KR20120048139A - Automatic translation device and method thereof - Google Patents

Abstract

PURPOSE: An automatic translation apparatus and a method thereof are provided to increase the coverage of a translation memory and to increase the performance of automatic translation without extension of an additional translation memory. CONSTITUTION: A database searching unit(230) searches for a similar sentence of a translation memory based on a generated keyword. A similarity calculating unit(240) calculates the similarity of similarity sentences based on a weighted value of a word. The word is a word for substituting/deleting/adding to be modified into the input sentence. The weighted value is matched for substituting/deleting/adding a task.

Description

Automatic translation device and method

The present invention relates to an automatic translation apparatus and a method thereof, and more particularly, to an automatic translation apparatus and method capable of generating a high quality translation results with a wide coverage.

In general, the conventional translation memory adopted by the automatic translation system has the advantage of obtaining a high quality translation, but it has a limitation that only a limited sentence can be translated by outputting the translation only when it is completely matched with the input requiring translation. have.

On the other hand, the translation support system has the advantage of calculating not only completely identical sentences but also similarity of sentences and retrieving relatively similar sentences from the translation memory, but the final translation is generated by the translator. have.

Therefore, in order to utilize the translation memory used in the translation support system in the automatic translation system, additional work for generating an appropriate band sentence is required.

For example, a method of increasing the coverage of an existing translation memory by expanding the translation memory in advance has been proposed. However, there was still a problem for such a work, such as manual work or limited scalability.

Disclosure of Invention The present invention has been made to solve the problems of the general technology as described above, and an object thereof is to provide an automatic translation apparatus and method capable of increasing the coverage of a translation memory without an additional translation memory expansion.

Another object of the present invention is to provide an automatic translation apparatus and method that can improve the performance of the automatic translation.

As an aspect of the present invention for solving the above technical problem, an automatic translation device, an input sentence keyword generation unit for generating a keyword for the input sentence; A DB search unit searching for one or more similar sentences in a translation memory based on the generated keyword; A similarity calculation unit that calculates a similarity degree of the searched one or more similar sentences based on the parts of speech to be replaced / deleted / added to the input sentence and weights corresponding to substitution / deletion / addition operations; And a band sentence generation unit configured to generate a band sentence according to a substitution / deletion / addition operation on the searched one or more similar sentences.

Here, the keyword generation unit may include a part-of-speech such as a verb, a question mark, a conjunction, a preposition, and a noun or adjective immediately preceding such a part-of-speech or punctuation.

In addition, the similarity calculator may calculate the similarity by the sum of the products of the parts of speech to be replaced / deleted / added and the weights corresponding to the substitution / deletion / addition in order to correct the input sentence, and the calculated similarity. According to one or more similar sentences can be sorted according to.

Further, the band sentence generation unit determines whether words requiring substitution / deletion / addition are arranged or generally do not need to be translated according to the similarity order, and is based on the first similar sentence that satisfies the determination result. To generate a band statement.

In another aspect of the present invention, an automatic translation method includes: generating a keyword for an input sentence; Searching for one or more similar sentences in a translation memory based on the generated keywords; Calculating similarity of the retrieved one or more similar sentences based on the parts of speech to be replaced / deleted / added and the weights corresponding to the substitution / deleted / added operations to correct the input sentence; And generating a band sentence according to a substitution / deletion / addition operation on the searched one or more similar sentences.

Here, the keyword generating step includes a part of a verb, a question mark, a conjunction, a preposition such as a preposition, and a noun or adjective immediately preceding such a part of speech or punctuation.

In addition, the similarity calculation step, the similarity can be calculated by the sum of the product of the parts of the word to be replaced / deleted / added to the input sentence and the weight corresponding to the replacement / deletion / addition operation, the calculated The method may further include sorting the searched one or more similar sentences according to similarity.

Furthermore, the band sentence generation step may include determining whether words requiring substitution / deletion / addition are arranged or generally not translated according to the similarity order, and satisfying the determination result. A band sentence can be generated based on the first similar sentence.

The present invention has the following effects.

First, there is provided an automatic translation apparatus and method that can increase the coverage of a translation memory without additional translation memory expansion.

Second, there is provided an automatic translation apparatus and method that can enhance the performance of automatic translation without additional translation memory expansion.

1 is a functional block diagram illustrating a process of generating a translation memory according to the present invention.
2 is a functional block diagram illustrating an automatic translation process according to the present invention.
3 is a diagram illustrating an embodiment of a translation memory DB according to the present invention.
4 is a flowchart showing an automatic translation method according to the present invention.

The following embodiments are a combination of elements and features of the present invention in a predetermined form. Each component or feature may be considered to be optional unless otherwise stated. Each component or feature may be implemented in a form that is not combined with other components or features. In addition, some of the elements and / or features may be combined to form an embodiment of the present invention. The order of the operations described in the embodiments of the present invention may be changed. Some configurations or features of certain embodiments may be included in other embodiments, or may be replaced with corresponding configurations or features of other embodiments.

Embodiments of the invention may be implemented through various means. For example, embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof.

For a hardware implementation, the method according to embodiments of the present invention may be implemented in one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs) , Field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, and the like.

In the case of an implementation by firmware or software, the method according to the embodiments of the present invention may be implemented in the form of a module, a procedure, or a function that performs the functions or operations described above. The software code may be stored in a memory unit and driven by a processor. The memory unit may be located inside or outside the processor, and may exchange data with the processor by various known means.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are intended to provide further explanation of the invention, and are not intended to limit the scope of the invention.

1 is a functional block diagram illustrating a process of generating a translation memory according to the present invention. Referring to FIG. 1, the translation memory tagging unit 120 performs tagging on an original text and a band text of a band corpus 110 of an unprocessed form. The word alignment unit 130 performs word alignment using the system translation dictionary 140 and a statistical method, and constructs a word alignment dictionary 150. The translation memory keyword generation unit 160 generates a keyword to be used as a DB key by using the tagging result of the original text. The translation memory DB generator 170 constructs a translation memory 180 using the generated keyword as a key.

Hereinafter, the generation of the translation memory to be used in the automatic translation apparatus and method according to the present invention will be described in detail with an example focusing on the functions performed in each component.

The band corpus 110 is comprised of the original text and the band text of an unprocessed form, as in the following example.

I'll call back in about five minutes. || I'll call you back in about five minutes.

We are having a meeting at ten tomorrow morning. ||

Split the salad. || Please share your salad.

Rainy days get me down. ||

How is your food? || How does the food taste?

Can i order the food in advance? || Can I order food in advance?

We have to have important meeting right now. || We need to have an important meeting now.

The translation memory tagging unit 120 morphologically analyzes words constituting each of the original text and the band sentence as described below, and simultaneously determines and tags the parts of speech of the words.

i / pronouns will / verb verb / verb circle back / verb in / preposition about / verb five / verb minutes / plural noun ./. || My + will + be back in + after about 5 + minutes.

we / pronouns are / verb present having / verbal action a / article meeting / singular noun at / preposition ten / investigation tomorrow / adverb morning / singular noun ./. || Tomorrow will open the meeting + at 10 + + + tomorrow morning.

The word alignment unit 130 constructs a word alignment dictionary 150 while performing word alignment using a statistical model such as IBM model 1 used in statistical machine translation.

Here, the word alignment refers to matching positions of the words constituting the original text and the band sentence in the corresponding sentences, and the word alignment dictionary corresponds to a word of the band sentence having a similar meaning with respect to one word in the original text. Refers to the database you placed.

1) One example of word sort:

i / pronouns will / verb verb / verb circle back / verb in / preposition about / verb five / verb minutes / plural noun ./. || My + will + be back in + after about 5 + minutes. || 1-1, 6-3, 7-4, 8-5, 9-9

we / pronouns be / verb present having / verbal action a / article meeting / singular noun at / preposition ten / investigation tomorrow / adverb morning / singular noun ./. || Tomorrow will open the meeting + at 10 + + + tomorrow morning. || 7-2, 5-5, 8: 9-1, 10-10

Here, 1-1 means that the first word in the original text and the first word in the band text correspond to each other.

2) An example of a word sort dictionary:

i-Me, my, ...

food-food, cooking, ...

meeting-meeting, ...

five-5, five, oh, ...

ten-10, ten, ten, ...

The translation memory keyword generated by the translation memory keyword generation unit 160 may be composed of verbs, interrogators, subordinate conjunctions, prepositions, and nouns or adjectives immediately before such a part of speech or punctuation.

For example, i will call back in aout five minutes. There is an original text, and the tagging result of this text is as follows.

i / pronouns will / verb verb / verb circle back / verb in / preposition about / verb five / verb minutes / plural noun ./.

Pronouns (i), verbs (will), adverbs (back, about), and rhetoric (five) in the original text are excluded because they are not included in the keyword generation. Thus, the keyword call, the preposition in, and the noun minutes immediately before the punctuation are included in the keyword, and the keyword is call in minutes.

In another example, We are having a meeting at ten tomorrow morning. There is an original text, and the tagging result of this text is as follows.

we / pronouns are / verb present having / verbal action a / article meeting / singular noun at / preposition ten / investigation tomorrow / adverb morning / singular noun ./.

Pronouns (we), articles (a), investigations (ten), and adverbs (tomorrow) in the original text are excluded because they are not included in the keyword generation. Therefore, the keywords are having, the meeting just before the preposition, the preposition at, and the morning just before the punctuation are included in the keyword, and the keyword is having meeting at morning.

An embodiment of the translation memory 180 constructed by the translation memory DB generator 170 is illustrated in FIG. 3. As shown in FIG. 3, the translation memory includes tagging results for original texts and translations, and word alignment results, and is stored in the same DB entry using the generated keyword as a DB key.

Through such components and / or processes, a translation memory to be used in the automatic translation apparatus and method according to the present invention is generated.

2 is a functional block diagram illustrating an automatic translation process according to the present invention. Referring to FIG. 2, when an original text to be translated is input, the input sentence tagging unit 210 performs tagging on the original text. The input sentence keyword generator 220 generates a keyword for the input sentence in the same manner as the translation memory keyword generator 160, and uses the generated keyword to generate the translation memory 180 in the DB search unit 230. Search. The similarity calculator 240 calculates a similarity degree for one or more similar sentences retrieved from the translation memory 180 by using a dynamic programming technique such as an edit distance, and the band sentence generator 250 satisfies a band sentence generation condition. A band sentence for the original text is generated using the word sort dictionary 150 and the translation dictionary 140 for the similar sentence.

Hereinafter, the automatic translation apparatus and method according to the present invention will be described in detail through an example of the English-Korean translation around the components. For example, suppose the sentence I'll call you back in about ten minutes.

The input sentence tagging unit 210 performs a morphological analysis on the words constituting the input sentence as described below, and simultaneously determines and tags the parts of speech of the words.

i / pronouns will / verb verb / verb round back / adverb in / preposition about / verb ten / verb minutes / plural noun ./.

The input sentence keyword generator 220 generates a keyword in the same manner as the method of the translation memory keyword generator 160. That is, the input sentence keyword may be composed of verbs, interrogatives, subordinate adjuncts, prepositions such as prepositions, and nouns or adjectives immediately before such a part of speech or punctuation.

Pronouns (i), modal verbs (will), adverbs (back, about), and rhetorical sentences (ten) are excluded because they are not included in keyword generation. Thus, the keyword call, the preposition in, and the noun minutes immediately before the punctuation are included in the keyword, and the keyword is call in minutes.

The DB search unit 230 searches for a DB entry having the same keyword in the translation memory 180 by using call in minutes which is a keyword of the generated input sentence. An example result of the present invention is shown in FIG.

The similarity calculator 240 calculates the similarity using a dynamic programming technique such as an edit distance. First, the minimum corrections necessary for correcting each of a plurality of sentences in the same keyword as an input sentence, that is, substitution / deletion Determine if further work is needed. In case of substitution, only the same parts of speech may be used. In case of different parts of speech, correction shall be made by deleting and adding.

The similarity calculation is performed based on the parts of parts of the word to be replaced / deleted / added and the weights corresponding to the parts of the parts / substituted / deleted / added. Same as

Parts of speech weight verb 10 Narrative adjective 10 preposition 8 Head noun 5 Etc 3

Corrective Actions weight same -3 Add 2 delete One substitution 2

The similarity calculation may be expressed as the sum of the product of the parts of the words to be substituted / deleted / added and the weights corresponding to the substitution / deleted / added operations. For example, it may be expressed as follows.

Here, Sim is similarity, pos_weight _i is the part-of-speech weight of the i-th word, and edit_cost _i is the modification work weight of the i-th word.

Similarity is calculated for each of one or more similar sentences under the same keyword searched by the DB search unit 230. For example, the similarity is calculated according to Equation 1 with respect to the sentences shown in FIG.

Sentence 1: may (delete) i will (add) call you (delete) back in about five (substitution) minutes? (Delete)

Similarity: 0-[3 + 3 (-3) + 3 * 2 + 10 (-3) + 3 + 3 (-3) +8 (-3) +3 (-3) + 3 * 2 + 5 (- 3) +3] = 75

Sentence 2: may (delete) i will (add) call you (delete) in about thirty (substitute) minutes? (Delete)

Similarity: 0-[3 + 3 (-3) + 3 * 2 + 10 (-3) + 3 + 8 (-3) +3 (-3) + 3 * 2 + 5 (-3) +3] = 66

Sentence 3: i (added) will (added) call again (substituted) in about 10 (substituted) minutes

Similarity: 0-[3 * 2 + 3 * 2 + 10 (-3) + 3 * 2 + 8 (-3) +3 (-3) + 3 * 2 + 5 (-3)] = 54

Sentence 4: i will call back in about five minutes

Similarity: 0-[3 (-3) +3 (-3) +10 (-3) +3 (-3) +8 (-3) +3 (-3) + 3 * 2 + 5 (-3) ] = 99

Sentence 5: i (added) will (added) call back in about ten minutes

Similarity: 0-[3 * 2 + 3 * 2 + 10 (-3) +3 (-3) +8 (-3) +3 (-3) +3 (-3) +5 (-3)] = 84

The similarity calculator 240 may also sort the searched one or more similar sentences according to the similarity in descending order according to the similarity according to the calculated similarity.

1: i will call back in about five minutes

2: i (added) will (added) call back in about ten minutes

3: may (delete) i will (add) call you (delete) back in about five minutes (?)

4: may (Delete) i will (Add) call you (Delete) in about thirty (Substitute) minutes? (Delete)

5: i (added) will (added) call again (substituted) in about 10 (substituted) minutes

The band sentence generation unit 250 determines whether or not the words that need to be replaced, deleted, or added to the searched one or more similar sentences generally do not need to be translated. For example, in the case of a word requiring substitution / deletion, it is determined whether the words are all sorted or correspond to words that are not generally translated in English-English translations such as a, an, the, your, and my. For words that require further work, it is determined whether they correspond to adverbs, punctuation, or words that are not normally translated in English-to-English translations such as a, an, the, your, my, and so on.

This determination may be performed sequentially on the one or more similar sentences arranged in descending order according to the similarity, and the band sentence generation may be performed on the first similar sentence that satisfies this determination condition.

For example, in the case of a word that needs to be deleted, the band word aligned in the band sentence of the similar sentence is removed, and in the case of the word that needs to be replaced, the band word for the word of the input sentence to be substituted is changed from the word sort dictionary or the system dictionary. After the selection, the bandwords aligned in the band statement are replaced with the selected bandword.

For words that require additional work, for example, a band word of an adverb to be added may be appropriately inserted into a band sentence, where the generation position is immediately before the band word if the next word of the adverb is sorted, and at the top of the sentence if it is not sorted. Create it earlier.

According to the present embodiment, a method of performing a determination on whether one or more similar sentences are arranged in descending order according to the similarity and sequentially satisfying the condition for generating a band sentence is applied to the present embodiment. It is determined whether the generation condition is satisfied.

i will call back in about five minutes || My + will call you again in about +5 minutes. || 1-1, 6-3, 7-4, 8-5

For words that require substitution, it is determined whether the words are all sorted or correspond to words that are not normally translated in English-Korean translations such as a, an, the, your, my, etc. Since 4 is aligned, the band generation condition is satisfied, and thus a band sentence is generated based on this. If five-4 is not aligned, it is determined whether the sentence generation condition is satisfied for the sentence 5 having the second order similarity.

Therefore, if you replace ten's band word "10" in the word sort dictionary (150) with the place of "5" in the band sentence, "the + will re-add + after about 10 + minutes." The result is a recombination of the separated morphemes, resulting in the final band statement "I'll call again in about 10 minutes."

4 is a flowchart showing an automatic translation method according to the present invention. Referring to FIG. 4, when a sentence to be translated is input by a user or a system, a keyword is generated for the input sentence (S400). Based on the generated keyword, one or more similar sentences are searched in the translation memory (S420). Subsequently, the similarity of the searched one or more similar sentences is calculated based on the parts of speech to be replaced / deleted / added and the weights corresponding to the replacement / deleted / added operations in order to correct the input sentence (S440). Finally, a band sentence is generated according to the substitution / deletion / addition operation on the searched one or more similar sentences (S460). Here, the description of the function and structure of the components of the automatic translation apparatus according to the present invention may also be employed as it is in the automatic translation method according to the present invention.

The following shows an example in which band sentences for each input sentence are generated according to the automatic translation apparatus and method according to the present invention.

(1) Example 1

Translation memory: We are having a meeting at ten tomorrow morning. || I will have a meeting tomorrow at 10 o'clock.

Input Sentence: We're having a meeting at 9 tomorrow morning.

Kim Moon-moon: I'll have a meeting tomorrow at 9:00.

(2) Example 2

Translation memory: Rainy days get me down. || On rainy days, I feel down.

Input sentence: Rainy days always get me down.

Band Moon: Always feels down on a rainy day.

(3) Example 3

Translation memory: What do you personally recommend? || What would you like personally?

Input sentence: What do you recommend?

Band Moon: What would you like?

(4) Example 4

Translation memory: When would be a convenient time to reach him? || When is the best time to speak with him?

Input Sentence: Well, when would be a convenient time to reach him?

Band Moon: Well, when is the best time to talk to him?

The invention can be embodied in other specific forms without departing from the spirit and essential features of the invention. Accordingly, the above detailed description should not be construed as limiting in all aspects and should be considered as illustrative. The scope of the invention should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of the invention are included in the scope of the invention. In addition, claims that do not have an explicit citation in the claims may be combined to form an embodiment or included in a new claim by amendment after the application.

The automatic translation apparatus and method of the present invention is applicable not only to the apparatus and method in which the original purpose is automatic translation, but also to any technical field that requires specific functions to be performed based on the automatic translation.

150: word alignment dictionary 180: translation memory
210: input sentence tagging unit 220: input sentence keyword generation unit
230: DB search unit 240: Similarity calculator
250: band sentence generation unit

Claims (1)

An input sentence keyword generator to generate a keyword for the input sentence;
A DB search unit searching for one or more similar sentences in a translation memory based on the generated keyword;
A similarity calculation unit that calculates a similarity degree of the searched one or more similar sentences based on the parts of speech to be replaced / deleted / added and weights corresponding to substitution / deletion / addition in order to correct the input sentence; And
And a band sentence generator for generating a band sentence in response to the substitution / deletion / addition of the searched one or more similar sentences.

Images