Translation quality determination method and device, electronic equipment and computer program product

Abstract

The invention discloses a translation quality determination method and device, electronic equipment and a computer program product. The method relates to the field of artificial intelligence and financial science and technology. The method comprises the steps that a source language text and a target language text obtained through translation are acquired; generating lexical element embedding representation vectors for the source language text and the target language text to obtain a first lexical element vector set and a second lexical element vector set; lexical element matching pairs between the first lexical element vector set and the second lexical element vector set are determined, M lexical element matching pairs are obtained, and each lexical element matching pair comprises a first lexical element vector and a second lexical element vector; determining the translation quality of the source language text according to the M matching degrees of the M lexical element matching pairs, wherein the matching degree of one lexical element matching pair is obtained based on the content similarity and the position similarity between the first lexical element vector and the second lexical element vector in one lexical element matching pair. Through the method and the device, the problem of inaccurate translation quality evaluation in related technologies is solved.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and fintech, and more specifically, to a method, apparatus, electronic device, and computer program product for determining translation quality. Background Technology

[0002] In banking customer service scenarios, customers may communicate using different languages ​​or dialects through hotlines, emails, or online channels. To facilitate understanding of customer needs, translation is required, and to ensure translation accuracy, translation quality assessment is necessary.

[0003] Translation quality assessment methods in related technologies have some problems: they cannot dynamically adapt to contextual semantics, affecting the relevance of assessment results to human judgment, leading to inaccurate translation quality assessment, and they cannot handle the identification of sensitive information in complex contexts.

[0004] There is currently no effective solution to the problem of inaccurate translation quality assessment in related technologies. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, electronic device, and computer program product for determining translation quality, in order to solve the problem of inaccurate translation quality assessment in related technologies.

[0006] To achieve the above objectives, according to one aspect of this application, a method for determining translation quality is provided. The method includes: acquiring source language text and target language text obtained by translating the source language text; generating word embedding representation vectors for the source language text to obtain a first set of word vectors, and generating word embedding representation vectors for the target language text to obtain a second set of word vectors; determining word matching pairs between the first and second word vector sets to obtain M word matching pairs, wherein each word matching pair includes a first word vector and a second word vector; and determining the translation quality of the source language text based on M matching degrees of the M word matching pairs, wherein the matching degree of a word matching pair is obtained based on the content similarity and positional similarity between the first and second word vectors in a set of word matching pairs.

[0007] Optionally, determining the word matching pairs between the first word vector set and the second word vector set to obtain M word matching pairs includes: for the M word vectors in the first word vector set, calculating the cosine similarity between each word vector and each word vector in the second word vector set to obtain M initial matching scores; adjusting the M initial matching scores based on the position of the word associated with each initial matching score in the source language text and the position in the target language text to obtain M sets of matching scores; determining the maximum matching score from the M sets of matching scores, and matching the maximum matching score with a target first word vector and... Match a target second word vector to obtain an initial word matching pair; clear the matching degree of the target first word vector with other second word vectors from the M sets of matching degrees, and clear the matching degree of the target second word vector with other first word vectors to obtain M-1 sets of updated matching degrees; determine the updated maximum matching degree from the M-1 sets of updated matching degrees, and match a target first word vector and a target second word vector associated with the updated maximum matching degree to obtain an initial word matching pair, until all M word vectors in the set of first word vectors have been traversed.

[0008] Optionally, the M initial matching scores are adjusted based on the positions of the associated lexical units in the source language text and the target language text, resulting in the M sets of matching scores. This includes: for each initial matching score in the M sets, determining the first and second lexical units associated with the initial matching score; determining the difference between the first position of the first lexical unit in the source language text and the second position of the second lexical unit in the target language text, thus obtaining the positional offset associated with the initial matching score; and for each initial matching score, adjusting the initial matching score based on the associated positional offset and the sentence length difference value, wherein the sentence length difference value is the difference in sentence length between the source language text and the target language text.

[0009] Optionally, for each initial matching degree, adjusting the initial matching degree based on the associated position offset and sentence length difference includes: calculating the ratio of the associated position offset to the sentence length difference to obtain the relative position between the first and second word units associated with the initial matching degree; increasing the initial matching degree when the relative position is less than a threshold to obtain the adjusted matching degree; decreasing the initial matching degree when the relative position is greater than the threshold to obtain the adjusted matching degree; and keeping the initial matching degree unchanged when the relative position is equal to the threshold.

[0010] Optionally, determining the translation quality of the source language text based on the M matching degrees of the M word matching pairs includes: determining the translation precision based on the M matching degrees, and determining the translation recall based on the M matching degrees; determining the translation score based on the harmonic mean of precision and recall, and determining the translation quality of the source language text based on the translation score, wherein the translation quality and the translation score are positively correlated.

[0011] Optionally, determining the translation precision based on the M matching degrees includes: calculating the sum of second weighted matching degrees based on the M matching degrees and the weights of the matched second word elements in the target language text; calculating the sum of the weights of all second word elements in the target language text; and calculating the ratio of the sum of second weighted matching degrees to the sum of the weights of all second word elements to obtain the translation precision. Determining the translation recall based on the M matching degrees includes: calculating the sum of first weighted matching degrees based on the M matching degrees and the weights of each first word element in the source language text; calculating the sum of the weights of all first word elements in the source language text; and calculating the ratio of the sum of first weighted matching degrees to the sum of the weights of all first word elements to obtain the translation recall.

[0012] Optionally, the target word is either the first word or the second word, and the weight of the target word is determined in the following ways: an initial weight is determined for the target word; if the target word is a noun or a verb, the initial weight is increased to obtain the weight of the target word; if the target word is a word in the financial terminology database, the initial weight is increased based on the frequency of the word in the reference sentence to obtain the weight of the target word.

[0013] To achieve the above objectives, according to another aspect of this application, a device for determining translation quality is provided. The device includes: an acquisition unit for acquiring source language text and target language text translated from the source language text; a generation unit for generating word embedding representation vectors for the source language text to obtain a first set of word vectors, and generating word embedding representation vectors for the target language text to obtain a second set of word vectors; a first determination unit for determining word matching pairs between the first and second set of word vectors to obtain M word matching pairs, wherein each word matching pair includes a first word vector and a second word vector; and a second determination unit for determining the translation quality of the source language text based on M matching degrees of the M word matching pairs, wherein the matching degree of a word matching pair is obtained based on the content similarity and positional similarity between the first and second word vectors in a set of word matching pairs.

[0014] In this embodiment, source language text and target language text translated from the source language text are obtained; a word embedding representation vector is generated for the source language text to obtain a first word vector set, and a word embedding representation vector is generated for the target language text to obtain a second word vector set; word matching pairs are determined between the first and second word vector sets to obtain M word matching pairs, where each word matching pair includes a first word vector and a second word vector; the translation quality of the source language text is determined based on the M matching degrees of the M word matching pairs, where the matching degree of a word matching pair is based on the content similarity and positional similarity between the first and second word vectors in a set of word matching pairs. By obtaining the matching degree of word matching pairs through the content similarity and positional similarity between the first word vector in the source language text and the second word vector in the target language text, and determining the translation quality of the source language text based on the multiple matching degrees of multiple word matching pairs, the technical effect of improving the accuracy of translation quality assessment results is achieved, thereby solving the problem of inaccurate translation quality assessment in related technologies. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0016] Figure 1 A hardware block diagram of a computer terminal for implementing a method for determining translation quality is shown.

[0017] Figure 2 This is a flowchart of a method for determining translation quality according to an embodiment of this application;

[0018] Figure 3 This is a schematic diagram of a translation quality determination device provided according to an embodiment of this application;

[0019] Figure 4 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations, providing users with corresponding operation entry points for them to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage. If the user chooses to agree, the user can view the purpose of data use in real time through the authorization interface and has the right to withdraw authorization or delete data at any time. After authorization is withdrawn, the system will terminate the relevant data processing within 24 hours.

[0023] Example 1

[0024] According to an embodiment of this application, an embodiment of a method for determining translation quality is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0025] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal for implementing a method to determine translation quality is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1(Illustrated using 102a, 102b, ..., 102n) Processor 102 (processor 102 may include, but is not limited to, a microprocessor (MCU) or a field-programmable gate array (FPGA) processing device), memory 104 for storing data, and transmission device 106 for communication functions. In addition, it may include: a display, input / output interface (I / O interface), Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), network interface, power supply, and / or camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0026] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0027] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the translation quality determination method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned translation quality determination method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0028] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0029] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0030] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for determining translation quality is shown. Figure 2 This is a flowchart of a method for determining translation quality according to an embodiment of this application.

[0031] Step S201: Obtain the source language text and the target language text obtained by translating the source language text.

[0032] In this context, source language text refers to the original text to be translated, such as a client's inquiry submitted in English. Target language text refers to the output of a machine translation system that converts the source language text into another language, such as translating English text into Chinese. Source language text serves as the benchmark for evaluating translation quality, while target language text is the object from which the translation quality is assessed.

[0033] Step S202: Generate word embedding representation vectors for the source language text to obtain the first word vector set, and generate word embedding representation vectors for the target language text to obtain the second word vector set.

[0034] In this context, a lexical unit is the basic building block of text, which can be a word or a phrase. Embedding representation is the process of mapping lexical units to a multi-dimensional vector space through a pre-trained neural network model (e.g., BERT (Bidirectional Encoder Representations from Transformers)). Embedding representation can capture the semantic information and contextual relationships of lexical units.

[0035] The first set of word vectors corresponds to the set of word embedding vectors in the source language, and the second set of word vectors corresponds to the set of word embedding vectors in the target language text. By converting words in the text into vector representations, semantic-level comparisons can be performed, rather than simply relying on surface-level character matching. This allows subsequent matching degree calculations to more accurately reflect the translation quality of the words.

[0036] Step S203: Determine the word matching pairs between the first word vector set and the second word vector set to obtain M word matching pairs, wherein each word matching pair includes a first word vector and a second word vector.

[0037] Lexical matching involves finding the most similar lexical pairs between the lexical embedding vectors of the source and target language texts based on similarity calculations, thus establishing a one-to-one correspondence between lexical units. Considering the semantic and positional relationships of lexical units in a sentence, the aim of establishing a one-to-one correspondence between lexical units in the source and target languages ​​is to identify the most similar pairings at the lexical level in translation, providing a data foundation for further evaluation of translation quality.

[0038] M token matching pairs refer to the set of all token pairs participating in the translation quality assessment. Each matching pair contains a token vector converted from the source language text and a token vector from the target language text. M is less than or equal to the total number of source language texts. The value of M depends on whether there is a suitable match.

[0039] Step S204: Determine the translation quality of the source language text based on the M matching degrees of the M word matching pairs, wherein the matching degree of a word matching pair is obtained based on the content similarity and positional similarity between the first word vector and the second word vector in a set of word matching pairs.

[0040] Matching degree is a quantitative indicator for evaluating the quality of word matching pairs. For each word matching pair, the matching degree is obtained by calculating the similarity between the first word vector and the second word vector. In this embodiment, the matching degree is affected by static and dynamic factors. Specifically, the calculation of the matching degree combines the content similarity of the word and the positional similarity of the word in the sentence. Content similarity refers to the semantic similarity of the word, while positional similarity refers to the positional relationship of the word before and after translation.

[0041] After collecting all the word matching pairs, the translation quality is quantified based on the matching degree of multiple matching pairs. By taking into account the similarity of content and position, the translation quality can be more comprehensively reflected even when the text in the banking scenario contains complex grammatical structures and professional terms in specific contexts.

[0042] The translation quality determination method provided in this application embodiment obtains source language text and target language text translated from the source language text; generates word embedding representation vectors for the source language text to obtain a first word vector set, and generates word embedding representation vectors for the target language text to obtain a second word vector set; determines word matching pairs between the first word vector set and the second word vector set to obtain M word matching pairs, wherein one word matching pair includes a first word vector and a second word vector; and determines the translation quality of the source language text based on the M matching degrees of the M word matching pairs, wherein the matching degree of one word matching pair is based on the content similarity and positional similarity between the first word vector and the second word vector in a set of word matching pairs, thus solving the problem of inaccurate translation quality assessment in related technologies. The matching degree of word matching pairs is obtained by using the content similarity and positional similarity between the first word vector in the source language text and the second word vector in the target language text. The translation quality of the source language text is determined based on the multiple matching degrees of multiple word matching pairs. This achieves the technical effect of improving the accuracy of translation quality assessment results and solves the problem of inaccurate translation quality assessment in related technologies.

[0043] To accurately determine the correspondence between word units in the source language text and the target language text, and thus more accurately assess translation quality, optionally, in the translation quality determination method provided in this application embodiment, determining word unit matching pairs between the first word unit vector set and the second word unit vector set to obtain M word unit matching pairs includes: for the M word unit vectors in the first word unit vector set, calculating the cosine similarity between each word unit vector and each word unit vector in the second word unit vector set to obtain M initial matching degrees; adjusting the M initial matching degrees based on the position of the word unit associated with each initial matching degree in the source language text and the position in the target language text to obtain M sets of matching degrees; and then adjusting the M sets of matching degrees from the M sets of matching degrees. The maximum matching degree is determined from the M sets of matching degrees, and a target first word vector and a target second word vector associated with the maximum matching degree are matched to obtain an initial word matching pair. From the M sets of matching degrees, the matching degree between the target first word vector and other second word vectors is cleared to zero, and the matching degree between the target second word vector and other first word vectors is cleared to zero, resulting in M-1 sets of updated matching degrees. From the M-1 sets of updated matching degrees, the updated maximum matching degree is determined, and a target first word vector and a target second word vector associated with the updated maximum matching degree are matched to obtain an initial word matching pair, until all M word vectors in the set of first word vectors have been traversed.

[0044] First, the cosine similarity is calculated between each of the M word vectors in the source language text (the first set of word vectors) and all word vectors in the target language text (the second set of word vectors), generating M sets of initial matching scores. Each set of initial matching scores includes the matching scores of one first word vector with each of the second word vectors. The cosine similarity measures the similarity between vectors, ranging from -1 to 1; a value closer to 1 indicates greater similarity between the two vectors.

[0045] Furthermore, the initial matching scores of group M are updated based on the positional relationship of word units in the source and target languages ​​to reflect the impact of positional similarity on word unit matching, making the evaluation results more in line with human intuitive understanding of translation quality. Specifically, if the positional shift of word units is within an acceptable range, the corresponding matching score is increased; otherwise, the matching score is decreased.

[0046] Then, the maximum matching degree (representing the best word-character correspondence) is selected from the M sets of matching degrees, and the corresponding word-character vectors are matched to form an initial word-character matching pair. To complete the matching of all words, for paired word-character vectors, their matching degree with other words is set to zero in subsequent matching degree calculations, i.e., they are removed from the matching degree matrix. Then, the maximum value is found again from the remaining matching degrees, and the above process is iterated until all word-character vectors are paired.

[0047] This embodiment effectively distinguishes grammatical errors and semantic deviations in translation by dynamically adjusting the matching degree based on word position information. By gradually eliminating determined matching pairs during the one-to-one matching process, it ensures that each word participates in the matching only once, thereby guaranteeing the uniqueness of the matching and the accuracy of the semantics. Moreover, it avoids information duplication or omission during the evaluation process, maintains the integrity and coverage of the matching, and thus improves the efficiency of the evaluation process and the reliability of the results.

[0048] To optimize the accuracy and relevance of translation evaluation and ensure that the evaluation results are closer to the actual translation quality, optionally, in the translation quality determination method provided in this application embodiment, adjusting M groups of initial matching degrees based on the position of each initial matching degree associated with the word in the source language text and the position in the target language text, to obtain M groups of matching degrees includes: for each initial matching degree in the M groups of initial matching degrees, determining the first word and the second word associated with the initial matching degree; determining the difference value between the first position of the first word in the source language text and the second position of the second word in the target language text, to obtain the position offset associated with the initial matching degree; for each initial matching degree, adjusting the initial matching degree according to the associated position offset and the sentence length difference value, wherein the sentence length difference value is the sentence length difference value between the source language text and the target language text.

[0049] Here, the first position refers to the specific location of the word element in the source language text, and the second position is the specific location of the corresponding word element in the target language text. The positional offset between the first and second positions reflects the changes in word element positions during the translation process. This can further determine whether the word elements in the matching pair have maintained their original positional relationships in the context during the translation process, thereby identifying possible word order adjustments or omissions during the translation process and providing basic information for subsequent adjustments to the matching degree.

[0050] Among them, the sentence length difference value is used to measure the difference in length between the source language text and the target language text. The sentence length difference may be caused by translation style, language characteristics or translation accuracy issues. By adjusting the matching degree related to the sentence length difference value, the influence of this factor can be taken into account when evaluating translation quality, thereby improving the accuracy of quality assessment.

[0051] For each of the M initial matching scores, the matching score is dynamically updated based on the calculated position offset and sentence length difference. For example, if the position offset exceeds a preset threshold, the matching score is reduced; otherwise, the matching score is increased appropriately. At the same time, the sentence length difference also affects the adjustment of the matching score; a larger sentence length difference reduces the matching score, and vice versa.

[0052] This embodiment introduces the calculation of position offset and sentence length difference value. By considering the position offset, it identifies and quantifies word order changes and grammatical errors in the translation process to ensure the accuracy of the evaluation results. By introducing the sentence length difference value, it helps to measure whether the translation style is suitable for the expression habits of the target language, and whether there is information omission or redundancy in the translation process. This ensures that the translation evaluation not only focuses on the surface similarity between word units, but also considers the performance of the translated content in terms of grammatical structure and semantic coherence in greater depth.

[0053] Optionally, in the method for determining translation quality provided in the embodiments of this application, adjusting the initial matching degree based on the associated position offset and sentence length difference value for each initial matching degree includes: calculating the ratio of the associated position offset to the sentence length difference value to obtain the relative position between the first word and the second word associated with the initial matching degree; increasing the initial matching degree when the relative position is less than a threshold to obtain the adjusted matching degree; decreasing the initial matching degree when the relative position is greater than the threshold to obtain the adjusted matching degree; and keeping the initial matching degree unchanged when the relative position is equal to the threshold.

[0054] Among them, positional offset refers to the difference between the position of a word in the source language text and the corresponding position of a word in the target language text, and sentence length difference is the difference between the lengths of the source language text and the target language text. Relative position is the ratio of positional offset to sentence length difference, used to measure the relative magnitude of word position change before and after translation. By calculating relative position, the degree of influence of word position change during translation can be quantified, thereby assessing the fluency and semantic consistency of the translation.

[0055] A relative position less than the threshold indicates that the word's position does not change much during translation, maintaining a good positional relationship with the source language text, resulting in high translation quality. The initial matching score should be positively adjusted by increasing it to obtain the adjusted matching score. A relative position greater than the threshold indicates that the word's position has changed significantly, potentially disrupting semantic coherence and reducing the naturalness of the translation. The initial matching score should be negatively adjusted by decreasing it to obtain the adjusted matching score. A relative position equal to the threshold indicates that the change in word position is within an acceptable range, requiring no adjustment to the initial matching score; its original value should be maintained.

[0056] By adjusting the matching degree based on relative position, this embodiment can more accurately quantify the similarity between the translation result and the source language text in terms of positional relationship, avoiding the inaccurate judgment that may be caused by simply relying on word overlap for evaluation, and optimizing the accuracy of translation quality evaluation.

[0057] To more comprehensively evaluate the translation results, optionally, in the method for determining translation quality provided in the embodiments of this application, determining the translation quality of the source language text based on the M matching degrees of M word matching pairs includes: determining the precision rate of the translation based on the M matching degrees, and determining the recall rate of the translation based on the M matching degrees; determining the translation score based on the harmonic mean of the precision rate and the recall rate, and determining the translation quality of the source language text based on the translation score, wherein the translation quality and the translation score are positively correlated.

[0058] Precision measures the proportion of correctly translated words in the translation result. In this embodiment, precision is determined by evaluating the matching degree of word matching pairs in the target language text, that is, the ratio of correctly translated words to all words participating in the matching. Recall evaluates the ratio of words in the source language text that are correctly translated by the translation system. In this embodiment, recall is also based on the matching degree of word matching pairs, focusing on the coverage of accurately translated words in the source language text.

[0059] The translation score is the harmonic mean of precision and recall, denoted as the F1 score, which is used to comprehensively evaluate translation quality. A higher F1 score indicates that the translation result has both high precision and high recall, meaning that the translation is not only accurate but also complete, without omitting any important information.

[0060] This embodiment calculates the translation score by considering not only the accuracy (precision) of word translation but also the completeness and retention of information (recall). This comprehensive metric reflects the balance between accuracy and completeness in the translation result. For text translation in banking scenarios containing a large amount of specialized terminology and sensitive information, assessing translation quality ensures that the translation result is both accurate and complete, improving customer service response efficiency and customer satisfaction. Furthermore, the translation score facilitates comparison with human translation standards, providing data support and improvement directions for the optimization of the translation system.

[0061] Optionally, in the method for determining translation quality provided in this application embodiment, determining the translation precision based on M matching degrees includes: calculating the sum of second weighted matching degrees based on the M matching degrees and the weights of the matched second word elements in the target language text, calculating the sum of the weights of all second word elements in the target language text, and calculating the ratio of the sum of the second weighted matching degrees to the sum of the weights of all second word elements to obtain the translation precision; determining the translation recall based on M matching degrees includes: calculating the sum of first weighted matching degrees based on the M matching degrees and the weights of each first word element in the source language text, calculating the sum of the weights of all first word elements in the source language text, and calculating the ratio of the sum of the first weighted matching degrees to the sum of the weights of all first word elements to obtain the translation recall.

[0062] Precision reflects the accuracy of word-by-word translation in the translation result. It measures the proportion of weighted words in the target language text that are correctly translated. The introduction of weights takes into account the importance of different words in semantic and grammatical structure, which helps to more comprehensively evaluate the translation quality. Precision is denoted as P, and the specific calculation formula is as follows:

[0063]

[0064] Where x is the set of word vectors in the source language text, and xi is the first word vector of the i-th word in the source language text. It is a set of word vectors of the target language text, x j It is the j-th second word vector in the set of word vectors of the target language text. The max function represents the source language text word vector x that is most similar to each word xi in the target language text. j , These are the word vectors xi and x. j The dot product between two lexical units represents the similarity between them. is the weight of the j-th second word element in the target language text. The numerator represents the weighted sum of the similarity of matching word vectors in the target language text, and the denominator represents the total weight of all word vectors in the target language text.

[0065] Recall measures whether a translation system has missed important information. It measures the proportion of words in the source language text that are successfully translated by the translation system. By calculating recall, it can be ensured that the translation covers all key content in the source language text, especially sensitive information such as professional terminology and amounts in financial scenarios. Recall is denoted as R, and the specific calculation formula is as follows:

[0066]

[0067] Where x is the set of word vectors in the source language text, and xi is the first word vector of the i-th word in the source language text. It is a set of word vectors of the target language text, x j It is the j-th second word vector in the set of word vectors of the target language text. The max function represents the source language text word vector x that is most similar to each word xi in the target language text. j , These are the word vectors xi and x. j The dot product between two terms represents the similarity between them, weight(x) i ) is the weight of the i-th first word element in the source language text. The numerator represents the weighted sum of the similarity of matching all word element vectors in the source language text, and the denominator represents the total weight of all word element vectors in the source language text.

[0068] This embodiment, by calculating precision and return rate, can more accurately reflect the actual performance of the translated content in a banking scenario when evaluating translation quality. The introduction of weights in calculating precision and return rate, especially the high weighting of key entities and terms, ensures that the evaluation process not only focuses on the surface matching of lexical units, but also emphasizes the accuracy and completeness of the information conveyed in the translation result.

[0069] To more accurately reflect the importance of lexical units and their role in sentences, optionally, in the translation quality determination method provided in this application embodiment, the target lexical unit is either a first lexical unit or a second lexical unit, and the weight of the target lexical unit is determined in the following ways: an initial weight is determined for the target lexical unit; if the target lexical unit is a noun or a verb, the initial weight is increased to obtain the weight of the target lexical unit; if the target lexical unit is a lexical unit in a financial terminology database, the initial weight is increased based on the frequency of the lexical unit in the reference sentence to obtain the weight of the target lexical unit.

[0070] The target word refers to the first word in the source language text or the second word in the target language text. Each target word is assigned a basic weight value as the basis for subsequent adjustments.

[0071] Nouns and verbs are core components of a sentence and have a significant impact on its meaning. When the target word is a noun or verb, its initial weight is adjusted by increasing its weight value to emphasize the importance of these words in translation quality assessment.

[0072] For target terms belonging to the financial terminology database, due to their special significance in banking contexts, their weights are adjusted based on their frequency of occurrence in reference sentences. First, it is checked whether the target term exists in the pre-built financial terminology database. If it does, the initial weight is dynamically adjusted based on the term's frequency of occurrence in the reference sentence set. The lower the frequency of a term, the higher its weight gain, because low-frequency words contain highly specific information, which is crucial for translation accuracy. The adjustment formula is: weight_adjusted = weight_initial + Delta_weight), where weight_initial is the initial weight of the target term, and Delta_weight is the weight increment based on part-of-speech or financial terminology attributes. Furthermore, when the target term is a term from the financial terminology database, the weight can also be calculated using the following formula: weight = -log((N+1) / M), where N is the number of reference sentences containing the term, and M is the total number of sentences.

[0073] This embodiment enables the translation system to pay more attention to the core components of sentences and key information in financial scenarios through dynamic adjustment of weights. In particular, when processing bank customer request forms containing multiple languages, dialects, and financial terminology, it can more accurately assess the translation quality.

[0074] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0075] Example 2

[0076] This application also provides a translation quality determination apparatus. It should be noted that the translation quality determination apparatus of this application can be used to execute the translation quality determination method provided in this application. The translation quality determination apparatus provided in this application will be described below.

[0077] According to an embodiment of this application, an apparatus for implementing the above-described method for determining translation quality is also provided. Figure 3 This is a schematic diagram of a translation quality determination device provided according to an embodiment of this application, such as... Figure 3 As shown, the device includes:

[0078] The acquisition unit 301 is used to acquire source language text and target language text obtained by translating the source language text.

[0079] The generation unit 302 is used to generate word embedding representation vectors for the source language text to obtain a first word vector set, and to generate word embedding representation vectors for the target language text to obtain a second word vector set;

[0080] The first determining unit 303 is used to determine word matching pairs between the first word vector set and the second word vector set, to obtain M word matching pairs, wherein a word matching pair includes a first word vector and a second word vector;

[0081] The second determining unit 304 is used to determine the translation quality of the source language text based on the M matching degrees of M word matching pairs, wherein the matching degree of a word matching pair is obtained based on the content similarity and positional similarity between the first word vector and the second word vector in a set of word matching pairs.

[0082] The translation quality determination apparatus provided in this application embodiment acquires source language text and target language text translated from the source language text through an acquisition unit 301; a generation unit 302 generates word embedding representation vectors for the source language text to obtain a first word vector set, and generates word embedding representation vectors for the target language text to obtain a second word vector set; a first determination unit 303 determines word matching pairs between the first word vector set and the second word vector set to obtain M word matching pairs, wherein one word matching pair includes a first word vector and a second word vector; and a second determination unit 304 determines the translation quality of the source language text based on the M matching degrees of the M word matching pairs, wherein the matching degree of one word matching pair is based on the content similarity and positional similarity between the first word vector and the second word vector in a set of word matching pairs, thus solving the problem of inaccurate translation quality assessment in related technologies. The matching degree of word matching pairs is obtained by using the content similarity and positional similarity between the first word vector in the source language text and the second word vector in the target language text. The translation quality of the source language text is determined based on the multiple matching degrees of multiple word matching pairs. This achieves the technical effect of improving the accuracy of translation quality assessment results and solves the problem of inaccurate translation quality assessment in related technologies.

[0083] Optionally, in the translation quality determination apparatus provided in this application embodiment, the first determination unit 303 includes: a calculation module, used to calculate the cosine similarity between each of the M word vectors in the first word vector set and each word vector in the second word vector set, to obtain M sets of initial matching scores; an adjustment module, used to adjust the M sets of initial matching scores based on the position of the word associated with each initial matching score in the source language text and the position in the target language text, to obtain M sets of matching scores; and an initial matching module, used to determine the maximum matching score from the M sets of matching scores, and to match a target first word vector associated with the maximum matching score with... A target second word vector is matched to obtain an initial word matching pair; the update module is used to clear the matching degree of the target first word vector with other second word vectors from the M sets of matching degrees, and clear the matching degree of the target second word vector with other first word vectors to obtain M-1 sets of updated matching degrees; the rematch module is used to determine the updated maximum matching degree from the M-1 sets of updated matching degrees, and match a target first word vector and a target second word vector associated with the updated maximum matching degree to obtain an initial word matching pair, until all M word vectors in the set of first word vectors have been traversed.

[0084] Optionally, in the translation quality determination device provided in the embodiments of this application, the adjustment module includes: a first determination submodule, used to determine a first word and a second word associated with each initial matching degree in M ​​groups of initial matching degrees; a second determination submodule, used to determine the difference value between the first position of the first word in the source language text and the second position of the second word in the target language text, to obtain the position offset associated with the initial matching degree; and an adjustment submodule, used to adjust the initial matching degree for each initial matching degree according to the associated position offset and the sentence length difference value, wherein the sentence length difference value is the sentence length difference value between the source language text and the target language text.

[0085] Optionally, in the translation quality determination device provided in this application embodiment, the adjustment submodule includes: a first calculation submodule, used to calculate the ratio of the associated position offset to the sentence length difference value to obtain the relative position between the first word and the second word associated with the initial matching degree; a first adjustment submodule, used to increase the initial matching degree when the relative position is less than a threshold to obtain an adjusted matching degree; a second adjustment submodule, used to decrease the initial matching degree when the relative position is greater than the threshold to obtain an adjusted matching degree; and a third adjustment submodule, used to keep the initial matching degree unchanged when the relative position is equal to the threshold.

[0086] Optionally, in the translation quality determination device provided in the embodiments of this application, the second determination unit 304 includes: a first determination module, configured to determine the precision rate of the translation based on M matching degrees, and to determine the recall rate of the translation based on M matching degrees; and a second determination module, configured to determine the translation score based on the harmonic mean of the precision rate and the recall rate, and to determine the translation quality of the source language text based on the translation score, wherein the translation quality and the translation score are positively correlated.

[0087] Optionally, in the translation quality determination device provided in this application embodiment, the first determination module includes: a second calculation submodule, used to calculate a second weighted matching degree sum based on M matching degrees and the weights of the matched second word elements in the target language text, calculate the weights of all second word elements in the target language text, and calculate the ratio of the second weighted matching degree sum to the weights of all second word elements to obtain the translation accuracy; and a third calculation submodule, used to calculate a first weighted matching degree sum based on M matching degrees and the weights of each first word element in the source language text, calculate the weights of all first word elements in the source language text, and calculate the ratio of the first weighted matching degree sum to the weights of all first word elements to obtain the translation recall.

[0088] Optionally, in the translation quality determination device provided in this application embodiment, the target word is a first word or a second word, and the weight of the target word is determined in the following ways: an initial weight is determined for the target word; if the target word is a noun or a verb, the initial weight is increased to obtain the weight of the target word; if the target word is a word in a financial terminology database, the initial weight is increased based on the frequency of the word in the reference sentence to obtain the weight of the target word.

[0089] It should be noted that the above-mentioned units and modules correspond to the steps in Embodiment 1, and the instances and application scenarios implemented by the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above-mentioned modules or units may be hardware components or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above-mentioned modules may also be part of a device and can run in the computer terminal 10 provided in Embodiment 1.

[0090] Example 3

[0091] Embodiments of this application may provide an electronic device. Figure 4 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 4 As shown, the electronic device may include: one or more ( Figure 4(Only one is shown) processor 1002, memory 1004, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0092] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0093] The processor can access information and applications stored in memory via a transmission device to perform the following steps: acquiring source language text and target language text translated from the source language text; generating word embedding representation vectors for the source language text to obtain a first word vector set, and generating word embedding representation vectors for the target language text to obtain a second word vector set; determining word matching pairs between the first and second word vector sets to obtain M word matching pairs, wherein each word matching pair includes a first word vector and a second word vector; determining the translation quality of the source language text based on the M matching degrees of the M word matching pairs, wherein the matching degree of a word matching pair is based on the content similarity and positional similarity between the first and second word vectors in a set of word matching pairs.

[0094] The processor can also invoke information and applications stored in the memory via a transmission device to perform the following steps: determining word matching pairs between the first word vector set and the second word vector set to obtain M word matching pairs, including: for the M word vectors in the first word vector set, calculating the cosine similarity between each word vector and each word vector in the second word vector set to obtain M initial matching scores; adjusting the M initial matching scores based on the position of the word associated with each initial matching score in the source language text and the position in the target language text to obtain M sets of matching scores; determining the maximum matching score from the M sets of matching scores, and matching the maximum matching score with the maximum matching score. Matching a target first word vector and a target second word vector associated with the match degree yields an initial word matching pair. From the M sets of match degrees, the match degree between the target first word vector and other second word vectors is reset to zero, and the match degree between the target second word vector and other first word vectors is also reset to zero, resulting in M-1 sets of updated match degrees. From the M-1 sets of updated match degrees, the updated maximum match degree is determined, and a target first word vector and a target second word vector associated with the updated maximum match degree are matched to obtain an initial word matching pair. This process continues until all M word vectors in the first word vector set have been traversed.

[0095] The processor can also invoke information and applications stored in the memory via the transmission device to perform the following steps: adjusting M sets of initial matching degrees based on the positions of the tokens associated with each initial matching degree in the source language text and in the target language text, to obtain M sets of matching degrees, including: for each initial matching degree in the M sets of initial matching degrees, determining the first token and the second token associated with the initial matching degree; determining the difference value between the first position of the first token in the source language text and the second position of the second token in the target language text, to obtain the position offset associated with the initial matching degree; for each initial matching degree, adjusting the initial matching degree according to the associated position offset and the sentence length difference value, wherein the sentence length difference value is the difference value between the sentence lengths of the source language text and the target language text.

[0096] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: For each initial matching degree, adjusting the initial matching degree according to the associated position offset and sentence length difference value includes: calculating the ratio of the associated position offset and sentence length difference value to obtain the relative position between the first word and the second word associated with the initial matching degree; if the relative position is less than a threshold, the initial matching degree is increased to obtain the adjusted matching degree; if the relative position is greater than the threshold, the initial matching degree is decreased to obtain the adjusted matching degree; if the relative position is equal to the threshold, the initial matching degree remains unchanged.

[0097] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: determining the translation quality of the source language text based on the M matching degrees of M word matching pairs includes: determining the precision of the translation based on the M matching degrees, and determining the recall of the translation based on the M matching degrees; determining the translation score based on the harmonic mean of precision and recall, and determining the translation quality of the source language text based on the translation score, wherein the translation quality and the translation score are positively correlated.

[0098] The processor can also invoke information and applications stored in the memory via the transmission device to perform the following steps: Determining the translation precision based on M matching degrees includes: calculating the sum of second weighted matching degrees based on the M matching degrees and the weights of the matched second word units in the target language text; calculating the sum of the weights of all second word units in the target language text; and calculating the ratio of the sum of second weighted matching degrees to the sum of the weights of all second word units to obtain the translation precision; Determining the translation recall based on M matching degrees includes: calculating the sum of first weighted matching degrees based on the M matching degrees and the weights of each first word unit in the source language text; calculating the sum of the weights of all first word units in the source language text; and calculating the ratio of the sum of first weighted matching degrees to the sum of the weights of all first word units to obtain the translation recall.

[0099] The processor can also call the information and application program stored in the memory through the transmission device to perform the following steps: the target word is the first word or the second word, and the weight of the target word is determined in the following way: an initial weight is determined for the target word; if the target word is a noun or a verb, the initial weight is increased to obtain the weight of the target word; if the target word is a word in the financial terminology database, the initial weight is increased based on the frequency of the word in the reference sentence to obtain the weight of the target word.

[0100] Those skilled in the art will understand that Figure 4 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs (tablet computers), and other terminal devices. Figure 4 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 4 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 4 The different configurations shown.

[0101] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0102] Example 4

[0103] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the translation quality determination method provided in Embodiment 1.

[0104] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0105] This application also provides a computer program product, which, when executed on a data processing device, is adapted to perform steps for determining translation quality.

[0106] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0107] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0109] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0110] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0111] If the integrated unit is implemented as 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 this application, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0112] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for determining translation quality, characterized in that, include: Obtain the source language text and the target language text obtained by translating the source language text; Generate a word embedding representation vector for the source language text to obtain a first word vector set, and generate a word embedding representation vector for the target language text to obtain a second word vector set; Determine the word matching pairs between the first word vector set and the second word vector set to obtain M word matching pairs, wherein each word matching pair includes a first word vector and a second word vector; The translation quality of the source language text is determined based on the M matching degrees of the M word matching pairs, wherein the matching degree of a word matching pair is obtained based on the content similarity and positional similarity between the first word vector and the second word vector in a set of word matching pairs.

2. The method according to claim 1, characterized in that, Determine the word matching pairs between the first word vector set and the second word vector set to obtain M word matching pairs, including: For each of the M word vectors in the first word vector set, calculate the cosine similarity between each word vector and each word vector in the second word vector set to obtain M initial matching scores. The M initial matching scores are adjusted based on the position of the lexical unit associated with each initial matching score in the source language text and the position in the target language text, to obtain the M sets of matching scores; The maximum matching degree is determined from the M sets of matching degrees, and a target first word vector and a target second word vector associated with the maximum matching degree are matched to obtain an initial word matching pair; From the M sets of matching scores, the matching score between the target first word vector and other second word vectors is cleared to zero, and the matching score between the target second word vector and other first word vectors is cleared to zero, resulting in M-1 sets of updated matching scores; The updated maximum matching degree is determined from the M-1 sets of updated matching degrees, and a target first word vector and a target second word vector associated with the updated maximum matching degree are matched to obtain an initial word matching pair, until the M word vectors in the first word vector set have been traversed.

3. The method according to claim 2, characterized in that, The M initial matching scores are adjusted based on the position of the lexical unit associated with each initial matching score in the source language text and its position in the target language text, resulting in M ​​sets of matching scores including: For each initial matching degree in the M groups of initial matching degrees, determine the first and second word elements associated with the initial matching degree; The difference between the first position of the first word element in the source language text and the second position of the second word element in the target language text is determined to obtain the position offset associated with the initial matching degree; For each initial match, the initial match is adjusted based on the associated position offset and the sentence length difference value, wherein the sentence length difference value is the difference in sentence length between the source language text and the target language text.

4. The method according to claim 3, characterized in that, For each initial match, adjusting the initial match based on the associated position offset and sentence length difference includes: Calculate the ratio of the associated position offset to the sentence length difference value to obtain the relative position between the first and second word units associated with the initial matching degree; If the relative position is less than a threshold, the initial matching degree is increased to obtain an adjusted matching degree; If the relative position is greater than the threshold, the initial matching degree is reduced to obtain the adjusted matching degree; If the relative position is equal to the threshold, the initial matching degree remains unchanged.

5. The method according to claim 1, characterized in that, Determining the translation quality of the source language text based on the M matching degrees of the M word matching pairs includes: The translation precision is determined based on the M matching degrees, and the translation recall is determined based on the M matching degrees. A translation score is determined based on the harmonic mean of the precision and the recall, and the translation quality of the source language text is determined based on the translation score, wherein the translation quality and the translation score are positively correlated.

6. The method according to claim 5, characterized in that, Determining the translation accuracy based on the M matching degrees includes: The second weighted matching degree is calculated based on the M matching degrees and the weights of the matched second word elements in the target language text. The weights of all second word elements in the target language text are then calculated, and the ratio of the second weighted matching degree to the weights of all second word elements is calculated to obtain the accuracy of the translation. Determining the translation recall based on the M matching degrees includes: The first weighted matching degree is calculated based on the M matching degrees and the weight of each first word element in the source language text. The weight of all first word elements in the source language text is then calculated, and the ratio of the first weighted matching degree to the weight of all first word elements is calculated to obtain the recall rate of the translation.

7. The method according to claim 6, characterized in that, The target word is either the first word or the second word, and the weight of the target word is determined in the following way: Determine the initial weights for the target lexical units; If the target word is a noun or a verb, the initial weight is increased to obtain the weight of the target word. When the target term is a term from a financial terminology database, the initial weight is increased based on the frequency of the term in the reference sentence to obtain the weight of the target term.

8. A device for determining translation quality, characterized in that, include: The acquisition unit is used to acquire source language text and target language text translated from the source language text; The generation unit is used to generate word embedding representation vectors for the source language text to obtain a first word vector set, and to generate word embedding representation vectors for the target language text to obtain a second word vector set; The first determining unit is used to determine word matching pairs between the first word vector set and the second word vector set, to obtain M word matching pairs, wherein a word matching pair includes a first word vector and a second word vector; The second determining unit is used to determine the translation quality of the source language text based on the M matching degrees of the M word matching pairs, wherein the matching degree of a word matching pair is obtained based on the content similarity and positional similarity between the first word vector and the second word vector in a set of word matching pairs.

9. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 7.

Images