Translation quality evaluation method and device, electronic equipment and storage medium

Abstract

The invention discloses a translation quality evaluation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a to-be-evaluated task; for each to-be-evaluated translated sentence, evaluating the target to-be-evaluated translated sentence based on a target reference translated sentence set associated with the target source text statement to obtain a sentence level evaluation attribute corresponding to the target to-be-evaluated translated sentence; evaluating each to-be-evaluated segmented word in the target to-be-evaluated translated sentence based on a reference translated word in the target reference translated sentence set to obtain a word level evaluation attribute corresponding to each to-be-evaluated segmented word; and according to the sentence level evaluation attribute of each to-be-evaluated translated sentence and the word level evaluation attribute of each to-be-evaluated segmented word, generating a translated text evaluation report corresponding to the to-be-evaluated task. The effect of more objectively and accurately evaluating the overall translation quality of the translated text and the translation quality of the proprietary entity is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of text translation, and in particular to a method, device, electronic device and storage medium for evaluating translation quality. Background Art

[0002] With the development of globalization, the demand for machine translation in the financial field has increased significantly, especially in professional text processing scenarios such as contracts and financial reports. Therefore, accurate translation based on machine translation is very important.

[0003] Currently, the translation quality of machine translation is mainly evaluated based on supervised learning models with manual annotation. However, this method relies on linguistic features and has accuracy and limitations. Alternatively, translation quality is evaluated through neural networks, but this evaluation method has the problem of inaccurate translation of proprietary entities in specific fields.

[0004] In order to solve the above problems, it is necessary to evaluate the translation quality of machine translation. Summary of the Invention

[0005] The present invention provides a translation quality evaluation method, device, electronic device and storage medium to solve the problem in the prior art that the translation quality evaluation of translated texts does not rely on manual annotation, and when proprietary entities exist in the source text, it is impossible to effectively identify whether the translated text of the proprietary entities meets the standards.

[0006] In a first aspect, an embodiment of the present invention provides a method for evaluating translation quality, comprising:

[0007] Obtaining a task to be evaluated; wherein the task to be evaluated includes a source text, a translation to be evaluated corresponding to the source text, and a reference translation set associated with each source text sentence in the source text, wherein the source text is composed of at least one source text sentence, the translation to be evaluated is composed of translations to be evaluated corresponding to each source text sentence, and the translation to be evaluated and the reference translations in the reference translation set correspond to the target language;

[0008] For each translation to be evaluated, evaluating the target translation to be evaluated based on a target reference translation set associated with the target source text sentence, and obtaining sentence-level evaluation attributes corresponding to the target translation to be evaluated; wherein the target translation to be evaluated corresponds to the target source text sentence;

[0009] Evaluate each to-be-evaluated segmented word in the target to-be-evaluated translation sentence based on the reference translated words in the target reference translated sentence set, and obtain word-level evaluation attributes corresponding to each to-be-evaluated segmented word;

[0010] A translation evaluation report corresponding to the task to be evaluated is generated according to the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each word to be evaluated.

[0011] In a second aspect, an embodiment of the present invention further provides a translation quality evaluation device, comprising:

[0012] a task acquisition module configured to acquire a task to be evaluated; wherein the task to be evaluated includes a source text, a translation to be evaluated corresponding to the source text, and a reference translation set associated with each source text sentence in the source text; the source text is composed of at least one source text sentence, the translation to be evaluated is composed of translations to be evaluated corresponding to each source text sentence, and the translation to be evaluated and the reference translations in the reference translation set correspond to the target language;

[0013] a sentence-level evaluation attribute determination module configured to evaluate each target translation to be evaluated based on a target reference translation set associated with a target source text sentence, and obtain sentence-level evaluation attributes corresponding to the target translation to be evaluated; wherein the target translation to be evaluated corresponds to the target source text sentence;

[0014] a word-level evaluation attribute determination module, configured to evaluate each to-be-evaluated segmented word in the target to-be-evaluated translation sentence based on reference translated words in the target reference translated sentence set, and obtain a word-level evaluation attribute corresponding to each to-be-evaluated segmented word;

[0015] The report generation module is configured to generate a translation evaluation report corresponding to the task to be evaluated based on the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each word to be evaluated.

[0016] In a third aspect, an embodiment of the present invention further provides an electronic device, including:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can perform the method for evaluating translation quality according to any embodiment of the present invention.

[0020] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a processor to implement the method for evaluating translation quality described in any embodiment of the present invention when executed.

[0021] In a fifth aspect, an embodiment of the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method for evaluating translation quality as described in any one of the embodiments of the present invention.

[0022] The technical solution of the embodiment of the present invention obtains a task to be evaluated; for each translation to be evaluated, the target translation to be evaluated is evaluated based on a target reference translation set associated with the target source text, obtaining sentence-level evaluation attributes corresponding to the target translation to be evaluated; each segmented word to be evaluated in the target translation to be evaluated is evaluated based on a reference translation in the target reference translation set, obtaining word-level evaluation attributes corresponding to each segmented word to be evaluated; and a translation evaluation report corresponding to the task to be evaluated is generated based on the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each segmented word to be evaluated. In this technical solution, a source text is translated using a machine translation model to obtain a translation to be evaluated, and a reference translation set corresponding to each source text sentence in the source text is retrieved from a pre-built corpus. The source text, the translation to be evaluated, and at least one associated reference translation set are used to generate a task to be evaluated. Furthermore, the task to be evaluated is input into a translation quality assessment model to evaluate the translation to be evaluated corresponding to the corresponding source text sentence based on the reference translation set corresponding to each source text sentence. It should be noted that when evaluating each translation to be evaluated, the evaluation is performed based on the sentence-level and word-level dimensions, thereby ensuring that the translation to be evaluated is consistent with the corresponding source text sentence at the semantic level, while ensuring the translation accuracy of the key word segmentation in the source text. On this basis, when generating a translation evaluation report, the sentence-level evaluation and word-level evaluation attributes of each translation to be evaluated are displayed, and when the translation quality of the translation to be evaluated does not meet the standards, or the translation quality of the word segmentation in the translation to be evaluated does not meet the standards, it is highlighted, thereby more intuitively displaying the translation quality of the translation to be evaluated to the source text. This solves the problem in the prior art that when evaluating the translation quality of the translated text, manual annotation is not required, and when there are proprietary entities in the source text, the translation text of the proprietary entity cannot be effectively identified as meeting the standards, thereby achieving a more objective and accurate evaluation of the overall translation quality of the translated text and the translation quality of the proprietary entity.

[0023] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 is a flowchart of a translation quality evaluation method provided according to the first embodiment of the present invention;

[0026] Figure 2 This is a flow chart of obtaining web page information provided by the first embodiment of the present invention;

[0027] Figure 3 Schematic diagram of constructing parallel corpus pairs provided in the first embodiment of the present invention;

[0028] Figure 4 is a flowchart of a translation quality evaluation method provided according to the second embodiment of the present invention;

[0029] Figure 5 2 is a schematic structural diagram of a translation quality evaluation device provided according to a third embodiment of the present invention;

[0030] Figure 6 The figure is a schematic diagram of the structure of an electronic device for implementing the method for evaluating translation quality according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention. The acquisition, transmission, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations. It should be noted that in the embodiments of this application, certain software, components, or models that already exist in the industry may be mentioned. They should be regarded as exemplary. Their purpose is only to illustrate the feasibility of the implementation of the technical solution of this application, but it does not mean that the applicant has or must use the solution.

[0032] It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present invention described herein can be practiced in an order other than that illustrated or described herein.

[0033] Example 1

[0034] Figure 1A flowchart of a translation quality assessment method is provided for a first embodiment of the present invention. This embodiment is applicable to generating a task to be assessed based on a source text, a translation to be assessed corresponding to the source text, and a reference translation set corresponding to each source text sentence in the source text. The reference translation set is used to assess the sentences to be assessed in the translation to be assessed at the sentence and word levels to accurately assess the translation quality of the translation to be assessed. Substandard translation quality in the translation to be assessed is highlighted in a translation assessment report to more intuitively display the translation quality of the translation to be assessed. The method can be performed by a translation quality assessment device, which can be implemented in hardware and / or software. The translation quality assessment device can be configured in a computing device capable of executing the translation quality assessment method.

[0035] like Figure 1 As shown, the method includes:

[0036] S110: Obtain tasks to be evaluated.

[0037] The task to be evaluated includes a set of source text sentences, a set of translated sentences to be evaluated in the corresponding target language, and a set of reference translated sentences associated with each source text sentence. The source text sentences in the source text sentence set correspond one-to-one to the translated sentences to be evaluated in the set of translated sentences to be evaluated. The language of the source text is different from the target language, and the reference translated sentences in the set of reference translated sentences are in the target language.

[0038] The source text refers to the text to be translated into a language, and the source text sentence set consists of all source sentences in the source text. The set of translations to be evaluated is composed of the translations to be evaluated after the machine translation model has translated each source sentence. The reference translation is generated by matching the translation in the corpus that is closest to each source sentence. The reference translation includes the reference translation set corresponding to all source sentences.

[0039] For example, in a Chinese-English translation, a Chinese source text is fed into a machine translation model to produce an English translation. The translation quality of the translation can be assessed based on factors such as lexical match, syntactical consistency, semantic fidelity, and textual coherence.

[0040] Based on this, before evaluating the translation to be evaluated, it is necessary to obtain at least one reference translation associated with the source text, so as to evaluate the translation quality of the translation to be evaluated based on each reference translation. The reference translation is an English text that matches the source text. When determining the reference translation of the source text, each source text sentence can be matched from a pre-built corpus to obtain a set of reference translations associated with each source text sentence. A reference translation is then obtained based on the reference translations of all source text sentences. The source text, the translation to be evaluated corresponding to the source text, and at least one reference translation associated with the source text are combined to generate a task to be evaluated.

[0041] Optionally, before obtaining the task to be evaluated, the method further includes: inputting the source text into a machine translation model to obtain a translation to be evaluated corresponding to the source text; for each source text sentence in the source text, determining at least one reference translation sentence associated with the current source text sentence from a pre-built corpus, and generating a reference translation sentence set corresponding to the current source text sentence based on all the reference translation sentences.

[0042] In a specific example, Figure 2 As shown, when the corpus is pre-built, multilingual text information can be obtained from web pages on the Internet in a legal and compliant manner. Specifically, in order to ensure the legitimacy of the data source, a web page driver is used to open the resource locator (Uniform Resource Identifier, URL) of the web page to check the web page acquisition protocol. If it is determined based on the web page acquisition protocol that the web page contains protocol content that prohibits acquisition, then the web page text information in the web page is prohibited from being acquired. On the contrary, for web pages that can be used to acquire web page content, the web page content can be parsed based on a web page content extraction tool to remove unnecessary scripts and tags, and only retain text files. For example, the mouse and keyboard are simulated to acquire web page content, and a web page analyzer is used to extract valid information from the web page. Among them, the web page is a multilingual web page. The so-called multilingual web page refers to a web page system that uses technical means to realize the presentation of the same website content in multiple language versions. Its core value lies in breaking through language barriers and providing a localized browsing experience for global users.

[0043] On this basis, in order to accurately evaluate the translation quality of the translation to be evaluated, the translation to be evaluated corresponding to each source text sentence can be evaluated.

[0044] It should be noted that before evaluating each translation to be evaluated, it is necessary to determine a reference translation set associated with each source sentence, so that the corresponding translation to be evaluated can be evaluated based on each reference translation set. Taking the current source sentence as an example, determining at least one reference translation associated with the current source sentence and generating a reference translation set corresponding to the current source sentence based on all reference translations involves: determining and ranking at least one candidate translation associated with the current source sentence based on the sentence similarity between the current source sentence and each sentence in the corpus; determining at least one reference translation associated with the current source sentence from the at least one candidate translation based on the ranking result, and generating a reference translation set corresponding to the current source sentence based on all reference translations.

[0045] Sentence similarity is determined based on the similarity between the sentence vector of the current source sentence and the sentence vectors of sentences in the corpus. Candidate translations can be understood as sentences in the corpus that are related to the current source sentence.

[0046] Taking the current source sentence as an example, the current source sentence and each sentence in the corpus are vectorized to obtain a first sentence vector corresponding to the current source sentence and a second sentence vector corresponding to each sentence in the corpus. A similarity algorithm is used to calculate the vector similarity between the first sentence vector and each second sentence vector. Based on this vector similarity, the sentence similarity between each sentence in the corpus and the current source sentence is determined.

[0047] It should be noted that the corpus in this technical solution is a pre-built data repository that stores bilingual parallel corpus pairs, financial entities, and all word importance assessment attributes. A bilingual parallel corpus pair can be understood as a text dataset that is strictly aligned between two languages ​​at the sentence, paragraph, or document level.

[0048] In a specific example, Figure 3 As shown, an unsupervised multilingual parallel language pair alignment tool is used to train on an existing unaligned multilingual database to obtain a preliminary aligned bilingual parallel corpus pair. Since the preliminary aligned bilingual parallel corpus still has a large amount of noise, the bilingual parallel corpus is screened again using tools such as text length, text content, bilingual similarity or text tagging specifications to obtain the final corpus. After obtaining the aligned bilingual parallel corpus pairs, a part-of-speech tagging tool is used to perform word segmentation and part-of-speech tagging on the bilingual parallel corpus pairs in the corpus. Furthermore, the word segmentation and part-of-speech tagging results can be used to screen out financial entities that frequently appear in the financial field. For example, financial entities include company names, financial product names, financial data, etc., and after obtaining the financial entities, the corresponding interpretations and accurate translations can be obtained.

[0049] Based on part-of-speech tagging, evaluating the translation quality of the translation to be evaluated requires not only assessing the overall translation quality of the translation, but also ensuring the accuracy of the translation of the key words in the translation to be evaluated. Therefore, the source text and reference translation in the corpus are segmented separately, and the TF-IDF algorithm is used to calculate the importance evaluation attribute of each segmentation and associate it with the corresponding segmentation.

[0050] On this basis, to further select reference translations that best match the current source sentence, the candidate sentences are ranked according to sentence similarity, and the top K candidate translations are selected as reference translations. Based on these K reference translations, a reference translation set corresponding to the current source sentence is generated. This arrangement has the advantage that, since the corpus stores a large amount of data, using all sentences in the corpus as reference translations for the current source sentence would be a significant amount of redundant work. Therefore, by using a preset similarity threshold, candidate translations that meet the threshold for sentence similarity with the current source sentence are retrieved from the corpus. Furthermore, based on the sentence similarity between the candidate translations and the source sentence, a reference translation set corresponding to the source sentence is determined from at least one candidate translation. This reduces the workload of evaluating the translation to be evaluated based on the reference translation set, improving evaluation efficiency.

[0051] Optionally, determining at least one candidate translation sentence associated with the current source sentence based on sentence similarity between the current source sentence and each sentence in the corpus includes: determining a source sentence vector corresponding to the current source sentence; calculating sentence similarity between the source sentence vector and sentence vectors corresponding to each sentence in the corpus based on a cosine similarity algorithm, and sorting the result; and determining at least one candidate translation sentence corresponding to the current source sentence based on the sorting result.

[0052] Exemplarily, a vector transformation is performed on the current source sentence to obtain a source sentence vector corresponding to the current source sentence. Simultaneously, a vector transformation is performed on each sentence in the corpus using the same vector transformation method to obtain a sentence vector corresponding to each sentence. Based on this, the cosine similarity between the current source sentence vector and each sentence vector in the corpus is calculated one by one to determine the sentence similarity between the current source sentence and each sentence. The sentences in the corpus are sorted according to the sentence similarity, and sentences with a similarity greater than a preset threshold are selected as candidate translations for the current source sentence.

[0053] Optionally, determining the source text sentence vector corresponding to the current source text sentence includes: performing word segmentation processing on the current source text sentence to obtain at least one word segmentation to be processed; determining at least one key word segmentation corresponding to the current source text sentence based on the importance evaluation attribute of each word segmentation to be processed; and obtaining the source text sentence vector corresponding to the current source text sentence based on the average value of the word segmentation vectors of all key word segmentations.

[0054] The importance assessment attribute can be used to characterize the uniqueness of the pending word in the current source text. A higher importance assessment attribute indicates a higher uniqueness of the pending word and a higher degree of importance within the document. Key words are pending words with high importance assessment attributes. For example, if there are 10 pending words and three are identified as key words, after calculating the importance assessment attributes for each pending word, the top three pending words will be considered key words.

[0055] For example, taking the current source text sentence "Please translate the source text into English from Chinese" as an example, the current source text sentence is segmented based on the word segmentation tool, and the segmented words to be processed are: please, yes, source text, perform, Chinese to English translation, and convert. Each segmented word to be processed is vectorized to obtain a segmented word vector corresponding to each segmented word to be processed. On this basis, the importance evaluation attribute of each segmented word to be processed is calculated using the TF-IDF algorithm, and each segmented word to be processed is sorted according to the importance evaluation attribute to obtain the topN key segmented words. Furthermore, each key segmented word is vectorized to obtain a segmented word vector corresponding to each key segmented word, and based on the average value of the segmented word vectors of all key segmented words, the source text sentence vector corresponding to the current source text sentence is obtained.

[0056] S120 : For each translation to be evaluated, evaluate the target translation to be evaluated based on a target reference translation set associated with the target source text sentence, and obtain sentence-level evaluation attributes corresponding to the target translation to be evaluated.

[0057] The target translation sentence to be evaluated corresponds to the target source text sentence. The target source text sentence is the source text sentence currently requiring translation quality evaluation, and the target reference translation sentence set refers to a collection of at least one reference translation sentence associated with the target source text sentence in the corpus. The sentence-level evaluation attribute can be used to characterize the translation quality of the target translation sentence to be evaluated corresponding to the target source text sentence. A larger sentence-level evaluation attribute indicates a better translation quality of the target evaluation translation sentence, while a smaller sentence-level evaluation attribute indicates a worse translation quality of the target evaluation translation sentence.

[0058] In this technical solution, when evaluating the translation quality of a translation to be evaluated, each translation sentence to be evaluated needs to be evaluated, and the translation sentences to be evaluated correspond one-to-one to the source text sentences in the source text. Therefore, the reference translation sentence set corresponding to the target source text sentence is also the reference translation sentence set corresponding to the target translation sentence to be evaluated.

[0059] Based on this, when evaluating the target translation, the quality of the target translation is assessed using a set of reference translations corresponding to the target source text. For example, the semantic similarity or sentence vector similarity between the target translation and each reference translation in the reference translation set is detected to evaluate the translation quality of the target translation.

[0060] S130 : Evaluate each to-be-evaluated segmented word in the target to-be-evaluated translation sentence based on the reference translated words in the target reference translated sentence set, and obtain word-level evaluation attributes corresponding to each to-be-evaluated segmented word.

[0061] Among them, word-level evaluation attributes can be used to characterize the translation quality of the word segmentation to be evaluated.

[0062] In practical applications, when translating a source text, it is necessary not only to ensure the overall translation quality of each source text sentence, but also to ensure the accuracy of the translation of each word in the source text sentence. Only when the overall translation quality of the translation sentence to be evaluated corresponding to the source text sentence and the translation quality of each word meet the translation requirements, the translation quality of the translation sentence to be evaluated is optimal.

[0063] S140 : Generate a translation evaluation report corresponding to the task to be evaluated based on the sentence-level evaluation attributes of each translation sentence to be evaluated and the word-level evaluation attributes of each word segment to be evaluated.

[0064] Specifically, after performing sentence-level and word-level evaluations on each translation to be evaluated, a translation evaluation report corresponding to the task to be evaluated may be generated to more intuitively demonstrate the translation quality of the translation to be evaluated.

[0065] The technical solution of the embodiment of the present invention obtains a task to be evaluated; for each translation to be evaluated, the target translation to be evaluated is evaluated based on a target reference translation set associated with the target source text, obtaining sentence-level evaluation attributes corresponding to the target translation to be evaluated; each segmented word to be evaluated in the target translation to be evaluated is evaluated based on a reference translation in the target reference translation set, obtaining word-level evaluation attributes corresponding to each segmented word to be evaluated; and a translation evaluation report corresponding to the task to be evaluated is generated based on the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each segmented word to be evaluated. In this technical solution, a source text is translated using a machine translation model to obtain a translation to be evaluated, and a reference translation set corresponding to each source text sentence in the source text is retrieved from a pre-built corpus. The source text, the translation to be evaluated, and at least one associated reference translation set are used to generate a task to be evaluated. Furthermore, the task to be evaluated is input into a translation quality assessment model to evaluate the translation to be evaluated corresponding to the corresponding source text sentence based on the reference translation set corresponding to each source text sentence. It should be noted that when evaluating each translation to be evaluated, the evaluation is performed based on the sentence-level and word-level dimensions, thereby ensuring that the translation to be evaluated is consistent with the corresponding source text sentence at the semantic level, while ensuring the translation accuracy of the key word segmentation in the source text. On this basis, when generating a translation evaluation report, the sentence-level evaluation and word-level evaluation attributes of each translation to be evaluated are displayed, and when the translation quality of the translation to be evaluated does not meet the standards, or the translation quality of the word segmentation in the translation to be evaluated does not meet the standards, it is highlighted, thereby more intuitively displaying the translation quality of the translation to be evaluated to the source text. This solves the problem in the prior art that when evaluating the translation quality of the translated text, manual annotation is not required, and when there are proprietary entities in the source text, the translation text of the proprietary entity cannot be effectively identified as meeting the standards, thereby achieving a more objective and accurate evaluation of the overall translation quality of the translated text and the translation quality of the proprietary entity.

[0066] Example 2

[0067] Figure 4 This is a flowchart of a translation quality evaluation method provided in a second embodiment of the present invention. Optionally, the target translation to be evaluated is evaluated based on a target reference translation set associated with a target source sentence, and sentence-level evaluation attributes corresponding to the target translation to be evaluated are refined.

[0068] like Figure 4 As shown, the method includes:

[0069] S210: Obtain tasks to be evaluated.

[0070] S220 : Based on semantic analysis, determine semantic evaluation attributes between the target translation to be evaluated and each reference translation in the target reference translation set.

[0071] The semantic evaluation attribute is used to represent the semantic similarity between the target translation and the sentences in the corpus. The larger the semantic evaluation attribute, the higher the similarity between the sentences.

[0072] In one specific example, a task to be evaluated includes a source text, a translation to be evaluated, and a set of reference translations associated with each source text sentence. The task to be evaluated is then fed into a pre-built translation quality assessment model, which then evaluates the task based on the model. The model performs semantic analysis on each sentence in the task to be evaluated and calculates the semantic similarity between the translation to be evaluated and each reference translation in the set of reference translations.

[0073] Taking the target translation sentence to be evaluated corresponding to the target source text sentence as an example, when evaluating the target translation sentence to be evaluated based on the translation quality assessment model, the target reference translation sentence set corresponding to the target source text sentence is retrieved, and semantic analysis is performed on the target translation sentence to be evaluated. At the same time, semantic analysis is performed on each reference translation sentence in the target reference translation sentence set, and the semantic similarity between the target translation sentence to be evaluated and each reference translation sentence is calculated.

[0074] S230 , calculating sentence-level evaluation attributes corresponding to the target translation to be evaluated based on the semantic evaluation attributes between the target translation to be evaluated and each reference translation.

[0075] Among them, the sentence-level evaluation attribute can be used to characterize the translation quality of the target translation sentence to be evaluated to the target source sentence.

[0076] Based on the above example, the interval of the sentence-level evaluation attribute is pre-set to 1-100, so as to evaluate the target translation sentence based on the translation quality evaluation model and obtain the sentence-level evaluation attribute corresponding to the target translation sentence, such as the sentence-level evaluation attribute is 90.

[0077] Exemplarily, the sentence-level evaluation attribute of the target translation to be evaluated may be determined by calculating an average value of at least one semantic evaluation attribute corresponding to the target translation to be evaluated, and using the average value as the sentence-level evaluation attribute of the target translation to be evaluated.

[0078] The advantage of this setting is that each reference translation in the target reference translation set is highly relevant to the target source text and has a high translation quality for the target source text. Therefore, calculating the sentence-level evaluation attributes of the target translation to be evaluated based on the mean of the semantic evaluation attributes of the reference translations can reduce individual errors and improve objectivity.

[0079] Alternatively, when calculating the sentence-level evaluation attribute, the target reference translation with the highest matching degree with the target source sentence in the target reference translation set is also determined, and the sentence-level evaluation attribute is obtained by calculating the semantic similarity between the target translation to be evaluated and the target reference translation.

[0080] The benefit of this setup is that the higher the semantic similarity between the target translation and the reference translation, the higher the semantic similarity between the target translation and the target source text, and accordingly, the better the translation quality of the target translation. Therefore, calculating sentence-level evaluation attributes using this method can speed up sentence evaluation and improve the overall efficiency of evaluating the translation.

[0081] S240 : Evaluate each to-be-evaluated segmented word in the target to-be-evaluated translation sentence based on the reference translated words in the target reference translated sentence set, and obtain word-level evaluation attributes corresponding to each to-be-evaluated segmented word.

[0082] At the same time, in order to ensure the accuracy of the translation of each word in the translation to be evaluated, it is also necessary to evaluate each word in the translation to be evaluated based on the translation quality evaluation model to obtain the corresponding word-level evaluation attributes.

[0083] On this basis, the range of word-level evaluation attributes is set to 1-5, and each to-be-evaluated segmented word is evaluated based on the translation quality evaluation model. Optionally, each to-be-evaluated segmented word in the target to-be-evaluated translation is evaluated based on the reference translated words in the target reference translated sentence set to obtain word-level evaluation attributes corresponding to each to-be-evaluated segmented word, including: performing segmentation processing on the target source text sentence to obtain at least one source text segmented word, and determining the to-be-evaluated segmented word corresponding to each source text segmented word in the to-be-evaluated translation sentence; and evaluating each to-be-evaluated segmented word according to the target reference translated sentence set to obtain word-level evaluation attributes corresponding to each to-be-evaluated segmented word.

[0084] It should be noted that if the source text segmentation in the target source text sentence is a financial entity to be translated, the financial entity translation to be evaluated corresponding to the financial entity to be translated is determined from the translation to be evaluated; a unique financial entity translation corresponding to the financial entity to be translated is determined from a pre-constructed corpus, and the financial entity translation to be evaluated is matched based on the unique financial entity translation; and based on the matching results, the word-level evaluation attribute corresponding to the financial entity to be translated is determined.

[0085] Financial entities refer to various institutions and organizations that play a role in the financial system. For example, financial entities can be financial business institutions, insurance institutions, or securities institutions. The financial entity to be translated refers to the financial entity in the target source text. The financial entity translation to be evaluated refers to the translation text corresponding to the financial entity to be translated in the translation to be evaluated.

[0086] Generally speaking, the translation of a financial entity is unique, meaning that each financial entity corresponds to a unique translation. Therefore, when translating a source text that contains a financial entity to be translated, the accuracy of the translation must be ensured.

[0087] For example, the target source text contains the financial entity to be translated, "A Bank." During the Chinese-to-English translation process, "A Bank" corresponds to the unique financial entity translation, "A BANK." This means that when translating "A Bank," not only does the name need to be translated, but all translated English letters must also be capitalized. Based on this, the corresponding financial entity translation for "A Bank" is first determined from the translation to be evaluated. Only when the translation is "A BANK" is the translation determined to match the entity to be evaluated. It should be noted that even if the translation is "A bank," and even if the translation has the same semantics as the entity to be translated, differences between the translation and the unique translation indicate that the translation does not match the entity to be evaluated.

[0088] Based on this, when the translation of the financial entity to be evaluated matches the financial entity to be evaluated, the word-level evaluation attribute corresponding to the financial entity to be translated is greater; conversely, when the translation of the financial entity to be evaluated does not match the financial entity to be evaluated, the word-level evaluation attribute is lower.

[0089] S250 : Generate a translation evaluation report corresponding to the task to be evaluated based on the sentence-level evaluation attributes of each translation sentence to be evaluated and the word-level evaluation attributes of each word segment to be evaluated.

[0090] In this technical solution, in order to more intuitively present the translation quality of the translation to be evaluated, when generating a translation evaluation report, a translation evaluation report corresponding to the task to be evaluated is generated based on the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each segmented word to be evaluated, including: for each translation to be evaluated, if the sentence-level evaluation attribute corresponding to the translation to be evaluated is less than the preset sentence-level evaluation attribute, then the translation to be evaluated and the translation-related information corresponding to the translation to be evaluated are highlighted; and if the word-level evaluation attribute of the segmented word to be evaluated in the translation to be evaluated is less than the preset word-level evaluation attribute, then the segmented word to be evaluated and the segmented information corresponding to the segmented word to be evaluated are highlighted.

[0091] The sentence-related information includes the sentence-level evaluation attributes corresponding to the sentence to be evaluated, as well as a set of reference sentences corresponding to the sentence to be evaluated. The word-related information includes the word-level evaluation attributes corresponding to the word to be evaluated, as well as a set of reference translated words corresponding to the word to be evaluated.

[0092] In practical applications, after obtaining the sentence-level evaluation attributes corresponding to each translation sentence to be evaluated and the word-level evaluation attributes of each word to be evaluated, a translation evaluation report corresponding to the task to be evaluated can be generated to facilitate review by quality inspectors.

[0093] On this basis, if the sentence-level evaluation attributes of all the translations to be evaluated and the word-level evaluation attributes of all the word segmentations to be evaluated are displayed, it will cause users to be unable to grasp the key points when they are looking up, and they will need to spend a lot of time to check the translation quality of the translation to be evaluated. Therefore, this technical solution pre-sets preset sentence-level evaluation attributes and preset word-level evaluation attributes in the translation evaluation report. When the sentence-level evaluation attribute of the translation to be evaluated is less than the preset sentence-level evaluation attribute, it indicates that the translation quality of the translation to be evaluated is poor. At this time, the translation to be evaluated is marked as abnormal; correspondingly, when the word-level evaluation attribute of the word segmentation to be evaluated is less than the preset evaluation attribute, it indicates that the translation quality of the word segmentation to be evaluated is poor. At this time, the word segmentation to be evaluated is marked as abnormal.

[0094] Based on this, the translation evaluation report can highlight the sentences or words to be evaluated that are marked as abnormal, and display the sentence-related information of the sentences to be evaluated and the word-related information of the words to be evaluated, so as to assist quality inspectors in further evaluating the translation quality of the displayed sentences to be evaluated and word-related information of the words to be evaluated.

[0095] The technical solution of the embodiment of the present invention obtains a task to be evaluated; for each translation to be evaluated, the target translation to be evaluated is evaluated based on a target reference translation set associated with the target source text, obtaining sentence-level evaluation attributes corresponding to the target translation to be evaluated; each segmented word to be evaluated in the target translation to be evaluated is evaluated based on a reference translation in the target reference translation set, obtaining word-level evaluation attributes corresponding to each segmented word to be evaluated; and a translation evaluation report corresponding to the task to be evaluated is generated based on the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each segmented word to be evaluated. In this technical solution, a source text is translated using a machine translation model to obtain a translation to be evaluated, and a reference translation set corresponding to each source text sentence in the source text is retrieved from a pre-built corpus. The source text, the translation to be evaluated, and at least one associated reference translation set are used to generate a task to be evaluated. Furthermore, the task to be evaluated is input into a translation quality assessment model to evaluate the translation to be evaluated corresponding to the corresponding source text sentence based on the reference translation set corresponding to each source text sentence. It should be noted that when evaluating each translation to be evaluated, the evaluation is performed based on the sentence-level and word-level dimensions, thereby ensuring that the translation to be evaluated is consistent with the corresponding source text sentence at the semantic level, while ensuring the translation accuracy of the key word segmentation in the source text. On this basis, when generating a translation evaluation report, the sentence-level evaluation and word-level evaluation attributes of each translation to be evaluated are displayed, and when the translation quality of the translation to be evaluated does not meet the standards, or the translation quality of the word segmentation in the translation to be evaluated does not meet the standards, it is highlighted, thereby more intuitively displaying the translation quality of the translation to be evaluated to the source text. This solves the problem in the prior art that when evaluating the translation quality of the translated text, manual annotation is not required, and when there are proprietary entities in the source text, the translation text of the proprietary entity cannot be effectively identified as meeting the standards, thereby achieving a more objective and accurate evaluation of the overall translation quality of the translated text and the translation quality of the proprietary entity.

[0096] Example 3

[0097] Figure 5 This is a schematic diagram of the structure of a translation quality evaluation device provided by the third embodiment of the present invention. Figure 5 As shown, the apparatus includes: a task acquisition module 310 , a sentence-level evaluation attribute determination module 320 , a word-level evaluation attribute determination module 330 and a report generation module 340 .

[0098] The task acquisition module 310 is configured to acquire a task to be evaluated. The task to be evaluated includes a source text, a translation to be evaluated corresponding to the source text, and a reference translation set associated with each source text sentence in the source text. The source text is composed of at least one source text sentence, and the translation to be evaluated is composed of translations to be evaluated corresponding to each source text sentence. The translation to be evaluated and the reference translations in the reference translation set correspond to the target language.

[0099] A sentence-level evaluation attribute determination module 320 is configured to evaluate each target translation sentence based on a target reference translation set associated with the target source text sentence, and obtain sentence-level evaluation attributes corresponding to the target translation sentence; wherein the target translation sentence corresponds to the target source text sentence;

[0100] A word-level evaluation attribute determination module 330 is configured to evaluate each to-be-evaluated segmented word in the target to-be-evaluated translation based on the reference translated words in the target reference translated sentence set, and obtain a word-level evaluation attribute corresponding to each to-be-evaluated segmented word;

[0101] The report generation module 340 is configured to generate a translation evaluation report corresponding to the task to be evaluated based on the sentence-level evaluation attributes of each translation sentence to be evaluated and the word-level evaluation attributes of each word segment to be evaluated.

[0102] The technical solution of the embodiment of the present invention obtains a task to be evaluated; for each translation to be evaluated, the target translation to be evaluated is evaluated based on a target reference translation set associated with the target source text, obtaining sentence-level evaluation attributes corresponding to the target translation to be evaluated; each segmented word to be evaluated in the target translation to be evaluated is evaluated based on a reference translation in the target reference translation set, obtaining word-level evaluation attributes corresponding to each segmented word to be evaluated; and a translation evaluation report corresponding to the task to be evaluated is generated based on the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each segmented word to be evaluated. In this technical solution, a source text is translated using a machine translation model to obtain a translation to be evaluated, and a reference translation set corresponding to each source text sentence in the source text is retrieved from a pre-built corpus. The source text, the translation to be evaluated, and at least one associated reference translation set are used to generate a task to be evaluated. Furthermore, the task to be evaluated is input into a translation quality assessment model to evaluate the translation to be evaluated corresponding to the corresponding source text sentence based on the reference translation set corresponding to each source text sentence. It should be noted that when evaluating each translation to be evaluated, the evaluation is performed based on the sentence-level and word-level dimensions, thereby ensuring that the translation to be evaluated is consistent with the corresponding source text sentence at the semantic level, while ensuring the translation accuracy of the key word segmentation in the source text. On this basis, when generating a translation evaluation report, the sentence-level evaluation and word-level evaluation attributes of each translation to be evaluated are displayed, and when the translation quality of the translation to be evaluated does not meet the standards, or the translation quality of the word segmentation in the translation to be evaluated does not meet the standards, it is highlighted, thereby more intuitively displaying the translation quality of the translation to be evaluated to the source text. This solves the problem in the prior art that when evaluating the translation quality of the translated text, manual annotation is not required, and when there are proprietary entities in the source text, the translation text of the proprietary entity cannot be effectively identified as meeting the standards, thereby achieving a more objective and accurate evaluation of the overall translation quality of the translated text and the translation quality of the proprietary entity.

[0103] Optionally, the translation quality evaluation device further includes: a translation to be evaluated determination submodule, configured to input the source text into the machine translation model before obtaining the task to be evaluated, to obtain a translation to be evaluated corresponding to the source text;

[0104] The reference translation set determination submodule is configured to determine, for each source sentence in the source text, at least one reference translation associated with the current source sentence from a pre-built corpus, and generate a reference translation set corresponding to the current source sentence based on all the reference translations.

[0105] Optionally, the reference translation set determination submodule includes: a candidate translation determination unit, configured to determine and sort at least one candidate translation associated with the current source sentence based on sentence similarity between the current source sentence and sentences in the corpus;

[0106] The reference translation set determining unit is configured to determine at least one reference translation associated with the current source sentence from at least one candidate translation according to the ranking result, and generate a reference translation set corresponding to the current source sentence based on all the reference translations.

[0107] Optionally, the candidate translation sentence determination unit includes: a vector determination subunit, configured to determine a source text sentence vector corresponding to the current source text sentence;

[0108] A first sorting unit is configured to calculate sentence similarities between the source text sentence vector and the sentence vectors corresponding to each sentence in the corpus based on a cosine similarity algorithm, and sort the sentences;

[0109] The candidate translation determination subunit is configured to determine at least one candidate translation corresponding to the current source text sentence according to the sorting result.

[0110] Optionally, the vector determination subunit is used to perform word segmentation processing on the current source text sentence to obtain at least one word segmentation to be processed; determine at least one key word segmentation corresponding to the current source text sentence based on the importance evaluation attribute of each word segmentation to be processed; and obtain the source text sentence vector corresponding to the current source text sentence based on the average value of the word segmentation vectors of all key words.

[0111] Optionally, the sentence-level evaluation attribute determination module includes: a semantic evaluation attribute determination submodule, configured to determine, based on semantic analysis, the semantic evaluation attributes between the target translation sentence to be evaluated and each reference translation sentence in the target reference translation sentence set;

[0112] The sentence-level evaluation attribute determination submodule is used to calculate the sentence-level evaluation attribute corresponding to the target translation to be evaluated based on the semantic evaluation attributes between the target translation to be evaluated and each reference translation.

[0113] Optionally, the word-level evaluation attribute determination module includes: a to-be-evaluated segmentation determination submodule, configured to perform segmentation processing on the target source text sentence to obtain at least one source segmentation, and determine the to-be-evaluated segmentation corresponding to each source segmentation in the to-be-evaluated translation sentence;

[0114] The first word-level evaluation attribute determination submodule is used to evaluate each to-be-evaluated segmented word according to the target reference translation sentence set, and obtain a word-level evaluation attribute corresponding to each to-be-evaluated segmented word.

[0115] Optionally, the word-level evaluation attribute determination module further includes: a financial entity translation determination submodule for determining a financial entity translation to be evaluated corresponding to the financial entity translation to be translated from the translation to be evaluated if the segmented word in the target source text sentence is a financial entity to be translated;

[0116] A matching submodule is used to determine a unique financial entity translation corresponding to the financial entity to be translated from a pre-built corpus, and match the financial entity translation to be evaluated based on the unique financial entity translation;

[0117] The second word-level evaluation attribute determination submodule is used to determine the word-level evaluation attribute corresponding to the financial entity to be translated based on the matching result.

[0118] Optionally, the report generation module includes: a first display submodule configured to highlight, for each translation to be evaluated, the translation to be evaluated and its corresponding sentence-related information if the sentence-level evaluation attribute corresponding to the translation to be evaluated is less than a preset sentence-level evaluation attribute; wherein the sentence-related information includes the sentence-level evaluation attribute corresponding to the translation to be evaluated and a reference translation set corresponding to the translation to be evaluated; and

[0119] The second display submodule is configured to highlight the segmentation to be evaluated and segmentation association information corresponding to the segmentation to be evaluated if the word-level evaluation attribute of the segmentation to be evaluated in the translated sentence to be evaluated is less than a preset word-level evaluation attribute; wherein the segmentation association information includes the word-level evaluation attribute corresponding to the segmentation to be evaluated and a reference translated word corresponding to the segmentation to be evaluated.

[0120] The translation quality evaluation device provided in the embodiment of the present invention can execute the translation quality evaluation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0121] Example 4

[0122] Figure 6A schematic diagram of the structure of an electronic device 10 of an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0123] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0124] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0125] Processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any other suitable processor, controller, or microcontroller. Processor 11 executes the various methods and processes described above, such as the translation quality assessment method.

[0126] In some embodiments, the translation quality assessment method can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the translation quality assessment method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the translation quality assessment method in any other appropriate manner (e.g., via firmware).

[0127] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0128] Computer programs for implementing the translation quality assessment methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0129] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0130] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0131] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0132] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0133] Example 5

[0134] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the method for evaluating translation quality provided in any embodiment of the present application.

[0135] The computer program product may be implemented by writing computer program code for performing the operations of the present invention in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0136] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0137] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for evaluating translation quality, characterized in that: include: Obtaining a task to be evaluated; wherein the task to be evaluated includes a source text, a translation to be evaluated corresponding to the source text, and a reference translation set associated with each source text sentence in the source text, wherein the source text is composed of at least one source text sentence, the translation to be evaluated is composed of translations to be evaluated corresponding to each source text sentence, and the translation to be evaluated and the reference translations in the reference translation set correspond to the target language; For each translation to be evaluated, evaluating the target translation to be evaluated based on a target reference translation set associated with the target source text sentence, and obtaining sentence-level evaluation attributes corresponding to the target translation to be evaluated; wherein the target translation to be evaluated corresponds to the target source text sentence; Evaluate each to-be-evaluated segmented word in the target to-be-evaluated translation sentence based on the reference translated words in the target reference translated sentence set, and obtain word-level evaluation attributes corresponding to each to-be-evaluated segmented word; A translation evaluation report corresponding to the task to be evaluated is generated according to the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each word to be evaluated.

2. The method according to claim 1, characterized in that Before obtaining the task to be evaluated, the following steps are also included: Inputting the source text into a machine translation model to obtain a translation to be evaluated corresponding to the source text; For each source sentence in the source text, at least one reference translation sentence associated with the current source sentence is determined from a pre-built corpus, and a reference translation sentence set corresponding to the current source sentence is generated based on all the reference translation sentences.

3. The method according to claim 2, characterized in that The determining of at least one reference translation associated with the current source sentence, and generating a reference translation set corresponding to the current source sentence based on all the reference translations, includes: Determining at least one candidate translation sentence associated with the current source sentence based on sentence similarity between the current source sentence and sentences in the corpus, and sorting the candidate translation sentences; According to the ranking result, at least one reference translation associated with the current source sentence is determined from at least one candidate translation, and a reference translation set corresponding to the current source sentence is generated based on all the reference translations.

4. The method according to claim 3, characterized in that The step of determining at least one candidate translation sentence associated with the current source sentence based on sentence similarity between the current source sentence and sentences in the corpus includes: Determining a source text sentence vector corresponding to the current source text sentence; Calculate the sentence similarity between the source text sentence vector and the sentence vector corresponding to each sentence in the corpus based on the cosine similarity algorithm, and sort them; At least one candidate translation sentence corresponding to the current source sentence is determined according to the sorting result.

5. The method according to claim 4, characterized in that Determining the source text sentence vector corresponding to the current source text sentence includes: Performing word segmentation processing on the current source sentence to obtain at least one word segmentation to be processed; Determining at least one key word corresponding to the current source text sentence according to the importance evaluation attribute of each to-be-processed word; Based on the average value of the word segmentation vectors of all key word segmentations, a source text sentence vector corresponding to the current source text sentence is obtained.

6. The method according to claim 1, wherein The step of evaluating the target translation sentence to be evaluated based on the target reference translation sentence set associated with the target source text sentence to obtain sentence-level evaluation attributes corresponding to the target translation sentence to be evaluated includes: Based on semantic analysis, determining semantic evaluation attributes between the target translation to be evaluated and each reference translation in the target reference translation set; According to the semantic evaluation attributes between the target translation to be evaluated and each of the reference translations, a sentence-level evaluation attribute corresponding to the target translation to be evaluated is calculated.

7. The method according to claim 1, characterized in that The step of evaluating each to-be-evaluated segmented word in the target to-be-evaluated translation sentence based on the reference translated words in the target reference translated sentence set to obtain a word-level evaluation attribute corresponding to each to-be-evaluated segmented word includes: Performing word segmentation processing on the target source text sentence to obtain at least one source text segmentation, and determining the segmentation to be evaluated corresponding to each source text segmentation in the translation sentence to be evaluated; Each to-be-evaluated segmented word is evaluated according to the target reference translation set to obtain a word-level evaluation attribute corresponding to each to-be-evaluated segmented word.

8. The method according to claim 7, characterized in that Also includes: If the segmented word in the target source text sentence is a financial entity to be translated, determining a financial entity translation to be evaluated corresponding to the financial entity to be translated from the translation to be evaluated; Determining a unique financial entity translation corresponding to the financial entity to be translated from a pre-built corpus, and matching the financial entity translation to be evaluated based on the unique financial entity translation; According to the matching result, a word-level evaluation attribute corresponding to the financial entity to be translated is determined.

9. The method according to claim 1, characterized in that Generating a translation evaluation report corresponding to the task to be evaluated based on the sentence-level evaluation attributes of each translation sentence to be evaluated and the word-level evaluation attributes of each word segment to be evaluated includes: For each translation to be evaluated, if the sentence-level evaluation attribute corresponding to the translation to be evaluated is less than a preset sentence-level evaluation attribute, the translation to be evaluated and the translation-related information corresponding to the translation to be evaluated are highlighted; wherein the translation-related information includes the sentence-level evaluation attribute corresponding to the translation to be evaluated and a reference translation set corresponding to the translation to be evaluated; and If the word-level evaluation attribute of the to-be-evaluated segmented word in the to-be-evaluated translated sentence is less than a preset word-level evaluation attribute, the to-be-evaluated segmented word and segmented word association information corresponding to the to-be-evaluated segmented word are highlighted; wherein the segmented word association information includes the word-level evaluation attribute corresponding to the to-be-evaluated segmented word and a reference translated word corresponding to the to-be-evaluated segmented word.

10. A translation quality evaluation device, characterized in that: include: a task acquisition module configured to acquire a task to be evaluated; wherein the task to be evaluated includes a source text, a translation to be evaluated corresponding to the source text, and a reference translation set associated with each source text sentence in the source text; the source text is composed of at least one source text sentence, the translation to be evaluated is composed of translations to be evaluated corresponding to each source text sentence, and the translation to be evaluated and the reference translations in the reference translation set correspond to the target language; a sentence-level evaluation attribute determination module configured to evaluate each target translation to be evaluated based on a target reference translation set associated with a target source text sentence, and obtain sentence-level evaluation attributes corresponding to the target translation to be evaluated; wherein the target translation to be evaluated corresponds to the target source text sentence; a word-level evaluation attribute determination module, configured to evaluate each to-be-evaluated segmented word in the target to-be-evaluated translation sentence based on reference translated words in the target reference translated sentence set, and obtain a word-level evaluation attribute corresponding to each to-be-evaluated segmented word; The report generation module is configured to generate a translation evaluation report corresponding to the task to be evaluated based on the sentence-level evaluation attributes of each translation to be evaluated and the word-level evaluation attributes of each word to be evaluated.

Images