JPH04153878A - Pre-editing assisting processor of machine translation device - Google Patents

Abstract

PURPOSE:To select a proper equivalent by displaying a sentence which is so rewritten that the structure is made clear as to a predetermined sentence pattern in a source sentence as a result of a sentence analysis and passing the rewritten sentence to a machine translating process. CONSTITUTION:An analytic part 2 for document structure analyzes information regarding the constitution of the whole document which is inputted and sends the result to a sentence analytic part 3 together with the input sentence. The sentence analytic part 3 analyzes the input sentence and transfers part of the analytic result to a structuring process part 4, which converts the received analytic result into a sentence whose structure is clear and displays it at a display part 5. Over a look at this display, a user receives or reject the rewriting result selectively according to the display matches his or her thinking interpretation. Consequently, when the source is vague and possibly translated differently, and even when the user does not know the target language, the proper equivalent can be selected.

Description

[Detailed Description of the Invention] [Summary] This invention relates to a pre-editing support processing device in a machine translation device that supports a user in the machine translation device.

The purpose is to make it easier for users to rewrite the original text into a form that is unambiguous, such as when the original text has ambiguity and there is a possibility that multiple translations may be obtained as a result of translation.

Regarding the analysis results by the sentence analysis unit in the machine translation device, we have a structuring processing unit that rewrites sentences of a predetermined pattern into a form that makes the structure clearer, and displays the processing results by the structuring processing unit. Configure it to do so.

[Industrial Field of Application] The present invention provides support for users in machine translation devices.
This invention relates to a pre-editing support processing device for machine translation equipment.

With current machine translation devices, it is difficult to analyze the structure of a single sentence without error due to the lack of semantic information and the inadequate descriptive ability of the parser. Therefore, in order to obtain correct translation results, processing (pre-editing) is required to unambiguously define the meaning of the original text.

[Conventional technology]

Currently, the pre-editing process takes a lot of effort and is an impediment to improving the efficiency of translation work using machine translation. The following reasons can be considered for this.

1) The editor has poor operability and takes time to rewrite.

2) Since the processing content of the translation system is not clearly defined, it is not clear how to modify the original text to achieve translation.

Particularly in the case of long sentences, the success rate of translation for sentences that are not subjected to 5-pre-editing is extremely low. Furthermore, even when translation was successful, the translated text output by machine translation systems was often extremely difficult to read. In this case, the two workers had to collect useful partial translations from the system's output and construct their own translations, resulting in a decrease in work efficiency.

That is, in order to obtain a correct translation from a conventional machine translation device, it was necessary for a user to rewrite the original text into a form that the system could correctly interpret. However, this work requires a lot of effort and effort by people who are familiar with system processing, and l! This greatly increased the cost of the entire translation process using machine translation.

Automatic sentence segmentation systems have already been announced as systems that automatically perform this process, but they have had many problems with processing accuracy and have not been practical.

Furthermore, little consideration has been given to the readability of the translated text, which also increases the effort required to check the translation results, which in turn causes an increase in the cost of translation.

[Problem to be solved by the invention]

When a beginner uses the system, it is difficult to know how to modify the sentence to be translated in order to obtain a correct translation result. Therefore, if you apply it to the system without modification,
As a result, there was a problem that a considerable number of errors were mixed into the translation. On the other hand, even for those who are familiar with the processing logic of the system, pre-editing is extremely inefficient as it requires almost complete rewriting of one sentence.

Furthermore, the translation results of machine translation systems are not necessarily easy to read, making it difficult to understand the translation results or correct the translation results to obtain the final translation.

The present invention aims to make it easier for users to rewrite the original text into an unambiguous form in cases where the original text has ambiguity and there is a possibility that multiple translations will be obtained as a result of translation.

[Means for Solving the Problems] FIG. 1 shows a diagram of the basic configuration of the present invention. Reference numeral 1 in the figure is an input document to be translated. 2 is a sentence structure analysis unit which converts the input document 1 into, for example, a title, a paragraph 1, a bar graph 2, and so on. 3 is a sentence analysis unit that analyzes the input sentence. 4 is a structuring processing unit which converts the received analysis result into a sentence with a clear structure.

A display section 5 displays the conversion results. Reference numeral 6 represents a separate result search unit that obtains another analysis result obtained in the sentence analysis unit 3.

In the present invention, (1) when there is ambiguity in the original text when using a machine translation device, the possible analysis is shown by rewriting the original text, and the system allows beginners to modify the original text in a way that allows them to obtain a correct translation. It is simplified so that rewriting can be performed simply by selecting from among the shown rewriting possibilities. In addition, the readability of the translation result is improved by rewriting the sentence so that the structure of the sentence is clearly shown. along with the input sentence.

0 Sentence analysis unit 3 sends text to sentence analysis unit 3 The sentence analysis unit 3 analyzes the input sentence and transfers a part of the analysis result to the structuring processing unit 4. The structuring processing section 4 converts the received analysis result into a sentence whose structure can be clearly understood, and displays it on the display section 5. The user, looking at the display, selects whether to accept or reject the rewritten result depending on whether or not it matches his/her own interpretation.

If rejected, the system transfers yet another analysis result from the sentence analysis section 3 and entrusts it to the processing of the structuring processing section 4 again. If accepted, the analysis results will be output as is. At this point, the interpretation of the ambiguous part has been made clear and approved by the user.

Of course, the above case is a case of processing whose purpose is to resolve ambiguity in translation through interaction with the user. If the aim is simply to improve the readability of translation results, the separate result search unit 6 for dialogue may be omitted.

〔Example〕

FIG. 2 shows the configuration of an embodiment of the structuring processing section, and FIG. 3 shows an example of its processing. In FIG. It represents a processed body.

The constituent word string analyzer 41 analyzes the words making up the input sentence and determines the part of speech corresponding to each word.

The structuring processing unit 42 extracts a predetermined sentence pattern and converts the sentence having the sentence pattern into a clearer sentence.

The processing example shown in FIG.

This shows an example of converting a sentence with a sentence pattern given by the conjunctive form of the verb + the final particle ``de'' into the form of the final verb + ``,'' + ``and''. In the case of ``Processing Example 1,''``delete'' is conjugated in five lines, and the input sentence ``Delete the file and finish the process.'' means ``Delete the file.''

Then the process is finished. ” is converted into the form.

In the case of ``Processing Example 2,''``delete'' is conjugated, and the input sentence ``Delete the file and finish the process.'' means ``Delete the file.''

Then the process is finished. ” is converted into the form.

In this process, 1, for example, performs rA, B, and C. ” If they are consecutive, use rA. And do B. And do C. ”.

FIG. 4 shows another embodiment of the structure of the structuring processing section, and FIG. 5 shows an example of its processing. In FIG. 4, reference numeral 4 is a structuring processing unit, 43 is a morphological analysis unit 544 that scans a specific word string and assigns an evaluation value (hereinafter abbreviated as evaluation value assignment unit), 45
46 represents a structured processing body, and 46 represents a structured processing control body.

In the case shown in Figure 4, based on the obtained word string,
Sentences in which the evaluation value assigning body 44 is 1, for example, three or more predicates are examined. Then, the construction processing unit 4 converts the sentence into a sentence having two or less predicates only when there are three or more predicates. That is, the evaluation value giving unit 44 gives the evaluation value ``number of predicates J'' to the control unit 46, and the control unit 46 checks whether the condition of ``3 or more predicates'' is met, and the structuring processing unit 45 'on' indicates whether or not to convert the original text.
10ff control signal.

The processing example shown in FIG. 5 represents an example in which "sentence segmentation processing is performed" when "three or more predicates exist".

In the case of illustration (1), since there are two predicates in the sentence "Delete the file and end the process," the two-sentence segmentation process is not performed. In contrast, one diagram is shown (
In the case of 2), save the data, delete the file, and finish the process.In the sentence J, there are 3 predicates, so 9 sentence division processing is performed.In this example, this is the case. is a line declension conjugation that ``evacuates'' and ``eliminates''.
Since it is a suru line declension conjugation, it is translated into the form ``Saving the data and deleting the file.And finishing the process.''

FIG. 6 shows still another embodiment of the structuring processing unit, FIG. 7 shows an example of structuring using sentence structure information, and FIGS.
The figures show examples of processing results.

In FIG. 6, reference numeral 4 represents a structuring processing unit, 43 represents a morphological analysis body, 47 represents a sentence structure analysis body, and 48 represents a structuring processing body.

In the case shown in FIG. 6, as will be explained with reference to FIG. 7, a sentence structure analyzer 47 clarifies the hierarchical relationship within the sentence, and a structuring processor main body 48 converts the original text.

FIG. 7(A) shows a state in which the sentence structure of the original sentence ``The user deletes 5 files and finishes the process.'' is clarified. In the figure, S represents a sentence, Vp represents a predicate, and pp represents a clause. In FIG. 7(A), (i) r
The part J after deleting the file and the part (ii) "Finish the process." correspond to the same predicate "The user is." As a result, the result is as shown in Fig. 7(B) and Fig. 7(C).
, divided into two sentences, as in the case of Figure 7(C),
The E user in Figure (B) is
), ``He is trying to compensate for J.

Therefore, as shown in the processing example shown in FIG. 8, the original sentence "The user deletes the file and finishes the processing."

This would be translated as "The user deletes the file. J [and he finishes the process."]. The case shown in FIG. 9 is converted in the same way as the case shown in FIG. 8, but the part of speech of the word is also displayed.

〔Effect of the invention〕

As explained above, according to the present invention, it is possible to select an appropriate translation even when there is ambiguity in the original text and there is a possibility of multiple translations, and when the user does not know the target language. Become. Further, according to the present invention, even if the user has no knowledge of sentences within the range that a machine translation system can process and does not know the target language, it is possible to select an appropriate translation.

In the present invention, it goes without saying that only one of the configurations shown in FIGS. There may be.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the principle configuration of the present invention, FIG. 2 is an embodiment of the structuring processing section, FIG. 3 is an example of the processing shown in FIG. 2, and FIG. 4 is another implementation of the structuring processing section. For example, FIG. 5 is a processing example for the case shown in FIG. 4, and FIG. 6 is a further example of the structuring processing section,
FIG. 7 shows an example of structuring using sentence structure information. FIGS. 8 and 9 show examples of processing results, respectively. In the figure, 1 represents an input document 52, a sentence structure analysis section, 3 a sentence analysis section, 4 a structuring processing section, 5 a display section, and 6 a separate result search section. Structuring processing section Fig. 2 Fig. Processing example Fig. 5 Structuring processing section Fig. 6 Example of outputting a character string as the final result Fig. 8 Single l as the final result! Figure 9 of I1 outputting chestnuts

Claims (4)

[Claims] (1) For a given input document (1), the sentence structure analysis unit (2) performs sentence structure analysis, and for each sentence, the sentence analysis unit (3) performs sentence analysis including determining the part of speech of the word.
In a machine translation device that machine-translates the original text in the given input document (1), based on the result of the sentence analysis performed by the sentence analysis unit (3), a predetermined sentence pattern in the original text is analyzed. , comprises a structuring processing unit (4) that rewrites the structure to make the structure clearer, and a display unit (5) that displays the rewritten sentence as a result of the structuring processing unit (4). A pre-editing support processing device for a machine translation device, characterized in that a rewritten sentence whose structure has been clarified is taken over to machine translation processing. (2) The structuring processing unit (4) is characterized in that the sentence pattern having the conjunctive form of the verb and the final particle in the original sentence is converted into the form of the final form of the verb and the conjunction. A pre-editing support processing device in a machine translation device according to claim (1). (3) The structuring processing unit (4) includes a assignor (44) that outputs an evaluation value for a specific word string in the original text, and determines whether to classify the original text based on the evaluation value. 2. The pre-editing support processing device in a machine translation device according to claim 1, further comprising a control body (46) for instructing. (4) The structuring processing unit (4) examines the sentence structure of the original text, divides the original text, and supplements each or one of the divided sentences with clauses. A pre-editing support processing device in a machine translation device according to claim (1).