JPS6170660A - Polysemy displaying and selecting means of machine translation system - Google Patents

Abstract

PURPOSE:To facilitate equivalent correcting operation by using polysemy data of all equivalents which are replaceable for each translation and a table containing coordinate relation among respective phrases in an original, many equivalents in translations, and respective equivalents in the polysemy data. CONSTITUTION:Respective data on text character strings in the original and text character strings of its translation obtained by machine translation, polyse my data of all equivalents which are replaceable for every equivalent in transla tion text character strings, and table data on the coordinate relation among respective phrases in original text character strings, respective equivalents in translation text character strings, and respective equivalents in the polysemy data are read out of a file. When some equivalent in optional one of the original or translation is specified, the table data are referred to and respective corre sponding equivalents in the polysemy data is retrieved and displayed. The origi nal translation word part in the translation is replaced with a proper equivalents among the displayed equivalents.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a machine translation system, and particularly to the handling of polysemous words in proofreading of a translated text.

[Background of the invention]

Generally, when translating from a first language to a second language, a certain word in the first language has multiple meanings, and there are multiple corresponding words in the second language (hereinafter referred to as translated words). Situations in which one is forced to select one often occur, but in machine translation systems, it is extremely difficult to automatically determine the optimal translation from among such multiple translations. Although it is possible to select one of them according to some criterion, a criterion that always gives the optimum result is not yet known. Therefore, when humans proofread a translated text obtained by machine translation, if an inappropriate translated word is found, it is necessary to correct it to another suitable translated word. However, in order to correct the translation, the operator usually has to find a more appropriate translation using his/her own knife and input it, which is cumbersome. If such correction operations occur frequently, various equipment operations associated with the correction operations become troublesome, and the efficiency of the calibration process decreases. However, outputting all the corresponding translations of all polysemous words along with the translated text would be very complicated and would actually reduce work efficiency.

Regarding the current state of technology and issues related to machine translation, please refer to "Knowledge Engineering" (edited by Kokichi Tanaka, published by Asakura Shoten, 1984).
), Section 3.4 ``English-Japanese machine translation system using functional semantic processing'' (pp. 282-303).

[Purpose of the invention]

An object of the present invention is to simplify translation correction work performed during proofreading in a machine translation system and improve its efficiency.

[Summary of the invention]

The present invention is capable of substituting each data of a text string of an original text (first language), a text string of a translated text (second language) by machine translation, and each translated word in the translated text string, including itself. Polysemic data that collects all translated words, and chiful data that shows the correspondence between each word in the source text string, each translation word in the target text string, and each translated word in the polysemic data (in the example below). MPGTB
L and MLTTBL) from the file device, and when a word or phrase in any one of the text strings of the original text or the t/'i translation is specified, the table data is referred to and the correspondence in the ambiguous data is determined. search for each translated word and display the results on the display device.

When one of the displayed translated words is designated, the data of the corresponding original translated word in the translated text character string is replaced with the character string data of the designated translated word.

Translated words that change in conjugation are displayed in the conjugated form that corresponds to the conjugated form of the translated word in the translation text (for example, the same conjugated form if it is the same part of speech, and the final form if the translated word in the translated text is a noun). It is best to configure ambiguous data in a format that allows

In addition, when searching and displaying translated words in ambiguous data, when a translated word in the translated text is specified, the previous proofreading (
If the correspondence between the translated text character string data and other character string data no longer matches the correspondence indicated by the table data, as a result of deletion, movement, etc.), in order to prevent errors, It is better to cancel ambiguous display/selection processing.

Furthermore, following the translation word replacement processing described above, for the parts in the original text where the same translated word as the original translated word was adopted for the same word/phrase, a process of performing the replacement with the specified translated word in a batch. It is convenient to add.

[Embodiments of the invention]

FIG. 2 is a block diagram illustrating an example of an apparatus for implementing the present invention. The processor 1 includes a display device 2, an input device 3 such as a keyboard, a translation result file 4,
It is connected to a memory 5 and a screen/outputer 6, and while exchanging information with these, proofreading including translation correction and other data processing are performed. The translation result data is
The data extracted from the translation result file 4 and stored in the memory 5 to be displayed is displayed on the display device 2 after being stored in the screen part 7a6.

For instructions such as drop collection, ambiguity display, translation correction, etc., input 1
A corresponding process is applied by the processor 1 to the data in the memory 5 and the screen buffer 6. The data after the proofreading process is returned to the translation result file 4.

In the code of the translation result file 4, various data obtained by the translation process are recorded in units of sentences in a variable length record format as shown in FIG. 3(a). The translation process can be done using Japanese Patent Application Laid-Open No. 58-40684.
This can be realized by the method described in the publication. In FIG. 3(a), KEY is a key to identify the record (for example, sentence number), RLNG is the length of the record, PFLAG is a proofreading process management flag indicating whether proofreading has been performed, and E LNG is a The length of the English text data EDATA that immediately follows, MPGC is the number of arrays in the English-Japanese contrastive polysemy management table MPGTBL that immediately follows, MLTC is the number of arrays in the polysemy data management table MLTTBL that immediately follows, and L to ILNG is the number of arrays in the polysemy data management table MLTTBL that immediately follows. Length of ambiguous data MLTDATA that immediately follows, JLN
G is data indicating the length of the translated text data JDATA that immediately follows.

An example of the main contents of these data is shown in Part 3.
Shown in Figure (b). However, the format of each data in FIG. 3(b) is the format when the data is developed in the memory 5 in a table format. Assume that the English text and its translated text are respectively EDATA and JDATA shown in the figure. However, the △ symbol in EDATA represents a word space.

MPGTBL maintains, for each ID number assigned to each phrase in JDATA, the correspondence between the corresponding English sentence, a translated sentence, and ambiguous data for the translated sentence. for example,
The data in the line with ID=5 corresponds to the 5th clause in the translated sentence, and is the 11th character of EDATA (indicated by EH=11)
The data for 7 characters (indicated by EL=7) from
H = 19) to 4 characters (JL = 4), that is, it corresponds to "move", and the translated word corresponds to the 6th term (M H = 6) to 2 items (MC = 2) of ambiguous data. However, when counting the number of characters, a word space in EDATA is counted as one character, and one character in JDATA is (
(Because it requires 2 bytes) it is counted as 2 characters.

For each part corresponding to each ID number of MLTDATA//i, MPGTBL, there is a monomer in which all possible translations are arranged in order, MLTTBI, MLTI)ATA(
7) A table of position information for each translated word. Continuing the above example, the 6th term (ID=6) of MLTTBL is ML
21st character of TDATA (indicated by DH=21)
It refers to the translation of 4 characters (indicated by DL = 4), that is, "work", and the 7th term (ID = 7) is DH =
25. From DL=10, it refers to the translated word "to do work" in M L T DATA. Therefore, MPGTB
L (7) I D = 5 (7) Line Nio is Le M H
= 6.

MC=2(7) data is the English “TOWORKJ”
This shows that it has two translations: ``work'' and ``work''.

Here, recording of ambiguous data will be explained.

Ambiguous data is processed in the translation process before moving on to the proofreading process.
These are all translated words extracted from word-compatible dictionaries, and generally,
One or more translation words exist as ambiguous data for one word. By the way, for example, rWOR in FIG. 3(b)
``Work'' has been adopted as a translation corresponding to KJ, but since this translation is a conjugated word, dictionaries do not include it in Figure 3 (
As shown in C), the combination of the stem word ``work'' and the ending conjugation type ``2'' (representing the 5-step conjugation of force) is registered in the form ``work/2'', and the combination is used in the translation process. It is transformed into a body shape and changes from "work/2" to "work".

Another translation of l'-WOR, KJ registered in the dictionary is ``Job wo/Cl (rcJ represents subflexive conjugation). In this way, translated words that change in conjugation have a combination of the stem and the conjugation. M
When setting in LTDATA, the same conjugation form as the translated word adopted in the translation is made.

Therefore, in the case of FIG. 3(b), as another translation word in the ambiguous data corresponding to "WO 几K", "Job wo wo" is set based on "Job wo/C".

On the other hand, there are many parts of speech in the dictionary (having data for two or more parts of speech).
When a word registered as a part of speech is analyzed as a part of speech that does not change in conjugation during the translation process, if there is a translated word that changes in conjugation among the parts of speech that were not adopted, it is changed to the final form and then treated as ambiguous data. set. For example, if ``LIKEJ'' in Figure 3(C) is treated as a preposition in the translation process as in Figure 9(b), and the translation ``like'' is adopted in the translated sentence, rLIKE
The verb translations of J, ``Sukide/DJ'' (rDJ represents a special conjugation of ``aru'') and [ko/8J (r8J represents a 5-step conjugation of the M line), are conjugated to the final form, and are respectively `` The forms of ``I like it'' and ``I like it'' are sent as ambiguous data.

The translation proofreading process is performed according to the flow outlined in FIG.

(101) First, in order to retrieve the translation you want to proofread, use the keyboard 3 to KE of the record containing the translation.
Specify Y. In the case of multiple sentences, specify the KEY of the first sentence and the KEY of the last sentence.

(102) A record having a KEY that matches the designated KEY is read from the translation result file '4 and stored in the memory 5. At this time, the main data in the record takes the format shown in FIG. 3(b) and is developed as shown in FIG. 4(b). Accordingly, the calibration processing management table CNTL-TBL shown in FIG. 4(a) is created in the memory 5.

CNTL-TBL is for managing the proofreading process of each translated text, and for each record read from the translation result file 4, corresponding to the ID number assigned to it,
PFLAG, ELNG, MPGC.

Hold JLNG. PF'LAG (proofreading process management flag) indicates the correspondence between the translated text data after proofreading and the content of MP()TBL as a result of insertion, deletion, copying, and movement processes on the text in the proofreading process of translated text data. 1°' is set when there is a collapse, and 0°' is set at other times. Still, II II+ is sent to the screen display management point CP prepared in the memory 5.

(103) With reference to CN'l''L-TBL, the above C
The English data (EDATA) and translated data (JDATA) after the ID indicated by I) are sequentially taken out from the memory 5 and stored in the screen buffer 6. As a result, display device 2
A screen like that shown in FIG. 7(a) is displayed.

(104) Proofreading of the translated text is performed according to the flow shown in FIG. FIG. 6 will be referred to for the explanation of the processing from (201) to (211) below.

(201) The calibration process completion flag ENDPROC prepared in the memory 5 is set to 0°'.

(202) According to the operation command input from the keyboard 3, the process branches to various calibration processes. These proofreading processes are performed for each translated text (translated text stored in the record in FIG. 3(a)).

(203) to (206) Insert key, delete key.

Insert processing in response to copy key and movement key operations.

Deletion processing, copying processing, and moving processing are each executed.

These are the same processes as in well-known document processing devices, so detailed explanations will be omitted.

(207) When any of the above four types of processing is performed,
The corresponding JDATA expanded in the memory 5 is corrected to the translated text after proofreading, and the contents of the screen buffer 6 are also corrected accordingly, and the CNTL-TBL K
JLNG at is updated, and the same (PFLAO is set to 0'1'') and returns to (202).

(208), (209) When the up scroll key or down scroll key is operated, the CP value is increased or decreased by 1 accordingly, and the instructions at the beginning of the displayed English sentence and translated sentence are changed one sentence at a time. After raising or lowering (
1O5).

(210) When the end key is pressed, K sets the calibration process completion flag ENDPROC to °゛1'' and then (1
05).

(211) When an ambiguous key is operated, an ambiguous display/
The selection process is performed according to the flow shown in FIG. 1, and when it is completed, the process returns to (202).

Below, step 6 from (301) to (317)
The explanation is based on FIG.

(301) When the ambiguous key is operated in (202), the ID number in CNTL-TBL of the English text or translated text is determined based on the cursor position at that time. This can be determined from the correspondence between the positions of the screen buffer 6 and the memory 5.

(302) Check whether the cursor position is in the English text area or the translated text area on the screen, and if it is the former (3
03), and if it is the latter, the process moves to (304). The former is when specifying a word in an English sentence and displaying the corresponding translated word in polysemic data, and the latter is when specifying a translated word in a translated sentence and displaying the corresponding translated word in polysemic data. be. As will be clear from what will be described later, any method may be used, except when the translation has been corrected in appearance by a previous proofreading process.

(304) If the cursor position is in the translation area, CN
Whether PFLAG of TL-TBL is It l 11,
In other words, it is checked whether or not the corresponding translation has already undergone font proofreading, and if PFLAG = 1, the translated word may have been corrupted over time, and ambiguity display/selection processing by specifying the translated word is not possible. Since it is possible, the process returns to step (202). If PFLAG=O, the process moves to step (305).

(303) If the cursor is in the English text area, PFLAG
There is no need to consider the O value, and l'vlPGTBL is learned to find the ID corresponding to the phrase that includes the English word at the cursor position.

(305) When the cursor is in the translation area, PFLAG
If = 0, MPGTBI is checked to find the ID corresponding to the phrase that includes the translated word at the cursor position.

(306) i obtained in (303) or (305) above
Check the value of MC corresponding to the ID number in 'vlPGTBL, and if MC>1, go to (308), and if MC=1, go to (307).

(307) Check again whether PFLAG=1. The state in (307) is possible only when the polysemous word is 1. Here, PFLA( )=1 when the cursor position is in the English text area and the translated text has already been subjected to font proofreading processing. In this case, there is a possibility that the translation of the English word specified for displaying multiple meanings has been deleted from the translated text, so it is meaningful to display multiple g8 even if the number of multiple meanings is 1. Therefore, if PFLAG=1, the process moves to step (308). If PFL,AG=00, the process returns to step (202).

(308) English-Japanese contrastive ambiguity management table MP in memory 5
GTBL, ambiguous data management table 1VILTTBL,
The ambiguous data is displayed using the data of the ambiguous data MLTDATA. That is, with reference to FIG. 3(b), the MPGTBL obtained in (303) or (305)
Based on the ID number in , extract the MH and MC for it, refer to N, (LTTBL) to find the corresponding DH and DL, and use these DH and DL to calculate the MLT.
D. The corresponding portion of ATA is read and stored in a predetermined area of the screen buffer 6. As the display screen IIPIJ, the 7th
In figure (a), if you move the cursor to "WORKJ" and specify ambiguous display, "WORKK" will be displayed as shown in figure (b).
``Work'' and ``Work'' are displayed as ambiguous data translation words. At that time, rWO on MPGTBL
[T in the English sentence indicated by the ID as a phrase containing RKJ
Oν～DRJ (J and the corresponding “work” in the translation,
Conveniently, highlighting can be done by suitable well-known forms such as burr/li. In this case, EH and EL in MPGTBL.

EDATA and JDAT in memory 5 from JI-I and JL
Since the positions of these character strings in A can be determined, it is easy to perform highlighting processing such as bricking on the corresponding character strings in the screen buffer 6.

(309) The translation word to be adopted is specified from among the ambiguous data displayed on the screen using the function key.

(310) If PFLAG=O, it is certain that the original translated word to be replaced by the translated word specified in (309) exists at the original position in the translated text, so the process immediately moves to the replacement process (311). However, if PFLAC) =
1, the translation has already been corrected literally, and as a result, the original translation corresponding to the translation 1 word specified in (309) has been deleted, or even if it exists, its position is no longer in MPGTBL. Since it does not match the above position, it is best to import the specified translated word through the insertion process for the time being, and therefore the insertion position must be specified by the kernel. Figure 8 (a) shows that when "IN i'llE FUTUREJ" is specified in the English text area to indicate ambiguity display, the corresponding translated word "
Since "Mirai" has been deleted from the translated text, the cursor is set at the position of "Work" in the translated text, and the translated word "Mirai" is specified to be inserted immediately before it. For the above reasons, if PFLAG=1, the process moves to kernel position check processing (314).

(311) Since PFLA() = 0, (309
) will replace the corresponding translated word in the translated text. That is, with reference to MPGTBL and MLTTBL,
The corresponding translation in JDATA in memory 5 is replaced with the translation in 1VfL, Tl)ATA corresponding to the specified translation, the contents of screen buffer 6 are corrected accordingly, and the contents in 1vlPGTBL are replaced accordingly. Correct JH, JL and JLNG in CNTL-TBL.

This correction maintains the correspondence between the new translated text and the contents of MPGTBL, making it possible to perform ambiguous display and selection processing again later.

(312) After the above replacement process is completed, if the batch ambiguity change key is pressed, the process moves to (313); otherwise, the process returns to (202).

(313) Checks PF'LAG in CNTL-TBL, and for all sentences whose value is 'I OII, there is a translated word in JDATA that is the same as the original translated word before ambiguous selection, and If the corresponding English words in EDATA are the same as the English words corresponding to the original translated word, the same processing as (311) is performed for those translated words. Figures 10(a) and (b) show "5ISTE" in the second line of the English text.
Select "sister" in RJ's ambiguous data translation, replace "sister" in the second line of the corresponding translation, and replace it with the third translation.
This shows how the corresponding original translated word ``sister'' in the line was corrected all at once. When this process is completed, the process returns to (202).

(314) Since PFLAG = 1, the selected translated word will be inserted immediately before the cursor position. Then, the cursor position is checked, and if it is in the translation area, the process proceeds to step (315). When the cursor reaches the translation area (316), it waits for the cursor to move and moves to (317). If the send key is pressed here, the process returns to (314), but if the cancel key is pressed (2 ( Return to 12).

(315) Insert the translated word specified in (309) using the function key immediately before the kernel position. In other words, MLTDA is placed at the corresponding location in JDATA in memory 5.
Insert the corresponding translation in TA, correct the contents of the screen buffer 6 accordingly, and at the same time, CNTL-T
Correct JLNG in BL. Figure 8(b) is the same figure (
From state a), the translated word ``Mirai'' is inserted immediately before ``Work'' specified by the cursor. When the insertion process is completed, the process returns to (202>).

(105) As described above, when the up or down scroll key or the end key is pressed in the process step o-- of FIG. 6, the process returns to step (105) and the ENDP OC is checked. If ENDPROC=0, this means that the up or down scroll key has been pressed, so the process returns to step (103) and the contents of the screen buffer 6 are changed according to the updated contents of the CP. Furthermore, if ENDPROC=1, this means that the end key has been pressed, and the process proceeds to (106).

(106) Various data expanded in memory 5 and CN
Edit the contents of TL-TBL into the variable length record format shown in FIG. 3(a), store it in the translation result file 4,
This completes the calibration process.

According to the present fFIJ, the translated word in the ambiguous data is displayed in the same inflection form as the corresponding translated word in the translated text, making the work of correcting the translated word even easier.

If the correspondence indicated by M P G T B L has collapsed due to the previous temporary fixing process, refer to I) F L AG, and if the correspondence indicated by M P G T B L has collapsed due to the previous temporary fixing process, you can specify the words in the translated text to display/select ambiguous meanings. This prevents the display of an incorrect alternative translation and the replacement thereof.

Note that the processor 1 and memory 5 in FIG. 2 may also be used as a processor and memory for translation processing, or a separate processor and memory may be prepared and the display device 2. The input device 32 together with the screen buffer 6 may constitute an intelligent terminal for proofreading.

〔Effect of the invention〕

According to the present invention, it is possible to specify any word or phrase from either the original text or the translated text, display all alternative translations, and replace the original translation with a desired translation among them. , it is possible to easily and efficiently perform interactive proofreading of translated texts obtained by machine translation.

By adding a function for replacing equivalent translations at once, duplication of work can be eliminated and the efficiency of the proofreading process can be further improved.

[Brief explanation of drawings]

FIG. 1 is a flowchart corresponding to the main part of an embodiment of the present invention, FIG. 2 is a block diagram of a line of apparatus for carrying out the present invention, and FIGS. 3 and 4 are useful for the present invention. A schematic diagram of the data structure, FIG. 5 is a flowchart showing the overall outline of an embodiment of the present invention, FIG. 6 is a flowchart showing a part of FIG. 5 in more detail, and FIGS. 7 to 10 are FIG. 7 is a diagram showing rows of display screens accompanying processing by the 11th FIJ. DESCRIPTION OF SYMBOLS 1... Processor, 2... Display device, 3... Input device, 4... Translation result file, 5... Memory, 6...
...Screen buffer, 102...Data reading processing from the translation result file, 104...Proofreading processing, 211.
... Polymorphic display/selection processing, 301-308... Translation search/display processing in polysemic data, 309-311...
Translation word replacement processing, 312-313... Batch replacement processing.

Claims (1)

[Scope of Claims] 1. In an apparatus for proofreading a translated text obtained by machine translation, each data of a text string in a first language and a text string in a second language that is a translation thereof, as well as the 2
A process of reading from a file device polysemic data consisting of all translated words corresponding to each translated word in the language text string, and table data showing the mutual correspondence between each part of the text string data and the polysemic data, and both text strings. a process of searching and displaying each translated word in ambiguous data corresponding to a specified word in any one of the translated words while referring to the table data, and displaying the translated word in the displayed translated words; An ambiguous display/selection method comprising the step of replacing a corresponding part in second language text character string data with character string data. 2. In claim 1, the polysemy is characterized in that the translated word in the polysemic data is displayed in a conjugated form corresponding to the conjugated form of the corresponding translated word in the second language text string. Display/selection method. 3. In claim 1 or 2, during the process of searching and displaying translated words in the ambiguous data, the specified text string is a second language text string, and
An ambiguous display/selection method characterized in that if language text character string data does not match the contents of table data as a result of previous proofreading processing, subsequent processing is not performed. 4. In an apparatus for proofreading a translated text obtained by machine translation, each data of a text string in a first language and a text string in a second language that is its translation, as well as the second language.
A process of reading from a file device polysemic data consisting of all translated words corresponding to each translated word in the language text string, and table data indicating the mutual correspondence between each part of the text string data and the polysemic data, and both text strings. The process of searching and displaying each translated word in ambiguous data corresponding to any one of the designated words while referring to the table data and displaying the translated word on a display device, the character of the designated word in the displayed translated word; The process of replacing the corresponding part in the second language text string data with the string data, and the first language text string that is the same as and corresponds to the original translation word that is the target of the replacement is also that of the original translation word. An ambiguous display/selection method comprising the step of collectively replacing data of all the same second language text character string portions with character string data of the specified translated word.