Method for evaluating translation quality, translation checking method and device

Abstract

The application discloses a translation quality evaluation method, a translation error checking method and device. The bilingual translation quality evaluation method comprises the following steps: acquiring at least one picture corresponding to a source text and a translation text respectively; determining the similarity between each picture corresponding to the source text and each picture corresponding to the translation text respectively; and determining the translation quality score of the source text and the translation text according to the similarity between each picture corresponding to the source text and each picture corresponding to the translation text. The application uses information other than natural language, and is independent of bilingual vocabulary, bilingual sentence and other corpus, thereby greatly improving the robustness of bilingual term translation, accurately evaluating the translation quality of free translation bilingual terms, and avoiding various problems caused by the prior art of evaluating the bilingual term translation quality based on word translation probability.

Description

Technical Field

[0009]

[0001] The present invention relates to the technical field of translation, and particularly relates to a method for evaluating translation quality, a method for translation inspection, and a device therefor. Background Art

[0002] With the increasing overseas business, Alibaba Translate provides multilingual two-way translation services, such as supporting two-way machine and human translation work from English to other languages and from Chinese to multiple other languages.

[0003] Among them, machine translation is to translate text from one natural language into another natural language by means of a computer program.

[0004] One of the key difficulties in machine translation lies in the translation of terms. How to actively discover mis-translated terms and quickly correct them has great business value.

[0005] A term is a set of appellations used to represent concepts in a specific discipline field and has a certain degree of professionalism. A bilingual term refers to two terms that are mutually translated and composed of two languages. Of course, in the case of multilingual translation, the corresponding terms of every two languages can form a bilingual term. Traditional solutions usually calculate the translation probability of bilingual terms based on the lexical translation probability. The common process is as follows:

[0006] 1. Construct a bilingual word list. Using a large number of bilingual sentence pairs, for example: It's a nice day today → 今天天气很好, first segment the sentences, then perform word alignment based on the segmented bilingual sentence pairs, calculate the lexical translation probability, and finally obtain the word list entries, such as: “apple苹果0.8, 0.6” (indicating that the translation probability of the English word “apple” into the Chinese word “苹果” is 0.8, and the reverse translation probability is 0.6).

[0007] 2. Segment the source text and the translation of the term respectively, and design an algorithm to calculate the overall translation probability of the bilingual term by calculating the translation probabilities of all words in the source text and the translation.

[0008] The above method for evaluating the translation quality of bilingual terms based on word translation probability may have the following disadvantages: 1. It seriously depends on the bilingual word list (or bilingual sentence pairs). If the word list is not comprehensive enough, the effect is poor. Terms usually have only 2 - 4 words and are short in length. The lower the coverage ratio of the word list, the worse the effect. 2. Some bilingual terms are free translations. For example, the Chinese source text is 七分裤, and its corresponding English translation is “Capri pants”. There is no one-to-one correspondence between the two texts, so it is difficult to cover them using the word list and bilingual sentence pairs. Summary of the Invention

[0009] In view of the above problems, the present invention is proposed to provide a method for evaluating the quality of bilingual translation, a method for checking translation errors, and an apparatus for overcoming or at least partially solving the above problems.

[0010] In a first aspect, embodiments of the present invention provide a method for evaluating translation quality, comprising:

[0011] Obtain at least one image corresponding to both the source text and the translation;

[0012] Determine the similarity between each image in the source text and each image in the translation;

[0013] The translation quality scores of the source text and the translation are determined based on the similarity between each image corresponding to the source text and each image corresponding to the translation.

[0014] In one embodiment, at least one image corresponding to the source text and the translated text is obtained, including:

[0015] Based on the source text and the translation, search for images corresponding to the source text and the translation using a search engine;

[0016] From the search results, determine the images corresponding to the source text and the translation, respectively;

[0017] Generate the association data between the source text and the translated text and the corresponding image list.

[0018] In one embodiment, after determining the preset number of images corresponding to the source text and the translation, and before generating the association data between the source text and the translation and the corresponding image list, the method further includes:

[0019] Randomly delete some images from the preset number of images.

[0020] In one embodiment, the similarity between each image corresponding to the source text and each image corresponding to the translation is determined, specifically including:

[0021] Based on the correspondence between the source text and the translation, as well as the association data between the source text, the translation, and the corresponding image list, the association between the image list corresponding to the source text and the image list corresponding to the translation is determined.

[0022] Based on the determined association, a Cartesian product operation is performed on the image list corresponding to the source text and the image list corresponding to the translation to obtain at least one set of image pairs for the bilingual terms composed of the source text and the translation;

[0023] The similarity of the two images in each image pair is determined by performing similarity matching.

[0024] In one embodiment, the translation quality score of the source text and the translation is determined based on the similarity between the images corresponding to the source text and the images corresponding to the translation, specifically including:

[0025] The similarity values ​​of all image pairs corresponding to the bilingual term are smoothed and weighted to obtain the translation quality score of the bilingual term.

[0026] In one embodiment, the similarity matching includes one or more of the following matching methods:

[0027] Using a convolutional neural network, the similarity between the two images in the image pair is calculated to obtain the similarity value of the image pair;

[0028] The similarity between the two images in the image pair is calculated using scale-invariant feature transformation to obtain the similarity value of the image pair.

[0029] The two images in the image pair are vectorized to obtain the feature vectors corresponding to the images. The Euclidean distance between the feature vectors of the two images is calculated to obtain the similarity value of the image pair.

[0030] Secondly, embodiments of the present invention provide a translation checking method, including:

[0031] Based on the source and translated data, determine the corresponding bilingual terms in the source and translated data;

[0032] Determine the translation quality score for the bilingual terms;

[0033] The translation quality score of bilingual terms is compared with a preset passing score threshold.

[0034] When the translation quality score is lower than the preset passing score threshold, the translation of the bilingual terminology is determined to be incorrect.

[0035] The translation quality score for the bilingual terms is determined using the aforementioned translation quality assessment method.

[0036] Thirdly, embodiments of the present invention provide a translation lexicon, which includes at least one set of bilingual terms, wherein the bilingual terms include source text and corresponding translations;

[0037] The bilingual terms in the translation lexicon were obtained by evaluating the translation quality using the aforementioned translation quality assessment method.

[0038] Fourthly, embodiments of the present invention provide a translation quality evaluation device, comprising:

[0039] The acquisition module is used to acquire at least one image corresponding to the source text and the translation text, respectively.

[0040] The similarity determination module is used to determine the similarity between each image corresponding to the source text and each image corresponding to the translation.

[0041] The translation quality score determination module is used to determine the translation quality score of the source text and the translation text based at least on the similarity between each image corresponding to the source text and each image corresponding to the translation text.

[0042] Fifthly, embodiments of the present invention provide a translation checking device, comprising:

[0043] A bilingual terminology determination module is used to determine the corresponding bilingual terms in the source text and the translation based on the source text and the translation data.

[0044] The score determination module is used to determine the translation quality score of the bilingual terms;

[0045] The inspection module is used to compare the translation quality score of bilingual terms with a preset passing score threshold; when the translation quality score is lower than the preset passing score threshold, it is determined that the translation of the bilingual terms is incorrect.

[0046] The translation quality score for the bilingual terms is determined using the aforementioned translation quality assessment method.

[0047] Sixthly, embodiments of the present invention provide a translation lexicon, which contains at least one set of bilingual terms, wherein the bilingual terms include source text and corresponding translations;

[0048] The bilingual terms in the translation lexicon were obtained by evaluating the translation quality using the aforementioned translation quality assessment method.

[0049] In a seventh aspect, embodiments of the present invention provide a multilingual translation server, comprising: a memory and a processor; wherein the memory stores a computer program, and when the program is executed by the processor, it is capable of implementing the aforementioned translation quality assessment method and / or the aforementioned translation checking method.

[0050] Eighthly, embodiments of the present invention provide a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the aforementioned translation quality assessment method and / or are capable of implementing the aforementioned translation checking method.

[0051] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:

[0052] The translation quality assessment method, translation inspection method, and apparatus provided in this invention embodiment acquire at least one image corresponding to the source text and the translation that constitutes the bilingual terminology. Then, the similarity between the images corresponding to the source text and the translation is determined. Based at least on the determined similarity between the images corresponding to the source text and the translation, the translation quality score of the source text and the translation in the bilingual terminology is determined. The matching degree between the source text and the translation is evaluated by using the matching degree of the corresponding images of the source text and the translation using two different languages. This invention utilizes information beyond natural language, freeing it from dependence on bilingual vocabularies, bilingual sentences, and other corpora, greatly improving the robustness of bilingual terminology translation. It accurately evaluates the translation quality of paraphrased bilingual terms, avoiding the problems of serious dependence on bilingual vocabularies or bilingual sentences and inaccurate translation of terms that require paraphrasing, which are caused by the existing method of evaluating the translation quality of bilingual terms based on word translation probability.

[0053] The translation quality assessment method, translation inspection method, and apparatus provided in this invention utilize a search engine and a text-to-image search approach to extensively acquire a preset number of images corresponding to the source text and translation text. This ensures that the acquired images are from a wide range of sources, more accurately reflecting the precise meaning of the terminology and guaranteeing the accuracy of the final bilingual terminology translation quality assessment. After acquiring the preset number of images corresponding to the source text and translation text, a portion of them can be randomly deleted, further enhancing the robustness of the bilingual terminology translation quality assessment.

[0054] In this embodiment of the invention, after obtaining at least one set of image pairs for bilingual terms, the similarity value of the two images in the image pair is calculated. Based on the similarity values ​​obtained from these image pairs, a smoothing process is performed, followed by a weighted average to obtain the translation quality score of the bilingual terms. This avoids the influence of excessively large or small similarity values ​​on the final translation quality score, making the translation quality score of the bilingual terms relatively more accurate.

[0055] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0056] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0057] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0058] Figure 1This is a flowchart of the bilingual translation quality evaluation method in an embodiment of the present invention;

[0059] Figure 2 This is a flowchart illustrating the process of obtaining at least one image corresponding to the source text and the translation in bilingual terminology in an embodiment of the present invention.

[0060] Figures 3A-3D This is a schematic diagram of the images corresponding to the source text and the images corresponding to the translation in an embodiment of the present invention;

[0061] Figure 4 This is a flowchart illustrating the process of determining the similarity between each image corresponding to the source text and each image corresponding to the translation text in an embodiment of the present invention.

[0062] Figure 5 This is a flowchart of an example from an embodiment of the present invention;

[0063] Figure 6 This is a flowchart of a translation error checking method in an embodiment of the present invention;

[0064] Figure 7 This is a structural block diagram of the bilingual translation quality evaluation device in an embodiment of the present invention;

[0065] Figure 8 This is a structural block diagram of the translation error checking device in an embodiment of the present invention. Detailed Implementation

[0066] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0067] To accurately assess the translation quality of bilingual terminology in existing machine translation, proactively identify errors and inaccuracies in bilingual terminology translation, and create conditions for machine or human correction of machine translation errors, this invention provides an image-based translation quality assessment method, a translation inspection method based on the above assessment method, and related apparatus. The following detailed description is in conjunction with the accompanying drawings.

[0068] This invention provides a method for evaluating translation quality, referring to... Figure 1 As shown, it includes the following steps:

[0069] S11. Obtain at least one image corresponding to the source text and the translation respectively;

[0070] In this embodiment of the invention, bilingual terms refer to two terms that are mutually translated from two languages, such as apple (English) <> apple (Chinese), and capri pants (Chinese) <> capri pants (English). This embodiment of the invention does not limit the types and number of mutually translatable languages. Of course, this embodiment of the invention can be applied to scenarios of mutual translation between multiple languages. In the case of mutual translation between multiple languages, the processing method for evaluating the translation quality of bilingual terms between each pair of languages ​​can refer to this embodiment of the invention.

[0071] The bilingual terms in step S11 are the bilingual terms to be evaluated. These terms can be obtained from various sources, such as terms mined from various business scenarios (e.g., website titles, user-input keywords), or translations obtained through translation engines, human translation, or machine-assisted human translation. To obtain accurate translation results, the translated bilingual terms need to be evaluated for translation quality.

[0072] S12. Determine the similarity between each image in the source text and each image in the translation;

[0073] S13. At least based on the similarity between each image corresponding to the source text and each image corresponding to the translation, determine the translation quality score of the source text and the translation.

[0074] In one embodiment, step S11 above involves obtaining at least one image corresponding to the source text and the translation of the bilingual terminology, respectively, referring to... Figure 2 As shown, this can be achieved through the following steps:

[0075] S21. Search for images corresponding to the source text and the translation using a search engine, respectively.

[0076] S22. From the search results, determine the images corresponding to the source text and the translation, respectively;

[0077] S23. Generate the association data between the source text and the translation and the corresponding image list, respectively.

[0078] There are multiple ways to obtain at least one image corresponding to the source text and translation of bilingual terms. Any text-based image search capability can achieve this. For example, you can use the source text and translation of bilingual terms to search on various search engines, obtain a large number of images using keywords, and then download a preset number of images to your local machine.

[0079] The present invention does not limit the specific implementation method for realizing the ability to search for images by text, nor does it limit the number of images selected in the above step S22. It also does not limit the method of selecting a preset number of images from a large number of images obtained by the search engine. For example, the random selection method can be used, or the images can be continuously selected starting from the first image in the order provided by the search engine, or the selection can be made sequentially starting from any image, and so on.

[0080] Using a search engine, in the way of searching for images by text, a preset number of images corresponding to the source text and the translation are widely obtained, so that the obtained images corresponding to the source text and the translation have a wide range of sources, more accurately reflect the accurate meaning of the term, and ensure the accuracy of the final bilingual term translation quality evaluation.

[0081] After the above steps S21 and S22, the associated relationship data of the image list corresponding to the source text and the associated relationship data of the image list corresponding to the translation can be generated respectively.

[0082] Refer to Figure 3A and Figure 3B As shown in the example, the bilingual term: Hacksaw<>Pinch; the source text is Hacksaw, and the translation is "Pinch", Figure 3A As shown is an example of the image list corresponding to the Chinese "Hacksaw", Figure 3B As shown is an example of the image list corresponding to the translation "Pinch" of the Chinese "Hacksaw". The specific generated associated relationship data of the image list corresponding to the source text and the translation are shown in Table 1:

[0083] Table 1

[0084]

[0085] The above Figure 3A and Figure 3B The number of images in Table 1 is only an example.

[0086] In one embodiment, after the above step S22 and before the above step S23, the images corresponding to the source text and the translation obtained in step S22 can also be randomly deleted, which can further enhance the robustness of the bilingual term translation quality evaluation.

[0087] In one embodiment, in the above step S12, the similarity between each image corresponding to the source text and each image corresponding to the translation is determined respectively. Refer to Figure 4 As shown, it is realized through the following process:

[0088] S41. According to the corresponding relationship between the source text and the translation in the bilingual term, and the associated relationship data between the source text and the translation and the corresponding image lists, determine the associated relationship between the image list corresponding to the source text and the image list corresponding to the translation;

[0089] Referring again to the example of bilingual terms mentioned above, based on the correspondence between the bilingual terms "snake-mouth pliers" and "Pinch", and the association data of the image list of "snake-mouth pliers" and "Pinch" shown in Table 1, we can determine the association between images 1 to 6 corresponding to "snake-mouth pliers" in the source text and images 7 to 12 corresponding to "Pinch".

[0090] Because the embodiments of the present invention may evaluate the translation quality of multiple bilingual terms simultaneously during implementation, a number of sets of association data between source text and translation and corresponding image lists, similar to those shown in Table 1, will be generated. It is necessary to determine the association between the images corresponding to the source text and translation in each set of bilingual terms based on the correspondence between the source text and translation in each set of bilingual terms.

[0091] S42. Based on the relationship between the image lists corresponding to the source text and the translation in each bilingual term, perform a Cartesian product operation on the image list corresponding to the source text and the image list corresponding to the translation to obtain at least one set of image pairs for the bilingual term.

[0092] The Cartesian product is an existing mathematical method. The Cartesian product of two sets X and Y, also known as the direct product, is represented as X × Y. The first object is a member of X and the second object is one of the members of all possible ordered pairs of Y.

[0093] Suppose set A = {a, b} and set B = {0, 1, 2}, then the Cartesian product of the two sets is {(a, 0), (a, 1), (a, 2), (b, 0), (b, 1), (b, 2)};

[0094] In step S42, the purpose of the Cartesian product is to obtain all possible image pairs consisting of each image in the image list corresponding to the source text and each image in the image list corresponding to the translation. Of course, any other way to obtain all image pairs corresponding to the source text and translation is also feasible and is not limited to this.

[0095] Using the example shown in Table 1, by combining the 6 images in the image list corresponding to the pliers and the 6 images in the image list corresponding to the pinch, we can obtain 6*6=36 image pairs after performing a Cartesian product. For example, "Image 1, Figure 7 “Image 2, Image 10”, etc.

[0096] S43. Perform similarity matching on the two images in each image pair to determine the similarity value of the image pair.

[0097] The similarity between two images in each image pair can be calculated using various existing methods, such as one or more of the following:

[0098] 1. Using Convolutional Neural Networks (CNN), the similarity between the two images in the image pair is calculated to obtain the similarity value of the image pair;

[0099] For example, if a large number of positively matched images and a large number of dissimilar images are used as training data, and convolutional neural networks are trained on each, then when calculating image similarity, the two images in the image pair can be used as input to the trained convolutional neural network model, and the similarity value between the two images can be output.

[0100] This invention does not limit the type of convolutional neural network.

[0101] 2. Scale-Invariant Feature Transform (SIFT) is used to calculate the similarity between the two images in the image pair to obtain the similarity value of the image pair;

[0102] The essence of the SIFT algorithm is to find key points (feature points) in different scale spaces and calculate the orientation of the key points. The key points found by SIFT are some very prominent points that do not change due to factors such as lighting, affine transformation and noise, such as corner points, edge points, bright spots in dark areas and dark spots in bright areas.

[0103] A typical example of the SIFT algorithm includes the following steps:

[0104] a) Scale-space extremum detection: Search for image locations at all scales. Identify potential scale- and rotation-invariant points of interest using a Gaussian differential function.

[0105] b) Keypoint localization: At each candidate location, the position and scale are determined using a finely fitted model. Keypoints are selected based on their stability.

[0106] c) Orientation determination: Based on the local gradient direction of the image, one or more orientations are assigned to each keypoint location. All subsequent operations on the image data are transformed relative to the orientation, scale, and location of the keypoints, thus providing invariance to these transformations.

[0107] d) Keypoint Description: Within the neighborhood of each keypoint, the local gradient of the image is measured at a selected scale. These gradients are transformed into a representation that allows for relatively large local shape deformations and illumination variations.

[0108] The embodiments of the present invention do not limit the specific SIFT implementation method used.

[0109] 3. Vectorize the two images in the image pair to obtain the corresponding feature vectors. Calculate the Euclidean distance between the feature vectors of the two images to obtain the similarity value of the image pair.

[0110] For example, existing image quantization tools can be used to vectorize each image in a pair into 512-dimensional vectors, obtaining the corresponding feature vectors. Then, the similarity between the two images in each pair can be calculated using Euclidean distance, resulting in a value of (0, 1). It should be noted that Euclidean distance calculates the actual distance between two points in n-dimensional space. The smaller the Euclidean distance, the greater the similarity.

[0111] For all image pairs corresponding to the same bilingual term, the same image similarity algorithm can be used, or different image similarity algorithms can be used independently.

[0112] After processing through steps S41 to S43, the similarity value for each image pair can be obtained. Taking the example of bilingual terms shown in Table 1, 36 similarity values ​​can be obtained.

[0113] In one embodiment, after determining the similarity between each image corresponding to the source text and each image corresponding to the translation text in step S13, that is, the similarity value corresponding to each image pair, the translation quality score of the bilingual term can be calculated accordingly.

[0114] To avoid the final translation quality score being too high or too low, the similarity values ​​of all image pairs corresponding to the bilingual term can be smoothed first, and then a weighted average can be performed to finally fit the translation quality score of the bilingual term.

[0115] The final translation quality score of the fitted bilingual terms can be normalized to a score between 0 and 1 (which may include both extreme values). The higher the score, the higher the translation quality.

[0116] The similarity values ​​of all images are smoothed using various existing smoothing methods. For example, the addition-1 method (also known as Laplacian smoothing) can be used, where each numerator is incremented by 1 and then divided by the total number of samples to eliminate the 0 probability. Of course, other similar smoothing algorithms can also be used, and this embodiment of the invention is not limited to them.

[0117] The similarity values ​​of all image pairs obtained after smoothing are weighted and averaged to obtain the final translation quality score of the source and translation texts that make up the bilingual term.

[0118] As can be seen from the above scores, the scores of the translation quality of bilingual terms are positively correlated with the similarity between the corresponding pictures of the source text and the translation text therein.

[0119] In the specific implementation of the embodiments of the present invention, the two words in the bilingual terms can be used as the source text or the translation text. That is to say, the two groups of bilingual terms of "nipper" (source text) <> "Pinch" (translation text) and "Pinch" (source text) <> "nipper" (translation text) can be regarded as the same.

[0120] To better illustrate the above evaluation method for the translation quality of the language, an example is given below for illustration.

[0121] This example refers to Figure 5 the flowchart shown, and includes the following steps:

[0122] 1. Construct bilingual terms including the source text and the translation text. For example, T1: Capri Pants <> Capri Pants and T2: nipper <> pinch two groups of bilingual terms.

[0123] 2. Search for the pictures of the source text and the translation text in T1 and T2 respectively through a search engine such as Taobao website, referring to Figure 3A and Figure 3B (each picture corresponding to the source text and the translation text in T1), Figure 3C and Figure 3D (each picture corresponding to the source text and the translation text in T2) shown, wherein, Figure 3C what is shown is an example of the picture list corresponding to the Chinese "Capri Pants", Figure 3D what is shown is an example of the picture list corresponding to the translation text "Capri Pants".

[0124] 3. Obtain the picture lists corresponding to the source text and the translation text in T1 and T2 from the results of the search engine.

[0125] 4. For the picture lists corresponding to the source text and the translation text in T1 and T2, perform the Cartesian product on the pictures in the picture lists of the source text and the translation text in T1 and T2 respectively.

[0126] 5. Calculate the picture similarity, that is, calculate the similarity value between the two pictures in the picture pairs obtained through the Cartesian product.

[0127] Specifically, the pictures in the picture pairs can be vectorized first, and then after vector normalization processing, the Euclidean distance is calculated to obtain the similarity value between the pictures.

[0128] 6. Calculate the quality scores of the bilingual terms T1 and T2 according to the similarity values of the picture pairs.

[0129] The pinchers <> pinch has a final score of 0.1, and the Capri pants <> Capri Pants has a final score of 0.9.

[0130] Assuming that the passing score threshold is 0.6, it can be concluded that the translation of pinchers <> pinch is a very poor quality translation, while the translation of Capri pants <> Capri Pants is a very good quality translation.

[0131] Based on the前述 translation quality assessment method, the present invention also provides a translation inspection method, as shown in Figure 6 shown, including the following steps:

[0132] S61. Determine the corresponding bilingual terms according to the data of the source text and the translation.

[0133] S62. Determine the translation quality score of the bilingual terms.

[0134] This step can be obtained through the前述 translation quality assessment method.

[0135] S63. Compare the translation quality score of the bilingual terms with the preset passing score threshold; when the translation quality score is lower than the preset passing score threshold, execute the following step S64.

[0136] S64. Determine the term translation errors in the translation.

[0137] After step S64, after determining the term translation errors in the translation, they can be output through a preset method, such as in the form of a table or in a highlighted manner, and then corrected through machine translation or manual translation.

[0138] Based on the same inventive concept, the embodiments of the present invention also provide an evaluation device for bilingual translation quality, an inspection device for translation errors, a bilingual word library, a multilingual translation server, etc. Since the principles of the problems solved by these devices and servers are similar to the前述 translation quality assessment method and translation inspection method, the specific ways of each module to execute operations have been described in detail in the embodiments related to this method, and will not be elaborated here.

[0139] An evaluation device for translation quality provided by an embodiment of the present invention, as shown in Figure 7 shown, includes:

[0140] An acquisition module 71, configured to acquire at least one picture corresponding to the source text and the translation respectively;

[0141] A similarity determination module 72, configured to determine the similarity between each picture corresponding to the source text and each picture corresponding to the translation respectively;

[0142] The translation quality score determination module 73 is used to determine the translation quality score of the source text and the translation text based at least on the similarity between each image corresponding to the source text and each image corresponding to the translation text.

[0143] In one embodiment, the acquisition module 71 is specifically used to search for images corresponding to the source text and the translation text respectively through a search engine; determine a preset number of images corresponding to the source text and the translation text respectively from the search results; and generate association data between the source text and the translation text and the corresponding image list respectively.

[0144] In one embodiment, the acquisition module 71 of the above-mentioned bilingual translation quality assessment device is further configured to randomly delete some images from the preset number of images after determining the preset number of images corresponding to the source text and the translation text respectively.

[0145] In one embodiment, the similarity determination module 72 is specifically used to determine the association between the image list corresponding to the source text and the image list corresponding to the translation based on the correspondence between the source text and the translation in the bilingual terms, as well as the association data between the source text and the translation and the corresponding image list; based on the determined association, perform a Cartesian product operation on the image list corresponding to the source text and the image list corresponding to the translation to obtain at least one set of image pairs for the bilingual terms; and perform similarity matching on the two images in each image pair to determine the similarity value of the image pair.

[0146] In one embodiment, the above similarity matching may include one or more of the following matching methods:

[0147] Using a convolutional neural network, the similarity between the two images in the image pair is calculated to obtain the similarity value of the image pair;

[0148] The similarity between the two images in the image pair is calculated using scale-invariant feature transformation to obtain the similarity value of the image pair.

[0149] The two images in the image pair are vectorized to obtain the feature vectors corresponding to the images. The Euclidean distance between the feature vectors of the two images is calculated to obtain the similarity value of the image pair.

[0150] In one embodiment, the translation quality score determination module 73 is specifically used to smooth and weight the similarity values ​​of all image pairs corresponding to the bilingual terms to obtain the translation quality score of the bilingual terms.

[0151] The present invention also provides a translation checking device, with reference to Figure 8 As shown, it includes:

[0152] The bilingual terminology determination module 81 is used to determine the corresponding bilingual terms in the source text and translation text based on the source text and translation text data;

[0153] The score determination module 82 is used to determine the translation quality score of the bilingual terms;

[0154] The inspection module 83 is used to compare the translation quality score of the bilingual term with a preset passing score threshold; when the translation quality score is lower than the preset passing score threshold, it is determined that the translation of the bilingual term is incorrect.

[0155] The translation quality scores for the aforementioned bilingual terms can be determined using the aforementioned translation quality assessment methods.

[0156] This invention also provides a translation lexicon containing at least one set of bilingual terms;

[0157] The translation quality of this or these bilingual terms in the translation lexicon can be obtained by evaluating the translation quality using the aforementioned methods.

[0158] This invention also provides a multilingual translation server, comprising: a memory and a processor; wherein the memory stores a computer program, which, when executed by the processor, is capable of implementing the aforementioned translation quality assessment method and / or the aforementioned translation checking method.

[0159] This invention also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the aforementioned translation quality assessment method and / or the aforementioned translation error detection method.

[0160] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0161] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0162] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0163] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0164] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method of evaluating the quality of a translation, characterized in that, The method comprises the following steps: respectively acquiring at least one picture corresponding to the source text and the translation; respectively determining the similarity between the pictures corresponding to the source text and the pictures corresponding to the translation; determining the translation quality score of the source text and the translation according to the determined similarity between the pictures corresponding to the source text and the pictures corresponding to the translation; wherein, after searching for multiple pictures through a search engine, the method at least comprises: randomly selecting a preset number of pictures from the multiple pictures to obtain a search result; or, starting from an arbitrary picture in the multiple pictures, selecting the preset number of pictures in a preset order to obtain the search result; respectively acquiring at least one picture corresponding to the source text and the translation, comprising: randomly selecting at least one picture corresponding to the source text and the translation from the search result; wherein, determining the translation quality score of the source text and the translation according to the determined similarity between the pictures corresponding to the source text and the pictures corresponding to the translation, comprises: acquiring the total number of the similarity; determining the sum of the similarity between the pictures corresponding to the source text and the pictures corresponding to the translation and a preset constant, and determining the quotient between the sum and the total number; performing weighted average on the quotient corresponding to the pictures corresponding to the source text and the pictures corresponding to the translation to obtain the translation quality score; The method further comprises: performing normalization processing on the translation quality score to obtain a normalized translation quality score, wherein the normalized translation quality score is positively correlated with the translation quality of the translation.

2. The method of claim 1, wherein, The method further comprises: respectively generating the association relationship data between the source text and the translation and the corresponding picture list.

3. The method of claim 2, wherein, respectively determining the similarity between the pictures corresponding to the source text and the pictures corresponding to the translation, specifically comprising: determining the association relationship between the picture list corresponding to the source text and the picture list corresponding to the translation according to the corresponding relationship between the source text and the translation and the association relationship data between the source text and the translation and the corresponding picture list; performing Cartesian product operation on the picture list corresponding to the source text and the picture list corresponding to the translation according to the determined association relationship to obtain at least one picture pair of the bilingual term composed of the source text and the translation; performing similarity matching on the two pictures in each picture pair to determine the similarity value of the picture pair.

4. The method of claim 3, wherein, The similarity matching comprises one or more matching methods: using a convolutional neural network to calculate the similarity of the two pictures in the picture pair to obtain the similarity value of the picture pair; using a scale-invariant feature transform to calculate the similarity of the two pictures in the picture pair to obtain the similarity value of the picture pair; vectorizing the two pictures in the picture pair to obtain the feature vectors corresponding to the picture pair, and performing Euclidean distance calculation on the feature vectors of the two pictures to obtain the similarity value of the picture pair.

5. A translation checking method characterized by, The method comprises the following steps: determining the bilingual term corresponding to the data of the source text and the translation according to the data of the source text and the translation; determining the translation quality score of the bilingual term; comparing the translation quality score of the bilingual term with a preset qualified score threshold; determining that the translation of the bilingual term is incorrect when the translation quality score is lower than the preset qualified score threshold; the translation quality score of the bilingual term is determined by the translation quality evaluation method according to any one of claims 1-4.

6. A translation library, characterized by, the translation database comprises at least one set of bilingual terms, wherein the bilingual terms comprise a source text and a corresponding translation; the bilingual terms in the translation database are obtained by using the translation quality evaluation method according to any one of claims 1-4.

7. An evaluation device for the quality of a translation, characterized by comprising: an acquisition module configured to acquire at least one picture corresponding to the source text and the translation, respectively; a similarity determination module configured to determine the similarity between each picture corresponding to the source text and each picture corresponding to the translation, respectively; a translation quality score determination module configured to determine the translation quality score of the source text and the translation according to the similarity between each picture corresponding to the source text and each picture corresponding to the translation; wherein the device is further configured to perform at least one of the following steps: randomly selecting a preset number of pictures from a plurality of pictures to obtain a search result; or selecting the preset number of pictures in a preset order from any picture in the plurality of pictures to obtain the search result; and the acquisition module is configured to acquire the pictures by performing the following steps: randomly selecting at least one picture corresponding to the source text and the translation from the search result, respectively; wherein the translation quality score determination module is configured to determine the translation quality score by performing the following steps: acquiring the total number of similarities; determining the sum of the similarity between each picture corresponding to the source text and each picture corresponding to the translation and a preset constant, and determining the quotient between the sum and the total number; and performing weighted average on the quotient corresponding to each picture corresponding to the source text and each picture corresponding to the translation to obtain the translation quality score; the device is further configured to perform the following steps: normalizing the translation quality score to obtain a normalized translation quality score, wherein the normalized translation quality score is positively correlated with the translation quality of the translation.

8. A translation checking apparatus characterized by comprising: comprising: a bilingual term determination module configured to determine the corresponding bilingual term from the data of the source text and the translation; a score determination module configured to determine the translation quality score of the bilingual term; a checking module configured to compare the translation quality score of the bilingual term with a preset qualified score threshold; determining that the translation of the bilingual term is incorrect when the translation quality score is lower than the preset qualified score threshold; the translation quality score of the bilingual term is determined by the translation quality evaluation method according to any one of claims 1-4.

9. A multi-lingual translation server, characterized by comprising: a memory and a processor; wherein the memory stores a computer program, and the program is executed by the processor to implement the translation quality evaluation method according to any one of claims 1-4 and / or to implement the translation checking method according to claim 5. comprising: a memory and a processor; wherein the memory stores a computer program, and the program is executed by the processor to implement the translation quality evaluation method according to any one of claims 1-4 and / or to implement the translation checking method according to claim 5.

10. A computer readable storage medium having stored thereon computer instructions, wherein, The instruction, when executed by the processor, implements the method for evaluating the quality of translation and / or enables the method for checking translation as claimed in claims 1-4 and 5, respectively.

Images