure consistency, and semantic retention, and combines these factors into a first quality score between 0 and 1. For example, if the SRC-on ly indicator model believes that the translation to be evaluated is very close to the source text in all key aspects, it may give a higher first quality score, such as 0.95.

[0101] In some implementations, the SRC-on ly indicator model may also score the quality of the source text and the translation to be evaluated based on other factors. In addition to the SRC-on ly indicator model, other methods may be used to determine the first quality score, such as the Sequence-to-Sequence model of the attention mechanism, the BERT-based scoring model, or a statistical method, which is not specifically limited in the present application embodiment.

[0102] In some implementations, in order to quantify the accuracy, fluency and fidelity of the translation of the translation to be evaluated, the first target feature in the translation source text and the second target feature in the translation to be evaluated may be compared to determine their alignment index, and the first quality score of the translation to be evaluated may be determined according to the alignment index, thereby improving the accuracy of the translation to be evaluated. For example, step 102 may include:

[0103] (102.1) extracting a first target feature from the translation source text, wherein the first target feature includes at least one of a grammatical feature, a sentence feature, and a semantic feature;

[0104] (102.2) extracting a second target feature from the translation to be evaluated, wherein a feature attribute of the second target feature corresponds to a feature attribute of the first target feature;

[0105] (102.3) Calculate an alignment index between the first target feature and the corresponding second target feature, and determine a first quality score of the translation to be evaluated based on the alignment index.

[0106] The first target feature may be a key attribute extracted from the translation source text, including at least one of a grammatical feature, a sentence feature, and a semantic feature. For example, the first target feature may be a grammatical feature, or a grammatical feature and a sentence feature, and so on.

[0107] Among them, grammatical features can refer to the characteristics of sentence structure, including parts of speech, collocation of sentence components, tense, voice and other features that represent grammatical norms.

[0108] Among them, sentence features can be the length, complexity, and coherence of the sentence.

[0109] Among them, semantic features can be the expression of concepts, implicit meanings, cultural characteristics, etc.

[0110] Among them, the second target feature can be a feature extracted from the translation to be evaluated, and the second target feature corresponds to the first target feature. For example, if the first target feature is the grammatical feature of the translation source text, then the second target feature is the grammatical feature of the translation to be evaluated; for example, if the first target feature is the grammatical feature, sentence feature and semantic feature of the translation source text, then the second target feature is the grammatical feature, sentence feature and semantic feature of the translation to be evaluated.

[0111] The alignment index may be a quantitative indicator of the degree of alignment between the first target feature of the translation source text and the second target feature of the translation to be evaluated. The alignment index ranges from 0 to 1. The closer the value is to 1, the higher the degree of feature alignment is, that is, the better the quality of the translation to be evaluated is.

[0112] In some embodiments, the first target feature and the second target feature are not limited to grammatical features, sentence features, and semantic features, but may also be other features, as long as they do not deviate from the concept of the present application and conform to the actual translation text quality assessment scenario. The embodiments of the present application do not impose specific limitations on this.

[0113] Exemplarily, in an English-Chinese translation scenario (or a translation scenario of other languages), a specific grammatical structure (such as passive voice), sentence complexity (ratio of long sentences to short sentences), and a specific semantic unit (translation accuracy of idioms or proper nouns) in the translation source text are extracted as the first target feature, and the corresponding second target feature is determined from the translation to be evaluated. Then, the alignment index of these features is calculated using predefined rules or machine learning models (such as the SRC-on ly indicator model), such as by calculating the vocabulary overlap, the similarity of the syntactic structure, and the cosine similarity of the semantic vector, etc. to determine the corresponding alignment index.

[0114] In some embodiments, when there are multiple selected first target features and second target features, the alignment degree index of the first target feature and the corresponding second target feature can be calculated one by one, and the average is used to obtain the final first quality score. Alternatively, a comprehensive alignment degree index can be directly given as the first quality score through the SRC-on ly indicator model.

[0115] By extracting multi-dimensional target features such as grammatical features, sentence features and semantic features from the source text and the translation to be evaluated respectively, and making a one-to-one comparison, the translation quality of the translation to be evaluated can be comprehensively and accurately evaluated.

[0116] Step 103: scoring the quality of the translation to be evaluated by using the reference translation to obtain a second quality score.

[0117] In some implementations, in order to improve the comprehensiveness and accuracy of the evaluation of the translation to be evaluated, in addition to scoring the quality of the translation to be evaluated by translating the source text, the quality of the translation to be evaluated can also be scored by using the reference translation, so that the quality evaluation result is more comprehensive and objective.

[0118] The second quality score may be a REF-only indicator of the translation to be evaluated, that is, a score evaluated only based on the comparison between the reference translation and the translation to be evaluated during the evaluation of the translation to be evaluated.

[0119] Exemplarily, vector encoding may be performed on the reference translation and the corresponding translation to be evaluated respectively, and the semantic similarity between the reference translation and the translation to be evaluated is calculated based on the vector encoding, and the semantic similarity is used as the second quality score of the translation to be evaluated.

[0120] In some embodiments, in addition to the REF-on ly indicator, the second quality score of the translation to be evaluated can also be determined by evaluating the number of editing operations (such as insertion, deletion, replacement, etc.) required to improve the translation to be evaluated so that it matches the reference translation. Other evaluation techniques can also be used to score the quality of the translation to be evaluated through the reference translation to obtain the second quality score, and the embodiments of the present application do not specifically limit this.

[0121] By scoring the quality of the translation to be evaluated through the reference translation, the degree of fit between the translation to be evaluated and the standard reference translation can be evaluated, which enhances the objectivity and accuracy of the evaluation.

[0122] In some implementations, in order to more systematically evaluate the quality of the translation, the semantic similarity between the reference translation and the translation to be evaluated may be calculated to determine the second quality score, so as to improve the accuracy of evaluating the translation to be evaluated. For example, step 103 may include:

[0123] (103.1) performing vector encoding on the reference translation to obtain a first vector encoding;

[0124] (103.2) performing vector encoding on the translation to be evaluated to obtain a second vector encoding;

[0125] (103.3) Calculate the semantic similarity between the first vector encoding and the corresponding second vector encoding, and determine a second quality score of the translation to be evaluated based on the semantic similarity.

[0126] The first vector code may be a set of digital vectors obtained by converting the reference translation. The first vector code is used to characterize the semantic features of the reference translation, and each word or entire sentence of the reference translation has a corresponding first vector code in a multidimensional space.

[0127] The second vector code may be a set of digital vectors obtained by converting the translation to be evaluated. The first vector code is used to characterize the semantic features of the translation to be evaluated, and each word or entire sentence of the translation to be evaluated has a corresponding second vector code in a multidimensional space.

[0128] The semantic similarity may be a quantitative indicator for measuring the semantic similarity between the first vector codes and the second vector codes.

[0129] It is understandable that in order to make the translation to be evaluated have a comparable representation with the reference translation in the same vector space, the reference translation and the translation to be evaluated can be encoded using the same vector space model and method to provide a basis for subsequent calculation of semantic similarity.

[0130] Exemplarily, when calculating the semantic similarity between the first vector code and the second vector code, commonly used calculation methods include cosine similarity, Euclidean distance or more complex semantic matching models, such as the REF-on ly indicator model. By comparing the similarity between the first vector code and the second vector code, the degree of closeness between the translation to be evaluated and the reference translation in terms of retaining the original meaning, fluency of expression and contextual adaptability can be evaluated. High semantic similarity means that the translation to be evaluated has a high degree of match in meaning with the reference translation, which usually indicates a higher translation quality. Conversely, low similarity may indicate that the translation to be evaluated has problems such as inaccurate translation or unauthentic expression.

[0131] It can be understood that the first mass fraction and the second mass fraction are both between 0 and 1. If they are not within this range, they can be normalized.

[0132] By converting the reference translation and the translation to be evaluated into vector encoding and then calculating the semantic similarity between the two, a more objective and quantitative evaluation method for the translation to be evaluated can be provided, thereby improving the accuracy of the quality evaluation results.

[0133] Step 104, according to the reference quality of the reference translation, the weight of the first quality score is adjusted to obtain a first target quality score, and the weight of the second quality score is adjusted to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases.

[0134] In some implementations, when the reference translation is of high quality, it is more reliable as an evaluation benchmark, and the weight of the first quality score (based on the quality evaluation score of the translated source text) is reduced to ensure that the quality evaluation result is closer to the actual perception of the translation quality. Conversely, if the reference translation is of low quality, the weight of the first quality score can be increased to maintain the objectivity and accuracy of the evaluation.

[0135] The first target quality score may be a result of adjusting the first quality score obtained based on the evaluation of the translation source text after taking into account the influence of the reference translation quality.

[0136] The second target quality score can be the result of adjusting the second quality score obtained based on the reference translation evaluation after also considering the influence of the reference translation quality. Its adjustment logic is similar to that of the first target quality score, but it is adjusted by multiplying the second quality score by the complement of the target weight, which means that when the quality of the reference translation is high, the weight of the second quality score increases, and more emphasis is placed on the evaluation criteria of the reference translation.

[0137] For example, the target weight of the translated source text may be determined based on the reference quality of the reference translation, and then the target weight is multiplied by the original first quality score and the second quality score, respectively, to obtain a new first target quality score and a new second target quality score that reflect the influence of the reference translation quality.

[0138] By applying the reference quality to both the first quality score and the second quality score, a more accurate and comprehensive translation text quality evaluation can be achieved, thereby improving the practicality and flexibility of the evaluation system.

[0139] In some implementations, since the quality of the reference translation directly determines whether the reference translation can well represent the expected translation effect, the target weight can be determined according to the reference quality of the reference translation, and the first quality score can be adjusted by the target weight to more reasonably evaluate the quality of the translation to be evaluated based on reference translations of different qualities. For example, "adjusting the weight of the first quality score according to the reference quality of the reference translation to obtain the first target quality score" in step 104 may include:

[0140] (104.a1) Determine the target weight of the source text for translation based on the reference quality of the reference translation;

[0141] (104.a2) Obtain a first target quality score according to the product of the target weight and the first quality score.

[0142] The target weight may be an adjustment factor determined based on the reference quality of the reference translation. The target weight may be used to reflect the reliability of the reference translation and the importance it should be given when evaluating the translation to be evaluated.

[0143] Specifically, when the quality of the reference translation is high, the target weight tends to be smaller, which means that in the comprehensive evaluation, the influence of the first quality score directly based on the translated source text will be relatively reduced, and the reference translation will be more trusted. On the contrary, if the quality of the reference translation is low, the target weight may be larger to maintain the emphasis on the evaluation of the fidelity of the translated source text and ensure the objectivity of the evaluation.

[0144] Exemplarily, the target weight w can be calculated by the following formula:

[0145] w=α*q+β

[0146] Wherein, q represents the reference quality of the reference translation, α is the first weight coefficient, β is the second weight coefficient, α and β are linear coefficients and bias, and α and β can be set according to the indicators used in calculating the first quality score and the second quality score, such as the SRC-on ly indicator when calculating the first quality score and the REF-on ly indicator when calculating the second quality score, or α and β can be set according to actual conditions.

[0147] For example, the first target quality score score 1 It can be calculated by the following formula:

[0148] score 1 =w*s 1

[0149] Among them, w represents the target weight, s 1 Represents the first mass fraction.

[0150] By multiplying the target weight by the first quality score, the first target quality score is determined, which can ensure the rationality and accuracy of the evaluation results regardless of whether the reference translation is of high quality or low quality, and is conducive to accurately obtaining the quality evaluation results of the translation to be evaluated in the future.

[0151] In some implementations, since the quality of the reference translation directly determines whether the reference translation can well represent the expected translation effect, the target weight can be determined according to the reference quality of the reference translation, and the second quality score can be adjusted by the complement of the target weight, so as to more reasonably evaluate the quality of the translation to be evaluated based on reference translations of different qualities. For example, "adjusting the weight of the second quality score to obtain the second target quality score" in step 104 can include:

[0152] (104.b1) Get the complement of the target weight;

[0153] (104.b2) A second target quality score is obtained according to the product of the complement of the target weight and the second quality score.

[0154] The complement of the target weight may be a supplementary adjustment value made to the target weight in order to balance the adjustment of the first quality score when calculating the second quality score. The sum of the target weight and the complement of the target weight is 1.

[0155] It is understandable that setting the complement of a table weight can ensure that the adjustment of the first quality score and the second quality score can complement each other in the overall evaluation process, maintaining the balance and rationality of the quality evaluation. When the target weight is reduced due to the high quality of the reference translation, its complement will be used to increase the weight of the second quality score, strengthening the influence of the reference translation quality on the evaluation, and vice versa.

[0156] Exemplarily, the complement of the target weight may be determined by the difference between 1 and the target weight.

[0157] For example, the second target quality score score 2 It can be calculated by the following formula:

[0158] score 2 =(1-w)*s 2

[0159] Where w represents the target weight, (1-w) represents the complement of the target weight, and s 2 Represents the second mass fraction.

[0160] By multiplying the complement of the target weight and the second quality score to determine the second target quality score, it is possible to ensure the rationality and accuracy of the evaluation results regardless of whether the reference translation is of high quality or low quality, which is conducive to the subsequent accurate acquisition of the quality evaluation results of the translation to be evaluated.

[0161] In some implementations, in order to adaptively adjust the evaluation strategy among reference translations of different quality levels, when the quality of the reference translation is high, the target weight is reduced, and when the quality of the reference translation is low, the target weight is increased to ensure the objectivity and accuracy of the evaluation. For example, (104.a1) may include:

[0162] (104.a1.1) Obtain a first weight coefficient and a second weight coefficient;

[0163] (104.a1.2) obtaining a first intermediate coefficient by multiplying the reference mass by the first weight coefficient;

[0164] (104.a1.3) Based on the sum of the first intermediate coefficient and the second weight coefficient, a target weight of the translated source text is obtained.

[0165] The first weight coefficient may be a preset linear coefficient.

[0166] The second weight coefficient may be a preset bias.

[0167] The first intermediate coefficient may be a temporary coefficient generated during the calculation process.

[0168] Exemplarily, the target weight w can be calculated by the following formula:

[0169] w=α*q+β

[0170] Wherein, q represents the reference quality of the reference translation, α*q is the first intermediate coefficient, α is the first weight coefficient, β is the second weight coefficient, α and β are linear coefficients and bias, and α and β can be set according to the indicators used in calculating the first quality score and the second quality score, such as the SRC-on ly indicator when calculating the first quality score and the REF-on ly indicator when calculating the second quality score, or α and β can be set according to actual conditions.

[0171] By determining the target weight corresponding to the translated source text, that is, adjusting the target weight of the first quality score, the contribution of the first quality score and the second quality score can be adjusted, thereby achieving a more efficient, more accurate and more adaptable evaluation mechanism in translation quality evaluation, ensuring the flexibility and effectiveness of translation quality control.

[0172] Step 105: obtaining a quality evaluation result of the translation to be evaluated based on the first target quality score and the second target quality score.

[0173] In some implementations, in order to comprehensively evaluate the translation to be evaluated, the translation quality evaluation from two different perspectives may be comprehensively considered, that is, the first target quality score and the second target quality score are added together, thereby providing a more objective and adaptable comprehensive quality evaluation result for the translation to be evaluated.

[0174] The quality assessment result may be a value or grade obtained by comprehensive calculation based on the first target quality score and the second target quality score, and is used to comprehensively and objectively reflect the translation quality of the translation to be assessed.

[0175] For example, when the quality assessment result is a numerical value, the quality assessment result can be represented by a score. Specifically, the quality assessment result is calculated as follows:

[0176] score=w*s 1 +(1-w)*s 2

[0177] Among them, w is the target weight, s 1 represents the first mass fraction, s 2 Represents the second mass fraction.

[0178] Or, score = score 1 +score 2 , score 1 Indicates the first target quality score, score 2 Represents the second target quality score.

[0179] In some implementations, the first target quality score and the second target quality score may be added to obtain a value as a quality evaluation result of the translation to be evaluated; or the quality evaluation level of the translation to be evaluated, such as high quality, may be determined based on the value obtained by adding the first target quality score and the second target quality score. The embodiment of the present application does not limit the form of the quality evaluation result.

[0180] In the embodiment of the present application, a translation source text and a reference translation of a translation to be evaluated are obtained, and quality evaluation of the reference translation is performed based on the translation source text to obtain a reference quality of the reference translation; quality scoring of the translation to be evaluated is performed based on the translation source text to obtain a first quality score; quality scoring of the translation to be evaluated is performed based on the reference translation to obtain a second quality score; according to the reference quality of the reference translation, the weight of the first quality score is adjusted to obtain a first target quality score, and the weight of the second quality score is adjusted to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases; and based on the first target quality score and the second target quality score, a quality evaluation result of the translation to be evaluated is obtained. In this way, by considering the different qualities of different reference translations, the reference quality of the corresponding reference translation can be obtained, and then the first quality score corresponding to the translated source text and the second quality score corresponding to the reference translation can be adjusted according to the reference quality, so that when the reference quality of the reference translation is high, the weight of the first quality score is lower, and when the reference quality of the reference translation is low, the weight of the first quality score is higher, thereby dynamically adjusting the contribution of the reference quality to the quality evaluation of the translation to be evaluated. The present application can achieve accurate evaluation of the quality of the translation to be evaluated under reference translations of different qualities.

[0181] See also Figure 3 , this application combines Figure 3 The overall process of the translation text quality assessment method is introduced. Exemplarily, first, the reference quality of the reference translation can be calculated, and the SRC-on ly indicator can be selected to score the quality of the translation to be evaluated by translating the source text to obtain a first quality score; further, the REF-on ly indicator can be selected to calculate the quality score of the translation to be evaluated by the reference translation to obtain a second quality score. Furthermore, based on the reference quality of the reference translation, the target weight of the first quality score can be adjusted, the first quality score can be calculated to obtain a first target quality score, and the second quality score can be calculated to obtain a second target quality score, and finally the first target quality score and the second target quality score can be added together to obtain the quality assessment result of the translation to be evaluated. Alternatively, the quality assessment level of the translation to be evaluated can also be determined based on the value obtained by adding the first target quality score and the second target quality score.

[0182] In the embodiment of the present application, a translation source text and a reference translation of a translation to be evaluated are obtained, and quality evaluation of the reference translation is performed based on the translation source text to obtain a reference quality of the reference translation; quality scoring of the translation to be evaluated is performed based on the translation source text to obtain a first quality score; quality scoring of the translation to be evaluated is performed based on the reference translation to obtain a second quality score; according to the reference quality of the reference translation, the weight of the first quality score is adjusted to obtain a first target quality score, and the weight of the second quality score is adjusted to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases; and based on the first target quality score and the second target quality score, a quality evaluation result of the translation to be evaluated is obtained. In this way, by considering the different qualities of different reference translations, the reference quality of the corresponding reference translation can be obtained, and then the first quality score corresponding to the translated source text and the second quality score corresponding to the reference translation can be adjusted according to the reference quality, so that when the reference quality of the reference translation is high, the weight of the first quality score is lower, and when the reference quality of the reference translation is low, the weight of the first quality score is higher, thereby dynamically adjusting the contribution of the reference quality to the quality evaluation of the translation to be evaluated. The present application can achieve accurate evaluation of the quality of the translation to be evaluated under reference translations of different qualities.

[0183] See also Figure 4 The present application also provides a translation text quality assessment device, which can implement the above-mentioned translation text quality assessment method. The translation text quality assessment device includes:

[0184] An evaluation module 41 is used to obtain a translation source text and a reference translation of a translation to be evaluated, and to perform a quality evaluation on the reference translation based on the translation source text to obtain a reference quality of the reference translation;

[0185] A first scoring module 42 is used to score the quality of the translation to be evaluated by translating the source text to obtain a first quality score;

[0186] A second scoring module 43 is used to score the quality of the translation to be evaluated by using the reference translation to obtain a second quality score;

[0187] An adjustment module 44 is used to adjust the weight of the first quality score according to the reference quality of the reference translation to obtain a first target quality score, and to adjust the weight of the second quality score to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases;

[0188] The acquisition module 45 is used to obtain a quality evaluation result of the translation to be evaluated based on the first target quality score and the second target quality score.

[0189] The specific implementation of the translation text quality assessment device is basically the same as the specific implementation of the translation text quality assessment method described above, and will not be repeated here. On the premise of meeting the requirements of the embodiments of this application, the translation text quality assessment device can also be provided with other functional modules to implement the translation text quality assessment method in the above embodiments.

[0190] The present application also provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned translation text quality assessment method when executing the computer program. The computer device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0191] See also Figure 5 , Figure 5 The hardware structure of a computer device according to another embodiment is shown, and the computer device includes:

[0192] The processor 51 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0193] The memory 52 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 52 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 52, and the processor 51 calls and executes the translation text quality assessment method of the embodiment of this application;

[0194] Input / output interface 53, used to implement information input and output;

[0195] The communication interface 54 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WI FI, Bluetooth, etc.);

[0196] A bus 55 that transmits information between the various components of the device (e.g., the processor 51, the memory 52, the input / output interface 53, and the communication interface 54);

[0197] The processor 51 , the memory 52 , the input / output interface 53 and the communication interface 54 are connected to each other in communication within the device via a bus 55 .

[0198] The embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned translation text quality assessment method is implemented.

[0199] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0200] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0201] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0202] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0203] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0204] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0205] It should be understood that in the present application, "at least one (item)" and "several" refer to one or more, and "plurality" refers to two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0206] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0207] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0208] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0209] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0210] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for evaluating the quality of a translation text, characterized in that: The method comprises: Obtaining a translation source text and a reference translation of the translation to be evaluated, and performing a quality evaluation on the reference translation based on the translation source text to obtain a reference quality of the reference translation; Scoring the quality of the translation to be evaluated by using the translation source text to obtain a first quality score; Scoring the quality of the translation to be evaluated by using the reference translation to obtain a second quality score; According to the reference quality of the reference translation, the first quality score is weighted to obtain a first target quality score, and the second quality score is weighted to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases; A quality evaluation result of the translation to be evaluated is obtained based on the first target quality score and the second target quality score.

2. The method for evaluating translation quality according to claim 1, characterized in that: The step of adjusting the first quality score by weight according to the reference quality of the reference translation to obtain a first target quality score includes: Determining a target weight of the translated source text according to the reference quality of the reference translated text; A first target quality score is obtained according to the product of the target weight and the first quality score.

3. The method for evaluating translation text quality according to claim 2, characterized in that: The step of adjusting the weight of the second quality score to obtain a second target quality score includes: Obtaining the complement of the target weight; A second target quality score is obtained according to a product of the complement of the target weight and the second quality score.

4. The method for evaluating translation quality according to claim 2, characterized in that: The step of determining the target weight of the translated source text according to the reference quality of the reference translated text comprises: Obtain a first weight coefficient and a second weight coefficient; Obtaining a first intermediate coefficient by multiplying the reference mass by the first weight coefficient; A target weight of the translation source text is obtained based on the sum of the first intermediate coefficient and the second weight coefficient.

5. The method for evaluating translation text quality according to claim 1, characterized in that: The step of performing quality assessment on the reference translation based on the translated source text to obtain a reference quality of the reference translation includes: Acquire multiple translation source texts from a translation source text collection, and acquire multiple reference translation texts from a reference translation text collection; Comparing each of the translated source texts with the corresponding reference translations to obtain a consistency score for characterizing the quality of the reference translations; The sum of the consistency scores of the reference translations is obtained, and the reference quality of the reference translation is obtained by the ratio of the sum of the consistency scores to the number of the reference translations.

6. The method for evaluating translation text quality according to claim 1, characterized in that: The step of evaluating the quality of the reference translation based on the translated source text to obtain the reference quality of the reference translation further includes: Obtain multiple translation source texts from a translation source text collection, and obtain multiple reference translation texts from a reference translation text collection; For each of the reference translations, the reference translation is compared with the corresponding translation source text to obtain a consistency score for characterizing the quality of different reference translations; The consistency score is used as the reference quality of the corresponding reference translation.

7. The method for evaluating translation quality according to claim 1, characterized in that: The step of scoring the quality of the translation to be evaluated by using the translated source text to obtain a first quality score includes: Extracting a first target feature from the translation source text, wherein the first target feature includes at least one of a grammatical feature, a sentence feature, and a semantic feature; Extracting a second target feature from the translation to be evaluated, wherein a feature attribute of the second target feature corresponds to a feature attribute of the first target feature; An alignment degree index between the first target feature and the corresponding second target feature is calculated, and a first quality score of the translation to be evaluated is determined based on the alignment degree index.

8. The method for evaluating translation quality according to claim 1, characterized in that: The step of scoring the quality of the translation to be evaluated by using the reference translation to obtain a second quality score includes: Performing vector encoding on the reference translation to obtain a first vector encoding; Performing vector encoding on the translation to be evaluated to obtain a second vector encoding; The semantic similarity between the first vector code and the corresponding second vector code is calculated, and a second quality score of the translation to be evaluated is determined based on the semantic similarity.

9. A translation text quality assessment device, characterized in that: The device comprises: An evaluation module, used to obtain a translation source text and a reference translation of a translation to be evaluated, and to perform a quality evaluation on the reference translation based on the translation source text to obtain a reference quality of the reference translation; A first scoring module, configured to score the quality of the translation to be evaluated using the translation source text to obtain a first quality score; A second scoring module, used to score the quality of the translation to be evaluated by using the reference translation to obtain a second quality score; an adjustment module, configured to adjust the weight of the first quality score according to the reference quality of the reference translation to obtain a first target quality score, and to adjust the weight of the second quality score to obtain a second target quality score; wherein the weight of the first quality score decreases as the reference quality increases; An acquisition module is used to obtain a quality evaluation result of the translation to be evaluated based on the first target quality score and the second target quality score.

10. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the translation text quality assessment method according to any one of claims 1 to 8 when executing the computer program.

11. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for evaluating the quality of a translation text according to any one of claims 1 to 8 is implemented.

Images