JPH03127264A - Mechanical translation system for sentence including parentheses - Google Patents

Abstract

PURPOSE:To obtain a proper translation to an input original sentence including the parentheses with the natural extension of a conventional mechanical translation system by holding the information on the parentheses in an internal structure and scattering the processes of the parentheses to the stages of a translating process. CONSTITUTION:A mechanical translating device consists of an input device 20, an input control part 21, an editing control part 22, a display control part 23, a display device 24, a translation system 25, a text storage part 26, a print control part 27, a printer 28, and a dictionary system 29. A secondary process is provided at each stage of a translating process in response to the process for parentheses. Thus, the information on the parentheses are held as the attributes contained in an internal structure. Thus, it is possible to accurately analyze the roles of parentheses against a whole sentence. In such a constitution, natural translation is obtained to an original sentence including parentheses with the natural extension of a conventional mechanical translation system.

Description

[Detailed Description of the Invention] [Field of Industrial Application] This invention expands the conventional machine translation method to make it possible to obtain natural translated sentences for a large amount of original text including parenthesized sentences. It is related to machine translation methods.

[Prior Art] Various machine translation devices have been developed that use computers to translate one language into another. Various measures have been taken to deal with external linguistic phenomena.

For example, FIG. 4 is a diagram showing the parsing process and generation process for parentheses in a conventional machine translation device shown in Japanese Patent Application Laid-Open No. 63-136262, and FIG.
68 is a diagram of the citation symbol analysis process and generation process in the conventional machine translation device shown in No. 68. Moreover, FIG. 6 is a diagram of the processing and internal structure for reference symbols in Japanese Patent Application Laid-Open No. 113668/1983.

In FIG. 4(a), at the beginning of the dictionary lookup morphological analysis section, the character pointer is set to the first word (40)L, the parentheses are treated as words, and it is determined whether the word is a left parenthesis (41). L, if the word is a left parenthesis, count the number of words up to the right parenthesis and add it to the number of the previous word (42)
, if there are no parentheses, while performing dictionary lookup processing (43),
Processing (41) to processing (43) are repeated.

In FIG. 4(b), the generation unit performs generation processing (45) for one word, and stores the translated word in a buffer (45).
6) L, at this time, determine whether the word is a left parenthesis (47) L, if the word is not a left parenthesis, continue the normal generation process (45), and if the word is a left parenthesis, , stores the words up to the right parenthesis in the translation word buffer (48). The above process is repeated until the word to be generated is completed (49).

In FIG. 5, before the dictionary lookup morphological analysis section, the character pointer is set to 0 (50) L, it is determined whether there is an open quotation mark (52) L, and if there is an open quotation mark, Memorize the word position (53), search for the closing quotation mark (54)L, and determine whether there is a closing quotation mark (5
5) L, if there is a closed bow symbol, remember the word position (5, 6) L, return to process (51). If it is determined (52) that there is no opening quotation symbol, the quotation symbol is deleted according to the opening and closing quotation symbol word positions (57). If there is no closing quotation mark in judgment (55), clear the opening word position (
58) Do. When processing (57) or processing (58) is completed, dictionary lookup processing (59), morphological analysis (60), and conversion processing (62) are applied in the same way as for sentences that do not include quotation marks. In the generation unit, it is determined whether the word position of the phrase to be generated matches the word position stored in word position storage (53) and (56) (63)L, and if they match,
Set the flag to ON (64) L, open quotation symbol +
1i! (65) Do. If the open quotation mark generation process (65) is completed or the word position does not match in the judgment (63), a phrase is generated (66), and it is judged whether the flag is ON or not (67). If ON, a closing quotation symbol is generated (68). If the closing quotation symbol generation process (68) is completed or the flag is not ON in the process (68), determine whether all phrases have been generated (69)L. If all phrases have been generated, the process end. In judgment (69), if all phrases have not been generated yet, turn off the flag.
(70) and repeat from judgment (63).

FIG. 6(a) is a diagram of the processing process for quotation marks in English-Japanese translation. When English text is input, the input section (7
1), pre-editing section (72), morphological analysis section (73), quotation marks processing section (74), syntax analysis section I section (75), syntax analysis section (76), structure conversion section (77), translation section Generation part (7
8), a post-editing section (79), and an output section (80) to output Japanese text. The control unit (83) analyzes the analysis rules (82)
The operation display unit (87) visually displays various instructions given to the device by the operator and various instructions from the device to the operator. do. Figure 6(b) is the same as Figure 6(a).
This is a diagram of the internal structure for holding bracket information during processing, including sentence information (84) and block information (85).
), word information (86), and other various information obtained during analysis are separately held. In this conventional method, a quotation mark processing unit (74) is provided between the morphological analysis unit (73) and the first syntactic analysis unit (75), and the quotation marks are stored in the block information (85) as position information. By keeping it as , we are trying to make it possible to translate the brackets.

In this way, in conventional methods, a special processing process for brackets is provided either before morphological analysis, inside the dictionary lookup section, or between morphological analysis and syntactic analysis, and information on brackets is stored. The idea was to enable translation of sentences containing parentheses by storing them as word or position information.

[Problems to be Solved by the Invention] In the processing means for parentheses in such conventional machine translation devices, the part that handles parentheses within the processing process is limited to a few special processes, and Because similar information is treated as an independent word or stored separately from other information obtained by analyzing sentences, processing is inflexible and the structure of the translation method is complicated. It is not sufficient to deal with the complex usage of parentheses in There was a problem that.

This invention was made in order to solve the above-mentioned problems, and by retaining bracket information in an internal structure and distributing the processing for brackets at each stage of the translation process, This is a natural extension of the machine translation method, and it is possible to obtain an appropriate translation even for the input source text that contains parentheses, and it also improves processing flexibility and the overall processing process for future expansion of the translation method. The purpose is to obtain a machine translation method that maintains bias.

[Means for Solving the Problem] The machine translation method according to the present invention includes a means for cutting out one sentence from an input original sentence, and a means for cutting out one sentence from an input original sentence, and a means for cutting out a left parenthesis when a parenthesis is included in the one sentence. Detect pair with right parenthesis,
a means for determining the purpose of use; a means for providing information regarding the type of characters to characters other than the parentheses in a sentence; a means for independently analyzing the characters inside the parentheses and replacing them with an internal structure;
A means for storing the character sequences located outside the parentheses in a mixed manner with other unparsed character sequences, a means for parsing the character sequences located outside the parentheses for each component of the sentence and replacing it with an internal structure, A syntactic analysis means for analyzing the relationship between the internal structures of each constituent element, and as a subprocess thereof, the relationship between the internal structure for the characters inside the parentheses and the arrangement of other internal structures located outside the parentheses. A means for analyzing, a conversion process for converting the internal structure of the analyzed original text into the internal structure of the target text, a means for converting the internal structure for characters inside the parentheses as a sub-process, and a means for converting the internal structure of the target text into the target text. and, as a subprocess thereof, a means for generating a translated sentence with parentheses from the internal structure of the converted characters inside the parentheses.

[Operations] The machine translation method in this invention has multiple processing steps for processing brackets, and by retaining information regarding brackets in the internal structure, it can translate sentences containing brackets in input text. It becomes possible to make translation possible and maintain expandability of the entire translation process.

[Example 1] An example of the present invention will be described below with reference to the drawings. 1st
The figure shows an example of the processing process of the machine translation method according to the present invention. In Figure 1, (1) is a means for cutting out one sentence from the input original text, and (2) is the type of character for each character in one sentence cut out by this cutting means (1). At the same time, when parentheses are included, a means for detecting a pair of a left parenthesis and a right parenthesis, determining the purpose of use of the parentheses, and adding information corresponding to the purpose of use of the parentheses, (3) ) is a means to perform ungrammatical processing such as symbol strings before full-scale analysis, and (4) is a means to analyze the characters inside the parentheses and replace them with the internal structure, and other unparsed characters located outside the parentheses. (5)
) is a means to analyze the sequence of characters located outside the parentheses for each component of the sentence and replace it with an internal structure, and (6) is a method for analyzing the character sequence located outside the parentheses by using the parenthesis white character analysis means (4). The first method of parenthesis function analysis, which analyzes the relationship between the internal structure and the internal structure of words outside the parentheses, involves identifying parallel ranges and local analysis of syntactic elements prior to syntactic analysis. means, (8) is the first means of parenthesis function analysis that correlates the internal structure inside the parentheses with other syntactic elements, and (9) is a syntax that analyzes the relationships between the internal structures of each constituent element of the sentence. The analysis means (10) is a sub-process of the syntax analysis means (9), which is a parenthesis function analysis step that analyzes the relationship between the internal structure of the characters inside the parentheses and the arrangement of other internal structures located outside the parentheses. ■ Means, (11) is a post-analysis processing means that corrects the parsed internal structure, and (12) is a post-analysis processing means before conversion that further replaces the internal structure obtained through processing up to (11) with an intermediate internal structure. Processing means, (■3) is a structure conversion processing means that converts the internal structure of the analyzed original text into the internal structure of the translated text, and (14) is a structure conversion that independently converts the internal structure of the characters inside the parentheses. Processing means (13
), (15) is a generation pre-processing means that converts the internal structure of the translated text into a structure suitable for the language of the translated text, and (16) is a means for generating the syntax of the translated text from the internal structure of the translated text. (17) is a parenthetical acid processing means for generating a translated sentence with parentheses from the internal structure of the characters inside the parentheses converted by the syntax generating processing means (16); (18) is a parenthetical acid processing means for generating a translated sentence with parentheses; Morphological generation processing means (19) performs morphological processing on each word in the translation, and post-generation processing means (19) performs final correction of the translated text.

FIG. 2 is a block diagram showing the configuration of a machine translation device for realizing the machine translation method according to the above embodiment. Second
As shown in the figure, this machine translation device uses input devices such as a keyboard, mouse, and OCR (optical character reader).
20), an input control unit (21) that controls this human-powered device (20), an editing control unit (22) that manages the entire translation work and controls the editing process of the original text and translated text, and A display control unit (23) that controls display, and a display device (2) that displays content instructed by the display control unit (23).
4) and a translation system (25
), a text storage unit (26) that stores the original text and translated text, and a print control unit (27) that controls printing of the original text and translated text, etc.
, a printing device (28) that prints the content instructed by the print control unit (27), and a dictionary system (28) that stores dictionary information necessary for translation processing and has a function to search this dictionary information. 29).

FIG. 3 is a diagram showing an internal structure resulting from processing related to this embodiment. The internal structure of the translation method of this embodiment is composed of nodes and links as shown in (36), and each node and link is composed of a collection of attribute name=attribute value. (30) is the original text that is cut out from the original text in the text storage unit (26) by the character segmentation means (1) and sent to the translation system (25); (31) is the original text (30
) is broken down character by character and held in the form of a list of character codes, (32) is the character type information corresponding to each character code in the list (31) given by the character type information analysis means (2), ( 33) is maintained by mixing the internal structure (34) obtained by analyzing the characters inside the parentheses with the sequence of other unanalyzed character codes located outside the parentheses, obtained by the parentheses analysis means (4). (35) is a list of the internal structure of each constituent element of the sentence after the processing of the morphological analysis means (6) has been completed, and (36) is the internal structure after the processing of the syntactic analysis means (9) has been completed. , (37) is the internal structure after the processing of the syntax conversion processing means (13) has been completed, (38) is the part of the internal structure converted by the conversion means (14) in parentheses, and (39) is the translated text.

Next, the operation of the above embodiment will be explained with reference to the drawings. In FIG. 1, information regarding the type of each character is extracted from the original text in the text storage unit (26) to a list (31) of character codes in one sentence by the sentence extraction means (1). At the same time as adding the corresponding list format (32), when a left parenthesis is detected in a character, the corresponding right parenthesis is searched, and if a left parenthesis and a right parenthesis appear as a pair in one sentence, As shown in (32), the character types 1p and rp are assigned as parentheses, and when the left and right sides do not correspond, the character type as symbols is assigned. This makes it possible to assign character types depending on the purpose of parentheses in a sentence. If a symbol string or a specific expression is included in the original text, the pre-analysis processing means (3) performs extra-grammatical processing of the symbol string or the like prior to full-scale analysis.

Next, the analysis pre-processing means (3) applies the analysis means from means (5) to means (11) to analyze the characters inside the parentheses, creates an internal structure (34), and generates an internal structure. (34
) and other unparsed character sequences located outside the parentheses are mixed and held in a list format as shown in (35). In the internal structure (34) in parentheses, the −containing soil node has
By giving an attribute such as quote=paren, information is set so that it is possible to determine that it is an internal structure within parentheses in later processing.

The morphological analysis means (5) uses the morphological analysis grammar rules to search the dictionary system (29) for the sequence of character codes located outside the parentheses, as shown in the list (35).
Each component of the sentence is replaced with an internal structure consisting of attributes Cat, auxword, etc. At this time, the parenthesis function analysis part (6) compares the internal structure (34) obtained by analyzing the sequence of character codes inside the parentheses with the internal structure of each component of the sentence located outside the parentheses. The relationship is analyzed, and each controlling internal structure (34) in parentheses is connected with a link, and an attribute such as 5yn=paren is given to the link.

Prior to the processing of the syntactic analysis means (9), the syntactic analysis preprocessing means (7) identifies parallel ranges and locally analyzes syntactic elements, and at the same time performs parenthesis function analysis part (8) and subprocessing. , and analyze the relationship between the internal structure for the characters inside the parentheses and the sequence of other internal structures located outside the parentheses, which could not be processed in the parentheses function analysis method (6). Reflect in attributes. The syntactic analysis means (9) analyzes the internal structure of each constituent element of the sentence according to the syntactic analysis grammar rules of the translation system (25), and uses the information obtained through the analysis as attributes syn of nodes and links.

Save as sem. At that time, the parenthesis function analysis means (10) is called as a sub-process to further deeply analyze the relationship between the internal structure (34) for the characters inside the parentheses and other internal structures located outside the parentheses. and stored as an attribute of the internal structure. The post-analysis processing means (11) corrects the parsed internal structure, and the pre-conversion processing means (12)
further replaces the internal structure obtained by the analysis up to the post-analysis processing means (11) with an intermediate internal structure.

The structure conversion processing means (13) converts the internal structure (36) of the analyzed original text into the internal structure (37) of the translated text.

At this time, the parenthesized conversion means (14) is called as a sub-process, and the internal structure (34) for the characters inside the parentheses is independently converted to obtain the internal structure (38).

Further, the syntax generation processing means (16) generates the syntax of the translated text from the internal structure of the translated text, and the parenthetical acid processing means (17) is called as a subprocess by the syntax generation processing means (16),
Generating a translated sentence with parentheses from the internal structure of the characters inside the parentheses converted in the parentheses conversion means (14),
The morpheme generation processing means (18) performs morphological processing on each word in the translated text, and the post-generation processing means (19) performs final correction of the translated text to generate a translated text (39).

As described above, it is possible to appropriately translate an original text including parentheses into a translated text by a natural extension of the conventional translation process.

Although the above example describes a method for processing parentheses in machine translation, this method may also be used in cases related to the handling of parentheses and symbols in general text processing methods, and the same effect as in the above example can be achieved. play.

[Effects of the Invention] As described above, according to the present invention, by providing a sub-processing process corresponding to the processing of brackets at each stage of the translation process, and retaining information regarding the brackets as an attribute in the internal structure,
Since it is possible to properly analyze the role that parentheses play in a sentence as a whole, it is a natural extension of conventional machine translation methods to convert original sentences that contain parentheses to include parentheses while maintaining the equivalence of the translation system. It becomes possible to translate into a translated text.

[Brief explanation of the drawing]

Fig. 1 is a basic flowchart of a machine translation system according to an embodiment of the present invention, Fig. 2 is a block diagram of a machine translation device for realizing an embodiment of the invention, and Fig. 3 is an implementation of the invention. Diagrams showing the internal structure of the example processing process, FIGS. 4 and 5,
FIG. 6 is a diagram for explaining the conventional machine translation method. In the figure, (1) is a single sentence extraction processing means, (2) is a character and other information analysis means, (3) is an analysis pre-processing means, (4) is an analysis means in parentheses, (5) is a morphological analysis means, (6) is the first means of bracket function analysis, (7) is the syntax analysis preprocessing means, and (8)
is the first means of bracket function analysis, (9) is the syntax analysis means, (1
0) is the first means of parenthesis functional analysis, (11) is the post-analysis processing means, (12) is the pre-conversion processing means, (13) is the structural conversion processing means, (14) is the conversion means within the parentheses, and (15) is the Generation pre-processing means, (16) syntax generation processing means, (17) parenthetical generation processing means, (18) morpheme generation processing means, (
19) is a post-generation processing means. Note that the same reference numerals in the figures indicate the same or corresponding parts.

Claims (1)

[Claims] In a machine translation device that obtains a translated text from an original text, there is a means for cutting out one sentence from an input original text, and a means for cutting out a pair of left and right parentheses when this one sentence contains parentheses. Detect and determine whether the parentheses are used as parentheses in their original meaning or as symbols, and at the same time provide information regarding the type of characters for characters other than parentheses in the sentence. A means of adding information, parsing the characters inside the parentheses and replacing them with the internal structure,
A means for storing the character sequences located outside the parentheses in a mixed manner with other unparsed character sequences, a means for parsing the character sequences located outside the parentheses for each component of the sentence and replacing it with an internal structure, A syntax analysis means for analyzing the relationship between the internal structures of each constituent element, and as a sub-processing process of this structure analysis means, an arrangement of the internal structure for the characters inside the parentheses and other internal structures located outside the parentheses. a conversion processing means for converting the internal structure of the analyzed original text into the internal structure of the translated text, and a subprocess of this conversion processing means that converts the internal structure of the characters inside the parentheses. A means for generating a translated sentence from the internal structure of the translated sentence, and a means for generating a translated sentence with parentheses from the internal structure for the converted characters inside the parentheses as a sub-process of the translated sentence generating means. A machine translation method for parenthesized sentences.