JPH04336381A - Machine translation system - Google Patents

Abstract

PURPOSE:To proofread a translation into a natural translation in style corresponding to a user's translation editing instruction through easy operation without any omission. CONSTITUTION:A dictionary 3 contains equivalents of parts of speech corresponding to parts of speech of source words and equivalents corresponding to syntax variation on the side of an object language. A source sentence to be translated, its translation as the result of a translating process, and a translation structure which is a tree structure where modification, parts of speech, and inflection, conjugation, declension, etc., are added to nodes of translation paragraph units are held in a translation structure area 5. In the translating process 2, the translation of the source sentence obtained by referring to the dictionary 3 and translation structure data are generated and written in the translation structure area 5. In an editing process 4, the translation structure is edited by referring to the translation structure area 5 and dictionary 3 according to the user's editing instruction and written in the translation structure area 5.

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 for editing translated texts.

0002

[Prior Art] Conventional machine translation systems that translate a natural language into a target natural language generally use a dictionary, which is a database of grammar rules and vocabulary, to understand the syntax and meaning of a single sentence in the input language, which is the source text. It performs the analysis and generates the translated text as output according to the syntax of the target language. but,
Conventional machine translation systems have difficulty creating grammatical rules and dictionaries that completely regularize the syntactic and semantic features of natural language, and it is difficult to create grammatical rules and dictionaries that completely regularize the syntactic and semantic features of natural language. The quality of the translation is not up to par.

0003

[Problems to be Solved by the Invention] The conventional machine translation system described above replaces the syntax of the source language with the translated word to adjust the word order because there is a syntactic difference between the input language and the target language. Since a literal translation cannot be said to be a natural translation for the target language, a syntactic transformation unit is provided between analysis and generation to perform syntactic transformation that can be handled regularly.

[0004] However, even when a human translator is translating, the decision as to whether or not to change the phrasing must be made depending on the style and context of the intended sentence, so it often cannot be handled uniformly using grammatical rules. It has the following drawbacks. Examples of such phrasing changes include changing the subject,
This includes changes in character, changes in phrases and clauses, etc.

[0005] Note that if you try to make the wording sound like the target language, but the grammar rules are applied incorrectly, it will be difficult to match the original text, and it may end up requiring more work than correcting a literal translation. First, if the grammar rules are translated in a literal translation style according to the syntax of the original text, it is necessary to proofread the translated text with a text editor to adjust the phrasing, and it is necessary to change the phrasing of words that have many modifiers. When this happens, it is necessary to modify each of the modifying elements, but there is a problem in that the modification of the modifying elements may be forgotten.

[0006]

[Means for Solving the Problems] The machine translation system of the present invention includes a dictionary having not only translated words corresponding to the parts of speech of the source language but also translated words corresponding to syntactic changes in the target language, the original text to be translated, and the translated words. A bilingual structure area that holds a bilingual structure that is a tree structure in which the translated text of the processing result and information such as dependency, part of speech, conjugation, and inflection are added to the nodes of the target clause, and a translation that has translated the original text by referring to the dictionary. A translation process in which a resulting translation and bilingual structure data are created and written into the bilingual structure area, and a bilingual structure is edited by referring to the bilingual structure area and the dictionary according to editing instructions from the user, and the edited bilingual structure is written into the bilingual structure area. It is configured to include editing processing.

[0007]

Embodiments Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the machine translation system of the present invention.

As shown in FIG. 1, this embodiment is comprised of an instruction input section 1, a translation processing 2, a dictionary 3, an editing processing 4, a bilingual structure area 5, a bilingual output section 6, and a terminal device 7.

[0010] First, the instruction input unit 1 reads an original text from the terminal device 7 and provides it to the translation process 2. When a translation instruction is input from the terminal device 7, the instruction input unit 1 starts the translation process 2. Furthermore, when an editing instruction is input from the terminal device 7,
The user is inquired about the translated word to be edited, and editing processing 4 is started as the word to be edited.

FIG. 2 is a diagram showing an example of the dictionary 3 in FIG. 1. Further, FIG. 3 is a diagram showing an example of the contents of the bilingual structure area 5 of FIG. 1.

Therefore, the translation process 2 translates the original text input from the instruction input unit 1 by referring to the dictionary 3 shown in FIG. The bilingual structure data is created and written into the bilingual structure area 5.

Furthermore, as shown in FIG. 2, the dictionary 3 has not only translations of parts of speech corresponding to the parts of speech of the source language, but also translations corresponding to syntactic changes in the target language. It is referred to to determine the translation word, and in editing process 4, it is referred to as information for replacing the translated word and adjusting the usage.

On the other hand, the bilingual structure area 5, as shown in FIG. It is maintained as a bilingual structure, which is a tree structure attached to unit nodes. That is, the line connecting 33a and 34a of the bilingual structure, the line connecting 34a and 36a, and 35a
The line connecting 36a and 36a represents the respective dependency relationship. The direction from which the arrow points is called the dependent element, and the direction at the tip of the arrow is called the modifier element.

In the editing process 4, the bilingual structure is edited and written in the bilingual structure area 5 by referring to the bilingual structure area 5 and the dictionary 3 in accordance with the editing instruction inputted through the instruction input section 1.

FIG. 4 is a flowchart showing an example of the phrase conversion editing operation in editing process 4 of FIG. 1 when the target language is Japanese.

Next, an example of the operation of the editing process 4 will be explained with reference to FIG. As mentioned above, the bilingual structure in the bilingual structure area 5 created in the translation process 2 has information such as dependency, part of speech, conjugation, and particle for each phrase of the translated sentence.

First, in step 41, the content of the bilingual structure clause of the word to be edited is rewritten as follows. That is,
If the editing instruction is clause → clause, the content of the literal translation in Dictionary 3 is used as the translation word, and the part of speech is set as the nominal word. If the editing instruction is clause → clause, the content of the literal translation in Dictionary 3 is used as the translation word, and the part of speech is set as the predicate. do. Note that 36a and 36b in FIG. 3 are examples of this.

Furthermore, in step 42, since the part of speech of the word to be edited has changed, the particle information in the phrase to be edited is adjusted. That is, the dictionary 3 is searched using the particle information in the clause to be edited as a heading. The dictionary description contents when the part of speech in the dictionary 3 is a particle are shown at 25 and 26 in FIG.

[0020] When the editing instruction is clause → phrase, if there is formal noun information in the clause to be edited, the formal noun information is deleted because the particle in the clause to be edited is a particle attached to the nominal. . If there is no formal noun information in the phrase to be edited, the particle in the phrase to be edited is a particle attached to a predicate. The header of the particle attached to the nominal word in the dictionary 3, that is, the content of the front body translation information is used as new particle information.

[0021] In addition, when the editing instruction is phrase → clause, if the dictionary contains the heading of the particle attached to the predicate, that is, the pre-translation information, when searching by particle information, the contents of the pre-translation can be searched. This is new particle information. If there is no previous translation information, formal noun information "koto" is supplemented.

Next, in step 43, for each of the modifying elements modifying the word to be edited, the conjugation and translation are adjusted because the part of speech to which the word to be edited, that is, the modifying element depends, has changed. We focus on the fact that the use of modifier elements and translation adjustment processing only need to be done on the parts that directly depend on them. 33a in FIG. 3 is applied to 36a through 34a, so it is not applied directly. Therefore, in the example of FIG. 3, adjustment processing for "this" 33a is unnecessary.

In addition, in steps 44 and 45, when the modifier clause is dependent on a particle, the clause of the modifier element being processed is called the target modifier clause, and the particle of the target modifier clause is Replace the conjunctive or conjunctive translation of the translation. When the target modifier clause is dependent on a case relation that exists in the case frame information of the new translation target word replaced in step 41, this is called an essential case, and the case relation in the case frame information and the particle are From the correspondence table, 34 in Figure 3
As shown in a and 34b, the particles of the modified clause of interest are set.

On the other hand, in the case of non-essential case, if the editing instruction is clause → clause, the target modifier clause particle is replaced with the adnominal translation from Dictionary 3, and when the editing instruction is clause → clause, Replace the focused modifier clause particle with the conjunctive translation (25 in Figure 2)
).

[0025] Then, in steps 46 and 47, if the modifier depends on a dependent without a particle, and if it is a conjugated word, then, based on the conjugated form of the modifier, the modifier is changed to adjunct → adnominal or adjunctive → adjunctive. (See 35a and 35b in FIG. 3).

Next, in step 48, if the modifying phrase is dependent without a particle and is not a conjugated word, and if the editing instruction is clause → phrase, the modified phrase is selected using the adnominal translation in Dictionary 3. If the translation is replaced and the editing instruction is phrase → clause,
Replace the target modified bunsetsu translation with the conjunctive translation from Dictionary 3.

Finally, in step 49, the translated text is replaced by following the bilingual structure rewritten as described above.

Furthermore, the bilingual translation output section 6 extracts bilingual translation from the bilingual translation structure area 5 and outputs it to the terminal device 7.

[0029]

[Effects of the Invention] As explained in detail above, the machine translation system of the present invention includes a dictionary that has not only translations of parts of speech corresponding to the parts of speech of the source language but also translations that correspond to syntactic changes in the target language; Translation processing that creates a translated text and bilingual structure data of the translation result by referencing the original text and writing it in the bilingual structure area, the original text to be translated, the translated text of the translation processing result, and information such as dependency, part of speech, conjugation, inflection, etc. There is a bilingual structure area that holds the bilingual structure, which is a tree structure in which is added to the node of each target clause, and an editing area that refers to the bilingual structure area and dictionary according to editing instructions from the user, edits the bilingual structure, and writes it to the bilingual structure area. This feature has the effect of making it possible to proofread the translation into a perfectly natural translation with simple operations in a style that matches the user's translation editing instructions.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of a machine translation system of the present invention.

FIG. 2 is a diagram showing an example of the dictionary 3 in FIG. 1.

3 is a diagram showing an example of the contents of the bilingual structure area 5 in FIG. 1. FIG.

4 is a flowchart showing an example of phrase conversion editing operation in editing process 4 of FIG. 1; FIG.

[Explanation of symbols]

1 Instruction input section 2 Translation processing 3 Dictionary 4 Editing process 5 Bilingual structure area 6 Translation output section 7 Terminal device

Claims (1)

[Claims] [Claim 1] A dictionary having not only translations of parts of speech corresponding to parts of speech of the source language but also translations corresponding to syntactic changes in the target language, the original text to be translated, the translated text of the translation processing result, dependency, part of speech, conjugation, A bilingual structure area that holds a bilingual structure, which is a tree structure in which information such as inflection is added to the nodes of each target phrase, and a translated sentence and bilingual structure data of the translation result obtained by translating the original text with reference to the dictionary are created, and Machine translation comprising a translation process of writing in a bilingual structure area, and an editing process of editing a bilingual structure by referring to the bilingual structure area and the dictionary according to an editing instruction from a user, and writing the edited bilingual structure in the bilingual structure area. system.