JPH07244665A - Method and device for dictionary and rule learning for machine translation system - Google Patents

Abstract

PURPOSE:To obtain the dictionary and rule learning device for the machine translation system which can automatically change or add dictionaries and rules by the effective method with the highest practicality by extracting the dictionaries and rules to be changed or added by referring to the ideal translation with the highest possibility which has the least difference from a machine translation and is generated by the machine translation system. CONSTITUTION:This device is provided with an original and translation read means 1 which reads in an input sentence in a 1st natural language and an ideal translation in a 2nd natural language in original-translation relation, and a machine translation and ideal translation comparison means 5 which compares the machine translation generated by translating the input sentence in the 1st natural language with the ideal translation in the 2nd natural language. When plural kinds of ideal translation are given for one kind of input sentence as a result of the comparison, a minimum difference ideal translation extraction means 6 extracts the ideal translation with the least difference. On the basis of the obtained ideal translation having the least difference, a dictionary and rule change and addition means 7 changes or adds dictionaries and rules.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation system dictionary / rule learning method and a machine translation system dictionary / rule learning device, and more particularly to automatically learning a dictionary and rules used in a machine translation system. Relates to a dictionary rule learning method for a machine translation system and a dictionary / rule learning device for a machine translation system.

[0002]

2. Description of the Related Art Conventionally, when constructing a dictionary or rules for a machine translation system, a dictionary or rules for a machine translation system is manually constructed by referring to a dictionary or the like used by humans for translation. In addition, if an inappropriate translation is generated based on the dictionary and rules for the machine translation system and the translation result of the machine translation system constructed in this way, it causes the translation to be generated. Experts who have a good understanding of the internal processing of machine translation systems manually modify and add dictionaries and rules.

As a method for automating the work of adding and modifying the dictionary and rules, a set of a pair of source language sentences and a target language sentence (translation corpus) having a bilingual relationship is used to represent the source language and its equivalent. An improved method of extracting a pair of expressions of a target language to be used and changing and adding a dictionary or a rule using the pair (for example, “Machine translation method and system (Japanese Patent Laid-Open No. 2-903”
64) ”or a bilingual corpus, and compares the machine translation of the target language translated by the machine translation system with the ideal translation, and if they are different, use them in the process of generating the machine translation. One of the generated rules is investigated, and an improved method for changing and adding an incorrect generation grammar (for example, "Generation grammar automatic generation device (Japanese Patent Laid-Open No. 63-4)).
4276) "is known.

[0004]

However, in the above-mentioned conventional dictionary and rule construction method of the machine translation system,
Since a dictionary created for humans is used, there is a problem that there is no guarantee that a correct translation will be generated when used for machine translation.

In addition, in the method of changing and adding the dictionary and rules of the machine translation system, it is necessary to verify the translation result by an expert who has a good understanding of the internal processing of the translation system. The problem is that it is possible. In addition, since such work is performed manually, there is a problem that the time and economical cost for constructing the dictionary and the rules becomes high.

Further, in the case of the former improvement method of the above-mentioned conventional technique, since the dictionary and the rule are changed and added only from the information of the bilingual corpus without considering the translation result of the machine translation system, the existing dictionary and the rule are added. There is a problem that the translation corpus makes inappropriate changes and additions, which adversely affects the machine translation result.

In the latter improvement method, the rule to be changed and added is extracted in consideration of the translation result of the machine translation system. Therefore, the former improvement method "machine translation method and system" is used.
There is little adverse effect on the machine translation result, but if the source language sentences are the same in the bilingual corpus, but the ideal translation sentences are different, different target language expressions will be used for the same source language expression. Since the dictionary and rules for generating are changed and added, there is a problem that the dictionary and rules cannot be matched. For example, considering that the source language is Japanese and the target language is English, the Japanese sentence "Having a memory device" is added to the sentence "Strage units is to be provided." It is assumed that two types of ideal translations, "To be equipped with the strage units." In this case, "strage un" for one Japanese expression "memory"
Two kinds of dictionaries and rules for generating the target language expression "its" and the target language expression "the strage units" are generated, and for one Japanese expression "having", "~ is" There is a problem that two types of dictionaries and rules for generating the target language expression "to be provided." and the target language expression "To be equipped with ~" are generated and modified and added to the existing dictionary and rules. .

The present invention has been made in view of the above points, and solves the above-mentioned conventional problems and refers to not only the information of the bilingual corpus but also the machine translated text output by the machine translation system. If there are some sentences in the corpus that have the same source language but different ideal translations, add the change by referring to the ideal translation that is most likely to be generated by the machine translation system that has the least difference from the machine translation. Automatically modify and add dictionaries and rules by extracting the dictionaries and rules that should be used, without making inappropriate changes to existing dictionaries and rules, in the most feasible and effective way. It is an object of the present invention to provide a machine translation system dictionary / rule learning method and a machine translation system dictionary / rule learning device capable of performing the above.

[0009]

FIG. 1 is a diagram (No. 1) for explaining the principle of the present invention. A dictionary / rule learning method for a machine translation system of the present invention is a machine translation system having means for converting a sentence of a first natural language into a sentence of a second natural language, and a first natural language having a bilingual relationship with each other. Input sentence and the second ideal translated sentence of natural language are read (step 1), and the first natural language input sentence is translated into a second natural language machine translated sentence using a machine translation system (step 2), The machine translation of the second natural language and the ideal translation of the second natural language are compared (step 3), and a plurality of types of ideal translations are obtained for one type of input sentence according to the comparison result obtained in step 3. If it is added (step 4), the ideal translation with the smallest difference (ideal translation with minimum difference)
Is extracted (step 5), and the dictionary and rule are changed and added based on the comparison result of comparing the ideal translation and the machine translation and the minimum difference ideal translation (step 6).

FIG. 2 is a diagram (part 2) for explaining the principle of the present invention. Machine translation system dictionary of the present invention
According to a rule learning method, in a machine translation system having means for converting a sentence of a first natural language into a sentence of a second natural language, a first natural language input sentence and a second natural language which are in a bilingual relationship with each other. The bilingual sentence, which is a collection of ideal translated sentences, is read (step 11), and the same input sentence and the bilingual sentence corresponding to the input sentence are extracted from the bilingual sentence (step 1).
2), using a machine translation system, translate the input sentence into a machine translation of the second natural language (step 13), compare the machine translation and the ideal translation (step 14), and identify the same by the comparison result. Input sentence 1 extracted by input sentence extraction
With respect to the bilingual sentence to which a plurality of types of ideal translated sentences are assigned (step 15), the ideal translated sentence with the smallest difference is extracted (step 16), and the comparison result obtained in step 14 and step 16 The dictionary and rules are changed and added based on the obtained ideal translated sentence with the smallest difference (step 17).

FIG. 3 is a diagram (part 3) for explaining the principle of the present invention. Machine translation system dictionary of the present invention
According to a rule learning method, in a machine translation system having means for converting a sentence of a first natural language into a sentence of a second natural language, a first natural language input sentence and a second natural language which are in a bilingual relationship with each other. Read the ideal translation of (step 21),
When translating an input sentence into a machine translation of a second natural language using a machine translation system (step 22), a machine translation conceptual structure of the second natural language is generated (step 23), and the second translation from the ideal translation is performed. The ideal translation conceptual structure of the natural language is generated (step 24), the machine translation conceptual structure and the ideal translation conceptual structure are compared (step 25), and a plurality of input sentence types are obtained according to the comparison result. When types of ideal translation concept structures are given (step 26), the ideal translation concept structure with the smallest difference is extracted (step 27), and the comparison result and the minimum difference ideal translation concept structure are obtained. The dictionary and rules are changed and added based on the difference extraction result (step 28).

FIG. 4 is a view (No. 4) for explaining the principle of the present invention. Machine translation system dictionary of the present invention
According to a rule learning method, in a machine translation system having means for converting a sentence of a first natural language into a sentence of a second natural language, a first natural language input sentence and a second natural language which are in a bilingual relationship with each other. The bilingual sentence, which is a set of ideal translated sentences of, is read (step 31), and the same input sentence and the set of ideal translated corresponding to the input sentence are extracted from the bilingual sentence (step 3).
2), when translating the input sentence into a machine translation sentence of the second natural language using the machine translation system (step 33), the second
Generate a machine translation conceptual structure of the natural language of (step 3
4) The ideal translated sentence is analyzed to generate the ideal sentence conceptual structure of the second natural language (step 35), the machine translated conceptual structure and the ideal translated conceptual structure are compared (step 36), and a step is performed according to the comparison result. With respect to the bilingual sentence in which a plurality of types of ideal translation concept structures are added to one type of input sentence extracted in 32 (step 37), the ideal translation concept structure with the smallest difference is extracted (step 38), Based on the comparison result and the minimum difference ideal translation conceptual structure obtained by the machine translation ideal translation conceptual structure comparison processing, the dictionary and rule are changed and added (step 3).
9).

Further, the machine translation system dictionary / rule learning apparatus of the present invention is a machine translation system 2 having translation means for converting a first natural language sentence into a second natural language sentence. Parallel text reading means 1 for reading the first natural language input sentence and the second natural language ideal translated sentence, and the second natural language translated the first natural language input sentence using the machine translation system. Machine translation of No. 2
According to the machine translation ideal translation comparing means 5 for comparing the ideal translations of the natural language and the comparison result obtained by the machine translation ideal translation comparing means 5, a plurality of types of ideal translations are given to one input sentence. If there is a difference, the minimum difference ideal translation extracting means 6 for extracting the ideal translation having the smallest difference, and the comparison result and the minimum difference ideal translation extracting means 6 obtained by the machine translation ideal translation comparing means 5.
And a dictionary rule changing / adding means 7 for changing and adding a dictionary or a rule based on the ideal translated sentence having the smallest difference obtained in (4).

The machine translation system dictionary / rule learning device of the present invention has a bilingual relationship with each other in the machine translation system 2 having translation means for converting a first natural language sentence into a second natural language sentence. A bilingual sentence reading means 10 for reading a bilingual sentence which is a set of an input sentence of the first natural language and an ideal translated sentence of the second natural language, and a bilingual sentence corresponding to the same input sentence are extracted from the bilingual sentence. An input sentence extracting means 11, a machine translated sentence obtained by translating the input sentence into a second natural language sentence using the machine translation system 2, a machine translation ideal translation comparison means 5 for comparing an ideal translated sentence, and a machine translation ideal translation. According to the comparison result obtained by the comparison means 5, the ideal translation having the smallest difference from the bilingual text in which a plurality of types of ideal translations are provided for one type of the input sentence extracted by the same input sentence extracting means 11. To Minimum difference ideal translation extraction means 6 for out
And a dictionary rule change adding means 7 for changing and adding a dictionary or a rule based on the comparison result obtained by the machine translation ideal translation comparing means 5 and the ideal translation sentence having the smallest difference obtained by the minimum difference ideal translation extracting means 6. Including and

Further, the machine translation system dictionary / rule learning apparatus of the present invention has a bilingual relationship in the machine translation system 2 having a translation means for converting a sentence of the first natural language into a sentence of the second natural language. A bilingual sentence reading means 10 for reading an input sentence of a certain first natural language and an ideal translated sentence of a second natural language, and first when translating an input sentence into a machine translated sentence of a second natural language using a machine translation system. 2. Machine translation conceptual structure generating means for generating a machine translation conceptual structure of 2 natural languages, ideal translation conceptual structure generating means for generating a second natural language ideal translation conceptual structure from an ideal translated sentence, machine translation conceptual structure and ideal Machine translation ideal translation concept structure comparison means 5 for comparing translation concept structures,
According to the comparison result obtained by the machine translation ideal translation conceptual structure comparison means 5, when a plurality of types of ideal translation conceptual structures are given to one type of input sentence, the ideal translation conceptual structure with the smallest difference is extracted. Minimum difference ideal translation conceptual structure extraction means 6
And dictionary rule change adding means 7 for changing and adding a dictionary or rule based on the comparison result obtained by the machine translation ideal translation concept structure comparison means 5 and the difference extraction result obtained by the minimum difference ideal translation concept structure extraction means 6. Including and

A dictionary for the machine translation system of the present invention
The rule learning device has a machine translation system 2 having a translation means for converting a first natural language sentence into a second natural language sentence.
In the above, in the bilingual sentence reading means 10 for reading a bilingual sentence, which is a collection of the first natural language input sentence and the second natural language ideal translated sentence, which are in a bilingual relationship, and the same input sentence and input sentence from the bilingual sentence. The same input sentence extraction means 11 for extracting a set of ideal translated sentences corresponding to the above and a second natural language machine translation concept when the input sentence is translated into a second natural language machine translated text using a machine translation translation system. A machine translation conceptual structure generating means for generating a structure; and an ideal translation conceptual structure generating means for generating an ideal translation conceptual structure of the second natural language from the ideal translation.
Based on the comparison result obtained by the machine translation ideal translation concept structure comparison means 5 for comparing the machine translation concept structure and the ideal translation concept structure and the comparison result obtained by the machine translation ideal translation concept structure comparison means 5, the same input sentence extraction means 11 extracts. Minimal difference ideal translation concept structure extraction means 6 for extracting the ideal translation concept structure with the smallest difference from the bilingual sentence in which a plurality of types of ideal translation concept structures are given to one type of input sentence, and machine translation ideal Translation structure comparison means 5
The dictionary rule changing / adding means 7 for changing / adding a dictionary or a rule based on the comparison result obtained in step 1 and the difference extraction result obtained by the minimum difference ideal translation conceptual structure extracting means 6 is included.

[0017]

According to the present invention, in a machine translation system having means for converting a first natural language sentence into a second natural language sentence using a dictionary or a rule, a translation result of an input sentence (machine translated sentence), By comparing the ideal translations of input sentences prepared in advance, when adding and changing dictionaries and rules, if multiple types of ideal translations are given to the same input sentence, The ideal translation with the smallest difference is used.

More specifically, the present invention reads a bilingual sentence, reads an input sentence of a first natural language and an ideal translated sentence of a second natural language, which are in a bilingual relationship with each other, and executes the subsequent processing for a machine translation system. Dictionaries and rules are automatically changed and added.

First, in the process of comparing the ideal translated text and the machine translated text, the input sentence of the first natural language read by the parallel translation reading is translated into the machine translated text of the second natural language by using the machine translation system. The machine translated text and the ideal translated text are compared by, for example, using a dictionary for a machine translation system to find the correspondence between the words in the machine translated text and the words in the ideal translated text.

Next, when a plurality of types of ideal sentences are added to one input sentence by this comparison, for example,
The number of different words in the ideal translated text and the machine translated text is investigated, and the ideal translated text with the smallest difference is extracted by, for example, extracting the ideal translated text with the smallest number of words. When changing the dictionary rule, based on the result of comparing the ideal translated text and the machine translated text and the minimum difference ideal translated text,
Add and change dictionaries and rules. As a result, even if a plurality of types of ideal translated texts are given to one input sentence, the ideal translated text that is the closest to the machine translated text is used to change and add the dictionary and rule.

The present invention also reads a set of parallel translation sentences of a first natural language input sentence and a second natural language ideal translation sentence which are in a bilingual relationship. In the process of extracting the same input sentence, the same input sentence and a group of the ideal translated sentences are extracted, the ideal translated sentence and the machine translated sentence are compared, the ideal translated sentence with the smallest difference from the machine translated sentence is extracted, and one type of the input sentence is extracted. Even when a plurality of types of ideal translated texts are assigned to, the dictionary and rules are changed and added using the ideal translated text that is the closest to the machine translated text.

Further, according to the present invention, a machine translation conceptual structure and an ideal translation conceptual structure are generated, these two conceptual structures are compared, and a plurality of types of ideal translation conceptual structures are given to one type of input sentence. In this case, the ideal translation concept structure with the smallest difference is extracted, and the dictionary and rules are changed and added based on the smallest difference ideal translation concept structure. As a result, even when a plurality of types of ideal translated sentences are given to one type of input sentence, the dictionary and rules are changed and added using the ideal conceptual structure that is closest to the machine translation conceptual structure.

Further, according to the present invention, the extracted same input sentence 1
When multiple types of ideal translations are given for each type, the ideal translation conceptual structure that is close to the machine translation conceptual structure is compared,
Select the ideal translation of the ideal translation conceptual structure with the smallest difference and change and add the dictionary or rule.

[0024]

Embodiments of the present invention will now be described in detail with reference to the drawings. In this embodiment, the first natural language is Japanese and the second natural language is English.

[First Embodiment] FIG. 5 shows a basic block configuration of a machine translation dictionary / rule automatic learning system according to a first embodiment of the present invention. The system shown in the figure has a bilingual sentence reading unit 1 for inputting a first natural language sentence (input sentence) and a second natural language sentence (ideal translated sentence) in a bilingual relationship,
Machine translation system 2 for translating the input first natural language sentence into a second natural language sentence (machine translation) using rule 3 and dictionary 4, reading machine translation and parallel translation obtained by machine translation system 2. Based on the comparison result obtained by the ideal translation machine translation comparison unit 5 and the ideal translation machine translation comparison unit 5 that compare the ideal translation sentences of the second natural language obtained by the unit 1, a plurality of ideal translation sentences are generated by the first natural language. Comparison result and minimum difference ideal obtained by the minimal difference ideal translation extraction unit 6 that extracts the ideal translation sentence with the smallest difference from the bilingual sentence assigned to the input sentence and the ideal translation machine translation comparison unit 5. It is composed of a dictionary rule changing / adding unit 7 which additionally corrects the rule 3 and the dictionary 4 used in the machine translation system 2 based on the ideal translated sentence having the smallest difference obtained by the translation extracting unit 6.

FIG. 6 is a basic block configuration diagram of a machine translation system using a dictionary and rules to be additionally modified according to the first embodiment of the present invention. The system shown in the figure includes a first natural language input unit 8, a rule 3, a machine translation system 2, and a dictionary 4.
And a second natural language output unit 9.

The machine translation system 2 is a first language analysis unit 2 for converting a first natural language sentence into a conceptual structure of a first natural language.
1. A language conversion unit 22 for converting the concept structure of the first natural language generated by the first language analysis unit 21 into the concept structure of the second natural language, and the concept structure of the second natural language generated by the language conversion unit 22. Second language generation unit 2 for generating a second natural language sentence based on
It consists of 3.

Rule 3 is composed of a first language analysis rule 31 used in the first language analysis unit 21, a language conversion rule 32 used in the language conversion unit 22, and a second language generation rule 33 used in the second language generation unit 23. It

The dictionary 4 comprises a first language analysis dictionary 41 used in the first language analysis unit 21, a language conversion dictionary 42 used in the language conversion unit 22, and a second language generation dictionary 43 used in the second language generation unit 23. To be done.

FIG. 7 is a diagram showing an example of dictionary entries of the language conversion dictionary. The language conversion dictionary 42 includes the first part in the noun part.
English, which is the second natural language corresponding to each noun of the natural language Japanese, is entered, and when there are multiple translations for one noun, the priority indicating the priority of translations to be used, and Selection conditions consisting of noun phrases indicating the usage of translated words are entered. For example, for the noun "memory device", the translated words "memory" and "strage device" are given in this order of priority. In addition, not only each Japanese verb but also a verb particle is attached as a pattern to the verb part. For example, for the verb "having", "" (object), "no" 8 (Particle) ”and the particle pattern, the verb“ run ”,“ S ”(subject, noun)“ ga ”(particle)“ O ”(object, noun)
The particle pattern is written as "wo" (particle), and it is written so that the verb is used as a key to search for the translated word of the verb. For example, the verb “having” indicates that there is no restriction on the meaning of the case element shown in the particle pattern (O). For the verb "run," the meaning of the case element shown in the particle pattern (S is O)
When "S" has the meaning of a person as an animal and "O" has the meaning of a place, it is translated as "S run on O", and "S" has the meaning of "ship."
When "O" has the meaning of "place", it means to translate it as "S sail on O".

Next, the processing procedure of the machine translation dictionary / rule automatic learning system having the basic block configuration of FIG. 5 will be described. Here, the actual processing procedure will be described by taking the bilingual sentence shown in FIG. 8 as an example.

(1) First, the bilingual sentence reading unit 1 inputs a pair (bilingual sentence) of an input sentence of a first natural language and an ideal translated sentence of a second natural language which are in a bilingual relationship. This bilingual sentence is used for automatic learning of machine translation dictionaries and rules. By selecting a set of bilingual sentences (bilingual sentences) for learning of this dictionary / rule, the machine translation system is tuned for sentences in the same field where similar language expressions frequently appear in the learned bilingual documents. Is possible. In the example of FIG. 8, for the Japanese sentence (input Japanese sentence) “having a storage device” that is the first natural language, the ideal sentence translated from the second natural language English, which is a parallel translation, is Two types of "Strage device is to be provided." And "To be equipped with the strage device." Are added, and these are input as parallel translations. In human translation, when it is possible to translate with multiple types of expressions, a person who actually uses this translation result may be presented with multiple types of translation results, and the person may select the translation that he or she thinks is most suitable. Further, when one type of sentence is translated by a plurality of translators, the translation result may be different depending on the habits of the translators. Therefore, as in the example shown in FIG.
Since a plurality of types of second natural language sentences are frequently given to various types of first natural language input sentences, it is essential to consider such a phenomenon.

(2) Among the input bilingual sentences, the input Japanese sentence which is the sentence of the first natural language is a machine English translated sentence which is the second natural language sentence by the machine translation system 2 using the rule 3 and the dictionary 4. Be translated into The translation processing procedure by the machine translation system 2 will be described using the machine translation system 3 having the basic block configuration of FIG.

2-1) First, the input Japanese sentence, which is the first natural language sentence, is analyzed by the first language analysis unit 21 using the first language analysis rule 31 and the first language analysis dictionary 41,
Generate the Japanese conceptual structure. For example, for the input Japanese sentence in FIG. 8, a conceptual structure as shown in FIG. 9 is generated. In this conceptual structure, the verb "having" is modified by the noun "memory", and the modification relationship between the two is that of the particle "wo". This conceptual structure can be obtained by using the first language analysis dictionary 41 that describes the properties and part-of-speech of each word, and the first language analysis rule 31 that describes the modification relation between words.

2-2) Next, the conceptual structure of this Japanese language is
The language is converted by the language conversion unit 22 using the language conversion rule 32 and the language conversion dictionary 42, and the conceptual structure of English (machine translation)
To generate. For example, an English conceptual structure as shown in FIG. 10 is generated for the Japanese conceptual structure shown in FIG. This English conceptual structure shows that the noun "memory" modifies the verb "be" in the Subj relationship, and the verb "be provided" modifies it in the To-ph relationship. Here, the relationship of Subj indicates that the noun functions as the subject of the verb, and the relationship of To-ph is
Indicates that the verb modifies the verb through the To infinitive phrase.

This conceptual structure has a language conversion dictionary 42 in which the bilingual relations of words such as nouns and verbs as shown in FIG. 7 are described, and the relation between Japanese words is changed to the relation between English words. It is obtained by using the language conversion rule 32 which describes information to be converted, rules for selecting a translated word, and the like.

2-3) Next, this English conceptual structure generates a machine English translated sentence by the second language generation rule 33 and the second language generation dictionary 43. For example, for the English conceptual structure of FIG. 10, the machine English translation “Memory is to be provided.” Of FIG. 8 is generated. This machine English translation is obtained by using a second language generation dictionary 43 that describes information such as tense, numbers, and word changes depending on the person, and a second language generation rule 33 that describes rules for determining the word order of translated words. To be It should be noted that the conceptual structure and dictionary information that are referred to / generated at each stage of this machine translation system are stored so that they can be referred to in the subsequent dictionary / rule automatic learning processing.

For example, as shown in FIG. 8, information on a word unit of a machine translated sentence recognized by the machine translation process, or individual words obtained when searching the language conversion dictionary 42 shown in FIG. Information about other translation candidates is also saved.

2-4) The machine translated texts obtained as described above are compared by the ideal translated texts and the ideal translated machine translation comparing section 5, and the comparison result is obtained.

FIG. 11 is a flow chart of the ideal translation machine translation comparison processing of the first embodiment of the present invention. The following flowchart shows the operation of the ideal machine translation comparison unit 5.

Step 501) The machine translated text is divided for each word unit set at the time of machine translation. For example, in the case of the machine English translation of FIG. 8, it is divided into "Memory" and "is to be provided" based on the information stored in the process of machine translation.

Step 502) The ideal translated sentences are extracted one by one. In the ideal English translation shown in FIG. 8, "Strage device is to be provided." Is extracted first, and then "To be equipped with the strage device." Is extracted.

Step 503) The individual words of the machine-translated sentence divided in step 501 are taken out. In the machine English translation of Fig. 8, "memory" is first, and then "is to be provided"
Is taken out.

(Step 504) For the machine-translated word extracted in step 503, the dictionary entry of the language conversion dictionary used at the time of machine translation is searched for another translation word candidate. In FIG. 8, in the case of a machine-translated English word, another translation word candidate of “strage device” is extracted for “memory”, and no other translation word candidate is present for “is to be provided”. So it's not taken out.

Step 505) It is searched whether the machine translated word fetched in step 503 exists in the ideal translated sentence fetched in step 502.

Step 506) It is judged whether or not a machine translated word is present in the ideal translated sentence, and if it is present, a word of the ideal translated sentence, another translated word candidate, and a word in the matched ideal translated sentence are combined in step 512. Hold in. If it does not exist, the process proceeds to step 507.

Step 507) The individual words of other translated word candidates are taken out. In the case of the machine translated English words in Fig. 8, "memory"
For, the "strage device" is retrieved. Further, since there is no other translation word candidate for "isto be provided", it is not extracted and the process proceeds to step 511.

(Step 508) It is searched whether the other translated word candidates extracted in step 507 exist in the ideal translated sentence extracted in step 502.

(Step 509) Next, it is judged whether or not there is a word that matches another translated word candidate in the ideal translated sentence, and if there is a word, the process proceeds to step 512. If it does not exist, the process proceeds to step 510.

Step 510) It is judged whether or not the word in the ideal translated sentence has been searched for all the other translated word candidates, and if checked, the machine translated word and all other translated word candidates for the word are extracted in step 502. Since it indicates that the translated text does not exist in the translated ideal translated text, the process proceeds to step 511. If the investigation has not been completed for all other translation word candidates, the process proceeds to step 507, and the above processing is repeated for all other translation word candidates.

Step 511) Only the words of the ideal translated sentence and other translated word candidates are held in pairs, and the process proceeds to step 513.

Step 512) A machine translation word, another translation word candidate, and a word in the matched ideal translation text are held as a set.

Step 513) It is judged whether or not all the machine-translated words have been searched for in the ideal translated text, and if checked, the process proceeds to step 514. If the survey has not been completed, the process returns to step 502.

Step 514) It is judged whether or not all ideal translated sentences have been investigated. If the survey has not been completed, the process returns to step 502 and the above processing is repeated until the survey is completed for all ideal translated sentences. Further, when all the investigations have been completed, the data indicating the correspondence between the machine translated text obtained by the above process and the words of each individual ideal translated text is obtained as the comparison result of the ideal translated machine translation comparing section 5.

In the case of the example of FIG. 8, the comparison result of the ideal translation machine translation comparing section 5 is as shown in FIG. FIG. 12 (A)
For the "Strage device is to be provided" in the ideal English translation A, the machine translated English word "Memory" and other translated word candidates "strage device" are machine translated English words "~ is to be provided" And other translated word candidates, the machine translated English words "~ is to be provided" are shown to match.

In FIG. 12B, the machine translated English word "Memory" and the other translated word candidate "strage device" match the ideal English translated sentence "To be equipped with the strage device." , The machine-translated English word "... Is to be provided" and other translation word candidates are shown as "no matching word".

In the comparison result of FIG. 8, various dictionary information obtained at the time of machine translation is added. This is used when the dictionary rule change addition process is performed.

Further, although the processing of the ideal translation machine translation comparison unit 5 shown in the flowchart of FIG. 11 uses the dictionary information of the language conversion dictionary 42 to compare the ideal translation and the machine translation, other dictionary or A form of comparison using rules is also conceivable.

Further, the ideal translation machine translation comparing section 5 uses the second natural language analysis device to generate the ideal translation conceptual structure of the second natural language when comparing the ideal translation and the machine translation. However, it is also possible to obtain a comparison result between the ideal translated text and the machine translated text by comparing with the machine translation conceptual structure of the second natural language of the machine translation generated during the machine translation. The means for generating the ideal translation concept structure and the machine translation concept structure may be provided in the ideal translation machine comparison unit 5, or may be provided before the ideal translation machine comparison unit 5.

From the comparison result obtained as described above, the minimum difference ideal translation extracting unit 6 extracts the comparison result between the ideal translation and the machine translation having the minimum difference. FIG. 13 is a flowchart of the operation of the minimum difference ideal translation extraction unit according to the first embodiment of this invention.

Step 601) It is judged whether there are a plurality of ideal translated sentences. In the case of one type, the minimum difference ideal translation becomes that one type of ideal translated text, so this processing ends. If there is more than one, the process proceeds to step 602.

Step 602) The comparison result is extracted in the order of ideal translation. In the comparison result of FIG. 12, the comparison results corresponding to the two types of ideal English translated sentences are sequentially extracted.

Step 603) A variable for calculating the difference between the ideal translated text and the machine translated text extracted in step 602, the ideal translated machine translated distance is set to 0 and initialized.

Step 604) The comparison result of the extracted ideal translated texts is taken out in the order of words of the machine translated texts. In the comparison result of FIG. 12, information on the comparison result corresponding to two types of machine English translated words is extracted.

Step 605) It is judged whether the machine translated word and the ideal translated word match in the comparison result corresponding to the extracted machine translated word. If they match, the process proceeds to step 609. If they do not match, Go to step 606.

Step 606) It is judged whether or not the other translated word candidate matches the ideal translated word in the comparison result corresponding to the extracted machine translated word, and if they match, the process proceeds to step 607,
If they do not match, the process proceeds to step 608.

Step 607) 1 is added to the variable and the ideal machine translation, and the process proceeds to step 609.

(Step 608) 2 is added to the variable, ideal machine translation, and the process proceeds to step 609.

Step 609) It is judged whether or not the comparison result has been evaluated for all the machine-translated words. If the comparison result has been evaluated, the process proceeds to step 610. If not evaluated, the process returns to step 604, and the above process is repeated until all machine-translated words are completed.

Step 610) A variable and an ideal machine translation value are added to the ideal translated text for which the evaluation of the comparison result is completed.

Step 611) It is judged whether or not the comparison results have been evaluated for all the ideal translated texts, and if they have been evaluated, the processing moves to step 612. If not evaluated, the process returns to step 602 and the above processing is repeated until all the ideal translated sentences are completed.

Step 612) The comparison result of the ideal translated sentence having the shortest ideal translated machine translated distance among all the ideal translated sentences is output by this processing.

In the example of FIG. 12, the ideal English translation A "Strage de
"vice is to be provided." for "1", "To be e
"3" for "quipped with the strage device"
12, the comparison result of "Strage device is to be provided." In FIG. 12 is output by the above processing, as shown in the result of the minimum difference ideal translation extraction unit in FIG.

The minimum difference ideal translation extraction unit 6 shown in the flowchart of FIG. 13 compares the ideal translation and the machine translation using the dictionary information of the language conversion dictionary 42, but uses other dictionaries or rules. A form in which the minimum difference ideal translation is determined is also conceivable.

Further, the ideal translation machine translation comparison unit 5 uses the second natural language analysis device to generate the ideal translation conceptual structure of the second natural language from the ideal translation, and the ideal translation conceptual structure and the machine translation. A form in which the machine translation conceptual structure generated at the time of translation is compared and the result of this comparison is used to determine the minimum difference ideal translation is also conceivable.

Based on the comparison information of the minimum difference ideal translations obtained by the minimum difference ideal translation extraction unit 6, the dictionary rule change adding unit 7
By this, the information of the rule or the dictionary to be additionally corrected is extracted, and the additional correction is made according to the result.

FIG. 15 shows a dictionary according to the first embodiment of the present invention.
It is a flow chart which shows operation of rule change addition processing.

Step 701) Minimal difference ideal translation extraction unit 6
The result of is extracted in the order of words in the machine translation. For example, in FIG.
In the result of the minimum differential ideal translation extraction unit 6 shown in FIG.
"ory" information is retrieved, then "~ is to be provided"
Information is retrieved.

Step 702) Information of Japanese words, parts of speech and matching words is extracted from the extracted information.

Step 703) Using the Japanese word extracted in step 702 as a key, the language conversion dictionary 42 corresponding to the part of speech is searched and the dictionary information is extracted. For example, "memor
For "y", use the Japanese word "memory" as a key
The noun part of the language conversion dictionary 42 is searched for and the dictionary information is extracted. In addition, regarding "-is to be provided", the Japanese word "-having to have" is used as a key, and the language conversion dictionary 4 in FIG.
The second noun part is searched, and the dictionary information is extracted.

When the method of the flowchart of the ideal translation comparing section 5 of FIG. 11 described in the present embodiment is used, the dictionary information can be searched. For example, the ideal translation machine translation comparing section 5 is used.
Although there is no information in the dictionary, if the ideal translated text and the machine translated text are one word, the dictionary information may not be searchable if the ideal translated text is used as a matching word. obtain.

(Step 704) It is judged whether or not the dictionary information can be extracted, and if it is extracted, the routine goes to Step 705.

Step 705) The translation word selection conditions such as the priority are changed so that the translation word of the matching word has the highest priority. For example, in the example of FIG. 14, since the dictionary information can be extracted for both of the two types of words, the priority is set to “1” so that the translated word of the translated word “Strage device” has the highest priority for the dictionary entry of “storage device”. To the priority "2" of the translated word "memory". Also, regarding the dictionary entry “having”, since there is only one type of translated word and the translated word is the same as the matching word, the matching word has already been given the highest priority, so the dictionary information is not changed. In the example of the translated word of FIG. 14, the priority of the dictionary entry of the “storage device” is simply changed, but in some cases, information such as the word that modifies the word and the type of the modified word is selected as the translated word. A method of registering as a condition is also conceivable.

Step 706) A dictionary entry is newly created from the translated words of the Japanese word and the matching word and newly registered.

Step 707) It is judged whether or not all machine English words have been added / corrected in the dictionary, and if so, the processing of the dictionary rule change / addition unit 7 ends. If not, the process returns to step 701 and the above process is repeated until all machine English words are completed.

FIG. 16 shows the dictionary information of the language conversion dictionary 42 as a result of the dictionary / rule change addition processing using the result of the minimum difference ideal translation extraction unit of the first embodiment of the present invention.
As shown in the same figure, the priority of the translated word of "memory device" of the noun part is changed from the information of FIG.

The dictionary rule change addition unit 7 shown in the flowchart of FIG. 15 uses the dictionary information of the language conversion dictionary 42 to change and adds the dictionary information of only the language conversion dictionary 42, but other dictionaries and rules are added. It is also conceivable to use and to change and add the dictionary information of those dictionaries and rules.

Further, in the flowchart shown in FIG. 15, the dictionary / rule change adding section 7 automatically changes / adds the dictionary 4 and the rule 3, but at the time of this change addition,
It may be a form in which the change / addition information is displayed and the judgment of registration such as change / addition is left to a person.

When the input Japanese sentence is machine-translated again using the dictionary and rules learned by the above method, a machine-translated sentence reflecting the tendency of the translated sentence appearing in the bilingual corpus, which is the learning data, is obtained. For example, the input Japanese sentence in Fig. 8 is one of the ideal English translations, "Strage device is to be proive".
The translation of "d." is obtained as a machine English translation.

[Second Embodiment] Next, a dictionary / rule learning method for a machine translation system according to a second embodiment of the present invention will be described. In the second embodiment, a set of first natural language sentences (input sentences) and second natural language ideal translated sentences (parallel translated sentences) which are in a bilingual relationship with each other are used for learning, and the same from the parallel translated sentences is used. The difference from the first embodiment is that a set of input sentences and their ideal sentences are extracted.

In the first embodiment, it is assumed that a plurality of ideal translated sentences are added to the input sentence.
In the embodiment described above, when the same input sentence exists in the bilingual sentence used for learning, the ideal translated sentences are collected for each type of the input sentence and the learning process is performed. Therefore, in the learning process in the case where a plurality of types of ideal translated sentences are added to one type of input sentence after being summarized in this way, a plurality of ideal sentences are input to the input sentence in the first embodiment. The learning process can be performed in the same manner as in the case of giving it.

FIG. 17 shows a basic block configuration of a machine translation dictionary / rule automatic learning system according to a second embodiment of the present invention. In the figure, the same components as those in FIG. 5 are designated by the same reference numerals, and the description thereof will be omitted. The system shown in the figure is different from the system configuration shown in FIG. 5 in that a parallel translation sentence reading unit 10 for inputting a parallel translation sentence and a collection of the same input sentence from the parallel translation sentence input by the parallel translation sentence reading unit 10 and its ideal translation sentence The configuration of the same input sentence extraction unit 11 for reading is different, and the other system configuration is the same as the configuration of FIG.

The processing procedure of the machine translation dictionary / rule automatic learning system having the basic block configuration shown in FIG. 17 will be described. Here, the bilingual text shown in FIG. 18 will be described as an example.

The bilingual sentence reading unit 10 inputs a set of bilingual sentences of the first natural language and the second natural language (translated sentences). In the example of FIG. 18, the first sentence is included in the bilingual sentence.
There are two input sentences “having a memory device” in Japanese, which is a natural language, and at each location, “Strage device is to be provided. "and" To be equippped with the strage device. " In this way, when the same first language sentence exists in the same sentence, the translation result may differ depending on the mood of the translator at that time. As in the example of FIG. 18, it is frequently performed that a plurality of types of second natural language sentences are given to one type of first natural language sentence, so such a phenomenon is considered. There is a need.

(1) The same input sentence extraction unit 11 classifies the input bilingual sentences according to different input sentences. FIG.
As shown in FIG. 19 regarding the input Japanese sentence “having a storage device” in the bilingual sentence, the two types of ideal English translated sentences appearing in different places are classified into one place.

(2) Next, similar to the first embodiment, the same input sentence extraction unit 11 sorts out different input sentences of the parallel translated sentences classified by different input sentences and the corresponding pairs of ideal English translated sentences as in the first embodiment. In the machine translation system 2 as shown in FIG. 6, the rule 3 and the dictionary 4 are used to translate into a machine English translation of the second natural language. Regarding the input Japanese sentence “having a memory” shown in FIG. 19, a machine English translated sentence as shown in FIG. 8 of the first embodiment is generated.

The machine English translation obtained as described above is
The ideal English translation and the ideal machine translation are compared by the comparison unit 5,
The result of comparison between the ideal translated text and the machine translated text is obtained. For example, it is obtained by the ideal machine translation comparison process shown in the flowchart of FIG. 11 described in the first embodiment. If the processing of FIG. 11 is used, the Japanese sentence “having memory” shown in FIG.
Is obtained for the comparison result as shown in FIG.

Next, from the thus obtained comparison result of the ideal translation and the machine translation, the minimum difference ideal translation extracting unit 5 extracts the comparison result of the ideal translation and the machine translation having the smallest difference. For example, it is obtained by the minimum difference ideal translation extraction process shown in the flowchart of FIG. 13 described in the first embodiment. Using the example of FIG. 13, the input Japanese sentence “having a memory” shown in FIG. 19 is compared with the comparison result shown in FIG.
The minimum differential ideal translation extraction result as shown in FIG. 4 is obtained.

Finally, based on the comparison information of the minimum difference ideal translations obtained by the minimum difference ideal translation extraction unit 6, the dictionary rule change addition unit 7 extracts the rule or dictionary information to be additionally corrected. Additional corrections will be made according to the results. For example, rules and dictionaries are changed and added by the dictionary rule change addition process shown in the flowchart of FIG. 15 described in the first embodiment. By using the process of FIG. 15, the input Japanese sentence “having a memory” shown in FIG. 19 is converted into a language used in the machine translation system shown in FIG. 7 from the minimum difference ideal translated sentence extraction result shown in FIG. The dictionary information is changed as shown in FIG.

By repeating the learning process as described above for all different input Japanese sentences and their ideal English translations obtained by the same extraction unit 11, it becomes possible to improve the translation quality of the input bilingual sentence. Can build dictionaries and rules for machine translation systems.

The dictionary / rule change addition unit 7 may be provided with a display function so that when the rules 3 and the dictionary 4 are updated, the contents to be updated are displayed so that the user can confirm the displayed contents. It is also possible to check the displayed contents and make corrections if necessary.

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

[0103]

INDUSTRIAL APPLICABILITY The present invention is a translation result of an input sentence (machine translation)
When the language conversion dictionary and language conversion rules are changed and added by comparing the ideal translation of the input sentence prepared in advance with multiple types of ideal translations for the same input sentence , The ideal translated text with the least difference from the machine translated text is used.

According to the present invention, a bilingual sentence is read, an input sentence of a first natural language and an ideal translated sentence of a second natural language which are in a bilingual relation are read, and the dictionary / rule for the machine translation system is processed by the subsequent processing. Changes are added automatically. First,
In the process of comparing the ideal translated text and the machine translated text, the first natural language input sentence read by the parallel translation reading is translated into the second natural language machine translated text by using the machine translation system.
The machine translated text and the ideal translated text are compared by, for example, using a dictionary for a machine translation system to find the correspondence between the words in the machine translated text and the words in the ideal translated text. Next, when a plurality of types of ideal sentences are given to one input sentence by this comparison, for example, the number of different words in the ideal translated sentence and the machine translated sentence is investigated, and the ideal word with the smallest number of different words is examined. An ideal translated sentence with the smallest difference (ideal translated sentence with minimum difference) is extracted by extracting translated sentences. When changing the dictionary rule, the dictionary and the rule are changed and added based on the result of comparison between the ideal translated text and the machine translated text and the minimum difference ideal translated text. As a result, even if a plurality of types of ideal translated texts are attached to one input sentence, the ideal translated text that is the closest to the machine translated text is used to change and add a dictionary or rule, so that it will be translated later. The machine translation is closer to the ideal translation.

Further, according to the present invention, a machine translation conceptual structure and an ideal translation conceptual structure are generated, these two conceptual structures are compared, and a plurality of types of ideal translation conceptual structures are given to one type of input sentence. In this case, the ideal translation concept structure with the smallest difference is extracted, and the dictionary and rules are changed and added based on the smallest difference ideal translation concept structure. As a result, even when a plurality of types of ideal translated sentences are given to one type of input sentence, the dictionary and rules are changed and added using the ideal conceptual structure that is closest to the machine translation conceptual structure.

Further, according to the present invention, for one type of input sentence,
When a plurality of types of ideal translations are given, the dictionary and rules are changed and added by using the ideal translation concept structure close to the machine translation concept structure.

Further, according to the present invention, a set of parallel translation sentences of the input sentence of the first natural language and the ideal translation sentence of the second natural language which are in the parallel translation relation are read. In the process of extracting the same input sentence, the same input sentence and a set of its ideal translated sentences are extracted, the ideal translated sentence and the machine translated sentence are compared, and the ideal translated sentence with the smallest difference is extracted. Even if the ideal translated text of is added, the dictionary and the rule are changed and added by using the ideal translated text that is the closest to the machine translated text.

As described above, according to the present invention, when learning a bilingual sentence or a dictionary or a rule for machine translation using a bilingual sentence, a plurality of ideal translated sentences or a plurality of types for one type of input sentence. Even if the ideal translated sentence of is added, the ideal translated sentence closest to the machine translated sentence can be selected and learned. Therefore, it is possible to avoid registering dictionaries or rules that generate a plurality of different translated expressions for the same input sentence expression.

Further, the dictionary and the rule can be registered by referring to the ideal translated sentence which is most easily targeted by the machine translation system. Therefore, it is possible to automatically add and modify dictionaries and rules in the most feasible and effective way for machine translation systems.

[Brief description of drawings]

FIG. 1 is a diagram (No. 1) for explaining the principle of the present invention.

FIG. 2 is a diagram (part 2) for explaining the principle of the present invention.

FIG. 3 is a diagram (part 3) for explaining the principle of the present invention.

FIG. 4 is a diagram (No. 4) for explaining the principle of the present invention.

FIG. 5 is a basic block diagram of a machine translation dictionary / rule automatic learning system according to the first embodiment of this invention.

FIG. 6 is a basic block configuration diagram of a machine translation system using a dictionary and a rule that are additionally modified according to the first embodiment of this invention.

FIG. 7 is a diagram showing an example of dictionary entries of a language conversion dictionary.

FIG. 8 is a diagram showing an example of information about an input Japanese sentence, an ideal English translated sentence, a machine English translated sentence corresponding to the input Japanese sentence, and a word of the machine English translated sentence obtained when the input Japanese sentence is machine translated.

FIG. 9 is a diagram showing a conceptual structure of Japanese obtained by the first language analysis unit of the machine translation system.

FIG. 10 is a diagram showing a conceptual structure of English obtained by a language conversion unit of a machine translation system.

FIG. 11 is a flowchart of an ideal translation machine translation comparison operation according to the first embodiment of this invention.

FIG. 12 is a diagram showing a comparison result of an ideal translation machine translation comparison according to the first embodiment of this invention.

FIG. 13 is a flowchart of the operation of the minimum difference ideal translation extracting unit according to the first embodiment of this invention.

FIG. 14 is a diagram showing a result of the minimum difference ideal translation extracting unit according to the first embodiment of this invention.

FIG. 15 is a flowchart of an operation of dictionary / rule change addition processing according to the first embodiment of this invention.

FIG. 16 is a diagram showing dictionary information when a dictionary / rule change addition process is performed using the result of the minimum difference ideal translation extraction unit according to the first example of the present invention.

FIG. 17 is a basic block configuration diagram of a dictionary / rule learning system for machine translation system according to a second embodiment of the present invention.

FIG. 18 is a diagram showing an example of a parallel translation sentence according to the second embodiment of the present invention.

FIG. 19 is a diagram showing an example of a bilingual sentence modified by the same input sentence extracting step of the second example of the present invention.

[Explanation of symbols]

1 parallel translation reading unit, parallel translation reading unit 2 machine translation system 3 rules 4 dictionary 5 machine translation ideal translation comparison unit, machine translation ideal translation comparison unit, machine translation ideal translation concept structure comparison unit 6 difference minimum ideal translation extraction unit, difference Minimum ideal translation extraction means,
Minimum difference ideal translation conceptual structure extraction means 7 Dictionary rule change addition unit, dictionary rule change addition unit 8 First natural language input unit 9 Second natural language output unit 10 Bilingual sentence reading unit, Bilingual sentence reading unit 11 Same input sentence extracting unit , Same input sentence extraction means 21 first language analysis unit 22 language conversion unit 23 second language generation unit 31 first language analysis rule 32 language conversion rule 33 second language generation rule 41 first language analysis dictionary 42 language conversion dictionary 43th Bilingual dictionary

Claims (10)

[Claims] 1. 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, wherein the input sentence of the first natural language is provided. An ideal translated sentence of the second natural language having a parallel translation relationship with the second natural language machine translation sentence obtained by translating the input sentence of the first natural language by the machine translation system; When comparing the ideal translated sentences of the second natural language and if there are a plurality of types of ideal translated sentences for the same input sentence, the ideal translated sentence with the smallest difference from the machine translated sentence is used, A dictionary / rule learning method for a machine translation system, characterized in that the rules are additionally changed. 2. A machine translation system having means for converting a 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 a parallel translation with an input sentence of the first natural language. A parallel translation reading step of inputting an ideal translated sentence of the second natural language, and a machine translated sentence of the second natural language obtained by translating the input sentence of the first natural language using the machine translation system; The ideal translation machine translation comparison step of comparing the ideal translations of the second natural language and the comparison result obtained in the ideal translation machine translation comparison step gives a plurality of types of ideal translations to one input sentence. If so, the difference minimum ideal translation extraction step for extracting the ideal translation with the smallest difference, the comparison result obtained in the ideal translation machine translation comparison step, and the difference obtained in the minimum difference ideal translation extraction step Is the least Claim 1 machine translation dictionary rule learning method for system further comprising a dictionary rule changes additional step of changing add the dictionary specification and the rules based on the ideal translation are. 3. A machine translation system having means for converting a 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 a parallel translation with an input sentence of the first natural language. A bilingual sentence reading step of inputting a bilingual sentence which is a collection of related second natural language ideal translated sentences, and a bilingual sentence corresponding to the same input sentence of the first natural language is extracted from the bilingual sentence. An identical input sentence extraction step; an ideal translation machine translation comparison step of comparing a machine translation sentence of the second natural language translated by the machine translation system with an ideal translation sentence; and an ideal translation machine translation comparison step According to the comparison result obtained in step 1, the ideal translated sentence having the smallest difference is extracted from the bilingual sentence in which a plurality of types of ideal translated sentences are provided for one type of the input sentence extracted in the same input sentence extracting step. Min. Minimum ideal translated sentence extraction step, the comparison result obtained in the ideal translated machine translation comparison step, and the dictionary or the rule is changed based on the ideal translated sentence with the smallest difference obtained in the difference minimum ideal translated sentence extraction step. The dictionary / rule learning method for a machine translation system according to claim 1, further comprising a step of adding and changing a dictionary rule. 4. A machine translation system having means for converting a sentence of a first natural language which is a source language into a sentence of a second natural language which is a target language, wherein the input sentence of the first natural language is A parallel translation reading step of inputting an ideal translation of a second natural language having a parallel translation relationship, and a step of translating the input sentence of the first natural language into a machine translation of a second natural language using the machine translation system. A machine translation conceptual structure generating step for generating a second natural language machine translation conceptual structure; an ideal translation conceptual structure generating step for generating a second natural language ideal translation conceptual structure from the ideal translation; A machine translation ideal translation concept structure comparison step for comparing the concept structure and the ideal translation concept structure, and a plurality of types of ideals for one input sentence according to the comparison result obtained in the machine translation ideal translation concept structure comparison step. With translation concept structure If so, the minimum difference ideal translation concept structure extraction step for extracting the ideal translation concept structure with the smallest difference, the comparison result obtained in the machine translation ideal translation concept comparison step, and the minimum difference ideal translation concept structure The machine translation system dictionary / rule learning method according to claim 1, further comprising a dictionary rule change addition step of changing and adding the dictionary or the rule based on the difference extraction result obtained in the extraction step. 5. A machine translation system having means for converting a sentence of a first natural language, which is a source language, into a sentence of a second natural language, which is a target language, in which a first natural language is in a bilingual relationship with each other. And a step of inputting a bilingual sentence that is a collection of the second natural language ideal translated sentence, and the same input sentence that extracts the same input sentence and a set of ideal translated sentences corresponding to the input sentence from the bilingual sentence An extracting step, and a machine translation conceptual structure generating step of generating a machine translation conceptual structure of a second natural language when translating the input sentence into a machine translation sentence of a second natural language using the machine translation translation system, An ideal translation concept structure generating step for generating a second natural language ideal translation concept structure from the ideal translation sentence; a machine translation ideal translation concept structure comparison step for comparing the machine translation concept structure with the ideal translation concept structure; Machine translation Based on the comparison result obtained in the ideal translation conceptual structure comparison step, a difference is obtained from the bilingual sentence to which a plurality of types of ideal translation conceptual structures are added for one type of the input sentence extracted in the same input sentence extraction step. The difference obtained by the minimum ideal translation concept structure extraction step, which extracts the ideal translation concept structure with the smallest number, the comparison result obtained by the machine translation ideal translation concept structure comparison step, and the difference obtained by the minimum minimum ideal translation concept structure extraction step The dictionary / rule learning method for a machine translation system according to claim 1, further comprising a dictionary rule change addition step of changing and adding the dictionary or the rule based on the extraction result. 6. A machine translation system having means for converting a sentence of a first natural language, which is a source language, into a sentence of a second natural language, which is a target language, in which the first natural language has a bilingual relationship with each other. Input sentence and a parallel translation sentence reading means for inputting a parallel translation sentence which is a set of ideal translation sentences of the second natural language, and a second natural language obtained by translating the input sentence of the first natural language using the machine translation system. Based on the machine translation ideal translation comparing means for comparing the machine translation of the language and the ideal translation of the second natural language, and the comparison result obtained by the machine translation ideal translation comparing means, a plurality of input sentences can be obtained. When types of ideal translations are added, the minimum difference ideal translation extracting means for extracting the ideal translation having the smallest difference, the comparison result obtained by the machine translation ideal translation comparing means, and the minimum difference ideal translation extraction The smallest difference obtained by means Machine translation system dictionary rule learning device which comprises a dictionary rule changing additional means for changing add the dictionary specification and the rules based on the ideal translation. 7. A machine translation system having means for converting a sentence of a first natural language, which is a source language, into a sentence of a second natural language, which is a target language, in which the first natural language is in a bilingual relationship with each other. Input sentence and a parallel translation sentence reading means for inputting a parallel translation sentence which is a set of ideal translation sentences of the second natural language, and the same input sentence extraction for extracting the parallel translation sentence corresponding to the same input sentence from the parallel translation sentence Means, a machine translation ideal translation comparing means for comparing a machine translation translated the input sentence into a second natural language sentence using the machine translation system, and an ideal translation, and the machine translation ideal translation comparing means According to the comparison result obtained in
A minimum difference ideal translation extracting unit that extracts an ideal translated sentence having the smallest difference from a bilingual sentence in which a plurality of types of ideal translated sentences are given to one type of input sentence extracted by the same input sentence extracting unit; Dictionary rule change adding means for changing and adding the dictionary or the rule based on the comparison result obtained by the machine translation ideal translation comparing means and the ideal translated sentence with the smallest difference obtained by the minimum difference ideal translation extracting means, A dictionary / rule learning method for a machine translation system, which includes: 8. A machine translation system having means for converting a sentence of a first natural language, which is a source language, into a sentence of a second natural language, which is a target language, in which the first natural language has a bilingual relationship with each other. Input sentence and a parallel translation sentence reading means for inputting a parallel translation sentence which is a set of ideal translation sentences of the second natural language, and when translating the input sentence into the machine translation sentence of the second natural language using the machine translation system. A machine translation conceptual structure generating means for generating a second natural language machine translation conceptual structure, an ideal translation conceptual structure generating means for generating a second natural language ideal translation conceptual structure from the ideal translated text, and the machine translation A machine translation ideal translation concept structure comparison means for comparing the concept structure and the ideal translation concept structure, and a plurality of types of ideals for one type of the input sentence according to the comparison result obtained by the machine translation ideal translation concept structure comparison means. A translation concept structure is added In the case of the difference, the minimum difference ideal translation concept structure extraction means for extracting the ideal translation concept structure having the smallest difference, the comparison result obtained by the machine translation ideal translation concept structure comparison means, and the minimum difference ideal translation concept structure. A machine translation system dictionary / rule learning device comprising: a dictionary rule change adding unit for changing and adding the dictionary or the rule based on the difference extraction result obtained by the extracting unit. 9. A machine translation system having means for converting a sentence of a first natural language, which is a source language, into a second natural language, which is a target language, and an input sentence of a first natural language having a bilingual relationship with each other. And a second translated sentence reading means for inputting a translated sentence which is a set of ideal translated sentences of the second natural language, and the same input sentence extraction for extracting the same input sentence and a set of ideal translated sentences corresponding to the input sentence from the translated sentence. Means, a machine translation concept structure generating means for generating a machine translation concept structure when translating the input sentence into a machine translation sentence of a second natural language by using the machine translation translation system, and analyzing the ideal translation sentence. An ideal translation concept structure generating means for generating a second natural language concept structure, a machine translation ideal translation concept structure comparing means for comparing the machine translation concept structure with the ideal translation concept structure, and the machine translation ideal translation concept structure. Ratio obtained by comparison means According to the comparison result, the ideal translation conceptual structure having the smallest difference is extracted from the bilingual sentence in which a plurality of types of ideal translation conceptual structures are given to one type of the input sentence extracted by the same input sentence extracting means. Based on the comparison result obtained by the minimum differential ideal translation conceptual structure extraction means, the machine translation ideal translation conceptual structure comparison means, and the difference extraction result obtained by the minimum differential ideal translation conceptual structure extraction means, the dictionary or the dictionary A dictionary / rule learning device for a machine translation system, comprising: a dictionary rule changing / adding means for changing / adding a rule. 10. The dictionary / rule for a machine translation system according to claim 6, 7, 8 and 9, wherein said dictionary rule changing / adding means includes display means for displaying information for changing / changing said dictionary and said rule. Learning device.