JPH08305706A - Method and device for learning rule of dictionary for machine translation system - Google Patents

Abstract

PURPOSE: To provide a machine translation system dictionary rule learning method and device capable of correctly adding/correcting a dictionary and a rule without generating a difference even when machine translation concept structure and ideal translation concept structure expressing the same concept are apparently different from each other. CONSTITUTION: An input sentence and a previously prepared ideal translated sentence which are mutually in parallel translation relation are inputted from a translated sentence reading means 1, the ideal translation concept structure of an objective language is generated from the read ideal translated sentence by an ideal translation concept structure generating means 2, the machine translation concept structure of a machine translation sentence is generated by a machine translation concept structure generating means 31, and the structure of the machine translation concept is converted by a machine translation concept structure converting means 61 based upon a structure conversion rule 611 for absorbing a difference between a machine translation concept structure generating method and an ideal translation concept structure generating method. The structure-converted machine translation concept structure and the ideal translation concept structure are analyzed by a concept structure analyzing means 71 and a dictionary 5 or a rule 4 is added/corrected by a dictionary/rule adding/ correcting means 72 baed upon analyzed results.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dictionary and rules used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. TECHNICAL FIELD The present invention relates to a dictionary / rule learning method and device for a machine translation system that changes the.

[0002]

2. Description of the Related Art Conventionally, when constructing a machine translation system dictionary or rules, it is common to manually construct a machine translation system dictionary or rules by referring to a dictionary or the like used by humans for translation. Is. In addition, the dictionary and rules for the machine translation system constructed in this way are the cause of the translation when an inappropriate translation is generated based on the translation result of the machine translation system. Experts who have a good understanding of the internal processing of machine translation systems have manually added and modified the rules and rules.

As a method of automating the work of adding and modifying the dictionary and rules, a difference is extracted from the difference between the machine translation conceptual structure and the ideal translation conceptual structure, and the dictionary is classified by classifying the differences according to the solution method in the machine translation system. A method is known in which a rule is additionally corrected and the dictionary or rule that has been additionally corrected is used and repeated until an optimum dictionary or rule is registered.

[0004]

In the above-mentioned conventional dictionary for machine translation system and automatic addition / correction method of rules, since the machine translation conceptual structure generating means and the ideal translation conceptual structure generating means are different, they represent the same concept. Nevertheless, the expression format may be different. In that case, in the above conventional method, when the difference is extracted from the difference between the machine translation conceptual structure and the ideal translation conceptual structure, a difference occurs even though they represent the same concept, and a dictionary or rule is added in an incorrect form. There is a problem of being fixed.

The present invention has been made in view of the above,
The purpose is to make it possible to add and correct dictionaries and rules appropriately without causing a difference even if the machine translation conceptual structure expressing the same concept and the ideal translation conceptual structure are apparently different. It is to provide a system dictionary / rule learning method and device.

[0006]

In order to achieve the above object, the present invention according to claim 1 converts an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning method for a machine translation system for changing a dictionary and / or a rule used in a machine translation system having means for converting, the method being generated when the machine translation system translates the first natural language. A machine translation concept structure conversion step for converting a machine translation concept structure based on a structure conversion rule that absorbs a difference between the machine translation concept structure generation method and the ideal translation concept structure generation method; and a machine translation concept structure conversion step. Conceptual structure converted and the second
Conceptual structure analysis step of analyzing an ideal translated conceptual structure generated by analyzing the ideal translated sentence of the natural language, and dictionary / rule addition change for additionally modifying the dictionary or rule based on the result of the conceptual structure analysis step. The point is to have steps and.

The present invention according to claim 2 is used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning method for a machine translation system for changing a dictionary and / or a rule, comprising: a parallel translation reading step for reading the input sentence and an ideal translated sentence prepared in advance, which are in parallel with each other; An ideal translation conceptual structure generating step of generating an ideal translation conceptual structure of a language, a machine translation conceptual structure generating step of generating a machine translation conceptual structure of the machine translated sentence, and a machine translation conceptual structure generating method and an ideal of the machine translation conceptual structure. A machine translation concept structure conversion step for performing structure conversion based on a structure conversion rule that absorbs a difference in translation concept structure generation method, and a machine translation concept structure conversion step. A conceptual structure that has been structurally converted and the ideal translation conceptual structure are analyzed, and a dictionary / rule addition / modification step that additionally modifies the dictionary or rule based on the result of the conceptual structure analysis step. Is the gist.

Further, the present invention according to claim 3 provides the invention according to claim 2.
In the invention described above, the machine translation conceptual structure converting step includes a condition of the machine translation conceptual structure and the ideal translation conceptual structure in a structure conversion rule that absorbs a difference between the machine translation conceptual structure generating method and the ideal translation conceptual structure generating method. The main point is to include.

The present invention according to claim 4 is a dictionary used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language, and And / or a machine translation system dictionary / rule learning method for changing rules, wherein the ideal translation conceptual structure generated by analyzing the ideal translation of the second natural language is a machine translation conceptual structure generation method and an ideal translation. An ideal translation concept structure conversion step for performing structure conversion based on a structure conversion rule that absorbs a difference in concept structure generation method, a concept structure converted in the ideal translation concept structure conversion step, and the first natural language. A concept structure analysis step of analyzing a machine translation concept structure generated when the machine translation system translates, and the dictionary or rule is added based on the result of the concept structure analysis step. And summarized in that with a positive add dictionary rules changing step.

Further, the present invention according to claim 5 is used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning method for a machine translation system for changing a dictionary and / or a rule, comprising: a parallel translation reading step for reading the input sentence and an ideal translated sentence prepared in advance, which are in parallel with each other; An ideal translation conceptual structure generating step of generating an ideal translation conceptual structure of a language, a machine translation conceptual structure generating step of generating a machine translation conceptual structure of the machine translated sentence, and a machine translation conceptual structure generating method and an ideal of the ideal translation conceptual structure. An ideal translation concept structure conversion step for performing structure conversion based on a structure conversion rule that absorbs a difference in translation concept structure generation method, and an ideal translation concept structure conversion step. A concept structure analysis step of analyzing the concept structure that has been structurally converted and the machine-translated concept structure, and a dictionary / rule addition / modification step of additionally modifying the dictionary or rule based on the result of the concept structure analysis step. Is the gist.

Further, the present invention according to claim 6 provides the invention according to claim 5.
In the invention described above, the ideal translation conceptual structure converting step includes a machine translation conceptual structure and an ideal translation conceptual structure in a condition of a structural conversion rule that absorbs a difference between the machine translation conceptual structure generating method and the ideal translation conceptual structure generating method. The main point is to include both.

The present invention according to claim 7 is a dictionary used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language, and And / or a machine / translation system dictionary / rule learning device for changing rules, which inputs a bilingual sentence that is a pair of a first natural language sentence and a second natural language ideal translation sentence that are in a bilingual relationship. The sentence reading means, the ideal translation conceptual structure generating means for analyzing the ideal translated sentence of the second natural language input by the bilingual sentence reading means, and generating the ideal translation conceptual structure, and the parallel translation sentence inputting means. Machine translation conceptual structure generating means for generating a machine translation conceptual structure of a target language which is included in a process of machine translating a first natural language sentence, and a machine translation conceptual structure generating method and an ideal translation conceptual structure generating the machine translation conceptual structure. Machine translation concept structure conversion means for performing structure conversion based on a structure conversion rule that absorbs a difference in law, ideal translation concept structure generated by the ideal translation concept structure generation means, and machine translation concept structure conversion means The gist is to have a conceptual structure analyzing means for analyzing the conceptual structure thus obtained, and a dictionary / rule adding / changing means for additionally modifying the dictionary or rule based on the result analyzed by the conceptual structure analyzing means.

Further, the present invention according to claim 8 is used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning device for a machine translation system for changing a dictionary and / or a rule, wherein a bilingual sentence that is a pair of a first natural language sentence and a second natural language ideal translation sentence that are in a bilingual relationship is input. Parallel translation reading means, ideal translation conceptual structure generating means for analyzing the ideal translation of the second natural language input by the parallel translation reading means, and generating an ideal translation conceptual structure, and input by the parallel translation reading means. A machine translation concept structure generating means for generating a machine translation concept structure of a target language which is included in a process of machine translating the generated first natural language sentence, and a machine translation concept structure generation method and an ideal translation concept for the ideal translation concept structure. Structure An ideal translation concept structure conversion means for performing structure conversion based on a structure conversion rule that absorbs a difference in generation method, and a machine translation concept structure generated by the machine translation concept structure generation means and the ideal translation concept structure conversion means. The gist is to have a conceptual structure analyzing means for analyzing the obtained conceptual structure, and a dictionary / rule adding / changing means for additionally modifying the dictionary or rule based on the result analyzed by the conceptual structure analyzing means.

[0014]

According to the present invention of claim 1, the machine translation conceptual structure generating method and the ideal translation conceptual structure generating method are used for the machine translation conceptual structure generated when the first natural language is translated by the machine translation system. The structure is converted based on a structure conversion rule that absorbs the difference, and the idealized conceptual structure generated by analyzing the structurally converted conceptual structure and the ideal translated sentence of the second natural language is analyzed, and this analysis is performed. Dictionaries or rules are added or modified based on the results.

According to the present invention of claim 2,
The input sentence having a bilingual relationship and the ideal translated sentence prepared in advance are read, the ideal translated conceptual structure of the target language is generated from the ideal translated sentence, the machine translated conceptual structure of the machine translated sentence is generated, and the machine translated conceptual structure is machine translated. Structure conversion is performed based on the structure conversion rules that absorb the difference between the conceptual structure generation method and the ideal translation conceptual structure generation method, the conceptual structure and ideal translation conceptual structure that have undergone this structural conversion are analyzed, and based on this analysis result You have added or modified a dictionary or rules.

Further, according to the invention of claim 3,
In the machine translation concept structure conversion step, the conditions of the machine translation concept structure and the ideal translation concept structure are included in the structure conversion rule that absorbs the difference between the machine translation concept structure generation method and the ideal translation concept structure generation method.

According to a fourth aspect of the present invention, the ideal translation conceptual structure generated by analyzing the ideal translation sentence of the second natural language is classified into a machine translation conceptual structure generating method and an ideal translation conceptual structure generating method. The structure is converted based on the structure conversion rule that absorbs the difference, and
The machine translation conceptual structure generated when the machine translation system translates the natural language of is analyzed, and the dictionary or rule is added and modified based on the analysis result.

According to the present invention of claim 5,
The input translation and the ideal translation prepared in advance are read, the ideal translation conceptual structure of the target language is generated from the ideal translation, the machine translation conceptual structure of the machine translation is generated, and the ideal translation conceptual structure is converted to the machine translation concept. The structure is converted based on the structure conversion rule that absorbs the difference between the structure generation method and the ideal translation concept structure generation method, and the structure structure and the machine translation concept structure are analyzed, and the dictionary is based on the analysis result. Or the rule is added and modified.

Further, according to the present invention of claim 6,
In the ideal translation concept structure conversion step, both the machine translation concept structure and the ideal translation concept structure are included in the conditions of the structure conversion rule that absorb the difference between the machine translation concept structure generation method and the ideal translation concept structure generation method.

In the present invention according to claim 7, a bilingual sentence which is a pair of a first natural language sentence and a second natural language ideal translated sentence which are in a bilingual relation is input, and the input bilingual sentence is input. 2. The ideal translation sentence of 2 natural languages is analyzed, the ideal translation concept structure is generated, and the machine translation concept structure of the target language which is in the process of machine-translating the input first natural language sentence is generated, and the machine translation concept is generated. The structure is converted based on a structure conversion rule that absorbs the difference between the machine translation concept structure generation method and the ideal translation concept structure generation method, and the ideal translation concept structure and the concept structure obtained by the machine translation concept structure conversion means are converted. Analyze and add or modify dictionaries or rules based on the results of this analysis.

According to the present invention of claim 8,
The second translated text, which is a pair of a first natural language sentence and a second natural language ideal translation sentence, which are in a bilingual relationship, is input
The ideal translation conceptual structure of the target language is generated by analyzing the ideal translation sentence of the natural language, generating the ideal translation conceptual structure, and generating the ideal translation conceptual structure in the process of machine-translating the input first natural language sentence. Is transformed based on the structural transformation rule that absorbs the difference between the machine translation conceptual structure generation method and the ideal translation conceptual structure generation method, and the conceptual structure obtained by the machine translation conceptual structure and ideal translation conceptual structure conversion means is analyzed. Then, based on the result of this analysis, the dictionary or rules are added and modified.

[0022]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the structure of a machine translation system dictionary / rule learning apparatus according to an embodiment of the present invention. The machine translation system dictionary / rule learning device shown in the figure is used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a second natural language which is a target language. In a system for changing a dictionary or a rule, a parallel translation sentence reading unit 1 for inputting a parallel translation sentence which is a pair of a first natural language sentence and a second natural language ideal translation sentence having a parallel translation relationship, and a parallel translation sentence reading unit. The ideal translation conceptual structure generating means 2 for analyzing the ideal translation sentence of the second natural language input by 1 and generating the ideal translation conceptual structure and the sentence of the first natural language input by the parallel translation sentence reading means 1 Machine translation conceptual structure generating means 31 of the machine translation system 3 for generating a machine translation conceptual structure in the process of machine translation, and a machine translation conceptual structure generating means for converting the machine translation conceptual structure generated by the machine translation conceptual structure generating means 31. A machine translation conceptual structure converting means 61 for structural conversion based on the structure conversion rule 611 so as to absorb the difference 1 and the ideal translation conceptual structure generating means 2,
Conceptual structure analysis means 71 for analyzing the idealized translation conceptual structure generated by the idealized translation conceptual structure generation means 2 and the conceptual structure generated by the machine translation conceptual structure conversion means 61, and the conceptual structure analysis means 7
The dictionary / rule addition / modification unit 72 that additionally modifies the dictionary 5 and the rule 4 according to the result of 1.

Next, the operation of the dictionary / rule learning apparatus for machine translation system configured as described above will be described with reference to the flowchart shown in FIG.

In FIG. 2, the bilingual sentence reading means 1 reads the bilingual input sentence and the model translated sentence as an ideal translated sentence prepared in advance (step S1).
The ideal translation conceptual structure of the target language is generated from the ideal translation (step S2), the input sentence is translated into a target language sentence as a machine translation using a machine translation system, and the translated machine translation conceptual structure of the target language is generated. It is generated (step S3), the machine translation conceptual structure is structurally transformed based on a structural transformation rule that absorbs the difference between the machine translation conceptual structure producing method and the ideal translation conceptual structure producing method (step S41), and the structure is transformed. The conceptual structure and the ideal translation conceptual structure are analyzed (step S5), and the dictionary and rules are added and changed based on the result obtained in step S5 (step S6).

FIG. 3 is a block diagram showing the configuration of a machine translation system dictionary / rule learning apparatus according to another embodiment of the present invention. The machine translation system dictionary / rule learning device shown in the figure is used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a second natural language which is a target language. In a system for changing a dictionary or a rule, a bilingual sentence reading means 1 for inputting a bilingual sentence which is a pair of a first natural language sentence and a second natural language ideal translated sentence having a bilingual relationship, and a bilingual sentence reading means. An ideal translation concept structure generating means 2 for analyzing the ideal translation sentence of the second natural language input by 1 and generating an ideal translation concept structure;
Machine translation conceptual structure generating means 31 of machine translation system 3 that generates a machine translation conceptual structure in the process of machine translation of the first natural language sentence input by parallel translation sentence reading means 1, and ideal translation conceptual structure generating means 2 An ideal translation concept structure conversion means 62 for converting the ideal translation concept structure generated according to the structure conversion rule 621 so as to absorb the difference between the machine translation concept structure generation means 31 and the ideal translation concept structure generation means 2.
Concept structure analysis means 71 for analyzing the concept structure generated by the ideal translation concept structure conversion means 62 and the machine translation concept structure generated by the machine translation concept structure generation means 31, and a dictionary according to the result of the concept structure analysis means 71. 5 and rule 4 are added to the dictionary / rule addition / change means 72.

Next, the operation of the dictionary / rule learning apparatus for machine translation system configured as described above will be described with reference to the flowchart shown in FIG.

In FIG. 4, the parallel sentence reading means 1 reads an input sentence and a model translated sentence as an ideal translated sentence which are prepared in advance (step S1).
The ideal translation conceptual structure of the target language is generated from the ideal translation (step S2), the input sentence is translated into a target language sentence as a machine translation using a machine translation system, and the translated machine translation conceptual structure of the target language is generated. It is generated (step S3), the ideal translation conceptual structure is structurally transformed based on the structural transformation rule that absorbs the difference between the machine translation conceptual structure generating method and the ideal translation conceptual structure generating method (step S42), and the structural transformation is performed. The conceptual structure and the ideal translation conceptual structure are analyzed (step S5), and the dictionary and rules are added and changed based on the result obtained in step S5 (step S6).

FIG. 5 is a block diagram showing the configuration of a machine translation system dictionary / rule learning apparatus according to still another embodiment of the present invention. Machine translation system dictionary shown in the figure
The rule learning device is a more specific configuration of the embodiment shown in FIG. 1, and includes a parallel translation sentence reading unit 10 corresponding to the parallel translation sentence reading unit 1 and an ideal corresponding to the ideal translation conceptual structure generating unit 2. A translation concept structure generation unit 20, a machine translation system 30 corresponding to the machine translation system 3, a machine translation concept structure generation unit 310 provided in the machine translation system 30 and corresponding to the machine translation concept structure generation unit 31, the machine translation Machine translation conceptual structure conversion unit 610 corresponding to the conceptual structure conversion unit 61, structural conversion rule 611, learning unit 70, conceptual structure analysis unit 710 provided in the learning unit 70 and corresponding to the conceptual structure analysis unit 71, the dictionary A dictionary / rule addition / change unit 720 corresponding to the rule addition / change unit 72, a rule 4, and a dictionary 5 are included.

In FIG. 5, the parallel translation reading section 10 is shown.
Inputs a pair (translated sentence) of a sentence of a first natural language (Japanese) and a sentence of a second natural language (English) that have a bilingual relationship.
The ideal translation conceptual structure generation unit 20 generates an ideal translation conceptual structure from the input ideal translation sentence of the second natural language.

The machine translation system 30 uses the first input
The second natural language sentence is translated into a machine translated sentence using the rule 4 and the dictionary 5. The machine translation system 30 has a machine translation conceptual structure generation unit 310 in the translation process.
Machine translation conceptual structure generation unit 310 of machine translation system 30
Generates a machine translation conceptual structure during machine translation. The machine translation concept structure conversion unit 610 performs structure conversion of the machine translation concept structure based on the structure conversion rule 611 that absorbs the difference between the machine translation concept structure generation method and the ideal translation concept structure generation method.

The learning unit 70 additionally changes the dictionary 5 and the rule 4 from the bilingual sentence input by the bilingual sentence reading unit 10.
The learning unit 70 includes a conceptual structure analysis unit 710 and a dictionary / rule addition / change unit 720. The conceptual structure analysis unit 710 of the learning unit 70 analyzes the conceptual structure converted by the machine translation conceptual structure conversion unit 610 and the ideal translation conceptual structure generated by the ideal translation conceptual structure generation unit 20. Dictionary / rule addition change part 7
Reference numeral 20 additionally changes the dictionary 5 and the rule 4 according to the result of the conceptual structure analysis unit 710.

Next, the processing procedure of the machine translation dictionary / rule learning device configured as shown in FIG. 5 will be described. Here, the bilingual sentence shown in FIG. 6 will be described as an example.

First, the bilingual sentence reading section 10 inputs a pair (bilingual sentence) of a sentence of a first natural language and a sentence of a second natural language which have a bilingual relationship. This bilingual sentence is used for automatic learning of machine translation dictionaries and rules. By selecting a set of bilingual sentences (translated sentences) for learning this dictionary / rule, it is possible to tune the machine translation system for similar field sentences in which language phenomena similar to the learned bilingual sentence frequently appear. It will be possible. In the example of FIG. 6, the first
Japanese sentence (input Japanese sentence) which is the natural language of
"He bought the AV," which is the second natural language in English that is bilingual to "I bought the system."
"system." is entered as a bilingual sentence.

Of the input bilingual sentences, the input Japanese sentence which is the first natural language sentence is translated by the machine translation system 30 using the rule 4 and the dictionary 5 into the machine English translated sentence which is the second natural language sentence. To be done. In the above example sentence, "He bought the AV system." As shown in FIG. 6 is translated, and a machine translation conceptual structure as shown in FIG. 7 is generated during the machine translation process.

On the other hand, the ideal translation conceptual structure is generated by the ideal translation conceptual structure generating unit 20 of the input bilingual sentence. In the above example sentence, an ideal translation conceptual structure as shown in FIG. 7 is generated.

Although the conceptual structures shown in FIG. 7 represent the same concept, the ideal translation conceptual structure and the machine translation conceptual structure have different representation forms because the respective structure generation methods are different. In order to absorb the difference in the structure generation method, the machine translation concept structure conversion unit 610 converts the machine translation concept structure into one that meets the specifications of the ideal translation concept structure generation method. Here, the process relating to the conceptual structure conversion is shown in the flowchart of FIG. 8, and the structure conversion rule 611 used is shown in FIG.

The process related to the machine translation concept structure conversion and the structure conversion rule used in the example will be described with reference to FIG.

In step S601 of FIG. 8, first, the machine translation conceptual structure is extracted. In the above example, the machine translation conceptual structure shown in FIG. 7 is input to the machine translation conceptual structure converting unit 610. In step S602, if there is a machine translation conceptual structure input in step S601 that meets the conditions of the structure conversion rule, the process proceeds to step S603, and if there is no machine translation concept structure, nothing is done. In the above example, since the structural conversion rule shown in FIG. 9 is valid, the process proceeds to step S603. In step S603, the structure conversion rule is applied to the machine translation conceptual structure.

The structure conversion rule applied here will be described.

When a word is regarded as a variable, and a certain condition structure exists in the machine translation conceptual structure, the structure is converted to arrange the converted word according to the variable arrangement of the word indicated by the condition. decide. In the example shown in FIG. 9, if the nouns are grouped as words consisting of two words, they are divided into a modifier and a modified word. In the above example, "AV system" is analyzed as one word, so it is converted into the conceptual structure as shown in Fig. 10 by dividing it into "AV" and "system" and setting them as a modification relation. It

In the learning unit 70, the machine translation conceptual structure conversion unit 6
The concept structure analysis unit 710 analyzes the conceptual structure converted by 10 and the ideal translation conceptual structure generated by the ideal translation conceptual structure generation unit 20, and adds a dictionary or rule by the dictionary / rule addition / modification unit 720 according to the result. Fix it. In the above example, the concept structure converted by the machine translation concept structure conversion unit 610 is as shown in FIG. 10, and the ideal translation concept structure generated by the ideal translation concept structure generation unit 20 is as shown in FIG. So there is no additional modification of dictionaries or rules.

FIG. 11 is a block diagram showing the configuration of a machine translation system dictionary / rule learning apparatus according to another embodiment of the present invention. The dictionary / rule learning device for machine translation system shown in FIG. 3 is a more specific configuration of the embodiment shown in FIG. 3, and includes a parallel translation reading unit 10 corresponding to the parallel translation reading unit 1 and the ideal The ideal translation concept structure generation unit 20 corresponding to the translation concept structure generation unit 2, the machine translation system 30 corresponding to the machine translation system 3, and the machine translation concept structure generation unit 3 provided in the machine translation system 30.
1. The machine translation conceptual structure generating unit 310 corresponding to 1, the ideal translation conceptual structure converting unit 620 corresponding to the ideal translation conceptual structure converting unit 62, the structural conversion rule 621, the learning unit 70, the learning unit 70. It has a conceptual structure analysis unit 710 corresponding to the structure analysis unit 71, a dictionary / rule addition / change unit 720 corresponding to the dictionary / rule addition / change unit 72, a rule 4, and a dictionary 5.

In FIG. 11, the parallel translation reading unit 1
0 inputs a pair (translated sentence) of a sentence of a first natural language (Japanese) and a sentence of a second natural language (English) which are in a bilingual relationship. The conceptual structure generation unit 20 generates an ideal translated conceptual structure from the input ideal translated sentence of the second natural language. The machine translation system 30 translates the input first natural language sentence into a machine translated sentence of the second natural language using the rule 4 and the dictionary 5. The machine translation system 30 has a machine translation conceptual structure generation unit 310 in the translation process.

The machine translation conceptual structure generator 310 of the machine translation system 30 generates a machine translation conceptual structure during machine translation. The ideal translation conceptual structure conversion unit 620 structurally converts the ideal translation conceptual structure based on the structural conversion rule 621 that absorbs the difference between the machine translation conceptual structure generating method and the ideal translation conceptual structure generating method.

The learning unit 70 additionally modifies the dictionary 5 and the rule 4 from the bilingual sentence input by the bilingual sentence reading unit 10.
The learning unit 70 includes a conceptual structure analysis unit 710 and a dictionary / rule addition / change unit 720. The conceptual structure analysis unit 710 of the learning unit 70 analyzes the conceptual structure converted by the ideal translation conceptual structure conversion unit 620 and the machine translation conceptual structure generated by the machine translation conceptual structure generation unit 310. The dictionary / rule addition / change unit 720 uses the dictionary 5 according to the result of the conceptual structure analysis unit 710.
And rule 4 are added and changed.

Next, the processing procedure of the machine translation dictionary / rule learning device configured as shown in FIG. 11 will be described. Here again, the bilingual text shown in FIG. 6 will be described as an example.

First, the bilingual sentence reading unit 10 inputs a pair (translated sentence) of a first natural language sentence and a second natural language sentence that are in a bilingual relationship. In the example of FIG. 6, the second natural language English sentence (ideal, which is a bilingual relation to the first natural language Japanese sentence (input Japanese sentence) “He bought the AV system” English translation) "He bought the AV system." Is entered as a parallel translation.

Of the input bilingual sentences, the input Japanese sentence which is the first natural language sentence is translated by the machine translation system 30 using the rules 4 and the dictionary 5 into the machine English translated sentence which is the second natural language sentence. To be done. In the above example sentence, "He bought the AV system." As shown in FIG. 6 is translated, and a machine translation conceptual structure as shown in FIG. 7 is generated during the machine translation process.

On the other hand, the ideal English translation sentence structure is generated by the ideal translation concept structure generating unit 20 of the input bilingual sentence. In the above example sentence, an ideal translation conceptual structure as shown in FIG. 7 is generated.

Although the conceptual structure shown in FIG. 7 represents the same concept, the ideal translation conceptual structure and the machine translation conceptual structure have different expression forms because the respective structure generating methods are different. In order to absorb the difference in the structure generation method, the ideal translation conceptual structure conversion unit 620 converts the ideal translation conceptual structure into one that meets the specifications of the machine translation conceptual structure generating method. Here, the process related to the conceptual structure conversion is shown in the flowchart of FIG. 8, and the structure conversion rule 621 used is shown in FIG.

The process related to the ideal translation conceptual structure conversion and the structure conversion rule used in the example will be described with reference to FIG.

In step S601 of FIG. 8, first, the ideal translation conceptual structure is extracted. In the above example, the ideal translation concept structure shown in FIG. 7 is input to the ideal translation concept structure conversion unit 620. In step S602, if any of the ideal translation conceptual structures input in step S601 satisfies the condition of the structure conversion rule, the process proceeds to step S603, and if there is no condition, nothing is done. In the above example, FIG.
Since the structure conversion rule indicated by is hit, step S603
Move to. In step S603, the structure conversion rule is applied to the ideal translation conceptual structure.

The structure conversion rule applied here will be described.

If a word is regarded as a variable and a structure that satisfies a certain condition is present in the ideal translation conceptual structure, the structure is converted to arrange the converted word according to the variable arrangement of the word indicated by the condition. decide. In the example shown in FIG. 12, if two nouns are included in a noun phrase and they have a modifying relationship, the rule is that they are grouped as a word consisting of two words. In the above example, the noun phrase contains two nouns because they have the ideal translation conceptual structure shown in FIG. 7, and they have a modifying relation. Therefore, "AV" and "system" are collectively "AV system". Is converted into the conceptual structure as shown in FIG.

In the learning unit 70, the ideal translation conceptual structure conversion unit 6
20. The conceptual structure analysis unit 710 analyzes the conceptual structure converted by 20 and the machine translation conceptual structure generated by the machine translation conceptual structure generation unit 310, and the dictionary / rule addition / modification unit 720 adds a dictionary or rule according to the result. Fix it. In the above example, the concept structure converted by the ideal translation concept structure conversion unit 620 is as shown in FIG. 13, and the machine translation concept structure generated by the machine translation concept structure generation unit 310 is shown in FIG.
As shown in, there is no difference, so there are no additional modifications to the dictionary or rules.

Next, the structure conversion rule 611 in the basic block configuration of the machine translation dictionary / rule learning apparatus of the embodiment shown in FIG. 5 includes the machine translation conceptual structure and the ideal translation conceptual structure as shown in FIG. The case of a rule will be described. Here again, the bilingual text shown in FIG. 6 will be described as an example. Except for the structure conversion rule, the process is the same as that of the embodiment shown in FIG. 5, so that the process flow for the conceptual structure conversion shown in FIG.

In step S601 of FIG. 8, first, the machine translation conceptual structure and the ideal translation conceptual structure are extracted. In the above example,
The machine translation concept structure and the ideal translation concept structure shown in FIG. 7 are input to the machine translation concept structure conversion unit 610. Step S60
In step 2, if there is a machine translation conceptual structure and an ideal translation conceptual structure input in step S601 that satisfy the conditions of the structure conversion rule, the process proceeds to step S603, and if there is no such condition, nothing is done. In the above example, FIG.
Since the structure conversion rule indicated by is hit, step S603
Move to. In step S603, the structure conversion rule is applied to the machine translation conceptual structure.

The structure conversion rule applied here will be described.

If a word is regarded as a variable and a structure that satisfies a certain condition exists in the machine translation conceptual structure and a structure that satisfies another condition exists in the ideal translation conceptual structure, the structure is converted and indicated by the condition. The placement of the converted word is determined according to the placement of the variable of the word. In the example shown in FIG. 14, nouns are grouped as a word consisting of two words in the machine translation concept structure, and are also grouped in the machine translation concept structure as a noun phrase in the ideal translation concept structure. Is a rule that means that the machine translation conceptual structure is structurally transformed if it is included in a qualifying relation. In the example above, the "AV syste"
Since "m" appears as one word, and "AV" and "system" have a modifier relation in the ideal translation conceptual structure, the machine translation conceptual structure is structurally transformed to the conceptual structure shown in FIG. Becomes

In the learning unit 70, the machine translation conceptual structure conversion unit 6
The concept structure analysis unit 710 analyzes the conceptual structure converted by 10 and the ideal translation conceptual structure generated by the ideal translation conceptual structure generation unit 20, and adds a dictionary or rule by the dictionary / rule addition / modification unit 720 according to the result. Fix it. In the above example, the concept structure converted by the machine translation concept structure conversion unit 610 is as shown in FIG. 10, and the ideal translation concept structure generated by the ideal translation concept structure generation unit 20 is as shown in FIG. So there is no additional modification of dictionaries or rules.

It is also possible to use the rules shown in FIG. 14 in the structure conversion rule 621 of the machine translation dictionary / rule learning device of the embodiment shown in FIG. This need only change the machine translation concept structure conversion to the ideal translation concept structure conversion in the above description.

The present invention is not limited to the above embodiments, but various modifications and applications are possible within the scope of the claims.

[0064]

As described above, according to the present invention,
A first natural language sentence (input sentence) and a second natural language sentence (ideal translated sentence) that are in a bilingual relationship with each other are read, an ideal translation conceptual structure and a machine translation conceptual structure are generated, and a machine translation conceptual structure is generated. Analyzing the conceptual structure and ideal translation conceptual structure that have undergone structural transformation, and adding or changing dictionaries or rules based on the results of this analysis, or generating ideal translation conceptual structure and machine translation conceptual structure, and converting the ideal translation conceptual structure The concept structure and machine translation concept structure are analyzed, and the dictionary and rules are added and changed based on the result of this analysis. Even if this happens, you can correctly register the dictionary and rules from the machine translation and the ideal translation.

Further, by inserting the conditions of the machine translation concept structure and the ideal translation concept structure into the conditions of the concept structure conversion rule, when the machine translation concept structure and the ideal translation concept structure match before the structure conversion, It is possible to write rules so that the structure conversion is not performed, and it is possible to prevent the adverse effect of the concept structure conversion rule.

As described above, according to the present invention, the concept structure dependent on the concept structure generation method is not used as it is, but the difference between the machine translation concept structure generation method and the ideal translation concept structure generation method is absorbed. By changing the structure of the concept structure based on a simple structure conversion rule and matching the structures expressing the same concept, even if the difference is generated as it is even though the same concept is expressed, the dictionary or rule Can be correctly added and corrected.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a machine translation system dictionary / rule learning device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the embodiment shown in FIG.

FIG. 3 is a block diagram showing a configuration of a machine translation system dictionary / rule learning device according to another exemplary embodiment of the present invention.

FIG. 4 is a flowchart showing the operation of the embodiment shown in FIG.

FIG. 5 is a block diagram showing the configuration of a machine translation system dictionary / rule learning device according to still another embodiment of the present invention.

FIG. 6 is a diagram showing an example of bilingual sentences in the embodiment of FIG.

FIG. 7 is a diagram showing a conceptual structure of an ideal translated text and a machine translated text in the embodiment of FIG.

FIG. 8 is a flowchart showing a process related to conceptual structure conversion in the embodiment of FIG.

FIG. 9 is a diagram showing a structural conversion rule of the embodiment of FIG.

FIG. 10 is a diagram showing a conceptual structure after applying a structure conversion rule to the machine-translated conceptual structure of the embodiment shown in FIG. 5;

FIG. 11 is a block diagram showing the configuration of a machine translation system dictionary / rule learning device according to another embodiment of the present invention.

FIG. 12 is a diagram showing a structural conversion rule of the embodiment of FIG.

13 is a diagram showing a conceptual structure after applying a structure conversion rule to the ideal translation conceptual structure of the embodiment of FIG.

FIG. 14 is a diagram showing a structure conversion rule of still another embodiment.

[Explanation of symbols]

 1 parallel translation reading means 2 ideal translation conceptual structure generating means 3 machine translation system 4 rules 5 dictionary 31 machine translation conceptual structure generating means 61 machine translation conceptual structure converting means 62 ideal translation conceptual structure converting means 70 learning unit 71 conceptual structure analyzing means 72 Dictionary / rule addition / change means 611, 621 Structure conversion rule

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display location 8420-5L G06F 15/38 J 8420-5LP

Claims (8)

[Claims] 1. A machine for changing a dictionary and / or a rule used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning method for a translation system, wherein the machine translation conceptual structure generation method and the ideal translation conceptual structure generation method differ in the machine translation conceptual structure generated when the machine translation system translates the first natural language. A machine translation concept structure conversion step for performing a structure conversion based on a structure conversion rule that absorbs, and an analysis of the concept structure structure-converted in the machine translation concept structure conversion step and the ideal translated sentence of the second natural language. A conceptual structure analysis step for analyzing the ideal translation conceptual structure generated by the dictionary, and a dictionary / routine for additionally modifying the dictionary or rule based on the result of the conceptual structure analysis step. Machine translation system dictionary characterized by having
Rule learning method. 2. A machine for changing a dictionary and / or a rule used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning method for a translation system, which comprises: a step of reading a translated sentence that reads the input sentence and a previously prepared ideal translated sentence that are in a bilingual relationship; and generating an ideal translated conceptual structure of a target language from the ideal translated sentence. An ideal translation conceptual structure generating step; a machine translation conceptual structure generating step for generating a machine translation conceptual structure of the machine translation; and a difference between the machine translation conceptual structure generating method and the ideal translation conceptual structure generating method for the machine translation conceptual structure Machine-translated concept structure conversion step for performing structure conversion based on such a structure conversion rule, and A machine translation system comprising: a conceptual structure analyzing step of analyzing the ideal translation conceptual structure; and a dictionary / rule adding / modifying step of additionally modifying the dictionary or rule based on the result of the conceptual structure analyzing step. dictionary·
Rule learning method. 3. The machine translation conceptual structure converting step includes a condition of the machine translation conceptual structure and the ideal translation conceptual structure in a structure conversion rule that absorbs a difference between the machine translation conceptual structure generating method and the ideal translation conceptual structure generating method. The dictionary / rule learning method for a machine translation system according to claim 2, wherein the dictionary / rule learning method is included. 4. A machine for changing a dictionary and / or a rule used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning method for a translation system, wherein the ideal translation conceptual structure generated by analyzing the ideal translated sentence of the second natural language is differentiated between a machine translation conceptual structure generating method and an ideal translation conceptual structure generating method. When the machine translation system translates the ideal translation concept structure conversion step for performing structure conversion based on the absorbing structure conversion rule, and the concept structure and the first natural language that have undergone the structure conversion in the ideal translation concept structure conversion step. A conceptual structure analysis step of analyzing a machine translation conceptual structure generated in the above, and a dictionary / rule addition step for additionally modifying the dictionary or rule based on the result of the conceptual structure analysis step. Dictionary for machine translation system characterized by having an addition / change step
Rule learning method. 5. A machine for changing a dictionary and / or a rule used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning method for a translation system, which comprises: a step of reading a translated sentence that reads the input sentence and a previously prepared ideal translated sentence that are in a bilingual relationship; and generating an ideal translated conceptual structure of a target language from the ideal translated sentence. An ideal translation conceptual structure generating step; a machine translation conceptual structure generating step for generating a machine translation conceptual structure of the machine translation; and a difference between the ideal translation conceptual structure generating method and the ideal translation conceptual structure generating method for the ideal translation conceptual structure Ideal translation concept structure conversion step for performing structure conversion based on such a structure conversion rule, and a concept structure structure converted by the ideal translation concept structure conversion step A machine translation system comprising: a conceptual structure analysis step of analyzing the machine translation conceptual structure; and a dictionary / rule addition / modification step of additionally modifying the dictionary or rule based on the result of the conceptual structure analysis step. dictionary·
Rule learning method. 6. The ideal translation concept structure conversion step includes the conditions of the machine translation concept structure and the ideal translation concept structure in a condition of a structure conversion rule that absorbs a difference between the machine translation concept structure generation method and the ideal translation concept structure generation method. The dictionary / rule learning method for a machine translation system according to claim 5, characterized in that both are included. 7. A machine for changing a dictionary and / or a rule used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning device for a translation system, comprising: a bilingual sentence reading means for inputting a bilingual sentence, which is a pair of a first natural language sentence and a second natural language ideal translated sentence having a bilingual relationship, and the bilingual translation. An ideal translation conceptual structure generating unit that analyzes the ideal translation sentence of the second natural language input by the sentence reading unit and generates an ideal translation conceptual structure; and a sentence of the first natural language input by the parallel translation sentence reading unit. And a machine translation concept structure generating means for generating a machine translation concept structure of a target language having a machine translation process, and absorbing the difference between the machine translation concept structure generation method and the ideal translation concept structure generation method. Machine-translated conceptual structure conversion means for performing structural conversion based on such a structural conversion rule, and an ideal-translated conceptual structure generated by the ideal-translated conceptual structure generation means and a conceptual structure obtained by the mechanical-translated conceptual structure conversion means are analyzed. A dictionary / rule learning device for a machine translation system, comprising: a conceptual structure analyzing means; and a dictionary / rule adding / modifying means for additionally modifying the dictionary or rule based on a result analyzed by the conceptual structure analyzing means. . 8. A machine for changing a dictionary and / or a rule used in a machine translation system having means for converting an input sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language. A dictionary / rule learning device for a translation system, comprising: a bilingual sentence reading means for inputting a bilingual sentence, which is a pair of a first natural language sentence and a second natural language ideal translated sentence having a bilingual relationship, and the bilingual translation. An ideal translation conceptual structure generating unit that analyzes the ideal translation sentence of the second natural language input by the sentence reading unit and generates an ideal translation conceptual structure; and a sentence of the first natural language input by the parallel translation sentence reading unit. And a machine translation concept structure generating means for generating a machine translation concept structure of a target language having a machine translation process, and absorbing the difference between the machine translation concept structure generation method and the ideal translation concept structure generation method for the ideal translation concept structure. The ideal translation concept structure conversion means for performing structure conversion based on such a structure conversion rule, and the machine translation concept structure generated by the machine translation concept structure generation means and the concept structure obtained by the ideal translation concept structure conversion means are analyzed. A dictionary / rule learning device for a machine translation system, comprising: a conceptual structure analyzing means; and a dictionary / rule adding / modifying means for additionally modifying the dictionary or rule based on a result analyzed by the conceptual structure analyzing means. .