JPH10312382A - Similar example translation system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform standardized translation fast with simple constitution by a machine translation system which makes use of an example data base. SOLUTION: In the example data base 8 many translation sentences are stored and identical or similar sentences are extracted for translation. When no similar sentence is found, the original text is divided and identical or similar examples are extracted for the divided sentences for translation. When no similar sentence is found for a divided sentence, translation is carried out by words or idioms by performing retrieval from a dictionary data base 9. The translation result is postedited by referring to the example data base 8 and dictionary data base 9. The completed translation sentences are stored in the example data base 8 together with the original text. Each time translation is performed, example data are stored, so the example data base is automatically enriched. The standardized translation is performed fast only by using simple grammatical rules 3, 5, and 6, the example data base 8, and the dictionary data base 9.

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 suitable for network computing (NC) using the Internet, and more particularly, to searching for an exact matching sentence or a similar sentence from an example database and using a bilingual sentence thereof. And a similar example translation system that performs translation.

[0002]

BACKGROUND OF THE INVENTION The idea of translating by machine had existed before the invention of the computer, and attempts to translate using a computer began as soon as the computer was invented.

[0003] As the first stage machine translation system,
As shown in FIG. 11, there is a direct-type machine translation system for dictionary lookup of words and translated words. This is from the 1950s
Research and development took place in the 1960s. The translation processing technology at this stage analyzes the source sentence, recognizes a word or a set of a few words in the source sentence as a unit, and performs replacement from the source language to the target language in the recognized unit. ,
Finally, rearrange the sentences by rearranging the order of these units. At this time, the unit of recognition is of such a degree that it can be treated as equivalent to one word. Therefore, the translation process at this time is called word-to-word translation. The content of recognition is about parts of speech and extremely simple grammatical functions. Therefore, it is not much different from translating a word in the original sentence into a translated word. Such a translation processing method is called direct translation.

At this time, the number of words that the machine translation system had in the dictionary was at most several hundreds, and the contents of the dictionary were extremely simple. Therefore, it was considered to use the frequency of appearance of words in addition to grammatical information. In order to simplify the translation process, a pivot method described later has been considered.

[0005] From the viewpoint of using a machine translation system, pre-editing in which a human corrects an original sentence so as to facilitate translation processing, and post-editing in which a human corrects an error or an insufficient point in a translated sentence have been considered. As described above, most of the basic ideas regarding the processing method and the use method of the machine translation system which are seen today were already considered at this time.

[0006] The machine translation system in the second stage is shown in FIG.
A transfer-type machine translation system that performs document syntax conversion (grammar analysis and language generation) as shown in Fig. 12,
As shown in FIG. 13, there is a pivot type machine translation system that translates using an intermediate language. These are from the 1960s to 19
Research and development took place in the 1970s. This stage of translation processing technology was introduced by the MIT linguist N. Chom in the late 1950s.
It is based on the linguistic theory of transformation-producing phrase structure grammar theory proposed by sky and the parsing method based on it. That is, the result of grammatically analyzing the original sentence is represented as a syntax structure expression, the syntax structure expression is converted into a target language and a syntax structure expression, and a translated sentence is created from the converted syntax structure expression. As described above, since the translation is performed via an indirect intermediate expression called a syntax structure expression, such a translation processing method is called indirect translation. That is, the second-stage machine translation technology analyzes the structure of a sentence, unlike the simple one of the first-stage word replacement. Therefore, as language data, not only dictionaries but also grammars related to structures have been used.

[0007] The syntax structure expression depends on the grammatical characteristics of the language, and even if the sentences have the same contents, the syntax structure expression of the sentence will be different if the language is different. for that reason,
Translation requires a conversion of the syntax expression. Therefore, such a translation processing method is called a transfer method.

On the other hand, instead of the syntax structure expression, one expression means independent of the language characteristics is set, and the translated expression is directly converted from the intermediate expression described using the expression means without conversion. Is called a pivot method (intermediate language method). A description language that provides such an expression means is called a pivot language. There are various types of pivot languages, such as those representing syntactic information and those representing semantic information. The pivot type translation process is also a type of indirect translation because translation is performed via an indirect expression called an intermediate expression.

[0009] In any of the direct method, transfer method, and pivot method, in order to obtain a translation result of 70% or more, it is necessary to enrich the basic word dictionary, the technical term dictionary, the user dictionary, and the grammar rules. is there. With these translation methods, there are considerable difficulties in constructing the system, and despite the complexity and scale of the system, complete translation is difficult, requiring human post-editing and other corrections, and 100% Full translation is still not achieved. Therefore, recently, a translation method different from the conventional method has been reviewed. With the speeding up of computers and the increase in capacity of storage devices, it has become possible to store a large amount of example sentences in a database, search for example sentences that match or resemble the original sentence, and translate them using bilingual sentences. . Japanese Patent Application Laid-Open No. 8-221422 describes an example of a translation method using this example sentence database.
Issue, Satoshi Sato, "Machine Translation by Analogy" (19
April 1997, published by Kyoritsu Shuppan Co., Ltd.), Makoto Nagao, "Iwanami Koza Software Science 15 Natural Language Processing" (published by April, 1996, Iwanami Shoten Co., Ltd.), edited by Makoto Nagao and Takenori Makino, "Translate by Computer" ( Published by Kyoritsu Shuppan Co., Ltd. in February 1995).

A first example of the translation using example sentences is disclosed in Japanese Patent Application Laid-Open No. Hei 8-2214.
No. 22 discloses a machine translation system for translating by matching one sentence. In this machine translation system, as shown in FIG. 14, a machine translation apparatus searches a bilingual database connected to a communication network, and if there is a matching sentence, extracts the translated sentence to obtain a translated sentence.

A second example of the translation using example sentences is a translation method using a bilingual pattern, which is called <pattern translation>.
“Refer to Chapter 3 for the detai” for the original text “For more information, see Chapter 3.”
ls. ”is registered as a bilingual pattern. For the sentence“ If you want to know <X>, see <Y> ”,“ Refer to <Y> for <X>. ” Is output. <X> and <Y> are variable parts that contain various expressions. No matter what word you enter in the <X> or <Y> part, you will be able to output the translated text "Refer to <Y> for <X>." It is efficient when many similar translations appear.

A third example of the translation using an example sentence is a translation based on an example or an example-based machine translation.
Translation; EBMT). This was a method proposed in the mid-1980s, but was actually implemented on a computer in the late 1980s.
Example-based translation means that a translated "translation pair" is prepared in advance, and similar sentences are used as examples.
It is intended to be translated. However, since the whole sentence is not very similar, even if it is a part of the sentence, we find a similar part from the example and try to translate it based on it. At this time, the translation is not merely a replacement of the word, but depends largely on the environment in which the word is used.
A major issue is to define whether similarity is recognized as a meaningful unit for translation and used as a partial translation, and how similarity is determined.

At present, there are already 10 or more machine translation systems (Machine Translation, MT for short) in Japan alone.
Is provided to the user. A few years ago, many systems were expensive running on workstations that cost millions of yen. Therefore, users are limited to offices that translate large documents, such as translation companies and translation departments of companies. Recently, however, the number of personal users is increasing, as systems running on personal computers have been sold, and machine translation services have been provided on personal computer nets.

[0014]

A conventional machine translation system translates by performing a complicated syntax analysis or prepares and translates a large number of example sentences in advance. A system had to be built, which was costly and time consuming to translate. Alternatively, it is difficult to create a practical translation for a translation using a small-scale system, and the work of correcting the translation is a burden.

[0015] In the translation using the example database, it is necessary to prepare about one million sentence examples in order to perform practical translation. The Birmingham Corpus, created in collaboration with the University of Birmingham in the United Kingdom and Collins Publishing, a dictionary publishing system, is one of the traditional example data. Since then, it has become a corpus of 200 million words. The Oxford Dictionary, created by James Murray over the course of 50 years, has 5 million example data. However, both contain little example data from the natural sciences, medicine, biotechnology, nuclear power, information, aviation, and space. In order to translate a specialized field, a large amount of example data in each field must be prepared.

[0016]

In order to solve the above-mentioned problems, according to the present invention, an example database, a dictionary database, a perfect matching sentence translating means, a similar example translating means, a sentence dividing means, and a divided sentence perfecting means are provided. By providing a matching sentence translating unit, a segmented sentence similar example translating unit, and a new sentence translating unit, a sentence that is the same as or similar to a previously translated sentence is translated using an example database, thereby achieving a simple configuration. To enable fast and consistent translation. In addition, a full-text post-editing means using an example database and a dictionary database is provided so that post-editing by a human can be easily performed.

The example database automatically stores a translated sentence in which the translated original sentence and the translated sentence are paired, and accumulates the translated sentences. The perfect matching sentence translating means searches the example database for an example that completely matches the original sentence, and outputs the bilingual sentence. The similar example translating means searches the example database, extracts an example sentence in which one or a few words do not match the original sentence, and extracts a bilingual sentence word corresponding to the example sentence word that does not match the original sentence into the original sentence word. Perform the translation by making changes using the translation. The sentence dividing means searches the example database and, when it cannot find an example sentence in which more than half of the words match the original sentence, divides the original sentence by a comma or the like. The divided sentence perfect matching sentence translating means searches the example database and outputs a bilingual sentence of an example sentence that completely matches the divided original sentence. The divided sentence similar example translation unit searches the example database, extracts an example sentence in which one or a few words do not match the divided original sentence, and extracts a bilingual sentence corresponding to the example sentence that does not match the divided original sentence, The translation is performed by changing the translation of the words of the original segment. The new sentence translating means searches the example database and finds no example sentence in which more than half of the words match the divided original sentence, searches the dictionary database and translates word-by-word or idiom units. The full-text post-editing means includes means for displaying a related example sentence with reference to the example database and means for displaying a related word with reference to the dictionary database.

The example data of a specialized field is prepared by inputting a bilingual manual or a term dictionary in the field.

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION
An input device for inputting an original sentence to be translated, one sentence extracting means for cutting out the input original sentence in units of sentences and outputting the original sentence, and a dictionary database and an example database storing example sentences and their parallel translations in pairs. A database part, a perfect-matching sentence translating means for searching the example database and outputting a bilingual sentence of an example sentence that completely matches the original sentence, and extracting an example sentence that partially matches the original sentence by searching the example database and translating the translated sentence Similar example translating means for performing translation by changing the sentence according to the original sentence, and sentence dividing means for dividing the original sentence according to a predetermined rule when an example database is searched and an example sentence that matches the original sentence by a predetermined rate or more cannot be found. A sentence translation unit that searches the example database and outputs a bilingual sentence of an example sentence that completely matches the divided original sentence; To extract an example sentence that is partially inconsistent with the divided original sentence, and to translate the translated sentence by changing the translated sentence according to the divided original sentence. When a matching example sentence cannot be found, a new sentence translating means for searching and translating a dictionary database, a divided translation assembling means for assembling each translated sentence of the divided original sentence and outputting a translated sentence of the original sentence, This is a similar example translation system having the following feature. The same translated sentence is output to the previously translated original sentence.

The second aspect of the present invention is the first aspect.
In the similar example translation system described above, a full-text post-editing means having means for displaying a related example sentence by referring to an example database and means for displaying a related word by referring to a dictionary database is provided. Since the post-editing can be performed while referring to the translation, the translation sentence can be easily corrected.

Hereinafter, embodiments of the present invention will be described in detail with reference to FIGS.

(First Embodiment) In a first embodiment of the present invention, an example that completely matches the original sentence is searched from the example database, and the translated sentence is used as a translated sentence, and there is no example sentence that is completely matched. In this case, search for an example sentence that does not match one or a few words, change the bilingual sentence to a translated sentence, and if there is no example sentence that matches the original sentence with more than half the words, split the original sentence at a position such as a comma Then, the example database is searched for the divided original text, and if there is an example that completely matches the divided original text, the translated text is used as the translated sentence. If there is no exact matched example sentence, the example sentence that does not match one or several words is searched. If the translated sentence is changed to a translated sentence, and there is no example sentence in which more than half the words match the divided original sentence, a similar example translation system that searches the dictionary database for each word or phrase and outputs the translated word. is there.

Hereinafter, the first embodiment will be described with reference to FIGS. FIG. 1 is an apparatus configuration diagram of a similar example translation system according to the first embodiment of this invention.

The input device 1 inputs an original sentence to be translated via input means such as a keyboard and an OCR. The information may be input from a recording medium such as a floppy disk, or may be input via communication means such as a LAN and a facsimile. The input buffer 2 stores the input original text. One sentence extraction rule table 3 is referred to when one sentence is extracted from the original sentence input to input buffer 2. The editing buffer 4 contains the extracted 1
It stores the minutes. The output device 11 outputs a translation. The output unit may be a communication unit such as a LAN or a storage unit such as a disk. The replacement rule table 5 is a table storing rules for replacing words and idioms that do not match the original sentence and the example sentence. The sentence division rule table 6 is a table that stores rules for dividing a sentence. The CPU 10 is a processing unit that includes a calculation unit and a main storage unit and performs search and translation processing.

The database section 7 comprises an example database section 8 and a dictionary database section 9. The example database section 8 stores an example sentence and its translation in pairs.
The syntactic analysis data is stored for each sentence so that it can be immediately known to which part of the translated sentence a word or idiom of the original sentence corresponds. If there is no correspondence between the words in the original sentence and the translated sentence due to idiomatic expressions, the correspondence is left unclear. In such a sentence, there is no practical obstacle because the words do not change.

The example sentence is stored, for example, using a hash value of a character code as a key. A plurality of bilingual sentences may be present in one example sentence. Each word may be stored as tree-structured data linked by a pointer. Alternatively, if the information is stored as a relational database, the search can be easily performed.

Extracting all words of all example sentences,
A word dictionary is created, and an index is created in which each sentence is located in which example sentence. By using a full-text search method, it is possible to search for example sentences that partially match words at high speed. By compressing the example sentences using the word dictionary as a compression dictionary, the disk device and the main storage device can be reduced in size. A bilingual sentence can also be compressed using a compression dictionary.

Each example sentence is given data in a field such as a technical field.

The dictionary database 9 includes a word dictionary and Colloc
It consists of an ation dictionary and an Idiom dictionary. The word dictionary is almost the same as a normal English-Japanese dictionary. Collocation dictionary and Idiom
The dictionary is a dictionary containing collocations, idioms, and idioms.

FIG. 2 is a flowchart showing the procedure for translating one sentence. With reference to FIG. 2, each functional means and processing procedure of one-sentence translation will be described. (1) One sentence extraction unit First, one sentence is extracted from an original sentence according to a predetermined rule with reference to a one sentence extraction rule table. Normally, English text is cut out using a period as a delimiter. In addition, colons, semicolons and parentheses may be used as sentence delimiters. Whether to cut out a sentence with a colon or the like or to separate a colon or the like at the time of sentence division is determined according to the type of the target sentence. (2) Example database search means For the cut-out sentence, an example database is searched by a hash method or a full-text search method, and an example sentence that matches perfectly is searched. For example, find the hash value of the character code of the original text,
The example database is searched using that as a key. 100%
If a matched example sentence is found, the bilingual sentence is output to the completely matched sentence translating means. At this stage, all the partial sentence example sentences may be extracted, but the processing amount of searching for the partial match example sentence is much larger than the processing amount of searching for the perfect match example sentence. Performing only the search is less wasteful. (3) Perfectly matched sentence translation means Outputs a bilingual sentence of an example sentence that completely matches the original sentence. If there are a plurality of parallel translations, all are output as translation candidates.
At this time, the highest score is given to the translation sent immediately before, and the lowest score is given to the translation sent most recently. In addition, the degree of relevance in the technical field is added, and a translation with a high degree of relevance is output with priority. For example,
It is assumed that the information processing field has a high degree of relevance to the electrical field and has a low degree of relevance to the chemical field. (4) Matching word matrix creating means If a perfect matching sentence cannot be found, the example database is searched again by the full-text search method or the like. The result is output as a match word matrix in FIG.

For each word of the original sentence, the example database is searched, an example sentence containing a matching word is extracted, and an example sentence is written in accordance with the position of the corresponding word. Although any search method can be used, a full-text search method in which a word dictionary index is created and searched in advance is fast. If this matrix is created by filling in all the example sentences that match only one word, it is wasteful. Therefore, only the example sentences with a match rate equal to or more than a predetermined value or only the upper number of example sentences with a high match rate are used. Is good. The value of the predetermined value is determined according to the size of the example database. In the example of FIG. 3, the example sentence is shorter than the input sentence, but may be longer than the original sentence. In the case of an example sentence that covers all the example sentences, the matching rate becomes 100%, but it is not an exact match, so a similar example translation process is performed.

If there is an example sentence with a matching rate of 50% or more in the matching word matrix, similar example translation is performed. If there is no example sentence with a matching rate of 50% or more, the sentence is segmented. (5) Similar example translation means Translates a bilingual sentence of an example sentence that is partially inconsistent with the original sentence according to the original sentence.

First, an example sentence having the highest matching rate is extracted from the matching word matrix, and a mismatching word is examined. Refer to the substitution rule table to find differences in numbers and symbols, mismatches in pronouns, differences in proper nouns, variations in verbs, plural forms of nouns, search the dictionary, create translations, and translate. Change the statement and output.

If the mismatched word is the difference between the singular and the plural, the bilingual sentence is output as it is. If one word that does not match is a number, the corresponding number in the bilingual sentence is replaced with the number in the original sentence to create a translated sentence. If one word that does not match is a symbol,
Replace the corresponding symbol in the bilingual sentence with the symbol in the original sentence, and create a translated sentence. If the one word that does not match is the subject pronoun, the subject pronoun of the original sentence is translated, and the corresponding word of the bilingual sentence is replaced with it to create a translated sentence. In the case of other pronouns, translation processing is performed in substantially the same manner. If the one word that does not match is a proper noun, the proper noun of the original sentence is translated (or transliterated), and the corresponding word of the bilingual sentence is replaced with it, thereby creating a translated sentence. If one of the disagreement words is different in the tense of the verb, the translated sentence is created by changing the tense of the corresponding verb in the bilingual sentence to the tense of the verb in the original sentence. The same applies to other verb variations. If there is no unprocessed word by this process, the translated sentence is output as the end of translation.

If there is an unprocessed word, a translation is obtained for collocations, idioms, and idioms by referring to the Collocation dictionary and Idiom dictionary, and the bilingual sentence is changed. If there is no unprocessed word by this process, the translated sentence is output as the end of translation.

If there is an unprocessed word, a translated word is obtained by referring to the dictionary, and the bilingual sentence is changed. If there is no unprocessed word by this process, the translated sentence is output as the end of translation. If there is an unprocessed word, the original word is left as it is, and the translation ends.

In the case of an incomplete match, if all the unmatched words can be translated by changing the bilingual sentence, the fact that they do not completely match is displayed with a white diamond mark. Also, the percentage of words in the original sentence that matches the example sentence is displayed by the hit rate.

Each time the translation of one sentence is completed, the original sentence and its translated sentence are automatically stored in the example database as a pair of translated sentences. (6) Sentence division means The original sentence is divided according to the sentence division rule. Sentence division is an
This is performed in d, that, relative pronouns, commas, parenthesis processing, and the like. If there are no words or commas that can be divided, the sentence is not divided and the statement is not divided. At the time of division, the structure of a sentence is stored and used for assembling a translated sentence. Relative pronouns, pronouns, and definite articles are converted to pronouns, nouns, and indefinite articles,
It should be possible to return to the original when reconstructing the translation.

For each of the divided sentences, if there is an example sentence that matches 100% with reference to the matching word matrix, the sentence is completely translated. Matching rate of split statements is 50 to 99
If it is%, the same processing as the similar example translation is performed. If the matching rate of the divided sentences is 49% or less, a new sentence translation process is performed.
If sentence division is not possible, a new sentence translation process is performed. (7) Split sentence perfect match sentence translation means Searches the example database and outputs a bilingual sentence of an example sentence that completely matches the split original sentence. This translation processing is the same as that of the perfect matching sentence translation means. The translated sentence is passed to the divided translated sentence assembling means. (8) Example of translating a sentence similar to a divided sentence The example database is searched to extract an example sentence that partially disagrees with the divided original sentence, and translation is performed by changing the bilingual sentence according to the divided original sentence. This translation processing is the same as the similar example translation means. The translated sentence is passed to the divided translated sentence assembling means. (9) New Sentence Translation Means Referring to the dictionary database, translations are obtained in units of collocations, idioms, idioms, and words, rearranged according to grammatical rules, and supplemented with particles, etc., to create translated sentences. If there is no unprocessed word by this process, the translated sentence is output as the end of translation.
If the word is not in the dictionary, leave the original language and end the translation. If it is not possible to create a translated sentence, the translated or original words are arranged and displayed to that effect. In order to indicate that the translation was performed at the word level with reference to the dictionary, the information is output with a black diamond mark. At the same time, a hit rate, which is a match rate with the example sentence, is also displayed. The translated sentence is passed to the divided translated sentence assembling means. (10) Split translation sentence assembling means Each split translation is assembled according to the syntax data at the time of sentence splitting, and the translation is reconstructed.

An example of translating from English to Japanese will be described.

Consider a case where an original sentence "She is my sister." Is input. When searching by a hash value, a hash value is obtained by adding all character codes including a space code and a period, and all examples having matching values are extracted. From these, search for example sentences whose word order matches all the words, and if so, extract the corresponding translation. In the case of a relational database, [Sh
Search by the search expression e] * [is] * [my] * [sister]. If a hit is found, a search is made for answers that match the word order. If so, retrieve the corresponding translation.

An example of translating from an example sentence of partial match when there is no example sentence of perfect match will be described.

"The blonde girl who is wearing blue
Consider the case where you enter the original text “jeans is my sister.” The example database contains “The girl is my siste.”
r. "and" She is wearing blue jeans. "and" She is my
sister. ”is stored. In the case of a relational database, [The] * [b
londe] * [girl] *...
Search for answers that match the word order from the hit answers. When using the method of creating a word dictionary index in advance and performing a full-text search, obtain the number of an example sentence for each word of the original sentence, and if there is an example sentence number common to each word, check the word order for the example sentence, Extract matching example sentences. The result of searching the example database in this way is used as the matching word matrix in FIG.

In this example, since more than half of the words do not match, the original sentence is divided. “The blonde girl is my sist
Search the example database for "er." and "who is wearing blue jeans" to find similar sentences.
ng blue jeans ”is“ She is wearing blue j
eans. "is changed to a partial translation.
"The blonde girl is my sister."
"girl is my sister." and one word of "blonde" are matched, so the dictionary is searched for a translation of "blonde", and the bilingual translation is changed to a partial translation. Combine according to the data to create a translation.

As described above, in the similar example translation system according to the first embodiment of the present invention, a translation is quickly generated with a simple configuration by retrieving and using past translations from an example database. can do. Also, since the same translated sentence is generated for the same original sentence, the translated sentence can be unified.

Note that the original sentence is divided when more than half of the words of the example sentence do not match the original sentence, but the ratio at which the original sentence is divided may be appropriately determined according to the translation conditions. This ratio may be arbitrarily specified.

(Second Embodiment) In a second embodiment of the present invention, a similar example translation system is provided with a full-sentence post-editing means, and a related example sentence is displayed with reference to an example database, and a dictionary database is displayed. To display related words.

Hereinafter, the second embodiment will be described with reference to FIG.

When the similar example translation system finds a completely matched bilingual sentence, the most recently registered or referenced bilingual sentence is output together with the original sentence, so that the suitability of translation can be checked. If it is an appropriate translation, it will be adopted as it is and proceed to the next. If the translation is inappropriate, select the next candidate. If there is no appropriate translation, some words or idioms are corrected with the closest translation as a candidate. At this time, the example database can be searched using the word as a keyword, and a plurality of translation candidates can be displayed. Alternatively, a translation can be found by searching the dictionary database and correcting it. Alternatively, the original text can be corrected and translated again.

If the similar example translation system does not find a completely matched bilingual sentence, the translated sentence based on the similar sentence and the original sentence are output by noting which part is inconsistent and how it was changed. Therefore, the appropriateness of the translation can be easily checked. If the translation is inappropriate, the translated sentence is corrected in the same manner as in the case of perfect matching.

If the similar example translation system finds neither a completely matched bilingual sentence nor an incompletely matched bilingual sentence, it notes how the plurality of candidate translations by sentence segmentation were translated together with the original sentence. The translation is corrected by referring to them. In this case, since the translated sentence is often incomplete, an appropriate translated sentence is created while searching the example database and the dictionary database.

In any case, when the translated sentence is determined, it is automatically stored in the example database. When a completely matched bilingual sentence is adopted, the registration date and time are updated. Alternatively, attribute information such as a translator and a specialty field may be added, and all the same translated sentences may be registered. In other cases, a set of the original sentence, the translated sentence, and the syntax data is registered, and the key table is also updated.

(Third Embodiment) In a third embodiment of the present invention, a similar example translation system can be used via the Internet.

Hereinafter, a third embodiment will be described with reference to FIG.

FIG. 10 shows the machine translation system of the first and second embodiments which can be used via the Internet. A machine translation server is set up in one place so that a terminal can access and translate through the Internet. Since the terminal only needs to have a function of accessing the Internet, machine translation can be used from any terminal and from any place.

[0056]

According to the present invention, a translation can be obtained from a completely matching example sentence by searching the example database, or translation can be performed using a similar example sentence. In addition, since a previously translated sentence is used, redundant translation work can be omitted and translated sentences can be unified.

In addition, since the translated sentence is corrected with reference to the example database at the time of post-editing, editing can be easily performed.

Further, since translation processing can be performed via the Internet, a user can use machine translation from anywhere with only a small terminal.

[Brief description of the drawings]

FIG. 1 is a device configuration diagram of a similar example translation system according to a first embodiment of this invention.

FIG. 2 is a flowchart of a single sentence translation process according to the first embodiment of this invention;

FIG. 3 is a matched word matrix of an example sentence that is partially mismatched, which is created in the first embodiment of the present invention.

FIG. 4 is a flowchart of a complete matching sentence translation process according to the first embodiment of this invention;

FIG. 5 is a flowchart of a similar example translation process according to the first embodiment of this invention.

FIG. 6 is a flowchart of a sentence division process according to the first embodiment of this invention.

FIG. 7 is a flowchart of a new sentence translation process according to the first embodiment of this invention.

FIG. 8 is a flowchart of a divided translation assembly processing according to the first embodiment of this invention.

FIG. 9 is a block diagram of a second embodiment of the present invention.

FIG. 10 is a block diagram of a third embodiment of the present invention.

FIG. 11 is a conceptual diagram of conventional direct machine translation.

FIG. 12 is a conceptual diagram of conventional transfer-type machine translation.

FIG. 13 is a conceptual diagram of conventional pivot-type machine translation.

FIG. 14 is a block diagram of conventional one-sentence matching machine translation.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input device 2 Input buffer 3 Sentence extraction rule table 4 Editing buffer 5 Replacement rule table 6 Sentence division rule table 7 Database part 8 Example database 9 Dictionary database 10 CPU 11 Output device

Claims (2)

[Claims] 1. An input device for inputting an original sentence to be translated, (2) a single sentence extracting means for cutting out the input original sentence in units of sentences and outputting the original sentence, (3) a dictionary database and an example sentence And (4) a perfect matching sentence translating means for searching the example database and outputting a bilingual sentence of an example sentence that completely matches the original sentence, and (5) A similar example translating means for searching the example database to extract an example sentence that is partially inconsistent with the original sentence, and performing translation by changing the bilingual sentence according to the original sentence, and (6) searching the example database for the original sentence and a predetermined ratio. If a matching example sentence cannot be found, a sentence dividing unit that divides the original sentence according to a predetermined rule, and (7) a bilingual translation of an example sentence that completely matches the divided original sentence by searching the example database A sentence translation unit that outputs a sentence, and (8) a search unit that searches the example database to extract an example sentence that is partially inconsistent with the divided original sentence, and performs translation by changing the bilingual sentence according to the divided original sentence. (9) a new sentence translating means for searching and translating a dictionary database when an example sentence that matches the divided original sentence by a predetermined ratio or more cannot be found by searching the example database; A translation unit for assembling each of the divided original sentences to output a translated sentence of the original sentence. 2. A full-text post-editing means comprising means for displaying a related example sentence by referring to an example database and means for displaying a related word by referring to a dictionary database. A similar example translation system as described.