CN104933038A - Machine translation method and machine translation device - Google Patents

Abstract

The present invention provides a dynamic machine translation method and machine translation device capable of improving translation quality. A machine translation device according to an embodiment of the present invention includes: an input unit, which inputs a sentence to be translated; a calculation unit, which calculates the degree of similarity between the sentence to be translated and the source language sentence in the bilingual corpus; a selection unit, It selects a plurality of sentence pairs in the above-mentioned bilingual corpus based on the above-mentioned similarity as the training corpus; the training unit uses the above-mentioned training corpus to train the translation system; and the translation unit uses the above-mentioned translation system to perform the above-mentioned sentence to be translated translate.

Description

Machine translation method and machine translation device

technical field

The present invention relates to natural language processing technology, in particular, to a machine translation method and a machine translation device.

Background technique

The general process of a statistical machine translation system is to first determine the model (algorithm), then train the model parameters (translation knowledge) based on the training data, and finally use the trained model parameters to translate the input sentence.

Training data usually consists of large-scale aligned bilingual sentence pairs, which may come from different domains and have different sentence forms, and even a sentence with the same source language may have different target language translations. Similarly, the same word in a sentence in the source language may have different translations depending on the context.

In usual translation systems, once the training process is completed, the resulting translation model does not change. Afterwards, the sentence to be translated is translated using the generated translation model. However, due to the diversity of sentences to be translated, such a translation system that does not change after being generated usually cannot be applied to all sentences to be translated, thus resulting in low translation quality.

In this regard, some domain-adaptive methods are proposed to construct "dynamic" translation systems. Some methods first interpolate the in-domain data with the out-of-domain data, and then use the interpolated data to build translation models. Other methods first cluster the training data according to the domain, and then use the clustered subsets to train a separate translation sub-model. When translating, according to the domain of the sentence to be translated, the translation sub-model corresponding to the domain is selected for translation. .

Contents of the invention

The inventors of the present invention have found after researching the methods for adapting to the above fields that although these methods have certain adaptability, once the translation model or translation sub-model is generated through training, it will not change, that is, the translation generated after training The model is still "static", so the translation system has limited adaptability and still leads to poor quality translations.

In order to solve the above-mentioned problems in the prior art, embodiments of the present invention provide a dynamic machine translation method and a machine translation device capable of improving translation quality. Specifically, the following technical solutions are provided.

[1] A machine translation method comprising the steps of: inputting a sentence to be translated; calculating the similarity between the above-mentioned sentence to be translated and the source language sentence in the bilingual corpus; Sentence pairs are used as training corpus; using the above training corpus to train the translation system; and using the above translation system to translate the above sentence to be translated.

The machine translation method of this embodiment can construct a dynamic and targeted translation system by selecting corpus with high similarity with the sentence to be translated in the bilingual corpus, and constructing a translation system in real time based on the selected corpus , which can improve translation quality.

[2] According to the machine translation method of [1] above, the above selection step includes the following steps: sorting the sentences in the above bilingual corpus in order of the above similarity from large to small; and selecting the first N sentences after sorting Yes, as the above-mentioned training corpus, N is an integer greater than or equal to 1.

In the machine translation method of this embodiment, by selecting the first N sentences after sorting, when there are a large number of corpus with high similarity with the sentence to be translated in the bilingual corpus, the most similar and certain amount of corpus can be used to train the translation system , so that not only the translation quality can be guaranteed, but also the processing burden of training the translation system can be reduced.

[3] According to the machine translation method of [1] or [2] above, the selection step includes the following steps: selecting sentence pairs in the bilingual corpus whose similarity is greater than a predetermined threshold as the training corpus.

In the machine translation method of this embodiment, the translation system is trained by selecting corpus with a similarity greater than a predetermined threshold, and the corpus with a low similarity can be excluded, thereby avoiding the interference of the corpus with a low similarity on the translation system, and further ensuring translation. Accuracy.

[4] According to the machine translation method according to one of the above [1]～[3], the above step of calculating the similarity includes the following steps: using the edit distance calculation between the above sentence to be translated and the source language sentence in the above bilingual corpus similarity to the above.

[5] According to the machine translation method according to one of the above [1]～[4], the above step of calculating the similarity includes the following steps: calculating the syntactic structure between the above sentence to be translated and the source language sentence in the above bilingual corpus similarity.

[6] The machine translation method according to any one of [1] to [5] above, after the above translation step, further includes the following step: saving the above sentence to be translated and its translation result in a translation buffer.

[7] According to the machine translation method of [6] above, the following step is further included after the above input step: searching for the above sentence to be translated in the above translation buffer.

In the machine translation method of this embodiment, by saving the sentence to be translated and its translation result in the translation buffer, when the same sentence is translated next time, the translation result of the sentence can be directly obtained from the translation buffer, saving computing resources. Improved translation efficiency.

[8] The machine translation method according to any one of the above [1] to [7], after the above translation step, further includes the following step: adding the above sentence to be translated and its translation result to the above bilingual corpus.

In the machine translation method of this embodiment, by adding sentences to be translated and their translation results into the bilingual corpus, the corpus data of the bilingual corpus can be expanded, thereby improving the translation quality of subsequent translations.

[9] According to the machine translation method according to any one of the above [1]～[8], after the above translation steps, the following steps are also included: performing word alignment on the above sentences to be translated and their translation results; and adding the word alignment results to the above Bilingual corpus.

In the machine translation method of this embodiment, by adding the word alignment result to the bilingual corpus, it can not only expand the corpus data of the bilingual corpus, improve the translation quality of subsequent translations, but also improve translation efficiency.

[10] The machine translation method according to any one of [1] to [9] above, further includes the following step before the above step of calculating the similarity: adding training data related to users to the above bilingual corpus.

In the machine translation method of this embodiment, by adding user-related training data, such as aligned sentence pairs, context-related data, and word-aligned sentence pairs, etc., it can also achieve Purpose of User Adaptation.

[11] The machine translation method according to any one of the above [1] to [10], further comprising the following step after the translation step: calculating the confidence degree of the translation result by using the similarity degree.

In the machine translation method of this embodiment, by using the similarity to calculate the confidence of the translation result, the confidence of the translation result can be obtained at the same time as the translation result, so that there is no need to use another method to calculate the confidence, and the translation efficiency is improved.

[12] A machine translation device, comprising: an input unit, which inputs a sentence to be translated; a calculation unit, which calculates the similarity between the sentence to be translated and the source language sentence in the bilingual corpus; a selection unit, which is based on For the above similarity, a plurality of sentence pairs are selected in the above-mentioned bilingual corpus as the training corpus; the training unit uses the above-mentioned training corpus to train the translation system; and the translation unit uses the above-mentioned translation system to translate the above-mentioned sentences to be translated.

The machine translation device of this embodiment can construct a dynamic and targeted translation system by selecting corpus with high similarity with the sentence to be translated in the bilingual corpus, and constructing a translation system in real time based on the selected corpus , which can improve translation quality.

[13] The machine translation device according to [12] above, wherein the selection unit includes: a sorting unit that sorts the sentence pairs in the bilingual corpus in descending order of the similarity; the selection unit selects and sorts After the first N sentence pairs, as the above-mentioned training corpus, N is an integer greater than 1.

The machine translation device of this embodiment, by selecting the first N sentences after sorting, when there are a large number of corpus with high similarity with the sentence to be translated in the bilingual corpus, it can use the most similar and certain amount of corpus to train the translation system , so that not only the translation quality can be guaranteed, but also the processing burden of training the translation system can be reduced.

[14] The machine translation device according to [12] or [13] above, wherein the selection unit selects sentence pairs in the bilingual corpus whose similarity is greater than a predetermined threshold as the training corpus.

The machine translation device of this embodiment trains the translation system by selecting corpus with a degree of similarity greater than a predetermined threshold, and can exclude corpus with a low degree of similarity, thereby avoiding the interference of corpus with a low degree of similarity on the translation system, and further ensuring translation Accuracy.

[15] The machine translation device according to any one of [12] to [14] above, wherein the calculation unit calculates the similarity using an edit distance between the sentence to be translated and the source language sentence in the bilingual corpus.

[16] The machine translation device according to any one of [12] to [15] above, wherein the calculation unit calculates a similarity in syntactic structure between the sentence to be translated and the source language sentence in the bilingual corpus.

[17] The machine translation device according to any one of [12] to [16] above, further comprising: a storage unit, which stores the sentence to be translated and its translation result in a translation buffer.

[18] The machine translation device according to [17] above, further comprising: a search unit that searches for the sentence to be translated in the translation buffer after the sentence to be translated is input by the input unit.

The machine translation device of this embodiment, by saving the sentence to be translated and its translation result in the translation buffer, when translating the same sentence next time, can directly obtain the translation result of the sentence from the translation buffer, saving computing resources, Improved translation efficiency.

[19] The machine translation device according to any one of [12] to [18] above, further comprising: a sentence pair adding unit, which adds the above-mentioned sentences to be translated and their translation results into the above-mentioned bilingual corpus.

The machine translation device of this embodiment can expand the corpus data of the bilingual corpus by adding sentences to be translated and their translation results into the bilingual corpus, thereby improving the translation quality of subsequent translations.

[20] The machine translation device according to any one of [12] to [19] above, further comprising: a word alignment unit, which performs word alignment on the sentence to be translated and its translation result; and a word alignment result adding unit, which converts Word alignment results are added to the above bilingual corpus.

The machine translation device of this embodiment can not only expand the corpus data of the bilingual corpus, improve the translation quality of subsequent translations, but also improve translation efficiency by adding word alignment results to the bilingual corpus.

[21] The machine translation device according to any one of [12] to [20] above, further comprising: a training corpus adding unit, which adds training data related to the user to the bilingual corpus.

The machine translation device of this embodiment, by adding user-related training data, such as aligned sentence pairs, context-related data, and word-aligned sentence pairs, can also achieve Purpose of User Adaptation.

[22] The machine translation device according to any one of [12] to [21] above, further comprising:

A confidence calculation unit, which uses the similarity to calculate the confidence of the translation result. The machine translation device of this embodiment calculates the confidence degree of the translation result by using the similarity, and can obtain the confidence degree of the translation result at the same time as the translation result, so that there is no need to use another method to calculate the confidence degree, and the translation efficiency is improved.

Description of drawings

FIG. 1 is a flowchart of a machine translation method according to an embodiment of the present invention.

FIG. 2 is a block diagram of a machine translation apparatus according to another embodiment of the present invention.

Detailed ways

Various preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

machine translation method

The present embodiment provides a kind of machine translation method, comprises the following steps: Input the sentence to be translated; Calculate the similarity between the above-mentioned sentence to be translated and the source language sentence in the bilingual corpus 10; Based on the above-mentioned similarity in the above-mentioned bilingual corpus 10 A plurality of sentence pairs are selected as training corpus; using the above training corpus to train the translation system; and using the above translation system to translate the above sentence to be translated.

The details will be described below with reference to FIG. 1 . FIG. 1 is a flowchart of a machine translation method according to an embodiment of the present invention.

As shown in FIG. 1, first, in step S101, a sentence to be translated is input. In this embodiment, the sentence to be translated may be any sentence that needs to be translated known to those skilled in the art, and may be in any language, which is not limited in the present invention.

Next, in step S105, the similarity between the sentence to be translated and the source language sentence in the bilingual corpus 10 is calculated.

In this embodiment, the bilingual corpus 10 includes sentence pairs of a plurality of source languages and target languages, which can be any bilingual corpus known to those skilled in the art, such as English-Chinese corpus, English-German corpus, Japanese-Chinese corpus etc. This embodiment has no limitation on the bilingual corpus 10 . In addition, the bilingual corpus in this embodiment may be a corpus that only performs sentence alignment, or a corpus that performs word alignment on sentence pairs, and the present invention has no limitation on this.

In this embodiment, the similarity is a parameter indicating the degree of similarity between the sentence to be translated and the sentence in the source language, for example, a string-based similarity or a structured similarity may be used. When calculating the similarity, for example, the similarity can be calculated based on the edit distance between the sentence to be translated and the source language sentence in the bilingual corpus 10 . In addition, the similarity may also be calculated based on the syntactic structure between the sentence to be translated and the language sentences in the bilingual corpus 10 . In this embodiment, there is no limitation on the calculation method of the similarity, and any method known to those skilled in the art may be used.

Next, in step S110, a plurality of sentence pairs are selected from the bilingual corpus 10 based on similarity as training corpus.

When selecting a sentence pair, for example, according to the calculation result of the similarity, the sentences in the bilingual corpus 10 are sorted in order of similarity from large to small, and the first N sentences after the sorting are selected as the training corpus, N is an integer of 1 or more.

In addition, instead of sorting the sentence pairs, sentence pairs whose similarity in the bilingual corpus 10 is greater than a predetermined threshold may be selected as the training corpus. In this embodiment, there is no limitation on the method of selecting training corpus based on the similarity, which may be a combination of the above methods or any other method known to those skilled in the art.

Next, in step S115, the translation system is trained using the training corpus selected in step S110.

In this embodiment, the translation system may be any translation system known to those skilled in the art, as long as it can be used to translate the sentence to be translated. That is to say, any algorithm known to those skilled in the art can be used to train the translation system based on the training corpus.

Specifically, in the case that the bilingual corpus is a corpus that has only been aligned with sentences, first use any word alignment method known to those skilled in the art to perform word alignment on the selected training corpus, and then use the training corpus that has undergone word alignment to train the translation system. On the other hand, when the bilingual corpus is a corpus in which word alignment is performed on sentence pairs, the translation system is trained directly using the selected training corpus without performing the above-mentioned word alignment.

Preferably, the translation system in this embodiment includes a translation model (Translation Model, TM) and a language model (Language Model, LM). The translation model is trained using the bilingual sentence pairs and their word alignment information in the training corpus to describe the accuracy of conversion from the source language to the target language. The language model is trained using the target language sentences in the training corpus to describe the fluency of the translation results.

Finally, in step S120, the sentence to be translated is translated by using the translation system trained in step S115, and the translation result 20 is obtained.

In the machine translation method of this embodiment, by selecting the corpus with high similarity with the sentence to be translated in the bilingual corpus 10, and constructing a translation system in real time based on the selected corpus, a dynamic and targeted translation can be constructed system, thereby being able to improve translation quality.

In addition, in the machine translation method of this embodiment, by selecting the first N sentences after sorting, when there are a large number of corpus with a high similarity with the sentence to be translated in the bilingual corpus 10, the most similar and certain amount of corpus can be used Train the translation system so that not only the translation quality can be guaranteed, but also the processing burden of training the translation system can be reduced.

In addition, the machine translation method of this embodiment trains the translation system by selecting corpus with a similarity greater than a predetermined threshold, and can exclude corpus with low similarity, thereby avoiding the interference of corpus with low similarity on the translation system, and further Guaranteed translation accuracy.

Preferably, the machine translation method of this embodiment further includes the following step after step S120: saving the sentence to be translated and its translation result in the translation buffer. The translation buffer can be any memory known to those skilled in the art.

Furthermore, the machine translation method of this embodiment preferably further includes the following step after step S101: searching for the sentence to be translated in the translation buffer. If there is a sentence to be translated in the translation buffer, its translation result is directly obtained; if not, then step S105 is performed.

In the machine translation method of this embodiment, by saving the sentence to be translated and its translation result in the translation buffer, when the same sentence is translated next time, the translation result of the sentence can be directly obtained from the translation buffer, saving computing resources. Improved translation efficiency.

Preferably, the machine translation method of this embodiment further includes the following step after step S120: adding the above sentence to be translated and its translation result to the above bilingual corpus.

In the machine translation method of this embodiment, by adding sentences to be translated and their translation results into the bilingual corpus 10, the corpus data of the bilingual corpus 10 can be expanded, thereby improving the translation quality of subsequent translations.

Preferably, the machine translation method of this embodiment further includes the following steps before adding the sentence to be translated and its translation result: performing word alignment on the sentence to be translated and its translation result; and adding the word alignment result to the bilingual corpus 10 .

In this embodiment, any word alignment tool known to those skilled in the art can be used, such as the GIZA++ tool, to perform word alignment on the sentence to be translated and its translation result. In this embodiment, there is no limitation on the word alignment method.

In the machine translation method of this embodiment, by adding the word alignment result to the bilingual corpus, it can not only expand the corpus data of the bilingual corpus, improve the translation quality of subsequent translations, but also improve translation efficiency.

Preferably, the machine translation method of this embodiment further includes the following step before step S110: adding training data related to the user to the bilingual corpus 10 .

In the machine translation method of this embodiment, by adding user-related training data, such as aligned sentence pairs, context-related data, and word-aligned sentence pairs, etc., it can also achieve Purpose of User Adaptation.

Preferably, the machine translation method of this embodiment further includes the following step after the above-mentioned step S120: calculating the confidence of the above-mentioned translation result by using the above-mentioned similarity.

In this embodiment, the confidence level refers to the degree of credibility of the translation result. If the part most similar to the input sentence to be translated in the bilingual corpus 10 is not enough to learn translation knowledge, it can be known that the probability of using the entire bilingual corpus 10 to obtain the best translation is also very small. When selecting similar sentences, if the number of sentences whose similarity is higher than a certain value is lower than a predetermined threshold, it can be considered that the bilingual corpus 10 does not contain enough knowledge to translate the sentence to be translated. For example, when the number of sentences whose similarity is greater than a certain value is equal to or greater than a predetermined threshold, the degree of confidence may be considered high.

In the machine translation method of this embodiment, by using the similarity to calculate the confidence of the translation result, the confidence of the translation result can be obtained at the same time as the translation result, so that there is no need to use another method to calculate the confidence, and the translation efficiency is improved.

machine translation device

Under the same inventive conception, FIG. 2 is a block diagram of a machine translation device according to another embodiment of the present invention. The present embodiment will be described below with reference to this figure. For those parts that are the same as those in the previous embodiment, descriptions thereof are appropriately omitted.

The present embodiment provides a machine translation device 200, comprising: an input unit 201, which inputs a sentence to be translated; a calculation unit 205, which calculates the similarity between the sentence to be translated and the source language sentence in the bilingual corpus 10; Selection unit 210, which selects a plurality of sentence pairs in the above-mentioned bilingual corpus 10 based on the above-mentioned similarity, as training corpus; training unit 215, which utilizes the above-mentioned training corpus to train the translation system; and translation unit 220, which utilizes the above-mentioned translation system, Translate the above sentence to be translated.

The details will be described below with reference to FIG. 2 . As shown in FIG. 2, the input unit 201 inputs a sentence to be translated. In this embodiment, the sentence to be translated may be any sentence that needs to be translated known to those skilled in the art, and may be in any language, which is not limited in the present invention.

The calculation unit 205 calculates the similarity between the sentence to be translated and the source language sentence in the bilingual corpus 10 .

In this embodiment, the bilingual corpus 10 includes sentence pairs of a plurality of source languages and target languages, which can be any bilingual corpus known to those skilled in the art, such as English-Chinese corpus, English-German corpus, Japanese-Chinese corpus etc. This embodiment has no limitation on the bilingual corpus 10 . In addition, the bilingual corpus in this embodiment may be a corpus that only performs sentence alignment, or a corpus that performs word alignment on sentence pairs, and the present invention has no limitation on this.

In this embodiment, the similarity is a parameter indicating the degree of similarity between the sentence to be translated and the sentence in the source language, for example, a string-based similarity or a structured similarity may be used. When calculating the similarity, the calculation unit 205 may calculate the similarity based on, for example, the edit distance between the sentence to be translated and the source language sentence in the bilingual corpus 10 . In addition, the calculation unit 205 can also calculate the similarity based on the syntactic structure between the sentence to be translated and the language sentences in the bilingual corpus 10 . In this embodiment, there is no limitation on the calculation method of the similarity, and any method known to those skilled in the art may be used.

The selection unit 210 selects a plurality of sentence pairs from the bilingual corpus 10 based on similarity as training corpus.

The selection unit 210 preferably includes a sorting unit. When selecting a sentence pair, at first the sorting unit sorts the sentences in the bilingual corpus 10 in order of similarity from large to small according to the calculation result of the similarity, and then the selection unit 210 selects the top N sentence pairs after sorting as Training corpus, N is an integer greater than 1.

In addition, the selection unit 210 may not have a sorting unit for sorting the sentence pairs, but directly selects the sentence pairs in the bilingual corpus 10 whose similarity is greater than a predetermined threshold as the training corpus. In this embodiment, there is no limitation on the method of selecting training corpus based on the similarity, which may be a combination of the above methods or any other method known to those skilled in the art.

The training unit 215 uses the training corpus selected by the selection unit 210 to train the translation system.

In this embodiment, the translation system may be any translation system known to those skilled in the art, as long as it can be used to translate the sentence to be translated. That is to say, any algorithm known to those skilled in the art can be used to train the translation system based on the training corpus.

Specifically, in the case that the bilingual corpus is a corpus that has only been aligned with sentences, the training unit 215 also includes a word alignment unit, which utilizes any word alignment method known to those skilled in the art to perform word alignment on the selected training corpus, Then the training unit 215 uses the word-aligned training corpus to train the translation system. On the other hand, when the bilingual corpus is a corpus in which word alignment is performed on sentence pairs, the training unit 215 does not need to include a word alignment unit, and directly uses the selected training corpus to train the translation system.

Preferably, the translation system in this embodiment includes a translation model (Translation Model, TM) and a language model (Language Model, LM). The translation model is trained using the bilingual sentence pairs and their word alignment information in the training corpus to describe the accuracy of conversion from the source language to the target language. The language model is trained using the target language sentences in the training corpus to describe the fluency of the translation results.

The translation unit 220 uses the translation system trained by the training unit 215 to translate the sentence to be translated to obtain the translation result 20 .

The machine translation device 200 of this embodiment can construct a dynamic and targeted translation system by selecting corpus with high similarity with the sentence to be translated in the bilingual corpus 10, and constructing a translation system in real time based on the selected corpus. translation system, thereby being able to improve translation quality.

In addition, the machine translation device 200 of this embodiment, by selecting the first N sentences after sorting, when there are a large number of corpus with a high similarity to the sentence to be translated in the bilingual corpus 10, the most similar, certain number of sentences can be used. The corpus trains the translation system, so that not only the translation quality can be guaranteed, but also the processing burden of training the translation system can be reduced.

In addition, the machine translation device 200 of this embodiment trains the translation system by selecting corpus with a degree of similarity greater than a predetermined threshold, and can exclude corpus with a low degree of similarity, thereby avoiding the interference of corpus with a low degree of similarity on the translation system. Further guarantee the translation accuracy.

Preferably, the machine translation device 200 of this embodiment further includes a storage unit for storing sentences to be translated and their translation results in the translation buffer. The translation buffer can be any memory known to those skilled in the art.

Furthermore, the machine translation device 200 of this embodiment preferably further includes a search unit for searching the sentence to be translated in the translation buffer. If there is a sentence to be translated in the translation buffer, its translation result will be obtained directly; if not, the calculation unit 205 , the selection unit 210 , the training unit 215 and the translation unit 220 will be used for translation.

The machine translation device 200 of this embodiment, by saving the sentence to be translated and its translation result in the translation buffer, when translating the same sentence next time, can directly obtain the translation result of the sentence from the translation buffer, saving computing resources , improving translation efficiency.

Preferably, the machine translation device 200 of this embodiment further includes a sentence pair adding unit for adding the above-mentioned sentences to be translated and their translation results into the above-mentioned bilingual corpus.

The machine translation device 200 of this embodiment can expand the corpus data of the bilingual corpus 10 by adding sentences to be translated and their translation results into the bilingual corpus 10, thereby improving the translation quality of subsequent translations.

Preferably, the machine translation device 200 of this embodiment further includes a word alignment unit for word alignment of the sentence to be translated and its translation result, and a word alignment result adding unit for adding the word alignment result to the bilingual corpus 10 .

In this embodiment, any word alignment tool known to those skilled in the art can be used, such as the GIZA++ tool, to perform word alignment on the sentence to be translated and its translation result. In this embodiment, there is no limitation on the word alignment method.

The machine translation device 200 of this embodiment can not only expand the corpus data of the bilingual corpus, improve the translation quality of subsequent translations, but also improve translation efficiency by adding word alignment results to the bilingual corpus.

Preferably, the machine translation device 200 of this embodiment further includes a training data adding unit for adding training data related to the user to the bilingual corpus 10 . Optionally, the above-mentioned alignment result adding unit may also be used as a training data adding unit.

The machine translation device 200 of this embodiment, by adding user-related training data, such as aligned sentence pairs, context-related data, and word-aligned sentence pairs, can also be used in the case of insufficient training data. To achieve the purpose of user adaptation.

Preferably, the machine translation device 200 in this embodiment further includes a confidence calculation unit for calculating the confidence of the translation result by using the similarity. Optionally, the above calculation unit 205 may also be used as a confidence calculation unit.

In this embodiment, the confidence level refers to the degree of credibility of the translation result. If the part most similar to the input sentence to be translated in the bilingual corpus 10 is not enough to learn translation knowledge, it can be known that the probability of using the entire bilingual corpus 10 to obtain the best translation is also very small. When selecting similar sentences, if the number of sentences whose similarity is higher than a certain value is lower than a predetermined threshold, it can be considered that the bilingual corpus 10 does not contain enough knowledge to translate the sentence to be translated. For example, when the number of sentences whose similarity is greater than a certain value is equal to or greater than a predetermined threshold, the degree of confidence may be considered high.

The machine translation device 200 of this embodiment, by using the similarity to calculate the confidence of the translation result, can obtain the confidence of the translation result at the same time as the translation result, so that there is no need to use another method to calculate the confidence, and the translation efficiency is improved. .

Although the machine translation method and the machine translation device of the present invention have been described in detail through some exemplary implementations, the above implementations are not exhaustive, and those skilled in the art can implement various translations within the spirit and scope of the present invention. changes and modifications. Accordingly, the present invention is not limited to these embodiments, and the scope of the present invention is determined only by the appended claims.

Claims (10)

1. a machine translation apparatus, comprising:

Input block, it inputs sentence to be translated;

Computing unit, it calculates the similarity between the source language sentence in above-mentioned sentence to be translated and bilingualism corpora;

Selection unit, it selects multiple sentence pair, as corpus based on above-mentioned similarity in above-mentioned bilingualism corpora;

Training unit, it utilizes above-mentioned corpus, training translation system; And

Translation unit, it utilizes above-mentioned translation system, translates above-mentioned sentence to be translated.

2. machine translation apparatus according to claim 1, wherein, above-mentioned selection unit comprises:

Sequencing unit, it sorts to above-mentioned similarity order from big to small to the sentence in above-mentioned bilingualism corpora;

Above-mentioned selection unit, the top n sentence pair after selected and sorted, as above-mentioned corpus, N is the integer of more than 1.

3. machine translation apparatus according to claim 1, wherein, above-mentioned selection unit, selects the above-mentioned similarity in above-mentioned bilingualism corpora to be greater than the sentence pair of predetermined threshold value, as above-mentioned corpus.

4. machine translation apparatus according to claim 1, wherein, above-mentioned computing unit, utilizes the editing distance between the source language sentence in above-mentioned sentence to be translated and above-mentioned bilingualism corpora to calculate above-mentioned similarity.

5. machine translation apparatus according to claim 1, wherein, above-mentioned computing unit, calculates the similarity of the syntactic structure between the source language sentence in above-mentioned sentence to be translated and above-mentioned bilingualism corpora.

6. machine translation apparatus according to claim 1, also comprises:

Storage unit, it preserves above-mentioned sentence to be translated and translation result thereof at translation buffer; And

Search unit, it searches above-mentioned sentence to be translated at above-mentioned translation buffer after the above-mentioned sentence to be translated of above-mentioned input block input.

7. machine translation apparatus according to claim 1, also comprises:

Above-mentioned sentence to be translated and its translation result are added the sentence of above-mentioned bilingualism corpora to adding device; Or

The word alignment unit of word alignment is carried out to above-mentioned sentence to be translated and its translation result and word alignment result is added the word alignment result adding device of above-mentioned bilingualism corpora.

8. machine translation apparatus according to claim 1, also comprises:

Training data adding device, it adds and user-dependent training data in above-mentioned bilingualism corpora.

9. machine translation apparatus according to claim 1, also comprises:

Confidence computation unit, it utilizes the degree of confidence of the above-mentioned translation result of above-mentioned Similarity Measure.

10. a machine translation method, comprises the following steps:

Input sentence to be translated;

Calculate the similarity between the source language sentence in above-mentioned sentence to be translated and bilingualism corpora;

In above-mentioned bilingualism corpora, multiple sentence pair is selected, as corpus based on above-mentioned similarity;

Utilize above-mentioned corpus, training translation system; And

Utilize above-mentioned translation system, above-mentioned sentence to be translated is translated.