JP3353647B2 - Dictionary / rule learning device for machine translation system and storage medium storing dictionary / rule learning program for machine translation system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

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

[0002]

2. Description of the Related Art Conventionally, when constructing a dictionary or a rule for a machine translation system, it is general to manually construct a dictionary or a rule for the machine translation system with reference to a dictionary or the like used for translation or the like by a human. It is. In addition, the dictionary and rules for the machine translation system constructed in this way are, if an inappropriate translation is generated based on the translation result of the machine translation system, the cause of the generation of the translation. Experts who have a good understanding of the internal processing of machine translation systems manually add and modify dictionaries and rules.

As a first technique for automating the work of adding and modifying dictionaries and rules, a set (a bilingual corpus) of pairs of a source language sentence and a target language sentence (ideal translation sentence) having a bilingual relationship.
An improved method for adding and modifying dictionaries by analyzing the source language concept structure and the ideal translation concept structure by analyzing each of them (for example, "Machine Translator" (Japanese Patent Laid-Open No. Hei 4-205056)) ).

As a second technique, a pair of corresponding expressions is extracted from a source language sentence and a target language sentence in a bilingual corpus, and a translation rule candidate to be additionally corrected based on the result is extracted. An improved method of presenting to a user and adding / correcting a translation rule specified by the user (for example, “Translation rule learning method in machine translator”
No.)).

[0005] As a third technique, a difference is extracted from a difference between a machine translation concept structure and an ideal translation concept structure, and the difference is classified according to a solution method in a machine translation system to add or modify a dictionary or rule. , A method of repeating until an optimal dictionary or rule is registered using the dictionary or rule that has been added and corrected (“Dictionary / rule learning method for machine translation system and dictionary / rule learning device for machine translation system”)
(Japanese Patent Application No. 6-34476) is known.

[0006]

However, in the above-described method of constructing a dictionary or rule of the conventional machine translation system, a dictionary or the like created for humans is used. Is not guaranteed to be generated. Further, in order to add or modify the dictionary or rule of the machine translation system, it is necessary to verify the translation result by an expert who understands the internal processing of the machine translation system. Therefore, there is a problem that the dictionary and the rules cannot be additionally modified unless the user is an expert. Also,
Since this work is performed manually, there is a problem that time and economic costs for constructing a dictionary and rules are increased.

In the latter of the above conventional techniques, that is, in the improvement method, in the case of the first method, a dictionary or a rule can be obtained from the bilingual corpus information without considering the translation result of the machine translation system. However, there is a problem that an improper additional change is made to an existing dictionary or rule by a bilingual corpus, which adversely affects a machine translation result.

In the second method, since a manual operation by a user is assumed, there is a problem that time and economic costs for constructing a dictionary and rules are increased. Further, since there is only one learning process together with the first method and the second method, it is unclear whether the learning result is really correct. In the case of the third method, it is assumed that there is no means for generating an ideal translation concept structure, or there is a large number of ideal translations,
If there is not enough time and computer space to generate the ideal translation structure, it is impossible to solve the problem using this method. In addition, even if there is a means for generating an idealized translation concept structure, if the generated ideal translation concept structure has low reliability, the dictionary and rules are added and corrected by referring to the incorrect ideal translation concept structure. In some cases, effective additional correction cannot be performed efficiently.

SUMMARY OF THE INVENTION The present invention has been made in view of the above points, and solves the above-mentioned conventional problems. The present invention is not limited to the use of the ideal translation concept dictionary generating means, and is not limited to the use of the ideal bilingual corpus. Using not only the information but also the information of the machine translation sentence output by the machine translation system, referring to the bilingual dictionary and the dictionary of the second language which is the target language, the input sentence, the machine translation, and the ideal translation are three types of sentences. From the input word in the input sentence, and a set of three sets of a machine translation word in the machine translation sentence and the ideal translation word in the ideal translation, which are in parallel with the input word, are extracted. An object of the present invention is to provide a dictionary / rule learning device for a machine translation system for adding and modifying a simple dictionary and rules, and a storage medium storing a dictionary / rule learning program for a machine translation system.

[0010]

[Means for Solving the Problems]

[0011]

[0012]

[0013]

[0014]

[0015]

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

According to the present invention, a dictionary and rules used in a machine translation system 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 are additionally changed . A machine translation system dictionary / rule learning device,
An ideal translation input means for inputting an ideal translation of a second natural language having a bilingual relationship with the input sentence, and an ideal translation input means,
A second natural language machine translation sentence converted from the input sentence by the machine translation system, a comparing means for comparing the input sentence, and the comparing means, the machine translation word having a parallel translation with the input word in the input sentence. And, by extracting a set of three sets of ideal translation words in the ideal translation, either a dictionary or a rule,
Update means for adding and changing both.

Further, the present invention relates to input data having a bilingual relation with each other.
Bilingual reading that reads the literary sentence and the ideal translation of the second natural language
Compare the input sentence, the ideal translation and the machine translation
The input word in the input sentence and the input word
The machine translation in the machine translation and the ideal translation in the ideal translation
Extract a bilingual expression set, which is a set of bilingual expression sets
The extracted bilingual expression set extracting means and the bilingual expression
Based on the set of bilingual expression sets issued, a dictionary or rule
Includes dictionary / rule addition / change means for additional change.

Further, the bilingual expression set extracting means of the present invention, when extracting the bilingual expression set, uses the conversion history when the input sentence is converted into the machine translated sentence by the machine translation system and compares the input word in the input sentence with the input word. A bilingual machine translation word pair extracting means for extracting a machine translation word in the machine translation sentence having a bilingual relationship. Further, the bilingual expression set extraction means of the present invention includes a bilingual ideal translation word pair extracting means for extracting an ideal translation word in an ideal translation text having a bilingual relationship with the input word in the input sentence when extracting the bilingual expression set.

The bilingual ideal translation pair extracting means of the present invention uses the bilingual dictionaries of the first natural language and the second natural language, and extracts the ideal translation word to extract the input word in the input sentence. Using the key as a key, a bilingual dictionary is searched, and an input word translation candidate extracting means for extracting an input word translation candidate of a second natural language having a bilingual relationship with the input word, and comparing the input word translation candidate with an ideal translation. And an input word / ideal translation word pair extracting means for extracting a pair of a pair of an input word in an input translation and an ideal translation word in an ideal translation.

The bilingual ideal translation pair extraction means of the present invention uses the second natural language dictionary and the input word translation candidate extracted by the input word translation candidate extraction key as keys, Means for comparing the input word candidate candidate related word extracted by searching the dictionary with the ideal translation and extracting a set of a pair of the input word in the input sentence and the ideal translation in the ideal translation having a bilingual relationship .

The bilingual expression set extracting means of the present invention, when extracting a bilingual expression set set, extracts a corresponding machine translation word in a corresponding machine translation sentence and an ideal machine translation ideal word in an ideal translation sentence. Including pair extraction means. Further, the corresponding machine translated word ideal word pair extracting means of the present invention includes means for using a second natural language dictionary when extracting the machine translated word in the machine translated sentence and the ideal translated word in the ideal translated sentence.

The corresponding machine translated word ideal word pair extracting means of the present invention uses a second natural language dictionary when extracting the machine translated word in the machine translated sentence and the ideal translated word in the ideal translated sentence. the machine translation in translation as a key, comprising a means for comparing the machine translation related words and ideal translation extracted by searching the second natural language dictionary. Further, the corresponding machine translated word ideal word pair extracting means of the present invention uses the second natural language dictionary when extracting the machine translated word in the machine translated sentence and the ideal translated word in the ideal translated sentence. Means for comparing a machine-translated word extracted from the second natural-language dictionary by using the machine-translated word as a key and an ideal translated word;

The corresponding machine translated word ideal word pair extracting means of the present invention, when extracting the machine translated word in the machine translated sentence and the ideal translated word in the ideal translated sentence, translates between the first natural language and the second natural language. Using the dictionary, a machine translation reverse translation candidate that is a set of translations of the first natural language obtained using the machine translation in the machine translation as a key, and a first translation using the ideal translation in the ideal translation as a key Means for extracting a candidate for a reverse translation word, which is a set of translations of a natural language; and comparing the candidate for a reverse translation word for a machine translation with the candidate for a reverse translation word for a machine. It includes a translation and means for extracting an ideal translation in the ideal translation.

Further, the bilingual expression set extraction means of the present invention, when comparing the words of the second natural language, can use the machine language in the machine translation, the input word translation candidate, or the input word translation candidate related word or A means for creating a list of different machine-translated words, which are calculated separately for at least one of the machine-translated words, and a method for calculating the list of ideal-translated words or ideal-translated words in an ideal translation, or both. Means for creating a word list different from the ideal translation word, and means for comparing words in the word list different from the machine translation and words in the word list different from the ideal translation are included .

Further, the bilingual expression set extracting means of the present invention, when comparing the words of the first natural language, sums up at least one of the input word and the machine translation reverse-translated word candidate separately for each word. A means for creating a list of words that are different from the input word machine translation reverse translation candidate, a means for creating a list of words that are different from the ideal translation reverse translation word candidate by summing up the ideal translation reverse translation word candidates, and an input word machine translation reverse Means for comparing a word in the word list with a different candidate word and a word in a word list with a different ideal translation word candidate are included.

Further, the bilingual expression set extracting means of the present invention, when comparing the words of the first natural language and the words of the second natural language, whether the two words match or one of the words Means for distinguishing the degree of correspondence depending on whether the other word is included or a part of both words match is included. Further, the bilingual expression set extraction means of the present invention performs morphological analysis of the ideal translation and extracts information of the ideal translation morpheme structure using information of the ideal translation morpheme structure, and uses the information of the machine translation reverse translation word candidate and the ideal translation reverse translation. At least one of the words in the word list, unlike the ideal translation,
Means for comparing at least one of the words in the word list different from the machine translation, the words in the word list different from the candidate for machine translation reverse word, and the words in the word list different from the candidate word for ideal translation reverse translation are included.

A set of bilingual expression sets that can be extracted from a set of an input sentence, an ideal translated sentence, and a machine translated sentence from a set of bilingual expression sets extracted by the bilingual expression set extracting means of the present invention. There is further provided a bilingual expression combination extracting means for extracting a set of bilingual expression combinations. The bilingual table of the present invention
Current set extraction means, the translated expression set extraction unit, the individual translated expression of sets of translated expression set set to be extracted, imparting translated expression set accuracy is accuracy in accordance with the individual translated expression of sets of extraction process bilingual has a representation of sets accuracy imparting means, translated expression combination extraction means uses the translated expression set accuracy granted to individual translated expression of sets of translated expression set in, means for applying a probability to each translated expression combination The dictionary / rule addition / change means is provided by the bilingual expression combination extraction means.
Te, dictionaries and rules using the probability granted to each translated expression combinations of translated expression combination set in which is extracted
A means for adding and changing files .

Further, the present invention has a repetition means for repeating learning using dictionaries and rules obtained by a dictionary / rule learning device for a machine translation system. Further, the dictionary / rule addition / change means of the present invention includes a learning end determining means for determining the end of learning. Further, the present invention has a machine translation history holding unit that records in a machine translation history storage unit which is a history of a result of translating the input sentence using a dictionary or a rule obtained by a dictionary / rule learning device for a machine translation system. The learning end determining means includes means for determining that learning is to be ended when the recorded contents of the machine translation history storage means are in a steady state.

Further, the present invention is a dictionary rules additional changing means comprises display means for displaying the dictionary or information to change the rule. The present invention additionally modifies a dictionary and rules used in a machine translation system that converts an input sentence of a first natural language as a source language into a sentence of a second natural language as a target language.
To a storage medium storing a machine translation system for a dictionary-rule learning program, ideal translation input to enter the ideal translation of the second of natural language in the input sentence and the bilingual relationship steps
And a comparison step of comparing the input sentence with the machine translation sentence of the second natural language obtained by converting the input sentence by the machine translation system in the ideal translation inputting step , and a comparison step.
By-up by extracting the three sets of the set of ideal translation of the machine translation and the ideal in translation in the input word and translation relationship of the input sentence, adds change either or both dictionary and rules update Machine translation system consisting of steps
Stores dictionary / rule learning programs.

[0037] As described above, in machine translation dictionary rule learning device for the system of the present invention, the translation reads hand
Stage, a first natural language in the translation of each other (input sentence)
And the second natural language (ideal sentence). From now on, the dictionary and rules for the machine translation system are automatically created and added and corrected by the following means . The bilingual expression set extracting means compares the input sentence with the ideal translation, and finds the input word in the input sentence, the machine translation in the machine translation having a bilingual relationship with the input word, and the ideal translation in the ideal translation. A bilingual expression set set, which is a set of bilingual expression sets, is extracted. Dictionary rules additional hand
In the stage , the dictionary and rules are additionally changed based on the bilingual expression group set extracted by the bilingual expression group extracting means . As a result, a dictionary or a rule for a machine translation system can be automatically created or additionally changed from the input sentence and the ideal translation.

The bilingual expression set extracting means extracts a machine translation word in a machine translation sentence having a bilingual relationship with an input word in the input sentence by using a conversion history when the input sentence is converted into a machine translation sentence by a machine translation system. A bilingual machine translation word pair extracting means can be provided. As a result, it is possible to accurately associate a machine translation word in a machine translation sentence having a parallel translation with an input word in an input sentence, and to extract a set of machine translation words in a machine translation sentence having a parallel translation with the input word in the input sentence. .

Further, the bilingual expression set extracting means may be provided with a bilingual ideal translation pair extracting means for extracting an input word in the input sentence and an ideal translation in an ideal translation having a bilingual relationship.
This creates a dictionary and rules for machine translation systems,
Alternatively, it is possible to extract a pair pair serving as a base for additional change. Also, the bilingual ideal translation word pair extracting means uses the bilingual dictionaries of the first natural language and the second natural language to search the bilingual dictionary using the input word in the input sentence as a key, and has a bilingual relationship with the input word. An input word translation candidate extraction means for extracting an input word translation candidate of the second natural language can be provided. Thereby, the input word candidate and the ideal translation can be compared based on the bilingual dictionary, and the number of pairs of the input word that is the source of the matched input word translation candidate and the ideal translation having a bilingual relationship can be more accurately and more often. Can be extracted.

Further, using the input word translation candidate extracted by the input word translation candidate extraction means as a key, a second natural language dictionary is searched to compare the input word translation candidate candidate related word extracted with the ideal translation. Can be. As a result, more pairs of pairs of the input word in the input sentence and the ideal translated word in the ideal translated sentence having a bilingual relationship can be extracted. Further, the bilingual expression set extracting means may be provided with a corresponding machine translation ideal translation word pair extracting means for extracting a machine translation word in the corresponding machine translation sentence and an ideal translation word in the ideal translation sentence. This makes it possible to determine whether to add or change a dictionary or a rule by using information on a set of a machine translation word in a corresponding machine translation and an ideal translation in an ideal translation.

In the corresponding machine translation ideal translation pair extraction means , the second natural language dictionary can be used.
A second natural language dictionary is searched using a machine translation word in a machine translation key as a key to compare a machine translation related word extracted with an ideal translation word, or a second natural language using an ideal translation word in an ideal translation key as a key And the machine translation can be compared with the ideal translation related word extracted by searching the dictionary. This makes it possible to extract pairs of pairs of the machine translation words in the machine translation and the ideal translations in the ideal translation that have more correspondence.

Further, the corresponding machine translation ideal translation pair extracting means <br/> Oite, by utilizing the first natural language and a second natural language bilingual dictionary, the machine translation in the machine translation key and machine translation reverse引訳word candidate is a collection of the first natural language translation obtained Te, ideal translation reverse引訳language is a set of the first natural language translation obtained by the ideal translation in the ideal translation in key Candidates can be extracted and machine translation reverse translated word candidates can be compared. As a result, it is possible to extract more machine translation words in the corresponding machine translation and ideal translations in the ideal translation.

In the bilingual expression set extracting means , the second
When comparing the word to each other in the natural language of, machine translation and / or input word translation candidate and / or in machine translation, obtained by aggregating the input word translation candidate related words and / or machine translation related words for each different word Create a list of words that are different from the machine translation, and create a list of words that are different from the ideal translation and / or the related words that are related to the ideal translation. Different words in the word list can be compared. This allows the same translation to be repeated
It is avoided to compare.

[0044] In addition, in the bilingual representation of sets extracted detemir stage, the first
When comparing the natural language words of each other, the input word and / or the machine translation reverse translation word candidate are aggregated separately for each word, and the input word machine translation reverse translation word candidate different word list is created, and the ideal translation reverse translation word is created. Creates a list of words that differ in terms of ideal translations and reverse translations, and compares the words in the list of words that are different from those in the list of input translations. can do. This allows
Avoiding repeated comparisons of the same translation.

In the bilingual expression set extracting means , the first
When comparing natural language words and / or second natural language words, both words match, one word contains the other word, or a part of both words The degree of correspondence can be distinguished depending on whether or not. As a result, more bilingual expression sets can be extracted.

Further, the bilingual expression set extracting means may be provided with an ideal translation morphological analysis means for morphologically analyzing an ideal translation and extracting an ideal translation morpheme structure. Thus, the machine translation reverse candidate word candidate and the ideal translation reverse translation word candidate and / or the machine translation of the word of the ideal translation different word list using the information of the ideal translation morphological structure and / or the machine and / or the machine in the different word list are different. It is possible to compare the words in the word list different from the candidate reverse translation word candidate with the words in the word list different from the ideal translation reverse translation word candidate. As a result, a more accurate bilingual expression set can be extracted.

Further, the translated expression set set extracted by the translated expression set extraction unit, translated expression of extracting translated expressions combination set is translated expression of sets of combinations that can be extracted from a set of input text and the ideal translation and machine translation Combination extraction means may be provided. This makes it possible to extract a combination of bilingual expression sets that can be extracted from one set of the input sentence, the ideal translation, and the machine translation.

Further, the bilingual expression group extracting means can be provided with a bilingual expression group certainty providing means for giving a bilingual expression group certainty which is a certainty according to the extraction process of each bilingual expression group,
The bilingual expression combination extracting means assigns certainty to each bilingual expression combination using the bilingual expression set accuracy given to each bilingual expression set in the bilingual expression set set, and the bilingual expression combination changing means adds The dictionary or rule can be additionally changed using the certainty given to each of the bilingual expression combinations in the bilingual expression combination set extracted by the expression combination extracting means . As a result, it is possible to more accurately and efficiently create or add a dictionary or a rule for a machine translation system by using a bilingual expression combination having high accuracy in the bilingual expression combination set.

[0049] The dictionary and rules obtained by machine translation system dictionary rule learning device of the present invention can be repeated a learning dictionary rules for machine translation systems using. As a result, the created or added dictionary or rule can be verified, and a bilingual expression pair or rule that cannot be obtained in one learning process can be extracted.

The learning end judging means records, in the machine translation sentence storage unit, which is a history of the result of translating the input sentence by using the dictionary and rules obtained by the dictionary / rule learning apparatus for the machine translation system, and ends the learning. The learning can be terminated when the recorded content of the machine translation history storage unit obtained by the determination unit has reached a steady state. Thereby, the optimal dictionary and rules are registered.

Further, the dictionary / rule addition / change means may have a function of displaying information for changing the dictionary or the rule. As a result, the user can confirm the change of the dictionary or rule, and can create a more accurate dictionary or rule. Therefore, by using the above-mentioned means, a dictionary or a rule for a machine translation system is automatically created or additionally changed, and learning of a dictionary or a rule for a machine translation system becomes possible.

[0052]

[0053]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below.
This will be described with reference to the drawings. In this embodiment, the first natural language
Is Japanese, and the second natural language is English.FIG.
Is a machine translation dictionary / rule learning system according to an embodiment of the present invention.
It is a basic block diagram of a stem. For machine translation shown in the figure
The dictionary / rule learning system consists of a bilingual sentence reading unit 1,
Machine translation system part 2, rule 3, dictionary 4, bilingual expression group extraction
Output unit 5, bilingual expression combination extraction unit 6, dictionary / rule addition change
It is composed of a further part 7.

The bilingual sentence reading unit 1 inputs a pair (a bilingual sentence) of a first natural language sentence (input sentence) and a second natural language sentence (ideal translated sentence) having a bilingual relationship. The machine translation system unit 2 translates the input first natural language sentence into a second natural language machine translation using the rule 3 and the dictionary 4. The bilingual expression set extraction unit 5 includes a bilingual machine translation word extraction unit 51, a bilingual ideal translation word extraction unit 52, and a corresponding machine translation ideal translation word pair extraction unit 53.
A set of an input word, a machine translation word, and an ideal translation word, which is a corresponding expression group, is extracted from the input translation obtained from the machine translation sentence and the bilingual translation obtained by the machine translation system unit 2 and the ideal translation.

The bilingual machine translation word extracting unit 51 extracts a machine translation word having a bilingual relationship with the input word by using a conversion history when the machine translation system unit 2 converts it into a machine translation sentence. The bilingual ideal translation extraction unit 52 searches the bilingual dictionary using the input word as a key by using, for example, a bilingual dictionary of the first natural language and the second natural language, and finds an ideal translation having a bilingual relationship with the input word. Extract.

The corresponding machine translation ideal translation pair extraction unit 53 retrieves a machine translation related word or an ideal translation related word using, for example, a second natural language dictionary, or obtains the machine translated word as a key. By searching for a machine translation reverse translation candidate candidate that is a set of first natural language translations and an ideal translation reverse translation candidate that is a set of first natural language translations obtained using an ideal translation as a key, Extract the corresponding machine translation and ideal translation.

The bilingual expression combination extraction unit 6 is composed of a bilingual expression certainty imparting unit 61 and a bilingual expression combination creating unit 62, and is a set of sets of input words, machine translation words, and ideal translation words obtained by the bilingual expression combination extraction unit 5. Then, all combinations of pairs of input words, machine translations, and ideal translations, which have a consistent correspondence in the bilingual sentence, are extracted. Bilingual expression set accuracy giving unit 61
Assigns, to each bilingual expression set, a bilingual expression set accuracy which is a certainty according to the extraction process of each bilingual expression set.

The bilingual expression combination creating unit 62 outputs an input word, a machine translated word,
All combinations of ideal translation word pairs are created, and the accuracy is given for each combination of the translation expression pairs based on the result of the translation expression pair accuracy giving unit 61. The dictionary / rule addition / change unit 7 adds / changes a dictionary or a rule based on the bilingual expression set obtained by the bilingual expression combination extracting unit 6.

FIG . 2 shows an example of a basic block configuration of a machine translation system using a dictionary or rule which is additionally changed according to an embodiment of the present invention. The machine translation system shown in FIG. 1 includes a machine translation system unit 2, a rule 3 used in the machine translation system unit 2, a dictionary 4 used in the machine translation system unit 2, and a first natural language sentence to be translated. It comprises a natural language input unit 8 and a second natural language output unit 9 for outputting the second natural language translated by the machine translation system unit 2.

The machine translation system 2 converts a first natural language sentence into a first natural language conceptual structure, and a first natural language generated by the first natural language analyzer 21. A language conversion unit 22 that converts a conceptual structure of a language into a conceptual structure of a second natural language; and a second natural language sentence that generates a second natural language sentence based on the conceptual structure of the second natural language generated by the language conversion unit 22. It is composed of a bilingual generation unit 23.

The rule 3 includes a first language analysis rule 31 used by the first language analysis unit 21, a language conversion rule 32 used by the language conversion unit 22, and a second language generation rule 33 used by the second language generation unit 23. You. The dictionary 4 includes a first language analysis dictionary 41 used by the first language analysis unit 21 and a language conversion unit 2
2 includes a language conversion dictionary 42 used in the second language generation dictionary 43 and a second language generation dictionary 43 used in the second language generation unit 23.

FIG . 3 shows an example of entries in the language conversion dictionary according to one embodiment of the present invention. As the language conversion dictionary 42, for example, there is a noun conversion dictionary.
If English is used as the second natural language for each Japanese noun that is a natural language, and there are multiple translations for one noun, the priority indicating the priority of the translation to be used, A selection condition or the like consisting of a noun phrase indicating how the translated word is used is entered. For example, a translation of “topographic chart” is assigned to “topographic map”, and a translation of “company” and “office” are assigned to “company” in this order of priority.

FIG . 4 shows an example of an entry of a language conversion rule according to an embodiment of the present invention. The language conversion rule 32 includes, for example, a verb conversion rule.
Not only each verb in Japanese but also the particle of the verb is attached as a pattern. That is, for example, for the verb "write", ("S" (subject, noun) "ga" (particle) "O
A particle pattern of "1" (object, noun) "wo" (particle) is entered, and a search is made to search for a translation of the verb using the verb as a key including the verb. Further, a selection condition indicating the meaning of each element of the verb is described. For example, for the verb “write”, “O1” has no semantic restriction, but “S write” when “S” has a subject.
This indicates that there is a restriction to apply O1 ″.

Next, the processing procedure of the machine translation dictionary / rule learning system having the basic block configuration shown in FIG. 1 will be described. Here, the bilingual sentence shown in FIG. 5 will be described as an example.
First, the bilingual reading unit 1 inputs a pair (a bilingual sentence) of a first natural language sentence and a second natural language sentence having a bilingual relationship. This bilingual sentence is used for automatic learning of a dictionary and rules for machine translation. By selecting a set of bilingual sentences (bilingual sentences) for learning this dictionary / rule,
It is possible to tune the machine translation system for sentences in the same field where similar language phenomena frequently appear in the learned translated sentences. In the example of FIG. 5 , the first natural language Japanese sentence (input Japanese sentence) "He writes a topographic map at the company." (Ideal English translation) "He draw a toppographical map in a office."

The input Japanese sentence, which is the first natural language sentence, of the input translated sentence is converted into the second natural language sentence, a machine English translated sentence, by the machine translation system unit 2 using the rule 3 and the dictionary 4. Be translated. Figure 2 shows the procedure of this machine translation process .
A description will be given using a machine translation system having the basic block configuration of FIG. First, an input Japanese sentence, which is a first natural language sentence, is analyzed by a first language analysis unit 21 using a first language analysis rule 31 and a first language analysis dictionary 41 to generate a Japanese conceptual structure. I do. The generated Japanese conceptual structure is, for example, a language conversion dictionary 42 such as the noun conversion dictionary shown in FIG.
Alternatively, the language conversion unit 22 converts the language using a language conversion rule 32 such as the verb conversion rule shown in FIG. 4 to generate an English conceptual structure. The generated conceptual structure of English is described in the second language generation rule 33 and the second language generation dictionary 4.
3, a machine English translation is generated. The generated English sentence is output by the second natural language output unit 9. The concept structure and the dictionary information referred and generated at each stage of the machine translation system are stored so that they can be referred to in the dictionary / rule automatic learning process. For example, the information of each other translated word candidate obtained when the language conversion dictionary 42 of FIG. 3 is searched is also stored.

By using the noun conversion dictionary 42 shown in FIG. 3 and the verb conversion rule 32 shown in FIG. 4 as an example in the machine translation system as described above, the input Japanese sentence shown in FIG. a topographic chart in a company. "

Next, the bilingual expression set extraction unit 5 will be described. The bilingual expression set extraction unit 5 converts the input sentence obtained by the bilingual sentence reading unit 1, the ideal translation and the machine translation obtained by the machine translation system unit 2 into a bilingual machine translation word extraction unit 51, a bilingual ideal translation word extraction unit 52, and a corresponding machine. Translated word ideal translation word pair extraction unit 53
Thus, a set of an input word, a machine translation word having a bilingual relationship with the input word, and an ideal translation word are extracted.

The bilingual machine translation word extraction unit 51 extracts an input word and a machine translation word having a bilingual relationship with the input word by using the concept structure and dictionary information referred to and generated by the machine translation system unit 2. For example, regarding the input Japanese sentence and the machine translation sentence of FIG. 5 , by referring to the noun conversion dictionary 42 shown in FIG. 3 , "he" and "he", "topographic chart" and "topographic chart", "company" And "company" are extracted. Further, "Write" and "write" are extracted by referring to the verb conversion rule 32 and the like shown in FIG.

The bilingual ideal translation word extracting section 52 will be described with reference to the flowchart shown in FIG. 6 which is an embodiment of the present invention. FIG. 6 is a flowchart of a process for extracting a bilingual ideal translation word according to an embodiment of the present invention. Step 5201)
The bilingual ideal translation extraction unit 52 includes a bilingual machine translation extraction unit 51.
The pairs of the input word and the machine translation word are sequentially extracted from the result of the above.
In the example of FIG. 5 , "he" and "he" are extracted first,
For the sake of explanation, it is assumed that "company" and "company" have been extracted.

Step 5202) Next, from the results of the bilingual sentence reading unit 1, the ideal translation words in the ideal translation having a bilingual relationship with the input sentence are sequentially extracted. In the example of FIG. 5 , “he” is extracted first, but it is assumed that the process has proceeded to the point where “office” is extracted for convenience of explanation. (Step 5203) The conversion dictionary is searched using the input word extracted in Step 5201 as a key, and a set of input word translation word candidates is extracted. In the above example, when the conversion dictionary is searched by using the input word “company” as a key, “company” is obtained from the noun conversion dictionary 42 shown in FIG.
"And" office "are extracted.

Step 5204) In the set of input word translation candidate extracted in step 5203, step 5
It is determined whether or not there is a word that matches the ideal translation extracted in step 202, or a word in which one of the words includes the other word. As a result of the determination, if there is a word that matches the ideal translation word or one of the words includes the other word (Yes), the process proceeds to step 5205; otherwise (No), the process proceeds to step 5206. .
In the above example, since the input word translation candidate "office" matches the ideal translation "office", it is determined that the input word "company" and the ideal translation "office" have a bilingual relationship, and this determination is Yes. , To step 5205.

Step 5205) With reference to the result of the bilingual machine translation word extraction unit 51 and the degree of matching in step 5204, the input word, the machine translation word, the ideal translation word, and the correspondence method are output. In the above example, “company”, “company”, “office”, and “matching other translation candidate” shown in FIG. 7 are output.

Step 5206) It is determined whether all ideal translations have been checked for the current input word / machine translation pair. If not (No), step 52
02, if all have been inspected (Yes), 5207
Move to. In the above example, since the ideal translation “office” is the last word of the ideal translation, it is determined that all have been checked, and the process proceeds to step 5207.

Step 5207) It is determined whether or not all pairs of input words and machine translation words have been extracted and inspected. If not (No), the process proceeds to step 5201, and if all have been inspected (Yes), this process ends. In the above example, the bilingual expression set shown in FIG. 7 is finally extracted. The corresponding machine translation ideal translation extraction unit 53 will be described with reference to the flowcharts shown in FIGS . 6 and 9 , which are one embodiment of the present invention.

FIG . 8 is a flowchart (part 1) of a process for extracting a corresponding machine translation ideal translation pair according to an embodiment of the present invention. Step 53101) corresponding machine translation ideal translation pair extraction unit 53 extracts a pair of input words and machine translation in order from the results of the translation machinery translation Extraction section 51. In the example of FIG. 5 , first, "he" and "he" are extracted, but for convenience of explanation, it is assumed that the process proceeds to the point where "topographic map" and "topographic chart" are extracted.

Step 53102) Next, from the result of the bilingual sentence reading unit 1, the ideal translation words in the ideal translation having a bilingual relationship with the input sentence are sequentially extracted. In the example of FIG. 5 , first, "he"
Is extracted, but for the sake of explanation, it is assumed that "topographical" has been extracted. Step 531
03) The conversion dictionary is searched using the input word extracted in step 53101 as a key, and a set of input word translation word candidates is extracted. In the above example, when the conversion dictionary is searched using the input word “topographic map” as a key, the noun conversion dictionary 4 shown in FIG.
2, a "topographic chart" is extracted.

Step 53104) Step 5310
The second language generation dictionary 43 is searched using the machine translation word extracted in step 1, the input word translation candidate extracted in step 53103, and the ideal translation word extracted in step 53102 as keys, and a set of machine translation related words and ideal translation related words. Get the set of In the above example, since the machine translation word is “topographic chart”, the English words (topographic) and “chart” are separated, and then the English dictionary (language conversion dictionary 42) is searched, and the ideal translation word is “topographic” as the key. Dictionary (language conversion dictionary 4
2) Search. As a result, “topographic” and “topogr
Regarding Aphical ", for example, the result shown in FIG. 9 shows an example of a second language generation entry of an embodiment of the present invention.

Step 53105) Step 5310
It is checked whether there is a match between the set of machine translation related words extracted in step 4 and the set of ideal translation related words, or whether one of the words includes the other. If there is a match, or if one of the words includes the other (Yes), the process proceeds to step 53106; otherwise (No), the process proceeds to step 53107. In the above example, the noun form matches “topography” or the adverb form matches “topo” in FIG.
graphically ”, so the machine translation“ topogr
aphic "and the ideal translation word" topographical "correspond to each other.
Shift to 06.

Step 53106) By referring to the result of the bilingual machine translation word extraction unit 51 and the degree of matching in step 53105, the input word, the machine translation word, the ideal translation word, and the correspondence method are output. In the above example, "topographic map", "to
“pographic chart”, “topographical”, and “related word match” are output and added to the result of the bilingual ideal translation word extraction unit 52, as shown in FIG . FIG. 10 shows how to correspond to the bilingual expression sets obtained as a result of FIG. 8 according to one embodiment of the present invention.

Step 53107) It is determined whether all ideal translations have been checked for the current pair of the input word and the machine translation. If not (No), step 5
The flow shifts to 3101. If all the inspections have been completed (Yes), the flow shifts to step 53108. In the above example, since the word next to the ideal translated word “topographical” is “map”, the determination is No (No), and the flow shifts to step 53102.

Step 53108) It is determined whether or not all pairs of input words and machine translation words have been extracted and inspected. If not (No), the process proceeds to step 53101, and if all have been inspected (Yes), this process ends. In the above example, the bilingual expression set shown in FIG. 10 is finally extracted. Next, the corresponding machine translation ideal translation word pair extraction unit 53 will be described based on the flowchart shown in FIG. 11 which is an embodiment of the present invention.

Step 53201) The corresponding machine translated word ideal translation word pair extraction unit 53 sequentially extracts pairs of the input word and the machine translation word from the result of the parallel machine translation word extraction unit 51. In the example of FIG. 5 , “he” and “he” are first extracted, but it is assumed that the process has proceeded to the point where “topographic chart” and “topographic chart” are extracted for convenience of explanation. Step 53202) Next,
From the result of the bilingual sentence reading unit 1, the ideal translation words in the ideal translation having a bilingual relationship with the input sentence are sequentially extracted. In the example of FIG.
First, it is assumed that "he" is extracted, but for the sake of explanation, the process proceeds to the point where "map" is extracted.

Step 53203) Step 53201
The conversion dictionary (language conversion dictionary 42) is reversely searched using the machine translation word extracted in step 53202 and the ideal translation word extracted in step 53202 as a key to obtain a set of machine translation reverse translation word candidates and a set of ideal translation reverse translation word candidates. In the above example, the machine translation "topographi
The English words “topographic” and “ch
The search is performed by reversely searching the conversion dictionary (language conversion dictionary 42) using the key word “art” and the ideal translation word “map.” As a result, for example, reverse lookup is performed on “chart” and “map” as shown in FIG. Translated word candidates are obtained.

Step 53204) Step 53203
Check whether there is a word match between the set of machine translation reverse-translated word candidates extracted in step 2 and the set of ideal translation reverse- translated word candidates, or whether one of the words contains the other. . If there is a match, or if one of the words contains the other, (Ye
s) The process moves to step 53205; otherwise (No), the process moves to step 53206. In the above example, since FIG. 12 shows that the machine translation word reverse translation word candidate “figure” is included in the ideal translation reverse translation word candidate “map” or “picture”, “chart” and “map” are It is determined that they correspond, and the determination is Yes, and the flow shifts to step 53205.

Step 53205) The input word, the machine translation word, the ideal translation word, and the correspondence are output with reference to the result of the bilingual machine translation word extraction unit 51 and the degree of matching in step 53204. In the above example, the “topographic map”, “topographic chart”, “map”, and “reverse translated word partial match” are output, and the bilingual ideal translation word extraction unit 52 and the corresponding machine translation ideal translation word pair extraction processing shown in FIG. Figure added to the result of
As shown in FIG. FIG. 13 shows an embodiment of the present invention .
FIG. 12 is a diagram showing how to correspond to a bilingual expression set obtained as a result of FIG. 11 .

Step 53206) It is determined whether or not all ideal translations have been searched for the current pair of the input word and the machine translation. If not (No), step 53
The process proceeds to step 202, and if all have been inspected (Yes), the process proceeds to step 53207. In the above example, the ideal translation "map"
Since the next word is "in", the determination is (No), and the flow shifts to step 53202.

Step 53207) It is determined whether or not all pairs of input words and machine translation words have been extracted and inspected. If not (No), the process proceeds to step 53201, and if all have been inspected (Yes), this process ends. In the above example, the bilingual expression set shown in FIG. 13 is finally extracted. Next, the bilingual expression combination extraction unit 6 will be described. The bilingual expression combination extracting unit 6 assigns certainty to each bilingual expression set according to the extraction process of each bilingual expression set by the bilingual expression group accuracy giving unit 61 based on the result of the bilingual expression set extracting unit 5. The expression combination creation unit 62 creates all combinations of input words and machine translation words having a consistent correspondence in the bilingual sentence, and assigns the accuracy of each combination based on the result of the translation expression combination accuracy assignment unit 61. I do.

In the above example, certainty is given by referring to the finally obtained bilingual expression sets shown in FIG .
For example, another word candidate match is 6 points, related word match is 5 points, reverse
Accuracy is given such as 4 points for bilingual word candidate matching, 3 points for other translated word candidate partial matching, 2 points for related word partial matching, and 1 point for reverse translated word partial matching. In addition, one possible combination that can be made from FIG. 13 is “topographic map”, “topographical ca
Only pairs of "hrt", "company", "company", and "office" are created.

Next, the dictionary / rule addition changing unit 7 will be described. Based on the result of the bilingual expression combination extracting unit 6, the dictionary / rule addition changing unit 7
The dictionary 4 and the rule 3 for the machine translation system are created or additionally changed by using the bilingual expression combination having the highest accuracy.
In the above example, since the result of the bilingual expression combination extraction unit 6 is one, there is a bilingual relation with the input word "topographic map" and the bilingual relation with the input word "company" is "topographical map". Is determined to be "office". As a result, “topographical map” is added as a priority of “topographical map” as a translation of “topographic map”, and priority of “office” is changed to 1 as a translation of “company”, and the noun conversion dictionary 42 shown in FIG.
(Language conversion dictionary).

When the words in the second natural language are compared with each other and / or the words in the first natural language are compared with each other in the bilingual expression set extraction unit 5, the words to be compared differ. It is also conceivable to add a method of separately compiling and comparing and extracting a bilingual expression set. Further, a method is also conceivable in which the bilingual expression set extracting unit 5 performs a morphological analysis of the ideal translation and uses the information of the ideal morphological structure to extract a more accurate bilingual expression set.

As described above, one process of the machine translation dictionary / rule automatic learning system having the basic block configuration of FIG. 1 is completed. However, if there is a possibility that there is something that can be dealt with using the result of learning, and if the dictionary 4 or the rule 3 that has been added and changed considers all the sentences to be translated, the quality of the translation will be worse than before the additional change. There is a possibility.
In the above example, in one learning process, the machine translated word "w
There is no correspondence between "rite" and the ideal word "draw".
Therefore, it is necessary to set a learning end condition and repeat the machine translation dictionary / rule automatic learning process of FIG. 1 until the end condition is satisfied.

The repetitive processing will be described with reference to the flowchart of FIG. 15 of the embodiment of the present invention. Step 1001) The bilingual sentence reading unit 1 reads bilingual sentences in order. Step 1002) Translate by the machine translation system unit 2 and save the result.

Step 1003) Bilingual expression set extraction unit 5
To extract a bilingual expression set. Step 1004) The bilingual expression combination extracting unit 6 assigns certainty to each bilingual expression set, and creates all possible bilingual expression combination combinations in one sentence. Step 1005) By the dictionary / rule changing unit 7,
The combination with the highest probability among the results of step 1004 is saved.

Step 1006: It is determined whether all the target sentences in the translation target sentence have been read. If not (No), the process proceeds to step 1001, and if all data has been read (Yes), the process proceeds to step 1007. In the example of FIG. 5 , since there is only one sentence, step 100
Move to 7. Step 1007) The dictionary / rule addition / change unit 7 adds / changes the dictionary 4 and the rule 3 in consideration of the bilingual expression sets of all sentences. In the example of FIG.
No. conversion dictionary 42 (language conversion dictionary) shown in FIG.

Step 1008) The translation is performed again using the dictionary 4 and the rule 3 added and changed in step 1007. In the above example, when the translation is performed again using the noun conversion dictionary 42 shown in FIG. 14 , the machine translation sentence "He write a topographical map in a office" shown in FIG. 16 is obtained.

Step 1009) Compare with the stored machine translation result to check whether there is any change. If there is a change (Yes), go to Step 1001. If there is no change (No), this learning is performed. The process ends. In the above example, since there is a change as compared with the stored machine translation result, the determination here is Yes and step 10
Move to 01.

In the above example, the second step 1003
For example, if the unreachable word is one word in the machine translation and one word in the ideal translation, it is determined by the processing of the bilingual expression group extraction unit 5 that there is a correspondence. As shown in FIG. 17 , the input word “write”, the machine translation “write”, and the ideal translation “draw” as a bilingual expression set as shown in FIG.
Moreover, how to take the correspondence it is extracted as a match to the remainder. Thus, in step 1007, "draw" is added to the verb conversion rule 32 (language conversion rule) shown in FIG. 4 as a translation of "write", and becomes the verb conversion rule 32 shown in FIG . However, the conversion condition is generated from the correspondence between the machine translation word and the ideal translation word with reference to, for example, the conceptual structure stored in the machine translation system unit 2. In the example above,
The selection condition corresponding to "S" is "he" and "subject"
And "O1" corresponds to "toporographic
al map "becomes" topographic map ".

As described above, the present invention provides an input sentence which is a first natural language, a machine translation sentence which is a second natural language obtained by translating the input sentence by a machine translation system, and a second translation which has a bilingual relationship with the input sentence. An input word, a machine translation word, and an ideal translation word which are in correspondence with each other are extracted from the ideal translation sentence, which is a natural language, and the dictionary 4 and the rule 3 for machine translation can be additionally changed from the result. Therefore, in the machine translation system, when the first natural language sentence is input and translated, the dictionary 4 and the rule 3 that have been added and changed as described above are used, so that the second translation that approximates the ideal translation is performed. It is possible to output a natural language sentence.

Further, the first natural language sentence is translated by the machine translation system using the dictionary 4 or the rule 3 which has been added and changed, and when there is no change compared with the previous translation result, this learning is performed. Upon completion, the dictionary 4 and the rule 3 are registered in an optimal state. The dictionary / rule addition / change unit 7 is also provided with a display function, and presents the contents of the dictionary 4 or the rule 3 to be added / changed to the user so that the user can confirm and update the dictionary 4 or the rule 3. Is also possible.

[0100] In the above embodiment has been described based on the configuration shown in FIGS. 1 and 2, without being limited to this example, FIG. 1 and translated sentence reading unit 1 that is described in Figure 2 , translated expression set extraction unit 5 constructs a translated expression combination extraction section 6 and the dictionary rules additional changing unit 7 as a program, portable Stories such as a flexible disk or a CD-ROM
Stored in 憶 medium or stored on a medium of a disk device or the like, by processing the information output from the machine translation system unit 2 to start each program, be adapted to add change rules and dictionaries Is also possible.

The present invention is not limited to the above-described embodiment, but can be variously modified and applied within the scope of the claims.

[0102]

According to the present invention, a sentence in a first natural language (input sentence) and a sentence in a second natural language (ideal translation) that are bilingually related to each other are read, thereby providing a dictionary for a machine translation system.・ Rules can be automatically added and changed. This is a combination of bilingual input words, machine translations, and ideal translations by referring to information such as the conceptual structure generated during translation by the machine translation system and information in the dictionary used by the machine translation system. Is extracted, and the dictionary and rules for machine translation are added and changed from the result. As a result, the dictionary and rules of the machine translation system can be tuned for the translation target field.

Further, it is possible to provide a display function to the dictionary / rule addition / change step, and to present the contents of the dictionary or rule to be added / changed to the user. Thereby, it is also possible to update the dictionary and the rules after the user confirms. By repeating the dictionary / rule learning method using the dictionary / rule obtained in the dictionary / rule addition / change step, the registered dictionary / rule can be verified, and cannot be obtained in a single learning process. Dictionaries and rules can be extracted.

The learning end determination step automatically determines whether to end learning. As a result, the learning is automatically terminated without a person watching the learning result. Machine translation results are automatically saved. As a result, when the machine translation is in a steady state, the learning is automatically terminated, and the optimal dictionary and rules are registered. Therefore, by executing all of the above steps, the machine translation system dictionary and rules are automatically tuned to the translation target field, and the machine translation system dictionary and rules can be learned.

As described above, according to the present invention, instead of using only the information of the bilingual corpus, the dictionary and rules are added and changed by using the information referred to and generated by the machine translation system. By repeating until the optimal rule is registered using the changed dictionary or rule, a dictionary or rule of the machine translation system for the translation target field obtained from the bilingual corpus can be extracted.

By using the dictionary and rules registered in this way, the machine translation system can generate a translated sentence close to the ideal translation, and can use the bilingual corpus used for learning even if the translation target sentence is different. If a first natural language sentence in the same field is input, a translated sentence closer to the ideal translation can be generated.

[Brief description of the drawings]

FIG. 1 is a dictionary / rule for machine translation according to an embodiment of the present invention .
It is a basic block configuration diagram of a dynamic learning system.

FIG. 2 is a diagram illustrating a dictionary and a rule to be added and changed according to an embodiment of the present invention;
FIG. 1 is a basic block diagram of a machine translation system using a file.

FIG. 3 shows an example of an entry in a language conversion dictionary according to an embodiment of the present invention;
It is a figure showing an example.

FIG. 4 shows an entry of a language conversion rule according to an embodiment of the present invention .
It is a figure showing the example of.

FIG. 5 is a diagram showing an example of a bilingual sentence according to an embodiment of the present invention.
You.

FIG. 6 is a diagram illustrating a process for extracting a bilingual ideal translation according to an embodiment of the present invention;
It is a flowchart of a process.

FIG. 7 is a bilingual table obtained as a result of FIG. 6 according to an embodiment of the present invention;
It is a figure which shows how to take correspondence with the present group.

FIG. 8 shows a corresponding machine translation ideal translation pair according to an embodiment of the present invention .
It is a flowchart (the 1) of the process which extracts.

FIG. 9 illustrates an entry of a second language generation dictionary according to an embodiment of the present invention .
FIG. 6 is a diagram illustrating an example of a file.

FIG. 10 is a translation obtained from the result of FIG. 8 according to an embodiment of the present invention;
It is a figure which shows how to take correspondence with an expression group.

FIG. 11 extracts a corresponding machine translation word pair according to an embodiment of the present invention .
10 is a flowchart (No. 2) of the process of FIG.

FIG. 12 is a diagram illustrating a case where a language conversion dictionary according to an embodiment of the present invention is stored in a second directory
In the figure which shows the example of the entry when the reverse is done with the language as the key
is there.

FIG. 13 shows the resulting pair of FIG. 11 for one embodiment of the present invention.
It is a figure which shows how to take correspondence with a translation expression group.

14 is a diagram showing an example of the noun conversion dictionary of FIG. 3 according to an embodiment of the present invention; FIG.
Name obtained as a result of adding and changing the dictionary from the entry
It is a figure showing the example of a lyric conversion dictionary.

FIG. 15 is a diagram illustrating a process of repeatedly performing learning according to an embodiment of the present invention;
It is a flowchart of FIG.

FIG. 16 is a diagram illustrating a process of repeatedly performing learning according to an embodiment of the present invention;
It is a flowchart of FIG.

FIG. 17 shows a second example in FIG. 15 of one embodiment of the present invention.
Correspondence between bilingual expression sets obtained as a result of learning processing
FIG.

FIG. 18 shows a verb conversion rule of FIG. 4 according to an embodiment of the present invention;
Obtained as a result of adding and changing rules from the entry
FIG. 9 is a diagram showing an example of a verb conversion rule.

[Explanation of symbols]

 1 bilingual sentence reading unit 2 machine translation system unit 3 rule 4 dictionary 5 bilingual expression set extraction unit 6 corresponding expression combination extraction unit 7 dictionary / rule addition / change unit 8 first natural language input unit 9 second natural language output unit 21 1 language analyzer 22 language converter 23 second language generator 31 first language analysis rule 32 language conversion rule 33 second language generation rule 41 first language analysis dictionary 42 language conversion dictionary 43 second language generation dictionary 51 bilingual machine translation Extraction unit 52 Bilingual ideal translation word extraction unit 53 Bilingual machine translation ideal translation word pair extraction unit 61 Bilingual expression group accuracy giving unit 62 Bilingual expression combination creating unit

Continuation of the front page (56) References JP-A-7-244664 (JP, A) Zero-pronoun anaphoric analysis using semantic attributes in Japanese-to-English translation systems, Transactions of the Information Processing Society of Japan, August 15, 1993 Day, Vol. 34, no. 8, p. 1705-p. 1715 Automatic recognition of zero pronouns and their referents in a Japanese-English bilingual corpus, IPSJ Research Report 98-NL-123, January 19, 1998, Vol. 98, No. 1, p. 33-p. 40 Text-to-Sentence Correlation Using a Japanese-English Bilingual Corpus Automatic Acquisition of Zero-Pronoun Parsing Rules, Information Processing Society of Japan 55th (Early 1997) National Congress 45-p. 46 (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/28 JICST file (JOIS)

Claims (22)

(57) [Claims] 1. An input sentence of a first natural language which is a source language is
Machine translation that translates into a second natural language sentence that is the target language
The dictionaries and rules used by the systemAdditional changesMachine translation
Dictionary and rule learning for translation systemA device, An ideal translation of a second natural language in parallel with the input sentence
An ideal translation input means for inputting The input sentence input by the ideal translation input means is
Of the second natural language converted by the machine translation system
Comparing means for comparing the machine translated sentence with the input sentence, The comparing means compares the input word in the input sentence
Three sets of the machine translation and the ideal translation in the ideal translation
By extracting a set, the contents of the dictionary and the rules
Update means to add or change both or both.
Dictionary and rule learning equipment for machine translation systems
Place. 2. The input sentence having a bilingual relationship with each other and
Read the ideal translation of the second natural languageRead bilingual sentence
Means,  Comparing the input sentence, the ideal translation and the machine translation,
The input word in the input sentence and the input word having a bilingual relationship with the input word
3 of the machine translation in the machine translation and the ideal translation in the ideal translation
Extract the bilingual expression set that is the set of bilingual expression sets
ToBilingual expression set extraction means,  The bilingual expression set extractionDepartureThe set of bilingual expressions extracted in the column
Add or change the dictionary or rule based on the
Dictionary / rule addition changemeansThe machine translation according to claim 1, comprising:
Dictionary and rule learning for translation systemapparatus. 3. Extracting the bilingual expression setMeans are  When extracting the bilingual expression set, the input sentence
Conversion history when converting to the machine translation by machine translation system
The machine having a bilingual relationship with the input word in the input sentence using
Bilingual Machine Translation Pair Extraction to Extract Machine Translation Words in Translated TextmeansTo
2. A dictionary and rules science for a machine translation system according to claim 1.
Learningapparatus. 4. Extraction of the bilingual expression setMeans are  When extracting the bilingual expression set, the ideal translation having a bilingual relationship with an input word in the input sentence
Extraction of ideal translation word pairsIncluding means
A dictionary / rule learning device for a machine translation system according to claim 2.
Place. 5. Extraction of the bilingual ideal translation pairMeans are Bilingual dictionary of the first natural language and the second natural language
When, It is used when extracting the ideal translation word, and is used in the input sentence.
Using the input word as a key, the bilingual dictionary is searched, and the input unit is searched.
Input word translation candidate of the second natural language that has a bilingual relationship with the word
Input word translation word extraction means for extracting  Comparing the input word translation candidate with the ideal translation,
The input word in the input sentence and the
Input word / ideal translation bilingual extraction to extract pairs of ideal translations
Out5. The dictionary for a machine translation system according to claim 4, comprising:
-Rule learning device. 6. Extraction of the bilingual ideal translation pairMeans are A second natural language dictionary;  Extraction of the input word translation candidatemeansThe input unit extracted by
Using the word translation candidate as a key, the second natural language dictionary is
Input word translation candidate candidate related words extracted by searching and the ideal
The input in the input sentence having a bilingual relation is compared with the translated word.
Extract pairs of words and ideal translations in the ideal translationmeans
6. The dictionary and rule for a machine translation system according to claim 5, wherein
Learning device. 7. Extracting the bilingual expression setMeans are  When extracting the bilingual expression group set,
Machine translation in the machine translation and the ideal translation in the ideal translation
Extraction of corresponding machine translation ideal word pairs to be extractedClaim 2 including means
The dictionary for a machine translation system according to any one of claims 1 to 4,
Rule learning device. 8. The corresponding machine translated word ideal word pair extracting means , when extracting the machine translated word in the machine translated sentence and the ideal translated word in the ideal translated sentence, uses the dictionary of the second natural language.
8. A dictionary for a machine translation system according to claim 7, comprising means.
Learning device. 9. The corresponding machine translated word ideal word pair extracting means, when extracting a machine translated word in the machine translated sentence and an ideal translated word in the ideal translated sentence, using the second natural language dictionary, It said machine machine translation in translation to the key of claim 7, including means for comparing the ideal translation with machine translation related words are extracted by searching the second natural language dictionaries machine transliteration
Dictionary / rule learning device for translation system. 10. The corresponding machine translated word ideal word pair extracting means.
, There is provided a method for calculating the ideal translation in the ideal translation and machine translation in the machine translation, using the second natural language dictionaries, the machine translation in the machine translation as a key, said 8. The machine translation according to claim 7, further comprising means for comparing said ideal translation with a machine translation related word extracted by searching a second natural language dictionary.
Dictionary / rule learning device for translation system. 11. The corresponding machine translated word ideal word pair extraction.means
Is  The machine translation in the machine translation and the ideal translation in the ideal translation
Extracting the first natural language and the second natural language
Using a bilingual dictionary of languages, machine translation words in the machine translation sentence
Is the set of first natural language translations obtained using
Machine translation reverse translation candidate and the ideal translation in the ideal translation
A set of first natural language translations obtained as keys
Means for extracting an ideal translation reverse-translated word candidate;
By comparing, the corresponding machine in the machine translation
Extract the machine translation and the ideal translation in the ideal translationMeans
The dictionary and rules science for a machine translation system according to claim 7.
Learning device. 12. The bilingual expression set extractionMeans are  When comparing words of the second natural language,
The machine language in the translation, or the input word translation candidate, or
The input word translation word candidate related word or the machine translation word related
Machine translations aggregated separately for at least one of the words
Create different word listMeans to  Ideal translation in the ideal translation or the ideal translation related word
Ideal translations that are aggregated separately for either or both
Create list of different wordsMeans to Said Unlike the machine translation, the words in the word list,SaidDifferent ideal translation
Compare words in word listClaims 2 and 3, including means
8. The machine translation system according to any one of 4 or 7,
Dictionary / rule learning device. 13. The bilingual expression set extracting means, when comparing the words of the first natural language, sums up at least one of the input word and the machine translation reverse-translated word candidate separately for different words. and means for creating the means for creating an input word machine translation reverse引訳word candidate different word list, the ideal translation reverse引訳word candidate Unlike word list obtained by aggregating the ideal translation reverse引訳word candidates by different words, the input word Compare the words in the list of words with different machine translation reverse-translated words to the words in the list of words that are different from the ideal translation reverse-translated words.
8. The method of claim 2, 3, 4 or 7, including means for performing
A dictionary / rule learning device for a machine translation system according to claim 1.
Place. 14. Extraction of the bilingual expression setMeans are  Words of the first natural language, words of the second natural language
When comparing two words that match, or that one word contains the other
Or the degree of correspondence depends on whether or not both words match.
Distinguish9. The method of claim 2, 3, 4, or 7, including means.
Dictionaries and rules science for machine translation systems according to any one of the above.
Learning device. 15. The extraction of the bilingual expression setMeans are  Morphological analysis of the ideal translation and extraction of ideal translation morphological structure
When using the information of the ideal morphological structure,
The machine translation reverse translation word candidate, the ideal translation reverse translation word candidate,
Different from the ideal translation, at least one of the words in the word list and the
A word in the word list that is different from the machine translation, and the machine translation reverse translation word candidate
Different words in the word list, different from the ideal translation reverse candidate word
Compare at least one of the words in the word listIncluding means
A machine according to any one of claims 2, 3, 4, or 7.
Dictionary and rule learning device for machine translation system. From the translated expression set set 16. is extracted by the translated expression set extraction unit, translated expressions the input sentence and the ideal translation is a combination of translated expression set which can be extracted from a set of said machine translation 3. The system according to claim 2, further comprising a bilingual expression combination extracting means for extracting a set of combinations .
Any one of 3, 4, 7, 13, 14, or 15
Dictionary / rule learning device for machine translation system described. 17.The bilingual expression set extracting means, The bilingual expression set extracting means The bilingual extracted by
Each bilingual expression set in the expression set
Gives the accuracy of the bilingual expression group corresponding to the extraction process
Bilingual expression group accuracy assignmentHaving means,  Extraction of the bilingual expression combinationMeans are  Pairs assigned to each set of bilingual expressions in the set of bilingual expressions
Using the translation expression combination accuracy, the individual translation expression combination
AccuracyHaving means to  Dictionary / rule addition changeMeans are  Extraction of the bilingual expression combinationBy meansExtracted
Individual bilingual expression combinations in the bilingual expression combination set
The dictionary or the rule using the accuracy given to the
Add or change17. The machine translation according to claim 16, comprising means.
System dictionary / rule learning device. 18. The apparatus according to claim 1, further comprising a repeating unit for repeating learning using the dictionary and the rules obtained by the dictionary and rule learning device for the machine translation system.
The dictionary for machine translation system according to any one of claims 17 to 17
Learning device. 19. The dictionary / rule addition changeMeans are  Learning end judgment to judge learning end19. The method of claim 18, including means.
Dictionary / rule learning device for machine translation system described. 20. A dictionary / rule for the machine translation system
LearningEquipmentUsing the dictionary and the rules
Machine translation history, which is the history of the result of translating the input sentence
Machine translation history stored in storageHaving means,  The end of learningMeans are  The recorded content of the machine translation history storage unit is in a steady state.
When learning endsClaims including means
19. The dictionary / rule learning device for a machine translation system according to item 19. 21. The dictionary / rule addition / changeMeans are  Display information for changing the dictionary or the rule
display20. Machine translation according to claim 17 or claim 19 including means.
System dictionary / rule learning device. 22. An input sentence of a first natural language which is a source language
Translation that translates a sentence into a second natural language sentence that is the target language
Dictionary and rules used in translation systemAdditional changesMachine
Dictionary and rule learning for translation systemsStored program
Storage media What  An ideal translation of a second natural language in parallel with the input sentence
Enter ideal translationSteps and  Input ideal translationStepsBy the input
The second natural language converted from a sentence by the machine translation system
Comparison between the machine translation of a word and the input sentenceSteps
WhenThe comparisonStepsWith the input word in the input sentence
3 of the machine translation and the ideal translation in the ideal translation
By extracting a set of sets, the dictionary or the rule
Update that adds or changes one or both ofFrom step
A machine learning dictionary and rule learning program
The storage medium in which it was stored.