JPH0421176A - machine translation device - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention is directed to I! which performs natural language translation. This relates to machine translation devices.

BACKGROUND ART Conventional machine translation devices take the following measures when the translated word is a technical term that is not well recognized.

That is, (1) Use the translation as is.

(2) Do not use the word or phrase, but substitute the corresponding word or phrase from the original text.

(3) After the machine translation is finished, the user can
c Modify independently of the translation.

Such methods were used.

Problems to be Solved by the Invention In conventional machine translation devices that translate an original text into a translated text, there are cases where the words in the original text, especially technical terms, cannot clearly determine the corresponding words in the translated text, or even if they have been determined, they are insufficient. If consensus has not been obtained or if it is thought that many people do not understand the terminology, the machine translated text cannot be effectively utilized as is.

On the other hand, in general translated specialized books, sufficient information is provided to the reader by inserting the word or phrase in the original text that corresponds to the word or phrase in the translated text in parentheses. There is. For example, in English-Japanese translation, the 'natural language process that existed in the original text
The expression s-sing J was translated as "natural language processing," but in order to clarify its meaning for readers who are familiar with English expressions, it was translated as "natural language processing."
natural language process
g) It is often done to complete the original sentence by putting it in parentheses, such as ``.

However, with conventional machine translation technology, when a user wants to create a document in this format, the user has to insert it into the translated text using an editor, which is a problem that is very inefficient. there were.

In view of the above problems, it is an object of the present invention to make it possible to efficiently insert words and phrases from a corresponding original text into parentheses into a translated text.

Means for Solving the Problems To achieve this object, the present invention provides a source text buffer that holds the source text to be translated, and a source sentence structure buffer that holds the source sentence parse tree that is the result of parsing the source text. ,
A translation structure buffer that holds a target parse tree that is the result of converting a source parse tree, a translation buffer that holds a translation generated from the target parse tree, a translation editor that proofreads the contents of the translation buffer, and a translation editor that proofreads the contents of the translation buffer. Correspondence holding means that stores the correspondence with the phrase structure buffer, the correspondence between the original phrase structure Hanwha and the translated phrase structure buffer, and the correspondence between the translated phrase structure buffer and the translated sentence buffer, and the a specification means for allowing a user to specify a specific range in a translated text based on a syntax tree of the translated text, an extraction means for extracting a range of the original text corresponding to the range specified by the specification means, and a means for placing the extracted range of the original text in parentheses. and an insertion means for inserting the text immediately after the specified range in the translated text.

Effects In the above configuration, the present invention provides correspondence between ranges in the original text and nodes of the original text parsing tree that are the analysis results of the original text, correspondence between the parsing tree of the original text and the parsing tree of the translated text that has been converted, and By establishing a correspondence between the nodes of the parse tree and the ranges in the translated text, and further specifying the ranges in the translated text, the user can
Corresponding words and phrases from the original text can be automatically inserted into the translated text.

EXAMPLE Hereinafter, an example of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a machine translation device according to the present invention.

In FIG. 1, 1 is an input document to be input into a machine translation device, 2 is an original text buffer that holds the input document, 3 is a parsing unit that generates a syntax tree from a human document, and 4 is a parsing unit that generates a syntax tree from a human document. 5 is a conversion unit that converts the syntax tree of the original sentence into a syntax tree of the translated sentence; 6 is a translated phrase structure buffer that holds the converted syntax tree of the target sentence; 7 is held by the translated phrase structure buffer 6 8 is a translation buffer that holds the translation generated by the translation generation unit 7, and 9 is a translation for correcting the translation held by the translation buffer 8. 10 is a document that is the translation result; 11 is a translation node table for making correspondence between the translation syntax tree and the translated text; 12 is a notebook for the translation syntax tree corresponding to the words in the translation into which the user should insert the original text; Reference numeral 13 denotes an original text extraction unit that extracts words from the original text buffer 2, and 14 an original text insertion unit that inserts the original text extracted by the original text extraction unit 13 into the translation buffer 8.

The meaning of each arrow in FIG. 1 is as shown in the table below.

(Blank below C) Next, the operation of an embodiment of the machine translation device according to the present invention will be described.

In the process of machine translation, an input document 1 is first stored in an original text buffer 2. The syntax analysis unit 3 uses the original text buffer 2
The information is analyzed to generate a syntax tree of the original sentence, and the generated syntax tree is sent to the original phrase buffer 4. At this time, the contents of the original text buffer 2 are not destroyed and are retained as they are. Also, every node in the syntax tree has a pointer to the corresponding range in the original text buffer 2. This situation is shown in FIGS. 2(a) and 2(b).

In FIG. 2, 15 is a node of the syntax tree in the original sentence structure buffer 4, 16 is a word node at the lowest layer of this syntax tree,
17 is a band-like representation of the contents of the original text buffer 2 (second
(see figure (b)), 18 is a position in the original text buffer 2. Information about the range that the word occupies in the original text buffer 2 is added to the word node 16 in the lowest layer (21st EI
(see (a)).

When a certain node in the syntax tree is specified, the range of all word nodes controlled by that node is examined, and the minimum range that includes all of them is extracted from the source text buffer 2, thereby corresponding to the specified node. The portion can now be extracted from the original text buffer 2.

Next, based on the syntax tree sent to the original phrase structure buffer 4 in FIG.

When creating a syntax tree for this translated sentence, the original phrase structure buffer 4
The contents of are not destroyed, and pointers are generated from each node in the translated phrase structure buffer 4 to the corresponding node in the original phrase structure buffer 4. This situation is shown in FIG.

In FIG. 3, 19 is a node of the target sentence syntax tree generated from the source sentence syntax tree by the conversion unit, 20 is the lowest node that directly corresponds to a word, and 21 is a node of the target sentence syntax tree which corresponds to which source sentence syntax tree node at the time of generation. A pointer 22 indicating whether it has been converted to a node is a dummy node when there is no corresponding node in the original syntax. The pointer 21 is extended in the direction from the target sentence syntax tree to the original sentence syntax tree. As a result, once a node of the source sentence syntax tree is determined, the corresponding node of the target sentence syntax tree can be determined immediately.

The translated sentence generation unit 7 generates a translated sentence based on the syntax tree of the translated sentence stored in the translated sentence structure Hanwha 6, and stores it in the translated sentence buffer 8. At this time, a translated text node table 11 is prepared in which each node in the translated text structure buffer 6 and the position of the corresponding translated text in the translated text buffer 8 are registered. The correspondence between the translation syntax tree and the translation via the translation node table 11 is shown in FIG.

In FIG. 4, 23 represents the contents of the translated text buffer 8 in a band shape, and 24 indicates the position of the translation string within the translated text buffer. A node number is added to the dummy node 22 of the translation syntax tree, and the translation node table 11 is a table containing the range in the translation buffer 8 that corresponds to the node number.

By doing this, the range in the translation buffer 8 corresponding to the node of the translation syntax tree can be specified in the same way as the second time except that the node number is used. Conversely, if position 24 in the translation buffer 8 is specified, by referring to the translation node table 11,
A dummy node 22 of the translation syntax tree can be selected. To create this translated text node table 11, when storing the word corresponding to each node in the translated text buffer 8, the position and node of the word in the translated text may be registered in the translated text node table 11.

The user proofreads the translated text in the translated text buffer 8 using the translated text editor 9. At this time, by pressing a specific key, the machine translation device can be instructed to complete the original text. When this key is pressed, the node determination unit 12 starts processing. The node determining unit 12 consults the translation node table 11 to determine the lowest dummy node 22 of the translation syntax tree corresponding to the current position in the translation buffer 8 of the system. At the same time, the range in the translated text buffer 8 described in the translated text node table 11 is displayed in reverse video to notify the user that the original text corresponding to this range will be complemented. The user is
This range can be changed to a wider one by pressing another key.

The details of the above processing of the node determination unit 12 will be explained using FIG. 4.

Assume that the cursor is now at position n+4.

When the user instructs complementation of the original text, the translated text node table 11 is looked up, node A is determined as the corresponding node, and the corresponding range a in the translated text buffer 8 is highlighted. Next, when a key for specifying expansion of the range is pressed, the node B that controls the lowest layer node A in the currently determined translation buffer 8 is determined as a new range. Then, by using all the nodes controlled by node B as keys, look up the translation node table 11 in the opposite direction, and note
Find the range corresponding to , and highlight the corresponding part. As enlargement is further specified, the nodes are changed one after another in the order of C, D, and E, and accordingly, portions c, d, and e in the buffer become targets for inverted display.

At this time, if the node of the source syntax tree corresponding to that node is the dummy node 22, that node is skipped. The user can also specify range reduction by pressing another key. When reduction is specified, the state returns to the state before enlargement. If you specify further reduction, it returns to the previous state. This can be achieved by storing the node history of a given syntax tree on a stack.

The designated node is sent to the original text extraction unit 13. The original text extraction unit 13 first follows the pointer 21 in FIG. 3 and determines the node of the original text syntax tree. Next, as explained in FIG. 2, the original text bar 1. Determine the corresponding range of file 2, copy that part of the string, and insert it into original text insertion section 1.
Pass it to 4.

The original text insertion unit 14 places the sent string in parentheses as it is and inserts it into the first specified translation buffer 8 immediately after the range specified by the user. Since the word immediately after the insertion position is shifted back by the inserted amount, the range of the translation node table 11 is rewritten. Furthermore, the inserted range in the translation buffer 8 and the node of the translation syntax tree targeted for insertion are registered in the system.

If the inserted range is expanded or reduced, the inserted range is deleted from the translation buffer 8, the translated text node table 11 is returned to its original state, and then a new node is inserted.

When the insertion is completed, control returns to the translation editor 9, and the user can continue proofreading. If the user deletes, adds, or changes part of the buffer during proofreading, the translation node table 11 is rewritten to maintain the correct correspondence for the parts that were not affected by the processing. . This allows you to use the insert function at any time.

As described in the above embodiments, the corresponding portion of the original text is inserted into the translated text buffer 8 by a simple key operation during the user's translation.

To insert a word, use the translation buffer 8.
Move the cursor over the word or phrase you want to specify,
Just hit the key. Inserting a wider range of original text often requires only an additional - or double key press.

There is no need to specify the range for each character or the insertion position for each cursor, just place the cursor at any position on the target word and press the key and the range and insertion position will be automatically set. can be determined. Having such a function is extremely useful when translating documents that use many technical terms and proper nouns, and facilitates proofreading work.

Effects of the Invention As described above, the present invention provides an original text buffer that holds the original text to be translated, an original phrase structure buffer that holds the original text parse tree that is the result of parsing the original text, and a text buffer that holds the original text syntax tree that is the result of parsing the original text. A translation structure buffer that holds a translation syntax tree, a translation buffer that holds a translation generated from the translation syntax tree, a translation editor that proofreads the contents of the translation buffer, and the correspondence between the original text buffer and the original text structure buffer. , a correspondence holding means for holding the correspondence between the original phrase structure buffer and the translated phrase structure buffer, and the correspondence between the translated phrase structure buffer and the target sentence buffer, respectively, and a specifying means for causing a user to specify a specific range in the translated text; an extracting means for extracting a range of the original text corresponding to the range specified by the specifying means;
The apparatus is configured to include an insertion means for inserting the extracted range of the original text into parentheses immediately after the range in the specified translated text.

With this configuration, it is possible to efficiently insert words and phrases from the original text into parentheses into the translated text.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the machine translation device according to the present invention, and FIG. 2 is a conceptual diagram explaining the operation of the machine translation device in the embodiment of the present invention with regard to the correspondence between the original text buffer and the original sentence structure buffer. 3 is a conceptual diagram illustrating the operation of a machine translation device in an embodiment of the present invention and the correspondence between an original phrase structure buffer and a translated phrase structure buffer, and FIG. 4 is a conceptual diagram illustrating the operation of a machine translation device in an embodiment of the present invention. FIG. 3 is a conceptual diagram illustrating the operation of the translation buffer, the correspondence between the translation buffer and the translation phrase structure buffer via the translation node table, and the operation when the user changes the range specification. 1...Input document, 2...Original text buffer, 3...Syntax analysis section, 4...Original sentence structure buffer, 5... Conversion unit, 6... Translation structure buffer, 7... Translation generation unit, 8...
...Translation buffer, 9...Translation editor,
10...Output document, 11...Translation node full, 12...Node determination unit, 13...
...Original text extraction part, 14...Original text insertion part, 1
5...node, 16...word node,
17... Original text buffer, 19... Node, 20... Node, 21... Pointer, 22... Dummy node, 23...・・・・・・
Original text buffer, 24... position. Name of agent: Patent attorney Shigetaka Awano

Claims (1)

[Claims] An original sentence buffer that holds the original sentence to be translated, an original phrase structure buffer that holds the source sentence syntax tree that is the result of parsing the original sentence, and a translated phrase that holds the target sentence syntax tree that is the result of converting the source sentence syntax tree. a structure buffer, a translation buffer that holds a translation generated from the translation syntax tree, a translation editor that proofreads the contents of the translation buffer, a correspondence between the original text buffer and the original phrase structure buffer, and the original phrase structure buffer. and a correspondence holding means for respectively holding the correspondence between the translated phrase structure buffer and the translated phrase structure buffer, and the correspondence between the translated phrase structure buffer and the translated phrase buffer, and a translated sentence based on the syntax tree of the translated sentence during the proofreading work by the translated sentence editor. a specifying means for causing the user to specify a specific range within the text; an extracting means for extracting a range of the original text corresponding to the range specified by the specifying means; and a specified translated text with the extracted range of the original text in parentheses. A machine translation device comprising: insertion means for inserting immediately after the middle range.