JP3477308B2 - Machine translation equipment - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device for converting a document in one language into a document in another language.

[0002]

2. Description of the Related Art Most of the document data handled by computers such as word processors and computers, including machine translation devices, are not limited to the character information that is the character data itself.
It includes information such as document structure, layout and character modification information, such as ruled line addition and typeface change information.

Several standards have already been proposed as a method of adding these pieces of information (hereinafter referred to as additional information), and SGML for describing the logical structure of a document and hypertext documents for describing. HTML and the like are known.

In such a situation, even in the field of machine translation, a portion of a normal document is translated without damaging the additional information contained in the data of the original document as much as possible. The translation result document translated by this device can be output in almost the same format as the original document data.

In such a machine translation device, consider a process of processing a sentence in which additional information is added to a specific section by using symbols called start tags / end tags.

Time flies <itaric> like an arrow </ it
aric>. In the above sentence, “<itaric>” and “</ itaric>” are the start and end tags, respectively. Both of these tags output the section sandwiched between these two tags in italics. It means that.

This sentence is converted into a translated sentence having the same additional information by the processing procedure described below. Description will be given based on the processing block diagram of FIG. 7. First, the tag separation unit 2
Then, the tag appearing in the input document 1 is identified, and which word the tag information is added to is stored. FIG. 8 is a diagram showing a storage state of the tag information (original text tag information 6). After storing the tag information, the tag is deleted from the original text.

Next, the translation unit 3 translates the sentence. The correspondence information between the original sentence and the translated word obtained in the translation process is stored. FIG. 9 is a diagram showing an example of translated word correspondence information (translated word correspondence information 7).

Further, the tag restoration unit 4 newly tags the translated sentence based on the translation result obtained by the translation unit 3, the word correspondence information, and the original sentence tag information stored in the tag separation unit. Is given. It is a figure which shows the translated text tag information from which FIG. 10 is obtained. A tag is added to the translated text based on this, and the output document 5
Is output.

The translation result obtained in this example is as follows.

Time is like <itaric> arrow </ itaric>
Fly

[0012]

Here, it is assumed that the following sentence is translated in the above-mentioned method.

<Tag1> Time flies </ tag2> like an <tag
3> arrow </ tag3>. In this sentence, the tag correspondence is incorrect. Therefore, the effective range of <tag1> and <tag2> is not fixed. If the tag represents information regarding the display format of characters, the display device (output device) that receives this sentence notifies the user that there is an error in the tag and interrupts the process, or Interpret the tag as invalid, or automatically change the tag so that it does not conflict and process it.

However, generally, as a user who tries to translate this sentence, if the tag becomes invalid, the tag information will be lost. This may cause not only the sentence but also the display format of the document to be significantly different from the original document data.

For example, in the above example sentence, it is assumed that the effective range of the start tag <tag1> is just before </ tag2> where an incorrect end tag appears, and the end tag does not have a corresponding start tag.
When </ tag2> is disabled, the translation result is as follows.

<Tag1> time </ tag1> is <tag3> arrow </ tag
3> like <tag1> fly </ tag1>.

If all the tags with incorrect correspondence are invalid, the time jumps like <tag3> arrow </ tag3>.

[0018]

In both of the above two results, the tag information of the original text is not a little missing. Therefore, as described above, there is a possibility that the problem that the display format changes greatly occurs.

In the present invention, even if such an erroneous tag is included, the translation device accurately reflects the tag information of the original text without interpreting the effective range of the tag arbitrarily. The purpose is to obtain the output result of the translation process in the form.

[0021]

According to claim 1 of the present invention, character data which is an original sentence described in a first language,
A pair of tags indicating the start position and end position
Input sentence consisting of additional information that adds various information between
And an input unit for inputting the character data and the additional information.
The relationship is stored as original text tag information, and
The tag separating unit for separating the additional information and the tag separating unit.
And translate the rest of the input text, excluding the tags, into the desired second language.
Translated, and the correspondence between the first language and the second language as translated word correspondence information
The translation unit to be stored and the second word obtained by the translation unit
In the sentence of the word, the tag temporarily separated by the tag separation unit is restored.
Tags are added by the original tag restoration unit and the tag restoration unit.
The translated text of the second language to a machine translation device equipped with an output unit.
Then, the tag correspondence inspection unit makes a pair in the sentence of the tag.
A pair of keys that indicate the start and end positions by inspecting the correspondence
Section that has an error in the correspondence
Is a pair of tags divided by the tag correspondence inspection unit.
Input sentence in the section that is judged to have an error in the correspondence,
Input sentence before the section and after the section
Characterized by the translation processing of the input sentence and
It

According to claim 2 of the present invention, the tag correspondence
The inspection unit determines that there is an error in the tag correspondence.
For the input sentence in the section, the translation unit performs translation processing.
Output the original text as it is in the first language, without prompting
It is characterized by

[0023]

[0024]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the present invention. The input document 1, the tag separation unit 2, the translation unit 3, the tag restoration unit 4, the output document 5, the original sentence tag information 6, and the translated word correspondence information 7 have the same configurations as those of the conventional technique shown in FIG. The description here is omitted.

The tag correspondence inspection unit 8 will now be described in detail. FIG. 2 is a diagram illustrating the tag correspondence inspection unit by disassembling it for each finer process. The character string buffer 81 is
It is a buffer that sets the input document as it is, that is, with the tag attached, for each sentence.

The tag identification unit 82 is a means for scanning the set character string buffer from the beginning and sequentially acquiring the appearance position and tag name of the tags included in the sentence. The tag acquired by the tag identification unit 82 is stored in the tag buffer 83.

The tag stack 84 is a last-in first-out stack for storing the state of the tag up to the current scanning position of the character string buffer. When the tag identification unit 82 detects the start tag, the contents and position of the start tag buffer are stored.

The tag comparing section 85 is means for comparing the end tag buffer with the start tag at the top of the tag stack 84 when the end tag is detected by the tag identifying section 82. That is, if there is a tag correspondence error in the sentence set in the character string buffer, the tag comparison unit 85 detects the error. The sentence division processing unit 86 is means for dividing a sentence immediately before and after the apologized tag when a tag error is detected by the tag comparison unit 85.

The divided result is the sentence first half buffer 86.
a, error tag buffer 86b, sentence latter half buffer 86c
Stored in each. The contents of the first sentence buffer 86a are translated according to the conventional translation process ([1] in the figure).
Processing). After outputting the translation result, the contents of the error tag buffer are output as they are (process [2] in the figure), then the contents of the latter half of the sentence buffer are set again in the character string buffer 81, and the process is repeated ([[] in the diagram]). 3]). When no tag correspondence error is detected in the sentence, the sentence division processing unit 8
6 does not perform any processing, and the content of the character string buffer 81 is directly sent to the translation processing (processing [4] in the figure).

Next, the flow of processing will be shown using the flowchart of FIG. First, the input sentence to the tag correspondence inspection unit 8 is stored in the character string buffer 81 in step S1. Next, in step S2, the character string buffer scanning position of the tag identifying unit 82 is initialized, and the scanning position pointer is set to zero.

Steps S3, S4, S5, S5, S8,
In S10, the tag identifying unit 82 cuts out a tag in a sentence. The character string buffer 81 is scanned sequentially from the front to determine whether the scanning position is the tag start symbol (“<” in the example) (step S3).

If the scan position is not the tag start symbol,
The tag pointer is advanced (step S4). Further, it is checked whether or not the result of moving the pointer has reached the end of the sentence (step S10). When the scanning position is the tag start symbol, the pointer (position) is set to the tag buffer 83.
(Step S4), the scanning is advanced to the tag end symbol (“>” in the example), and the tag name is stored in the tag buffer 83 (Step S5).

Here, it is checked whether or not the tag name starts with "/" (step S6). Tag name is "/"
If it does not start with, it is determined to be a start tag, and the pair of the tag name stored in the tag buffer and the value of the pointer at the tag start position is stored in the tag stack 84 (step S7).

If the tag name starts with "/", it is an end tag. Therefore, it is checked whether the tag stack 84 is empty (step S9). If it is not empty, the tag stack 84 is at the top. The contents of (the most recently stored one) are taken out, and the contents of the tag buffer 83 and the tag name are compared (step S14).

If the tag name extracted from the tag stack 84 and the tag name in the tag buffer 83 are the same, it is determined that the start tag and the end tag have the correct correspondence, and the process is continued (return to step S3). Here, an example of the scanning position pointer in the character string buffer is shown in FIG. 4, and an example of the tag stack 84 is shown in FIG. When the pointer comes to the position 24 shown in FIG. 4, it can be seen that the three tags are stacked in the tag stack 84 shown in FIG.

When the scanning position reaches the end of the sentence in the character string buffer by repeating the above process (step S10),
If the tag stack 84 is empty, it means that there is no tag correspondence error in the input sentence, and therefore the contents of the character string buffer are directly passed to the conventional machine translation process (step S13). .

There is an error in the tag correspondence as described in the following 3
Divided into patterns. One is when the end tag appears but the tag stack is empty.
This is because the end tag is stored in the tag buffer 83 in the tag comparison unit 85 (the end tag is detected by scanning).
This is an error detected when the tag stack 84 is empty during the operation. In the flowchart, it is detected in step S9.

In response to the detection of an error, the sentence division processing unit 86 causes the character string buffer 81 to include a portion immediately before the end tag, an end tag, and a portion after the end tag in the sentence front half buffer 86a and the error tag buffer, respectively. 86b and the latter half of the sentence buffer 86c are stored (step S17).

Then, the contents of the sentence first half buffer 86a are transferred to the translation process (step S18) . Its Re followed by outputting the contents of the error tag buffer 86b (step
S19) . Following that, the latter half of the sentence buffer 86
The content of c is output to the character string buffer 81 (step S20). In this way, the contents of the character string buffer 81 are rewritten. Then, the process returns to step S2.

Second, the tag name of the end tag that appears is
This is a case where the tag name of the start tag stored in the tag stack 84 does not match. This is because, in the tag comparison unit 85, the start tag stored in the tag stack 84 when the end tag is stored in the tag buffer 83 (the end tag is detected by scanning), and the end tag in the tag buffer 83. Is an error that is detected when does not correspond. In the flowchart, it is detected in step S14.

When an error is detected, the contents of the tag stack 84 are all taken out and the appearance position pointer of the start tag at the bottom of the stack is acquired (step S16). Upon detection of an error, the sentence division processing unit 86 divides the character string buffer into a portion before the tag, a portion after the tag, and a portion after the tag (the division is similar to the above-described processing), and each of them is divided into buffers 86a to. It is stored in 86c. Then, similarly, the contents of the latter half of the sentence are newly set in the character string buffer 81, and the process returns to step S2.

The third pattern is a case in which tags remain in the tag stack 84 at the time when scanning is completed up to the end of the buffer. This is an error detected when the tag identifying unit 82 scans the character string buffer 81 to the end of the sentence and there are tags remaining in the tag stack 84.
In the flowchart, it is detected in step S12.

When an error is detected, the sentence division processing unit 86
Fetches all the contents of the tag stack 84 and acquires the appearance position pointer of the start tag at the bottom of the stack (step S16). Upon detection of an error, the sentence division processing unit 86 divides the character string buffer into a portion before the tag, a portion after the tag, and a portion after the tag (the division is similar to the above-described processing), and each of them is divided into buffers 86a to. It is stored in 86c. Then, similarly, the contents of the latter half of the sentence are newly set in the character string buffer 81, and the process returns to step S2.

The processing up to this point will be described based on an actual input sentence. As the input sentence, the above <tag1> Time flies </ tag2> like an <tag3> arrow </ t
Use ag3>.

First, <tag1> is stored in the tag buffer 83, and the tag comparison unit 85 finds that it is a start tag and stores it in the tag stack 84. Next, </ tag2> is stored in the tag buffer 83, and the tag comparison unit 85 determines that the tag is the end tag.
Since it is <tag1> and does not match, the tag names do not match.
Therefore, since <tag1> is divided as an error tag, the sentence division processing unit 86 stores “” in the sentence first half buffer 86a.
(Empty) and “<tag1>” in the error tag buffer 86b,
In the latter half of the sentence buffer 86c, "Time flies </ tag2> lik
e an <tag3> arrow </ tag3>. ”is stored.

Then, the contents of the sentence first half buffer 86a are handed over to the translation process. In this case, since the empty (empty sentence) is passed, the translation result is also empty (empty sentence).

Then, the tag "<tag1>" stored in the error tag buffer 86b is output.

Next, "Time flies </ tag2> likean <tag3> arrow </ t stored in the sentence latter half buffer 86c.
“Ag3>.” is set again in the character string buffer 81, and the process is repeated. In this case, since the tag stack 84 is empty when </ tag2> is found, the start tag becomes insufficient.

The error tag becomes "</ tag2>", "Time flies" is stored in the sentence first half buffer 86a, "</ tag2>" is stored in the error tag buffer 86b, and the sentence latter half buffer 86c.
"Like an <tag3> arrow </ tag3>." Is stored in each.

The content of the sentence first half buffer is passed to the translation process, and "time flies" is obtained as the translation result. Following that, the error tag "</ tag2>" is output.
In addition, the “like an” stored in the sentence latter half buffer 86c
"<tag3> arrow </ tag3>." is returned to the character string buffer 81 and the process is repeated. In this case, since there is no tag correspondence error, the sentence is not divided and is passed as it is to the translation process. The translation result is “<tag3> arrow </ tag3>.
like. Is obtained and output.

That is, the following translated text will be finally obtained <tag1> Time flies </ tag2><tag3> Arrow </ tag
Like 3>.

Thus, in the translation result according to the present invention,
The translation result is output in the state where the context of the tag and the word match the original sentence, that is, without distorting the tag information.

When a tag error is detected, the sentence may be output as the original sentence. A flowchart of this process is shown in FIG. Steps S1 to S14 are the same as in FIG. Step S9, Step S12,
When the tag correspondence error is detected in any of the steps S14, the original sentence is output as it is without being processed in a step S21.

[0054]

With the apparatus according to the present invention, the correspondence between the start tag and the end tag is inspected and detected for each sentence of the input document.
For a sentence that has an error in the correspondence of the specified tag,
To split the tag before and after the translation process and return the tag
In addition, the translation result can be obtained without distorting the tag information of the original text.
You can

[0055]

By outputting the original sentence as it is with respect to the sentence having an incorrect tag correspondence, it is possible to prevent the tag information from being lost or distorted.

This is very effective when it is desired to obtain the translation result in the same state as the original text as the tag information, which has a serious influence on the document data, even if the translation quality is slightly deteriorated.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration in an embodiment of the present invention.

FIG. 2 is a block diagram showing a detailed configuration of a tag correspondence inspection unit 81 in the embodiment of the present invention.

FIG. 3 is a flowchart showing a flow of processing in the embodiment of the present invention.

FIG. 4 is a diagram showing an example of a character string buffer 81 and an example of a scanning position pointer in one embodiment of the present invention.

FIG. 5 is a tag stack 8 according to an embodiment of the present invention.
It is a figure which shows the example of FIG.

FIG. 6 is a flowchart showing a flow of processing in the embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of a conventional technique.

FIG. 8 is a diagram showing a storage example of tag information in a conventional technique.

FIG. 9 is a diagram showing an example of storage of translated word information in the related art.

FIG. 10 is a diagram showing a storage example of translated text tag information in a conventional technique.

[Explanation of symbols]

1 Input document 2 Tag separation unit 3 Translation Department 4 Tag restoration section 5 Output document 6 original text tag information 7 Translation information 8 tag compatible inspection department 81 string buffer 82 Tag identification section 83 Tag buffer 84 tag stack 85 Tag comparison section 86 sentence division processing unit 86a sentence first half buffer 86b error tag buffer 86c sentence latter half buffer

Claims (2)

(57) [Claims] 1. An input unit for inputting an input sentence consisting of additional information for adding various information in the section by character data which is an original sentence described in a first language, and a pair of tags indicating a start position and an end position. , A tag separating unit that stores the relationship between the character data and the additional information as original sentence tag information and separates the additional information from the input sentence, and the remaining input sentence excluding the tags is desired by the tag separating unit. A translation unit that translates into a second language and stores the correspondence between the first language and the second language as translation correspondence information; and a sentence in the second language obtained by the translation unit, temporarily in the tag separation unit. A tag restoring unit for restoring the separated tag, and a sentence in the second language tagged by the tag restoring unit.
In a machine translation device having an output unit for outputting, a tag correspondence inspection unit that inspects a correspondence relation in one sentence of the tag and detects a section in which a correspondence relation between a pair of tags indicating a start position and an end position has an error. The translation unit includes an input sentence in a section divided by the tag correspondence inspection unit and determined to have an error in tag correspondence, an input sentence in front of the section, and a section after the section. A machine translation device characterized by performing a translation process on a given input sentence. 2. The tag correspondence inspection unit uses the first language without causing the translation unit to perform a translation process on the input sentence in the section in which the correspondence relation of the tags is determined to be erroneous. The output as the original text is output.
The machine translation device described in.