JPH07253981A - Machine translation system - Google Patents

Abstract

PURPOSE:To improve the translation capability of a machine translation system by additionally registering an example of original-translation in an original-translation data base. CONSTITUTION:First and second language character strings are preliminarily generated as a translation example (original-translation). The first language character string is inputted to a first language character string input means 3, and the second language character string obtained by translating the first language character string to the second language is inputted to a second language character string input means 4. Inputted first and second language character strings are displayed on a display means 5. A correspondence means 6 makes first and second language character strings displayed on the display means 5 correspond to each other. A register means 7 registers first and second language character strings, which are made correspond to each other, in an original-translation data base 1. At the time of translation, a translating means 2 retrieves the first language character string, which is a similar or exact equivalent expression for an original character string, from the original-translation data base and translates the original character string to the second language based on the second language character string corresponding to the retrieved first language character string.

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, and more particularly to a machine translation system capable of improving translation ability by learning translation examples.

[0002]

2. Description of the Related Art Recent developments in computer technology have created machine translation systems that perform automatic translation. Most of the machine translation systems currently in practical use employ a translation method called a syntax conversion method. An outline of the processing of the machine translation system using the syntax conversion method will be described with reference to FIG.

For example, "Development of
It is assumed that the English sentence "..." is input to the machine translation system. The input English sentence is decomposed into words, and the part of speech (noun (n), transitive verb (vt), ...) Of the decomposed word is determined (morphological analysis). Then, the structure of the input English sentence is analyzed based on a predetermined grammatical rule, and the input English sentence is decomposed into noun phrases (NP), verb phrases (VP), etc. (syntactic analysis). The syntax parsed in this way is in another language (Japanese)
After being converted to, morpheme generation produces a Japanese sentence "The development of computer science and linguistics has paved the way for machine translation."

However, since the machine translation system based on the syntax conversion system translates only by the grammatical rules,
There is a problem that an expression that is far from the grammar rules cannot be made, and an unnatural translation is output as described above. Further, in order to improve the translation quality, it is necessary to add grammatical rules, but the grammatical rules interfere with each other, which causes a problem that the translation quality is deteriorated.

Therefore, in recent years, EBMT (Example-Based Machine Translatio) for performing translation based on actual examples (translation examples)
n) type machine translation system was proposed (Nagao, MAF
ramework of a Mechanical Translation between Japan
ese and English by Analogy Principle, in ARTIFICIA
L AND HUMAN INTELLIG-ENCE (Elithorn & Banerji, Ed
s.), Elsevier Science Publishers, pp 173-180, 1984).
This EBMT-based machine translation system searches for a translation example that is most similar to the original text to be translated and translates based on this translation example. Although no specific means for realizing the machine translation system based on the EBMT method has been devised, it is considered that the processing shown in FIG. 32 is performed.

In the machine translation system based on the EBMT method, in the syntax analysis process, the original text is syntactically analyzed (NP, VP, etc.) based on past translation examples, and the syntactical conversion is performed based on the syntactic analysis result. It produces a translation of the original text.

[0007]

The conventional machine translation system described above has various problems as described below.

First, it has been difficult for the user to improve the translation ability. In the machine translation system using the syntax conversion method, it is necessary to rewrite the grammar rules and the syntax rules in order to improve the translation ability. However,
Since the grammar rules and the syntax rules are incorporated in the machine translation system as a program, the system developers are limited to those who can rewrite the grammar rules and the syntax rules. Therefore, the user cannot improve the translation quality, and it is impossible for the user to cause the machine translation system to perform a desired translation. For example, when an undesired translation result is output, even if the user corrects the translation result, the undesired translation result is repeatedly output for the same original sentence. For this reason, the user has been forced to make excessive correction work.

Although the machine translation system based on the EBMT system has not been concretely realized, it is theoretically possible to improve the translation capability by additionally registering a translation example. However, the operator must perform grammatical analysis (NP, VP) of the additionally registered translation example, which causes a problem that the operator's work load increases.

Second, there is a limit to the improvement of translation quality. In a machine translation system using a syntactic conversion method, if grammatical rules and syntactic rules are added to improve translation quality, the number of rules increases, and the rules are likely to interfere with each other. Therefore, there is a limit to the improvement of the translation quality, and conversely, the addition of the rule may reduce the translation quality.

On the other hand, in the machine translation system based on the EBMT method, although it is theoretically possible to improve the translation quality by adding a translation example, the EBMT
The concrete realization method of the machine translation system by the method has not been devised yet.

Third, it was difficult to perform natural translation. Actual sentences are often created without being restricted by grammar rules. However, in the machine translation system of the syntax conversion method, since the translation is performed only according to the grammar rules and the syntax rules, the translation result is likely to be unnatural. For example, the English sentence “De
“Veloping of ...” should be translated as “the path to machine translation has been opened by the development of computer science and linguistics”, but “of computer science and linguistics” The development has paved the way for machine translation. "

A machine translation system based on the EBMT system is
Since the translation is performed based on the past translation examples, it is possible to output a relatively natural translation. However, even in the machine translation system based on the EBMT method, since the grammar analysis is performed based on the past translation examples, there may be a case where a natural translation cannot be output for an idiomatic expression far apart from the grammar.

The present invention has been made in view of the above problems, and an object of the present invention is to solve the above problems.

[0015]

The invention according to claim 1 is
A plurality of first language character strings represented in a first language, and a plurality of second language strings obtained by translating the plurality of first language character strings into a second language.
A first language character string that is most similar to or the same as the original character string represented in the first language is searched from the parallel translation database 1 in which the language character string is registered, and the searched first language is searched. In a machine translation system including a translation means 2 for translating an original character string into a second language based on a second language character string corresponding to a character string, a first language character string input means 3 for inputting a first language character string. A second language character string input means 4 for inputting a second language character string obtained by translating the first language character string into a second language, and a display means 5 for displaying the first language character string and the second language character string. The associating means 6 for associating the first language character string and the second language character string displayed on the display means with each other, and the associated first language character string and the second language character string are registered in the parallel translation database. With registration means 7 It is a machine translation system, wherein the door.

The invention according to claim 2 comprises a first language dictionary in which words in the first language are registered, and a second language dictionary in which words in the second language are registered. If the first language character string displayed on the display means includes a word that has not been registered in the first language dictionary, or if the second language character string displayed on the display means is the second language dictionary. The machine translation system according to claim 1, wherein, when the unregistered word is included in, the unregistered word is displayed on the display means.

The invention according to claim 3 is characterized in that the registration means according to claim 1 registers the translation number in the translation database together with the associated first language character string and second language character string. Machine translation system.

The invention according to claim 4 is provided with an interactive input means for pointing the first language character string and the second language character string respectively displayed on the display means, and the associating means is the designated first language character. The machine translation system according to claim 1, wherein the strings and the second language character strings are associated with each other.

According to a fifth aspect of the present invention, the first language character string input means and the second language character string input means according to the first aspect input character strings from different files. It is a translation system.

According to a sixth aspect of the present invention, the first language character string inputting means and the second language character string inputting means according to the first aspect input the character strings from the same file. System.

[0021]

In the invention according to claim 1, the first language character string and the second language character string are created in advance as a translation example (translation example). Then, the first language character string is input to the first language input means, and the first language character string is translated into the second language.
The language character string is input to the second language character string input means. The display means displays the input first language character string and second language character string. The associating unit associates the first language character string and the second language character string displayed on the display unit with each other. Then, the registration means registers the associated first language character string and second language character string in the parallel translation database. At the time of translation, the translation means searches the bilingual database for a first language character string that is most similar to or the same as the original character string, and creates a second language character string corresponding to the searched first language character string. Based on this, the original character string is translated into the second language.

Therefore, according to the present invention, the translation capability of the machine translation system can be improved by additionally registering the bilingual example in the bilingual database. Further, since the operator can register the parallel translation examples in the database while looking at the parallel translation examples displayed on the display means, the registration work can be easily performed.

In the invention according to claim 2, when the first language character string displayed on the display means includes a word which is not registered in the first language dictionary, or when the first language character string displayed on the display means is displayed. When the bilingual character string includes unregistered words in the second language dictionary, the display means displays these unregistered words. Therefore, the operator can easily understand that there is an unregistered word in the first language character string and the second language character string to be registered for parallel translation.

In the invention according to claim 3, the registration means according to claim 1 registers the parallel translation number in the parallel translation database together with the associated first language character string and second language character string. This facilitates the search for the registered first language character string and second language character string.

In the invention according to claim 4, the dialogue input means indicates the first language character string and the second language character string respectively displayed on the display means. That is, the user
It is possible to interactively input the first language character string and the second language character string to be registered for parallel translation while observing the display means. The work of parallel translation registration can be reduced.

In the invention of claim 5, the first language character string input means and the second language character string input means according to claim 1 can input character strings from different files. Therefore, it is possible to register and print the parallel translations written in different files depending on the translator or the like.

In the invention according to claim 6, the first language character string input means and the second language character string input means according to claim 1 can input character strings from the same file. In other words, parallel translations written in the same file can be registered in the parallel translation database.

[0028]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A machine translation system according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 2 is a block diagram showing a machine translation system according to an embodiment of the present invention. The machine translation system according to the present embodiment is an EBMT (Ex.
ample-Based Machine Translation) method. This machine translation system is composed of a plurality of workstations and the like. Each workstation is connected online via a communication system 2e such as Ethernet, and a database 24 such as a dictionary to be described later can be transmitted and received between the workstations. FIG. 3 shows a functional block diagram of the workstation. This workstation is provided with a mainframe 2a composed of a processor and a memory, an input unit 2b such as a keyboard or an OCR (optical reading device), a display unit 2c such as a display, a dictionary database 24, and a document file 25.

The mainframe 2a realizes the three functions of a parallel translation data registration means 21 for performing learning registration of translation examples, a translation means 22 for translation processing, and a dictionary management means 23 for dictionary management and database transmission / reception. is there. Hereinafter, these means will be described.

(1) Parallel translation data registration means 21 The parallel translation data registration means 21 is means for registering and learning translation examples (actual examples), and comprises the following components. Parallel translation data registration unit 211 As a translation example, the parallel translation data registration unit 211 associates and registers English sentences and Japanese sentences in the document file 25 (parallel translation data registration). In this way, by registering the English sentence and the Japanese sentence in association with each other, it becomes possible to output the same Japanese sentence to the same English sentence in the subsequent translation processing.

There are the following two methods for registering parallel translation data. The first is a method of registering Japanese and English sentences in the same file, and the second is a method of registering Japanese and English sentences contained in another file.

First, the first method for registering Japanese and English sentences in the same file will be described. The Japanese-English document file 251 is written in Japanese and English in pairs, and the mainframe 2a is written in the Japanese-English document file 251.
Read the paired Japanese and English sentences from. The read Japanese and English sentences are displayed on the display unit 2b as shown in FIG.

English sentence 41 "Selective ...
-In section 6.3.1. The Japanese sentence 42 for "" has been applied according to the calculation method in the section 6.3.1. "Has already been translated by the translator. The operator, while watching this display screen,
Mainframe 2 to register 1 and Japanese 42 as a pair
Instruct a. By this instruction, the parallel translation data registration unit 21
1 is a database management unit 21 with English 41 and Japanese 42
5 is registered in the parallel translation database 246. At this time, key data used for searching the parallel translation data during translation is also registered in the parallel translation database 246. Further, when the registration is instructed, “Registration.OK? [Y / n] at sentence number 3646” is displayed for the translation. When "y" is selected, a unique sentence number 3646 automatically generated internally is added, and this number is registered.

The parallel translation data registration unit 211 registers the parallel translation data of the English sentence 41 and the Japanese sentence 42 based on the English morphological analysis dictionary 243 and the Japanese morphological analysis dictionary 244 in which the information of words and parts of speech is registered. That is, if there is a word that is not registered in the English morphological analysis dictionary 243 in the English sentence 41 to be registered for parallel translation, this word is displayed in the English unregistered word display field 43. Similarly, if there is a word that is not registered in the Japanese morphological analysis dictionary 244 in the Japanese sentence 42 to be registered for parallel translation, this word is displayed in the Japanese unregistered word display field 44. As long as unregistered words are displayed in these unregistered word display fields 43 and 44, parallel translation cannot be registered. When the operator registers an unregistered word, the unregistered word display fields 43 and 44 become blank and the parallel translation can be registered.

The second method is a method of performing parallel translation registration of English and Japanese sentences stored in another file. The English file 252 stores English, and the Japanese file 253 stores Japanese, which is a parallel translation of English, in the Japanese file 253. As described above, when the English sentence and the Japanese sentence are stored in separate files, the correspondence between the English sentence and the Japanese sentence is unknown, and therefore the following processing is required. First, the mainframe 2a reads English sentences and Japanese sentences from the English sentence file 252 and the Japanese sentence file 253, and displays these English sentences and Japanese sentences on the same screen as shown in FIG. By inputting the command 53, the operator designates an English sentence and a Japanese sentence to be registered as parallel translation data. For example, if you specify Japanese 52 for English 51,
The parallel translation data registration unit 211 registers the parallel translation data including the English sentence 51 and the Japanese sentence 52 in the parallel translation database 246.
At this time, the bilingual sentence number and the key data are also registered in the bilingual database 246.

It is not always necessary to carry out parallel translation registration for each sentence, and it may be carried out for each word or for each sentence. In addition, parallel translation may be registered by associating one sentence with one word as a whole.

Parallel Translation Learning Unit 212 The parallel translation learning unit 212 performs parallel translation learning processing of the parallel translation data registered by the parallel translation data registration unit. The bilingual learning process is a process in which the English and Japanese sentences registered in the bilingual data are decomposed into parts, and the English parts and the Japanese parts are associated with each other. By performing the parallel translation learning process in this way, it is possible to search for an English sentence that is most similar to the translation target English sentence from among the English sentences registered for parallel translation, and perform a natural Japanese translation.

FIG. 6 shows an example of a display screen during the parallel translation learning process. This display screen shows 6
It is a screen when performing parallel translation learning of 1 and Japanese sentence 62. The parallel translation learning unit 212 decomposes the English sentence 61 into word units called parts, and generates English part information 66. Similarly, the parallel translation learning unit 212 generates Japanese parts information 67 from the Japanese sentence 62. The numbers before the ":" in the English parts information 66 and the Japanese parts information 67 represent the order of the parts in the English sentence 61 and the Japanese sentence 62.

The parallel translation learning unit 212 uses the English parts information 6
6. Inference is performed based on Japanese parts information 67 and past learning examples, and English sentences 61 and Japanese sentences 62 are automatically associated (linked) in parts units. For example, in the past bilingual learning process, [Selectively plated]
It is assumed that the part called "partial plating is applied" and the part called "partially plated" are linked. That is, it is assumed that both parts are registered as parallel translations in the parallel translation parts analysis dictionary 245. In this case, the parallel translation learning unit 212 uses [1:
Selective 2: Plated parts are [18; part 19: plated 20: applied 21: end 22:
23: Things] can be automatically associated. << 1: Select in the parts link information 63
ively2: plated 18; part 19: plating is 2
0: Enforced 21: Ended 22: Existence 23: Thing >> indicates that the two have been associated.

Similarly, in the past parallel translation learning process, [m
It is assumed that the part of [inimum of] is already associated with the [lowest] part, and both parts are registered in the bilingual part analysis dictionary 245. Bilingual learning unit 212
It becomes possible to automatically link the part [minimum of] in the English sentence 61 and the part of the Japanese sentence [minimum] in the Japanese sentence 62.

However, if there is a part that has not been associated in the past learning example in the English sentence 61 and the Japanese sentence 62, << 3: in the parts link information 63.
A display such as with *** >> is made. The operator can grasp the parts that are not associated with each other by looking at this display. The operator associates the unregistered parts with each other. For example, the operator inputs the sequence number "3" of the part of [3: wth], and the corresponding [16: thickness 17:
By inputting the sequence number "16, 17" of the part of [de] (reference numeral 68), the part of [3: wth] and the part of [16: thickness 17: de] can be associated with each other. . Similarly, enter the sequence number "13, 14" of the part of [13: measured 14: as], and the sequence number "3, 4 of [3: Measurement 4: Method 5:]" corresponding to this part.
By inputting "4, 5", [13: measure
ed14: as] parts can be associated with [3: Measurement 4: Method 5:] parts. The parts of [13: measured 14: as] and the parts of [3: measurement 4: method 5:] are registered in the bilingual part analysis dictionary 245.

In this way, English 61 and Japanese 62
The part link information 63 representing the correspondence of each part is generated. The generated parts link information 63 is in English 6.
It is registered in the parallel translation database 246 while being associated with 1 and the Japanese sentence 62. The parts link information 63 is used in the translation process described later.

In parallel translation learning, information about grammatical analysis of each part is also generated in addition to the part link information. 8 and 9, the English text "In the beginning
ing God created the heave
ns and the earth "and Japanese [God created heaven and earth] as an example.

[In the beginning] includes the preposition [In]. For this reason,
[In the beginning] is made to correspond to an adverbial part [in the beginning] including a particle [to] corresponding to [In]. Also, [God] is
Since it is used as a nominative in the sentence, it is associated with [God] based on the word link information 93. [Cr
“Eated” represents a motion and is associated with the passive form “created”. In addition, [thehea
vens and the earth] is a noun, but it is not a nominative case (because [God] is a nominative case)
Corresponds to [top and bottom] representing the object. The above grammar analysis result is registered in the parallel translation database 246 together with the parts link information 96. Then, in the later-described translation process, a translated sentence is generated based on these pieces of information.
That is, the nominative part is placed at the nominative position of the translated sentence, and the action part is placed at the predicate position of the translated sentence.

Registration Test Unit 213 The registration test unit 213 is for managing (maintaining) the parallel translation data registered on the parallel translation database 246. That is, the registration test unit 213 has functions such as displaying, deleting, adding, and searching the parallel translation data registered in the parallel translation database 246.

Learning Test Unit 214 The learning test unit 214 is for managing the display, deletion, addition, retrieval, etc. of the part link information registered in the parallel translation database 246. The display screen at the time of the learning test is shown in FIG. This display screen shows English morpheme 9
1, Japanese morpheme 92, word link information 93, English parts information 94, Japanese parts information 95, and parts link information 96 are shown. These pieces of information 91 to 96 are about the English sentence "In the ... the earth" and the Japanese sentence "Initially ... Created."

English morpheme 91 and Japanese morpheme 92
Is composed of information such as words and parts of speech. The English sentence "In the beginning God" shown in FIG.
created the heavens and t
"he earth" means [In preposition] [the article] [beginning noun] [God noun]
[Created transitive verb] [the article] [hea
vens noun] [and conjunction] [the article]
It can be decomposed into English morphemes in which words and parts of speech are paired, such as [earth noun]. Similarly, the Japanese sentence "God first created heaven and earth."
It can be decomposed into Japanese morphemes such as words and parts of speech, such as [,] [God: noun] [ha] [top and bottom: noun] [o] [creativity: name sa] [re] [ta]. is there.

The word link information 93 is the English-Japanese dictionary 241 out of the words and phrases in the English morphemes 91 and the Japanese morphemes 92.
And the same words as those registered in the Japanese-English dictionary 242. That is, the word link information 93
The phrase displayed in is, without using past learning examples,
This means that correspondence (linking) is possible only with the dictionaries 241 and 242. However, when the same parts as these words / phrases are already registered in the bilingual part analysis dictionary 245, they are associated with each other in the bilingual learning of these words / phrases based on the parts link information 93.

In the word link information 93, the [4: transitive verb created: 7] is the 4th (starting from 0th) morpheme [created transitive verb] of the array of English morphemes 91 is the 7th (0th) of the Japanese morpheme 92. It corresponds to the morpheme [creativity: name sa] of

English parts information 94, Japanese parts information 9
5. The parts link information 96 is the same as that described above. That is, the English parts information 94 and the Japanese parts information 95 represent words and phrases obtained by decomposing English sentences and Japanese sentences based on past learning examples and the like, and the parts link information 96 represents correspondence between the decomposed phrases. .

Database Management Unit 215 The database management unit 215 uses the parallel translation database 246.
Is for controlling access to various data with respect to. Therefore, all the data access to the parallel translation database 246 at the time of parallel translation data registration, parallel translation learning, and translation is performed via the database management unit 215. However, at the time of transmitting / receiving a database, which will be described later, data access to the parallel translation database 246 is performed by the dictionary database transmitting / receiving unit 236.

(2) Translating Means 22 The translating means 22 includes an English-Japanese translation section 221, which performs English-Japanese translation, a Japanese-English translation section 222, which performs Japanese-English translation, and an English-Japanese translation printing section, which prints out the English-Japanese translation result. 223, and a Japanese-English translation printing unit 224 that prints out the Japanese-English translation result.

English-Japanese translation unit 221 and Japanese-English translation unit 222 The English-Japanese translation unit 221 performs a process of translating an English sentence into a Japanese sentence based on the parts link information and key data registered in the parallel translation database 246. Further, the Japanese-English translation unit 222 performs a process of translating a Japanese sentence into an English sentence.

FIG. 7 shows an outline of Japanese-English translation processing. For example, when the original sentence 71 is given to the Japanese-English translation unit 222, the Japanese-English translation unit 222 searches the parallel translation data registered in the parallel translation database 246 for the Japanese sentence (translation number 3646) 73 most similar to the original sentence 71. The parallel translation (English translation) of the Japanese sentence 73 retrieved in this way is registered as the English sentence 72. The Japanese-English translation unit 222 reads the parts link information 74 of the parallel translation sentence number 3646 from the parallel translation database 246, and compares each part in the Japanese sentence in the parts link information 74 with each part in the original sentence 71.

The result of the comparison is illustrated in list 75. The display of <***> in the list 75 indicates that the two parts do not match. Of the parts link information 74, "1
Since the parts "50", "By the measuring method", "No.", "6.3.1" do not match the parts in the original sentence 71, these parts are marked with <****> . In this case, the parts to which <***> is added are not necessary in the original sentence, and are deleted, and the numbers are replaced and output as the translation result.

The search for a character string most similar to the original sentence (a character string having a high degree of similarity) is performed by the following procedure. For example, suppose that the original text [Progress] is given. The independent words in this text are [Progress] and [Writing]. Then, from the parallel translation database 246, an independent word [advancement]
Search for a character string that includes and [author]. Here, the following 3
Suppose one string is searched.

Character string 1: [Major progress in machine translation] Character string 2: [Remarkably advanced technology] Character string 3: [Technological progress and product price]

Next, these character strings are classified into the following sets. The similarity of the set A is the highest, and
It is assumed that the degree of similarity decreases in the order of set C.

Set A: Original text appearing in a character string. Set B: A character string that includes all the constituent words of the original sentence. Set C: A character string that includes a part of the entire machine language of the original text.

Sentence 1 corresponds to set A, and set B
The sentence 2 corresponds to. The sentence 3 corresponds to the set C. That is, sentence 1 has the highest similarity to the original sentence. When a plurality of character strings are classified into the same set, the similarity is further determined by the following method.

A character string in which a plurality of constituent words of the original sentence are continuously included without interposing other words is set as a character string having the highest similarity. Next, a character string having a high ratio of the number of constituent words of the original sentence to the number of independent words of the character string is set as a character string having the second highest similarity. Furthermore, the character string that includes more constituent words of the original sentence is set as the character string having the third highest similarity. The method of determining the degree of similarity is not limited to the method described above, and the degree of similarity may be determined using another method. Here, by the similarity calculation, the sentence numbers of a plurality of bilingual translations extracted from the bilingual database are arranged in order of similarity, and comparison is performed from the top.However, even if the bilingual translation is read out while calculating the similarity each time. Good.

English-Japanese translation stamp section 223, Japanese-English translation stamp section 224 The English-Japanese translation stamp section 223, Japanese-English translation stamp section 224 prints translated Japanese or English text data in a predetermined format. It is for outputting to 2d and printing.
An example of printing by the English-Japanese translation stamp section 223 is shown in FIG. 10, and an example of printing by the Japanese-English translation stamp section 224 is shown in FIG. In these figures, the sentence in the left column represents the original sentence, and the sentence in the right column represents the translated sentence. If the translation result is insufficient, the translation means 22
If is judged, the mark of "◇" is displayed at the beginning of the translated text.

(3) Dictionary management unit 23 The dictionary management unit 23 manages the English-Japanese dictionary management unit 231, the Japanese-English dictionary management unit 232, the English morphological analysis dictionary management unit 233, the Japanese morphological analysis dictionary management unit 234, and the bilingual part analysis dictionary management. It is composed of a unit 235 and a dictionary database transmitting / receiving unit 236.

The English-Japanese dictionary management unit 231 and the Japanese-English dictionary management unit 232 manage the English-Japanese dictionary 241 and the Japanese-English dictionary 242. That is, registration and deletion of words in the Japanese-English dictionary 241, English-Japanese dictionary 242 are performed by the English-Japanese dictionary management unit 231 and the Japanese-English dictionary management unit 232.

The English morpheme analysis dictionary management unit 233, the Japanese morpheme analysis dictionary management unit 234, the English morpheme analysis dictionary management unit 233, and the Japanese morpheme analysis dictionary management unit 234. Management (registration, display, deletion)
Is to do.

The bilingual part analysis dictionary management unit 235 manages (registers, displays, and deletes) the bilingual part analysis dictionary 245 in which the learned bilingual parts are registered.

Dictionary database transmitting / receiving unit 236 The dictionary database transmitting / receiving unit 236 transmits / receives various information in the dictionary database 24 to / from other workstations via the communication system 2e. 12
A conceptual diagram of transmission / reception of a dictionary database is shown in FIG. Items to be transmitted and received are various information in the dictionary database 24, that is, the English-Japanese dictionary 241, the Japanese-English dictionary 242, the English morphological analysis dictionary 243, the Japanese morphological analysis dictionary 244,
This is information on the bilingual part analysis dictionary 245 and the bilingual database 246. For example, various dictionaries and parallel translation databases transmitted from the workstation 12a to the workstation 12b are added to the dictionary database 24 on the workstation 12b side. Of the received information, only the information that is not registered in its own dictionary database 24 is written in the dictionary database 24. As a result, the writing operation to the dictionary database 24 can be performed in a short time.

Various dictionaries and parallel translation databases can be transmitted from the workstation 12b to the workstation 12a, or can be transmitted / received to / from other workstations not shown in the figure. . As a result, the dictionary database 24 of a plurality of workstations can be shared, and the efficiency of parallel translation learning can be improved.

(4) Dictionary database 24 As described above, the dictionary database 24 is the English-Japanese dictionary 24.
1, Japanese-English dictionary 242, English morphological analysis dictionary 243, Japanese morphological analysis dictionary 244, parallel translation parts analysis dictionary 245,
It is composed of a parallel translation database 246. The translated parts analysis dictionary 245 stores the learned translated parts. The bilingual database 246 stores bilingual data, English morpheme information, Japanese morpheme information, part link information indicating correspondence between bilingual parts, and key data for searching during translation.

Next, the operation of the machine translation system according to this embodiment will be described with reference to the flowchart.

(Main Flowchart) FIG. 15 is a main flowchart showing the operation of the machine translation system according to this embodiment. First, the operator selects a desired process from the menu screen displayed on the display unit 2c. When the operator selects the parallel translation data registration learning process (YES in S1), the parallel translation data registration means 2
1 executes the processing of registration learning of parallel translation data (S2).
When the parallel learning data registration learning process is not selected (NO in S1), the mainframe 2a determines whether the translation process is selected (S3). When the translation process is selected (YES in S3), the translation unit 22 executes the translation process (S4). When the translation process is not selected (NO in S3), the mainframe 2a determines whether or not the dictionary management process is selected (S5). When the dictionary management process is selected (YES in S5), the dictionary management unit 23 executes the dictionary management process (S6). If the dictionary management process is not selected (NO in S5), the process of the main flowchart ends.

(Learn parallel language data registration) FIG.
It is a flowchart showing the registration learning process (S2) of the parallel translation data in 5. The operator selects one of parallel translation data registration, parallel translation learning, a registration test, and a learning test while looking at the menu displayed on the display unit 2c. When the operator selects the parallel translation data registration process (S10)
If YES in 1), the parallel translation data registration unit 211 performs parallel translation data registration using translated English and Japanese sentences (S1).
02). The parallel translation data generated by this processing is stored in the parallel translation database 246. In the later translation process, if the same sentence (phrase) as the parallel translation data is present in the sentence to be translated, it is possible to output the same translated sentence as the parallel translation data.

If the parallel translation data registration process is not selected (NO in S101), the mainframe 2a determines whether or not the parallel translation learning process is selected (S10).
3). When the bilingual learning process is selected (YE in S103)
In S), the parallel translation learning unit 212 executes a parallel translation learning process (S104). That is, the parallel translation data is decomposed into parts, and the English parts and the Japanese parts are associated with each other. As a result, in a later translation process, it becomes possible to perform a natural translation on a sentence similar to the parallel translation data.

When the parallel translation learning process is not selected (NO in S103), the mainframe 2a determines whether or not the registration test process is selected (S10).
5). When the registration test process is selected (YES in S105), the registration test unit 213 executes the registration test process (S106). That is, the registration test unit 21
3 manages (maintains) registered bilingual data such as displaying, deleting, adding, and searching.

When the registration test process is not selected (NO in S105), the mainframe 2a determines whether or not the learning test process is selected (S10).
6). When the learning test process is selected (YES in S107), the learning test unit 214 executes the learning test process (S108). That is, the learning test unit 21
4 manages display, deletion, addition, search, etc. of registered part link information and the like. When the above process ends, the process returns to the main flowchart of FIG.

(Translation) FIG. 17 is a flowchart showing the translation process (S4) in FIG. The operator
English-Japanese translation while looking at the menu displayed on the display 2c,
Select one of Japanese-English translation processing. When the operator selects the English-Japanese translation processing (YES in S201), the English-Japanese translation unit 221 performs English-Japanese translation based on the bilingual translation data, the part link information and the like (S202). When the translation is completed, the operator determines whether to print the translation result (S203). When the operator determines that the print is to be made (YES in S203), the English-Japanese translation stamp section 223
Outputs the translation result to the printer 2d. As a result,
Printing is performed in the format shown in 0. On the other hand, when the operator determines not to print (NO in S203)
, The process returns to the main flow chart of FIG. 15 without printing.

When the English-Japanese translation process is not selected (NO in S201), the mainframe 2a determines whether the Japanese-English translation process is selected (S205).
When Japanese-English translation processing is selected (YES in S205)
First, the Japanese-English translation unit 222 performs Japanese-English translation based on the parallel translation data, the parts link information, and the like (S206). When the translation is completed, the operator determines whether to print the translation result (S207). If the operator determines that the print is to be made (YES in S207), the Japanese-English translated stamp section 22
4 outputs the translation result to the printer 2d. As a result, printing is performed in the format as shown in FIG. On the other hand, when the operator determines that the printing is not performed (N in S207,
In step O), the process returns to the main flow chart of FIG. 15 without printing.

(Dictionary Management) FIG. 18 is a flowchart showing the dictionary management processing (S6) in FIG. While looking at the menu displayed on the display unit 2c, the operator performs any of the processes of English-Japanese dictionary management, Japanese-English dictionary management, English morphological analysis dictionary management, Japanese morphological analysis dictionary management, parallel parts analysis dictionary management, and dictionary database transmission / reception. select. When the operator selects the processing of managing the English-Japanese dictionary (S30
If YES, the English-Japanese dictionary management unit 231 determines that the English-Japanese dictionary 2
41 is registered, displayed, deleted, and searched (S302).
When the processing for managing the English-Japanese dictionary is not selected (S301
If the answer is NO, the mainframe 2a determines whether the Japanese-English dictionary management process has been selected (S303). When the Japanese-English dictionary management process is selected (YES in S303), the Japanese-English dictionary management unit 232 causes the Japanese-English dictionary 242 to be selected.
Registration, display, deletion, search, etc. are performed (S304).

When the Japanese-English dictionary management process is not selected (NO in S303), the mainframe 2a determines whether the English morphological analysis dictionary management process is selected (S305). When the processing of managing the English morphological analysis dictionary is selected (YES in S305), the English morphological analysis dictionary management unit 233 registers the English morphological analysis dictionary 243,
Display, deletion, search, etc. are performed (S306). Similarly, when the processing of managing the Japanese morphological analysis dictionary is selected (S30
7 is YES), the Japanese morphological analysis dictionary management unit 234
Performs registration, display, deletion, search, etc. of the Japanese morphological analysis dictionary 244 (S308). Further, when the process of managing the bilingual parts analysis dictionary is selected (YES in S309),
The bilingual part analysis dictionary management unit 235 executes processes such as registration, display, deletion, and search of the bilingual part analysis dictionary 245 (S310).

When the dictionary database transmission / reception process is selected (YES in S311), the dictionary database transmission / reception unit 236 transmits / receives various dictionaries and bilingual databases (S312). That is, the dictionary database transmitting / receiving unit 236 stores various information on the English-Japanese dictionary 241, the Japanese-English dictionary 242, the English morphological analysis dictionary 243, the Japanese morphological analysis dictionary 244, the bilingual part analysis dictionary 245, and the bilingual database 246 as shown in FIG. Send to another workstation 12. This makes it possible to share the dictionary database of each workstation.

(Details of parallel translation data registration) FIGS. 20 and 21.
FIG. 16 is a flowchart showing details of parallel translation data registration (S102 in FIG. 16).

When the English sentence and the Japanese sentence to be registered as the parallel translation data exist in the same file (YES in S401)
The operator inputs a Japanese-English document file name having English and Japanese sentences (S402). Mainframe 2
a reads the English-Japanese document file 251 (S403),
A pair of English and Japanese sentences are displayed on the display unit 2 as shown in FIG.
It is displayed in c (S404). Parallel data registration unit 211
Judges whether or not there is a word not registered in the English morphological analysis dictionary 243 in the displayed English sentences (S40).
5). Similarly, the parallel translation data registration unit 211 determines whether or not the displayed Japanese sentence includes a word that is not registered in the Japanese morphological analysis dictionary 244 (S406). If there is an unregistered word (YES in S407), the unregistered word is displayed in the English unregistered word display column 43 and the Japanese unregistered word display column 44 (FIG. 4) (S408).

Thereafter, a predetermined command is input (S40
9), it is determined whether or not to register parallel translation data for other English and Japanese sentences (S410). When parallel translation data registration of other English and Japanese sentences is performed (YES in S410), a new English sentence and Japanese sentence are selected (S422), and the processes of S404 and thereafter are executed. On the other hand, when it is not necessary to change to another English sentence or Japanese sentence (NO in S410), the operator determines whether to further register the displayed English sentence and Japanese sentence as parallel translation data (S411). When parallel translation data registration is performed (YES in S411), the processing of S412 and subsequent steps is executed, and when parallel translation data registration is not performed (S411).
If NO), the processing from S417 is executed.

In S412, if there are unregistered words in English and Japanese (YES in S412), S4 is entered.
Returning to 08, the unregistered word is displayed. On the other hand, when there is no unregistered word in English and Japanese (NO in S412), a translation number, which is a unique number of English and Japanese, is generated and the translation number is added to the English and Japanese (S41).
3). Then, the parallel translation data registration unit 211 generates key data for searching the parallel translation data of English and Japanese (S414), and registers the English and Japanese sentences as parallel translation data in the parallel translation database 246 together with this key data (S415). After that, the parallel translation data registration unit 211 prepares other English and Japanese sentences in the Japanese-English document file 251 (S418) and displays them on the display unit 2c (S404). Hereinafter, the processing from S405 is repeated.

On the other hand, when the registration is not performed in S411, the operator determines whether or not to register the English unregistered word displayed in the English unregistered word display field 43 (S).
417). When it is determined that the English unregistered word is to be registered (YES in S417), the English unregistered word is registered in the English morphological analysis dictionary 243 (S418), and the process returns to S404. If it is determined not to register (NO in S417), it is further determined whether or not to register an unregistered word in Japanese (S419). If it is determined to register the Japanese unregistered word (YES in S419), the Japanese unregistered word is registered in the Japanese morphological analysis dictionary 244 (S420).
It returns to S404. When it is determined not to register (S419
If NO, further, it is determined whether or not the parallel translation data registration processing is to be ended (S421). If it is determined that the parallel translation data registration processing is not to be ended (NO in S421), S
When it is determined that the processes of 404 and thereafter are executed and the process of parallel translation data registration is completed (YES in S421), the process of parallel translation data registration is completed.

The processing after S431 is performed when the English and Japanese sentences to be registered for the parallel translation data are in different files (S
This is the process of NO in 401). The operator inputs an English file name and a Japanese file name from the input unit 2b to the mainframe 2a (S431, S432). Then, the mainframe 2a reads the English file 252 and the Japanese file 253 from the document file 25 (S433, S434). Read English file 2
The English sentence in 52 and the Japanese sentence in the Japanese sentence file 253 are displayed on the display unit 2c (S435).

For example, as shown in FIG. 5, the English sentence 51 "Selective ... In
section 6.3.1. , And the Japanese sentence 52 “6.
According to the calculation method in section 3.1 ... Are displayed as a pair. When the operator determines that the English and Japanese sentences thus paired are directly registered as the bilingual data (YES in S436), S43.
The processing after 7 is executed. On the other hand, if the operator determines that the English and Japanese sentences should not be registered as they are (S4
If NO in S36, the processes in and after S445 are executed.

In S437, the parallel translation data registration unit 21
1 checks whether or not there is a word in the English sentence 51 that is not registered in the English morphological analysis dictionary 243. Similarly, the parallel translation data registration unit 211 checks whether or not the Japanese sentence 52 has a word that is not registered in the Japanese morphological analysis dictionary 244 (S438). When there are unregistered words in the English sentence 51 and the Japanese sentence 52 (YES in S439), the parallel translation data registration unit 211 displays the unregistered word on the display unit 2b (S4).
40). Based on this display, the operator determines whether to register an unregistered word in English (S441). When it is determined to register an unregistered word in English (YES in S441)
In English, an unregistered word in English is registered in the English morphological analysis dictionary 243 (S442), and the process returns to S435. If it is determined not to register (NO in S441), the operator further determines whether to register an unregistered word in Japanese (S44).
3). If it is determined to register an unregistered word in Japanese (S
If YES in 443), the Japanese unregistered word is registered in the Japanese morphological analysis dictionary 244 (S444), and the process returns to S435. If it is determined not to register (NO in S443), the process returns to S435 without registering an unregistered word.

In S439, if the parallel translation data registration unit 221 determines that there is no unregistered word, the processing from S457 is executed. That is, the parallel translation data registration unit 221
Generates a bilingual translation number, which is a unique number for English and Japanese, and adds the bilingual translation number to the English and Japanese (S45).
7). Then, the parallel translation data registration unit 211 generates key data for searching the parallel translation data of English and Japanese (S458), and registers the English and Japanese sentences as parallel translation data in the parallel translation database 246 together with this key data (S459). After that, the operator selects another displayed English sentence or Japanese sentence (S460), and the processing from S435 is executed.

In S436, when the operator determines that the parallel translation of the displayed English sentence and Japanese sentence is not registered (NO in S436), another English sentence is selected by inputting the command 53 (S445, S446). Similarly, another Japanese sentence is selected by inputting the command 53 (S447, S448). When the pair of English sentence and Japanese sentence is not changed (NO in S445, NO in S447), the processes of S449 and thereafter are executed.

In S449, the operator determines whether to combine the English sentence 51 with another English sentence (S449).
When combining with another English sentence (YES in S449), the English sentence 51 and the English sentence 54 following the English sentence 51 are combined (S45).
0), the processing from S435 is executed. When the English sentence 51 is divided (YES in S451), the English sentence 51 is divided at the designated location (S452), and the processing of S435 and thereafter is executed.

The same processing is performed for Japanese sentences.
That is, when the operator determines to combine the Japanese sentence 52 and the Japanese sentence 55 following the Japanese sentence 52 (YE in S453).
In S), the Japanese sentence 52 and the Japanese sentence 55 are combined (S45).
4), the processing from S435 is executed. When the Japanese sentence 52 is divided (YES in S455), the Japanese sentence 52 is divided at the designated portion (S456), and the processes of S435 and thereafter are executed. Therefore, desired parts of the English sentence 51 and the Japanese sentence 52 can be registered in parallel translation, or a plurality of English sentences and a plurality of Japanese sentences can be combined and registered in parallel translation. When the English sentence 51 and the Japanese sentence 52 are not combined and divided (N in S449, S451, S453, S455)
In O), the parallel data registration process ends.

(Details of Parallel Translation Data Learning) FIG. 22 is a flowchart showing the details of parallel translation learning (S104 in FIG. 16). The operator inputs the bilingual sentence number to the mainframe 2a via the input unit 2b (S501). This bilingual sentence number is a unique number added to the sentence registered in the bilingual data. For example, as shown in FIG.
If 46 is input, the main frame 2a displays the English sentence 61 and the Japanese sentence 62 on the display unit 2c.

The bilingual learning unit 212 uses English 61 and Japanese 6 based on past learning examples (bilingual parts analysis dictionary 245).
2 is automatically associated with each part, and the part link information 63 is generated. That is, when the same part as the bilingual part registered in the bilingual part analysis dictionary 245 exists in the English sentence 61 and the Japanese sentence 62, the part link information 63 is generated based on the registered bilingual part. English 6
If there is an unregistered part in the bilingual parts analysis dictionary 245 in 1 and the Japanese sentence 62, the parts link information 63 is generated by using the words registered in the English-Japanese dictionary 241 and the Japanese-English dictionary 242 (S502). .

The generated parts link information 63 is displayed on the display section 2c (S503). Parts link information 63
If the operator determines that the is correctly generated (S5
(YES in 04), enter “y” in the display field 64. As a result, the parallel translation learning unit 212 automatically additionally registers the generated parts link information 63 in the parallel parts analysis dictionary 245 in the parallel translation database 24. (S511),
The bilingual learning process ends.

On the other hand, when the operator determines that the parts link information 63 is not correctly generated (NO in S504), the processes of S505 and thereafter are executed. S50
5, when the operator determines that the parts link list is displayed (YES in S505), the display column 6
5. Input "y" in 5. By this, a list of parts link information, English parts information, and Japanese parts information is output from the printer 2d. For example, English sentence 61 and Japanese sentence 62 are associated with each other in past learning examples. If there is a part that has not been checked, the parts link information 63
In, a display such as <<< 3: with *** >>> is made. By looking at this display, the operator can grasp the parts that have not been associated.

When the parts link list is not output (S
Even in the case of NO in 505), the English parts information 66 and the Japanese parts information 67 are displayed on the display unit 2c (S5).
07, S508). The operator grasps the correction items while looking at the displayed or printed list, and instructs the parallel translation learning unit 212 how to associate the English parts with the Japanese parts (S509). In the display example of FIG. 6, [3: with], [13: mea
[sured 14: as] is not associated.
Therefore, the operator inputs the sequence number "3" of the part of [3: with], and the sequence number "16, 17" of the part of [16: thickness 17:] corresponding to this part (reference numeral 68). Similarly, [13: measured 1
Enter the sequence number "13, 14" of the part of [4: as],
The sequence number "3, 4, 5" of [3: measurement 4: method 5:] corresponding to this part is input (reference numeral 69).

In this way, the process of S509 is repeatedly executed until the modification of the parts link information 63 is completed (YES in S510). After the correction of the displayed parts link information 63 is completed, the processes after S503 are repeated to display and register the corrected parts link information.

(Details of English-Japanese Translation) FIG. 23 is a flowchart showing the details of English-Japanese translation (S202 of FIG. 17). The operator inputs the English file name to be translated into the mainframe 2a via the input unit 2b (S6).
01). The main frame 2a reads the English text file into the work area of the internal memory (S602), and displays the read English text file on the display unit 2c (S603).

The operator inputs a command into the mainframe 2a (S604) and instructs the English-Japanese translation section 221 whether to translate each sentence or batch translation.
When the original sentence is translated for each sentence (YES in S605), the sentence is translated by the process of S606, and then S6 is performed.
The processing from 04 is repeated.

On the other hand, in the case of translating the original sentence in a batch (YES in S607), the operator starts the translation (translation start sentence number) and the translation end sentence number (translation end sentence number). ) And are sent to the English-Japanese translation unit 221 (S608). The English-Japanese translation unit 221 first translates one sentence of the translation start sentence number into Japanese (S609). Then, the English-Japanese translation unit 221 determines whether or not the translation of the English sentence with the translation end sentence number is completed (S610), and if it is completed (S61).
If YES, the process from S604 is repeated. When the translation of the English sentence with the translation end sentence number is not completed (NO in S610), the English sentence of the second sentence is displayed (S
611) and the processing from S609 are repeatedly executed. In this way, a plurality of English sentences from the translation start sentence number to the translation end sentence number can be collectively translated into a Japanese sentence.

When the operator determines in S607 that the batch translation is not performed (S607), it is determined whether or not the translation result is printed (S612). When printing the translation result (YES in S612), the operator instructs the English-Japanese translation stamp section 223 of the print start sentence number and the print end sentence number. The English-Japanese translation printing unit 223 prints the parallel translation (English and Japanese) of the first sentence of the printing start sentence number.
Then, the printer 2d prints the parallel translation of the first sentence in a predetermined format (S614). The English-Japanese translation stamp unit 223 continues printing until the printing of the parallel translation of the print end sentence number is completed (YES in S615). When the printing of the parallel translation of the print end sentence number is completed (YES in S615), the processing of S604 and thereafter is repeatedly executed.

When the operator determines in S612 that the printing is not performed (NO in S612), it is determined whether or not the translated sentence is output to the file (S617). When the operator determines that the translated text is to be output to the file (YES in S617), the translated text is output to the file,
Returning to the flowchart of FIG.

FIG. 24 is a flow chart showing the above-described one-sentence translation processing in S606 and S609. First, the English-Japanese translation unit 221 generates key data of an original sentence given as a translation target (S651). The key data is data for searching the parallel translation database 246. English-Japanese translation department 2
21 is a parallel translation database 246 according to this key data.
Of the above bilingual data, the bilingual data having an English sentence most similar to the original sentence is searched (S652), and this bilingual data is read (S653).

For example, as shown in FIG. 13, [Wh
en I fell into the river
he came to my aid. ] 1
Suppose 31 is given. The English-Japanese translation unit 221 searches the parallel translation database 246 for the parallel translation data 132 including the English sentence most similar to the original sentence 131. In other words, the English sentence [When it came to voting,
parallel translation data 132 including “he abtained”
Will be searched.

The English-Japanese translation unit 221 checks the parts of the parallel translation data 132 that match the parts of the original sentence 131. That is, it is determined whether or not there is a part that can be used for translating the original sentence 131 among the parts of the parallel translation data 132 (S654). As a result of the judgment, if usable parts exist in the parallel translation data 132 (YES in S654).
In S), this part is loaded into the work area on the internal memory (S655). Of the parts of the parallel translation data 132,
Available parts are [When] and [he].

Next, the English-Japanese translation section 221 determines whether or not all the parts in the original sentence 131 are prepared (S65).
7). At this time, among the parts of the original sentence 131, [If
ellinto my river] [came] [t
o my aid] parts are not available (not translated) (NO in S657), English-Japanese translation section 22
1 reads out from the bilingual translation database 246 the bilingual translation data 133 having an English sentence second most similar to the original sentence 131 (S658). In the parts link information of the bilingual data 133, two parts [came] [to my aid] match the parts of the original sentence 131 (S65).
4). Therefore, the English-Japanese translation unit 221 takes these two parts into the work area (S655), and further determines whether the parts of the original sentence 131 are complete (S65).
7). At this time, of the parts of the original sentence 131, [I fe
[ll into my aid] is not available (NO in S657), the English-Japanese translation unit 221 determines that the bilingual translation data 13 having an English sentence third similar to the original sentence 131.
4 is read (S658).

Of the parts link information of the parallel translation data 134 (FIG. 15), only [I] matches the parts of the original sentence 131, and [fell into my ai of the original sentence 131 is displayed.
The parts of d] have not been searched for a match yet.
Further, the English-Japanese translation unit 221 reads the parallel translation data 135 having an English sentence similar to the fourth sentence of the original sentence 131, and searches for usable parts. As a result, all parts of the original sentence 131 are gathered, and Japanese parts 136 corresponding to the English parts of the original sentence 131 are generated. English-Japanese translation department 221
Generates a translated sentence 137 by rearranging the Japanese parts 136 (S659, S660).

On the other hand, in S654, there is no usable part in the part link information about the parallel translation data read (NO in S654), and the original sentence 131
If there is no parallel translation data having an English sentence similar to (YES in S656), the Japanese translation is performed based on the English-Japanese dictionary 241 (S661). That is, the English-Japanese translation section 221
Reads, from the English-Japanese dictionary 241, a Japanese translation of a part (word) whose parallel translation has not been searched among the parts of the original sentence 131, and completes the translated sentence of the original sentence 131. By the above processing,
When the translated sentence of the original sentence 131 is completed, the process returns to the flowchart of FIG.

(Details of Japanese-English Translation) FIG. 25 is a flowchart showing the details of Japanese-English translation (S206 in FIG. 17). The processing of Japanese-English translation shown in this flowchart is
The process is substantially the same as the English-Japanese translation process shown in the flowchart of FIG.

That is, the operator inputs a Japanese file name to be translated into the mainframe 2a via the input unit 2b (S701). The mainframe 2a reads the Japanese file into the work area of the internal memory (S70
2) The read Japanese file is displayed on the display unit 2c (S703).

The operator inputs a command to the mainframe 2a (S704), and instructs the Japanese-English translation unit 222 whether to translate each sentence or batch translation.
When the original sentence is translated for each sentence (YES in S705), the sentence is translated by the process of S706, and then S7 is performed.
The processing from 04 is repeated.

On the other hand, in the case of translating the original sentence all at once (YES in S707), the operator starts the sentence number (translation start sentence number) and the translation end sentence number (translation end sentence number). ) And are sent to the Japanese-English translation unit 222 (S708). First, the Japanese-English translation unit 222 translates one sentence of the translation start sentence number into English (S709). Then, the Japanese-English translation unit 222 determines whether or not the translation of the Japanese sentence with the translation end sentence number is completed (S710), and if it is completed (S710).
If YES), the processing after S704 is repeatedly executed. If the translation of the Japanese sentence with the translated sentence number has not been completed (NO in S710), the Japanese sentence of the second sentence is displayed (S7).
11), the processing from S709 is repeatedly executed. In this way, a plurality of Japanese sentences from the translation start sentence number to the translation end sentence number can be collectively translated into English.

When the operator determines in S707 that the batch translation is not performed (S707), it is determined whether or not the translation result is printed (S712). When printing the translation result (YES in S712), the operator instructs the Japanese-English translated print section 224 with the print start sentence number and the print end sentence number. The Japanese-English translation printing unit 224 prints a parallel translation (English and Japanese) of the first sentence of the print start sentence number in the printer 2.
Then, the printer 2d prints the parallel translation of the first sentence in a predetermined format (S714). The Japanese-English translation stamp unit 224 continues printing until the printing of the parallel translation of the print end sentence number is completed (YES in S715). When the printing of the parallel translation of the print end sentence number is finished (YES in S715), the processes of S704 and thereafter are repeatedly executed.

In S712, when the operator determines that the printing is not performed (NO in S712), it is determined whether or not the translated sentence is output to the file (S717). When the operator determines that the translated text is to be output to the file (YES in S717), the translated text is output to the file,
Returning to the flowchart of FIG.

FIG. 26 is a flow chart showing the above-described one-sentence translation processing in S706 and S709. This process is substantially the same as the process of FIG. 24 described above. That is, first, the Japanese-English translation unit 222 generates key data of the original sentence given as the translation target (S751). The Japanese-English translation unit 222 searches the bilingual data on the bilingual database 246 according to the key data for bilingual data having a Japanese sentence most similar to the original sentence (S752), and reads this bilingual data (S753).

For example, as shown in FIG. 7, the original sentence 7
Suppose 1 is given. The Japanese-English translation unit 222 searches the parallel translation database 246 for the parallel translation data 73 including the Japanese sentence most similar to the original sentence 71. Japanese-English Translation Department 222
Examines the parts of the parallel translation data 73 that match the parts of the original sentence 71. That is, it is determined whether or not there is a part that can be used for translation of the original sentence 71 among the parts of the parallel translation data 73 (S754). If usable parts exist in the parallel translation data 73 (YE in S754
In S), this part is loaded into the work area on the internal memory (S755). In addition, unusable parts [15
0] [By measurement method] [of item] [6.3.1], a mark indicating unusable, eg <***>
Is added.

In this example, since all the parts of the original sentence 71 can be translated by the parts forming the parallel translation data 73, the result is output as the translation result. what if,
If all parts of original sentence 71 are not available (N in S757)
In O), the Japanese-English translation unit 222 stores the bilingual data having the Japanese sentence second similar to the original sentence 71 in the bilingual database 24.
It is read from 6 (S758). And Japanese-English translation department 22
In step 2, the processes from S754 onward are repeatedly executed to search the bilingual data for usable parts. When the bilingual data similar to the original sentence 71 is completely searched from the bilingual database 246 (YES in S756), the Japanese-English dictionary 2
Japanese translation is performed based on 42 (S761). Japanese-English dictionary 242
The personal name, tense, etc. of the word retrieved from are matched with the original sentence 71 (S762). The parts (words) thus searched are rearranged in a predetermined order, and the translated sentence 76 of the original sentence 71 is completed. The completed translated sentence 76 is displayed on the display unit 2c (S760). When the above process ends, the process returns to the process of FIG.

(Details of Dictionary Database Transmission / Reception) FIG.
30. FIG. 30 is a flowchart showing details of the process of transmitting / receiving the dictionary database 24 (S312 in FIG. 18).

In the flowchart of FIG. 27, the operator selects either transmission or reception of the dictionary database while looking at the menu displayed on the display unit 2c. When the transmission process is selected (YES in S801)
In S), the dictionary database transmission / reception unit 236 executes S803.
When the following processing is executed and the reception processing is selected (S8
If YES in 02), the dictionary database transmitting / receiving unit 236 executes the processing from S901 (FIG. 29).

Dictionary database transmission processing When the operator selects transmission processing (YES in S801)
In S), desired processing is further selected from the dictionary database transmission main menu shown in FIG. When the operator selects the processing for transmitting the English-Japanese dictionary (Y in S803)
In ES), the dictionary database transmitting / receiving unit 236 extracts one English-Japanese dictionary data from the English-Japanese dictionary 241 (S80).
4). Then, the dictionary database transmitting / receiving unit 236 writes the extracted data in the transmission file (S80).
5). In this way, until reading of all the dictionary data from the English-Japanese dictionary 241 is completed (YE at S806).
S) and S804 to S805 are repeatedly executed.
When reading of all the dictionary data is completed (YES in S806), the dictionary database transmission / reception unit 236 transmits the dictionary data in the transmission file to another workstation via the communication system 2e. With the above, the processing of transmitting the English-Japanese dictionary data ends, and the process returns to the flowchart of FIG.

When the operator selects the processing for transmitting the Japanese-English dictionary (YES in S810), the dictionary database transmitting / receiving unit 236 sets the Japanese-English dictionary data to 1 from the Japanese-English dictionary 242.
The case is taken out (S811). Then, the dictionary database transmitting / receiving unit 236 writes the retrieved data in the transmission file (S812). In this way, the Japanese-English dictionary 2
The processes of S811 to S812 are repeatedly executed until the reading of all the dictionary data from 42 is completed (YES in S813). When reading of all the dictionary data is completed (YES in S813), the dictionary database transmission / reception unit 236 transmits the dictionary data in the transmission file to another workstation via the communication system 2e. With the above, the processing of transmitting the Japanese-English dictionary data ends, and the process returns to the flowchart of FIG.

When the operator selects the transmission process of the English morphological analysis dictionary (YES in S821 of FIG. 28),
The dictionary database transmission / reception unit 236 takes in the data in the English morphological analysis dictionary 243 to the transmission file (S82).
2 to S824). Then, the dictionary database transmitting / receiving unit 2
36 transmits the dictionary data in the transmission file to another workstation, and returns to the flowchart of FIG. Thereafter, similarly, the transmission processing of the Japanese morphological analysis dictionary 244, the parallel translation parts analysis dictionary 245, and the parallel translation database 246 is executed. That is, when transmitting the Japanese morphological analysis dictionary 244, the dictionary database transmission / reception unit 236 performs S
When the processes of 826 to S830 are executed and the bilingual part analysis dictionary 245 is transmitted, the processes of S831 to S835 are executed. Further, when transmitting the parallel translation database 246, the dictionary database transmitting / receiving unit 236 performs S836.
~ The process of S840 is executed. As described above, various data in the dictionary database 24 can be transmitted to other workstations.

Receiving Process of Dictionary Database When the operator selects the receiving process (Y in S802)
In S), a desired process is selected from the dictionary database reception main menu. It is assumed that the operator selects the receiving process of the English-Japanese dictionary (YES in S901). After this,
Dictionary data is transmitted from another workstation to the workstation via the communication system 2e, and the dictionary data is stored in the English-Japanese dictionary reception file. The dictionary database transmitting / receiving unit 236 extracts one item of dictionary data from the English-Japanese dictionary reception file (S902), and determines whether the same dictionary data as the extracted dictionary data exists in the English-Japanese dictionary 241 (S903). The same dictionary data is English-Japanese dictionary 2
If it does not exist in 41 (NO in S903), the dictionary database transmitting / receiving unit 236 adds this dictionary data to the English-Japanese dictionary 241. On the other hand, if the same dictionary data exists in the English-Japanese dictionary 241 (YES in S903), the dictionary database transmitting / receiving unit 236 adds the dictionary data (S9).
The processing of S905 is executed without performing 04).

In step S905, the dictionary database transmission / reception unit 236 determines whether or not dictionary data remains in the English-Japanese dictionary reception file. If dictionary data remains (NO in step S905), the processing in steps S902 to S904 is performed. repeat. Then, when all the dictionary data in the English-Japanese dictionary reception file has been extracted (YES in S905), the process returns to the flowchart of FIG. 18 through the flowchart of FIG.

Thereafter, similarly, the receiving process of the Japanese-English dictionary 242, the English morphological analysis dictionary 243, the Japanese morphological analysis dictionary 244, and the bilingual parts analysis dictionary 245 is executed. That is, when receiving the Japanese-English dictionary 242, S906 to S906
When the processing of 910 is executed and the English morphological analysis dictionary 243 is received, the processing of S911 to S915 is executed. When receiving the bilingual parts analysis dictionary 245, the processes of S916 to S920 are executed, and when receiving the Japanese morphological analysis dictionary 244, S921 to S92.
Process 5 is executed.

When the bilingual database is received (S9
If YES in S26, the processes in S927 to S931 are executed. First, it is assumed that the parallel translation data or the like is transmitted from another workstation to the workstation via the communication system 2e, and these data are stored in the parallel translation database reception file. The dictionary database transmitting / receiving unit 236 extracts one item of data from the parallel translation database reception file (S927), and determines whether the same data as the extracted data exists in the parallel translation database 246 (S928). The same data is translated database 246
If it does not exist (NO in S928), the dictionary database transmitting / receiving unit 236 generates a parallel translation sentence number of this data (S929). Then, the dictionary database transmitting / receiving unit 236 adds the received data together with the bilingual sentence number to the bilingual database 246 (S930).

On the other hand, the same data is stored in the parallel translation database 2.
46 (YES in S928), the dictionary database transmitting / receiving unit 236 adds the data (S929,
The process of S931 is executed without performing S930). In S931, the dictionary database transmitting / receiving unit 2
36 determines whether or not data remains in the parallel translation database reception file, and if data remains (S9
If NO in S31, the processes of S927 to S930 are repeated. Then, when all the data in the parallel translation database reception file has been extracted (YES in S931),
It returns to the flowchart of FIG. 18 through the flowchart of 7.

By the above processing, it becomes possible to receive the dictionary registered in another workstation, the parallel translation data registered for learning, etc., and add them to the dictionary database 24. This makes it possible to share a dictionary database among a plurality of workstations.

The present invention is not limited to the above-mentioned embodiments, but can be carried out within the scope of the present invention. For example, a parallel processing type processor or the like may be used to search the parallel translation database.
As a result, the search can be performed in a short time, and the translation speed can be significantly improved. Further, by reading the translation examples using the OCR (optical reading device), a large number of translation examples can be efficiently input.

[0132]

As described above, according to the present invention, the following effects can be obtained.

First, the translation ability can be improved by the operator. That is, translation examples can be learned by registering parallel translation data and performing parallel translation learning (part link information generation). Therefore,
By allowing the user to learn a desired translation example in the machine translation system, it is possible for the user himself to improve the translation capability of the machine translation system. Further, since the learning registration can be automatically performed based on the past learning examples (translation parts), the work load on the operator in the learning registration can be reduced.

Secondly, according to the present invention, since the translation quality can be improved without adding the grammatical rule, it is possible to avoid the deterioration of the translation quality due to the interference between the added grammatical rules. Is possible.

Thirdly, it becomes possible to perform natural translation. That is, according to the present invention, since the translation example is performed based on the past translation example without depending on the grammar rule, it is possible to output a natural translation for an idiomatic expression or the like that is apart from the grammar rule. Is. Further, in the present invention, general E
Unlike the machine translation system by BMT method,
Each part of the original sentence is translated while comparing each example with a translation example similar to the original sentence. As a result, each part of the original sentence can be translated regardless of the grammatical rules, and a more natural translation can be output.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a machine translation system of the present invention.

FIG. 2 is a diagram showing an outline of a machine translation system according to an embodiment of the present invention.

FIG. 3 is a functional block diagram of a machine translation system according to an embodiment of the present invention.

FIG. 4 is a diagram showing a display screen during parallel translation registration according to an embodiment of the present invention.

FIG. 5 is a diagram showing a display screen during parallel translation registration according to an embodiment of the present invention.

FIG. 6 is a diagram showing a display screen during parallel translation learning according to an embodiment of the present invention.

FIG. 7 is a diagram showing an outline of a translation process according to an embodiment of the present invention.

FIG. 8 is a diagram showing an example of parallel translation according to an embodiment of the present invention.

FIG. 9 is a diagram showing an example of word link information and the like according to an embodiment of the present invention.

FIG. 10 is a diagram illustrating an English-Japanese translation example according to an embodiment of the present invention.

FIG. 11 is a diagram illustrating a Japanese-English translation example according to an embodiment of the present invention.

FIG. 12 is a conceptual diagram of transmission / reception of a dictionary database according to an embodiment of the present invention.

FIG. 13 is a diagram showing an outline of a translation process according to an embodiment of the present invention.

FIG. 14 is a diagram showing an outline of a translation process according to an embodiment of the present invention.

FIG. 15 is a main flowchart showing the operation of the machine translation system according to the embodiment of the present invention.

FIG. 16 is a main flowchart showing a bilingual learning registration process according to an embodiment of the present invention.

FIG. 17 is a main flowchart showing a translation process according to an embodiment of the present invention.

FIG. 18 is a main flowchart showing a dictionary management process according to an embodiment of the present invention.

FIG. 19 is a display example of a dictionary database transmission main menu according to an embodiment of the present invention.

FIG. 20 is a flowchart showing details of parallel translation data registration processing according to an embodiment of the present invention.

FIG. 21 is a flowchart showing details of parallel translation data registration processing according to an embodiment of the present invention.

FIG. 22 is a flowchart showing details of parallel translation learning processing according to an embodiment of the present invention.

FIG. 23 is a flowchart showing details of the English-Japanese translation processing according to the embodiment of the present invention.

FIG. 24 is a flowchart showing details of processing of English-Japanese translation according to an embodiment of the present invention.

FIG. 25 is a flowchart showing details of processing of Japanese-English translation according to an embodiment of the present invention.

FIG. 26 is a flowchart showing details of the processing of Japanese-English translation according to the embodiment of the present invention.

FIG. 27 is a flow chart showing details of a dictionary database transmission / reception process according to an embodiment of the present invention.

FIG. 28 is a flow chart showing details of a dictionary database transmission / reception process according to an embodiment of the present invention.

FIG. 29 is a flow chart showing details of a dictionary database transmission / reception process according to an embodiment of the present invention.

FIG. 30 is a flow chart showing details of a dictionary database transmission / reception process according to an embodiment of the present invention.

FIG. 31 is a diagram for explaining an outline of processing of a conventional machine translation system using a syntax conversion method.

[Fig. 32] Fig. 32 is a diagram for explaining the outline of the processing of a conventional EBMT machine translation system.

[Explanation of symbols]

2b input means (dialogue input means) 2c display means 21 parallel translation data registration means 22 translation means 24 dictionary database 25 document files (first language character string input means, second language character string input means) 211 parallel translation data registration unit (registration means ) 243 English morphological analysis dictionary (first language dictionary, second language dictionary) 244 Japanese morphological analysis dictionary (first language dictionary, second language dictionary) 246 Bilingual database

Claims (6)

[Claims] 1. A bilingual database in which a plurality of first language character strings represented in a first language and a plurality of second language character strings obtained by translating the plurality of first language character strings into a second language are registered. And most similar to the original string expressed in the first language, or
Search the same first language character string from the bilingual database,
In a machine translation system including a translation unit that translates an original character string into a second language based on a searched second language character string corresponding to the first language character string, a first language for inputting the first language character string Character string input means, second language character string input means for inputting a second language character string obtained by translating the first language character string into a second language, and display for displaying the first language character string and the second language character string Means, associating means for associating the first language character string and the second language character string respectively displayed on the display means, and the associated first language character string and the second language character string are registered in the parallel translation database. A machine translation system comprising: 2. A first language dictionary in which words in a first language are registered, and a second language dictionary in which words in a second language are registered.
When the first language character string displayed on the display means includes an unregistered word in the first language dictionary, or the second language character string displayed on the display means is 2. The machine translation system according to claim 1, wherein when the bilingual dictionary includes an unregistered word, the unregistered word is displayed on the display means. 3. The machine translation system, wherein the registration means according to claim 1 registers the parallel translation number in the parallel translation database together with the associated first language character string and second language character string. 4. The interactive input means for pointing the first language character string and the second language character string respectively displayed on the display means, and the associating means, the first language character string and the second language character pointed to. The machine translation system according to claim 1, wherein the columns are associated with each other. 5. The machine translation system according to claim 1, wherein the first language character string input means and the second language character string input means input character strings from different files. 6. The machine translation system according to claim 1, wherein the first language character string input means and the second language character string input means input character strings from the same file.