JPH05101095A - Method and device for detecting nonuniformity of translated word - Google Patents

Abstract

PURPOSE:To improve efficiency for the uniform of a translated word and the correction of spelling error by dividing a source text into words, also dividing a text edited after translation into word, drawing a dictionary concerning words in the source text and executing the extraction and display of words which corresponding relation can not be recognized. CONSTITUTION:The source text added a sentence number for each sentence is divided into words and stored in a first word division table 41. Next, an analysis processing processor 1 divides the translated text added a sentence number corresponding to that of the source text by one-to-one into words by drawing the dictionary while processing the end of the word or the like, and a processing program in a main memory 2 stores the respective divided words into a second word division table 42. Afterwards, the words in the first word division table 41 are extracted, and a corresponding relation decision table 43 with the second word division table 42 is prepared by using a translation dictionary 3. Then, the words not uniforming the translated word corresponding to the source word are extracted from this table 43, and an alarm information table 44 is prepared and displayed at a display device 7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a translation word inconsistency detection method and apparatus for checking translation word inconsistency associated with machine translation from a first language to a second language and manual post-editing work.

[0002]

2. Description of the Related Art Conventionally, a machine translation processing device from a first language to a second language generally comprises a dictionary, grammatical rules, and a processing device for applying the grammatical rules. However, since it is difficult to resolve the ambiguity mainly due to the polysemy of the first language, it is impossible to completely analyze the first language. Further, even if the analysis of the first language is successful, the generation of the second language does not always follow the natural language flow that humans use everyday. As a result, it becomes necessary for a human to post-edit the text translated by the machine translation processor. When the translation is performed by the machine translation processing device, the translation algorithm operates in the same manner for the same pattern in the original text, and thus the same translated word is applied to the same word in the original text. However, when a human post-edits, there is a possibility that the original word and the translated word that has been unified by the machine translation processing device may become inconsistent. Therefore, a method has been adopted in which the same correction is performed on multiple parts of the entire text so that the translated words do not become inconsistent even by human post-editing work. The post-editing work associated with the machine translation process is disclosed in Japanese Patent Application Laid-Open No. 58-151677. Japanese Laid-Open Patent Publication No. 58-151677 stores a text in a first language, a translated text in a second language, and a correspondence relationship between syntax element units of both texts, and a syntax element modified when the translation text is modified. It is disclosed that the character strings before and after the correction are stored, the same syntax element as the corrected syntax element is searched over the entire text, and the same correction is applied to this.

[0003]

In the current machine translation processing apparatus, the translation quality is not sufficient, and it is indispensable for the operator to perform post-editing work on the translated text. As a result, there is a possibility that the translated words of the same word in the original text may become inconsistent, and it is necessary to prevent the translation inconsistency from occurring. Even with the method disclosed as the conventional technique, there is a certain effect in preventing the occurrence of the word inconsistency, but it can be assumed that the worker uses the character correction function of a general word processor to perform post-editing work. There is no effect when performing such post-editing work. In view of the above circumstances, the purpose of the present invention is to
It is to automatically detect a translated word that is inconsistent in a text and also to automatically detect the occurrence of a spelling error associated with a post-editing operation, thereby increasing the efficiency of the unifying translation word and the proofreading operation of the misspelling.

[0004]

[Means for Solving the Problems] The above problem is that the first text in the first language is divided into words and stored in the first word division table, and the second text in the second language is divided into words. Stored in the second word division table, the words of the first text and the words of the second text are associated with each other using the bilingual dictionary and stored in the correspondence determination table, and the correspondence with the words of the second text is stored. The words of the first text with ununiform relations are extracted from the correspondence relation determination table, stored in the warning information table, and using this, the warning information of unconcordance translations is displayed on the display screen. It is achieved by doing. Further, the first text obtained by machine-translating the text in the first language into the second language is divided into words and stored in the first word division table, and the machine translation result is further edited to obtain the first text. The second text is divided into words and stored in the second word division table, the words of the first text and the words of the second text are associated and stored in the correspondence determination table, and the words of the second text are stored. The words of the first text having both the occurrence places with the correspondence relation and the appearance places without the correspondence relation are extracted from the correspondence relation determination table and stored in the warning information table, and the correspondence relations of the extracted words are not unified. This is achieved by displaying that

[0005]

In the method for detecting inconsistency of translated words of the present invention, first,
The original text is word-divided and stored in the first word division table, and the translated and edited text (hereinafter referred to as translated text) is word-divided and stored in the second word division table, and the first word division table. The word in the dictionary is looked up in the dictionary, and the second word is extracted for all the translations of the word registered in the dictionary.
The word division table is searched, the matching ones are extracted, and stored in the correspondence determination table together with the words in the first word division table. The word in this case is the minimum of the constituent element unit in the sentence. Words in the first word division table for which no match is found are treated as “??
? It is stored in the correspondence relationship determination table together with the character string “”. If there is a misspelling in a word in the translated text that should correspond to a word in the original text,
Of course, since a match with the translated word of the word in the dictionary cannot be found, the word corresponding to the word in the original text is "?
? ? Will be In this way, by referring to the correspondence relationship determination table, it is possible to know what translated word is assigned to the word in the original text. Next, referring to the correspondence between the original word and the translated word in the correspondence relationship determination table, the correspondence relationship in which a plurality of translated words are assigned to the same original word is extracted, stored in the warning information table, and stored. Display warning information about the correspondence on the display screen. Even if multiple words appear in the original text and some of them are misspelled, the corresponding word in the original text is "????".
And the translated words registered in the dictionary are mixed, which corresponds to the correspondence relationship in which a plurality of translated words are assigned to the same original word. Therefore, warning information is also displayed for the correspondence relationship in which such a spelling error has occurred. The text to be processed is
Similar effects can be obtained not only by machine translation processing but also by human translated text. Also,
In the translation word inconsistency detection method of the present invention, a machine translation text obtained by machine translation of an original text is divided into words, stored in a first word division table, and a translation obtained by further editing the machine translation result. The text is divided into words and stored in the second word division table, the words of the machine translation text and the words of the translation text are associated with each other and stored in the correspondence relationship determination table, and the appearance locations having the correspondence relationship with the words of the translation text are stored. , And the words of the machine-translated text that have both occurrences that do not correspond to each other are extracted from the correspondence determination table and stored in the warning information table, and it is displayed on the display screen that the correspondences of the extracted words are inconsistent. indicate.
This simplifies the process because the words in the first word division table can be directly searched from the second word division table without referring to the bilingual dictionary.

[0006]

[Embodiment] As an embodiment of the present invention, whether or not a translated word corresponding to a word in the original text is unified in the entire translated text in a machine translated from Japanese to English and manually post-edited The case of examining is explained. FIG. 1 is a block diagram showing an embodiment of the present invention. In FIG. 1, 1 is an analysis processor,
2 is a main memory having an internal processing table for analysis and the like, and a processing program, 3 is a dictionary memory, 4 is a translation word checking memory for checking inconsistency of translation words, 5 is a document file with text to be checked, and 6 is work. A keyboard used when the operator manually inputs the text to be checked, and 7 is a display device for the operator to confirm the check result. In the internal allocation of the translation word inconsistency check memory 4, 41 is a first word division table for storing the result of word division of the original text to be translated, 42 is a translation process, and is post-edited manually. A second word division table that stores the result of word division of the translated text, 43 is a correspondence relationship determination table that associates the words of the original text and the translated words of the translated text in a one-to-one correspondence, and 44 is the unified translated words in the translated text. 11 shows a warning information table for extracting and storing uncorresponding correspondences.

FIG. 2 is a flow chart showing the processing operation of an embodiment of the present invention. The processing operation will be described below with reference to FIG. The first text is divided into words by dividing the original text with a sentence number for each sentence input by the operator from the keyboard 6 or fetched from the document file 5 if it is in the document file 5 (step 11). The data is stored in the division table 41 (step 12). FIG. 3 is a flowchart showing the processing operation of (step 12).
The processing operation will be described in detail with reference to FIG. The analysis processor 1 divides the input translated text into words while referring to the bilingual dictionary 3 (step 121). In this case, the translated text is a text translated by a machine translation processing apparatus and manually post-edited, and a word is the smallest unit among the constituent elements of a sentence. As a method of word division, for example, Japanese Patent Laid-Open No. 62-1692
There is a publication No. 62, which can be used. The processing program in the main memory 2 of FIG. 1 stores the words divided by the analysis processor 1 in the first word division table 41 as shown in FIG. 4 (step 12).
2). When all the words in the text have been stored in the first word division table 42 (step 123), the processing operation ends. In the first word division table 41 of FIG. 4, 411 stores sentence numbers, and 412 stores all words in one sentence in the order of appearance. It is assumed that the number of rows 411 and the number of columns 412 of the first word division table 41 increase or decrease according to the size of the original text to be processed.

Next, the analysis processor 1 of FIG. 1 is input by a worker from the keyboard 6, or
If the translated text is included in the document file 5 (step 13), the translated text in which the sentence number corresponding to the sentence number of the original text in a one-to-one correspondence is added for each sentence is's. While performing processing such as'or'ed ', the dictionary is divided into words by the character string from blank to blank, and the processing program in the main memory 2 in FIG. It is stored in the second word division table 42 as shown in FIG. 5 by the same processing operation as the flowchart shown (step 14). Second of FIG.
In the word division table 42, 421 stores the number of each sentence, and 422 stores all the words in one sentence in the order of appearance. It is assumed that the number of rows 421 and the number of columns 422 of the second word division table 42 increase or decrease according to the size of the translated text to be processed.

FIG. 6 is a flow chart showing the processing operation (step 15) of creating the correspondence relationship determination table 43 of FIG. The processing operation will be described in detail with reference to FIG. In the first word division table 41 of FIG. 4, the 1st sentence of 411 and the 1st word of “412” “paper” are extracted (step 1
502), the bilingual dictionary 3 is searched for the word (step 1503). FIG. 7 shows the structure and contents of the bilingual dictionary 3. In FIG. 7, 31 is a Japanese word,
Reference numeral 32 indicates a part of speech of the word, 33 indicates a type of inflection when the part of speech of the word is a verb / adjective, and 34 indicates an English translation corresponding to the word. In the example of FIG. 7, a plurality of English translations of “paper” are registered as “paper” and “form”. First, the “paper”
"r" (step 1506), all words of the sentence 1 of 421 in the second word division table 42 of FIG. 5 are searched (step 1507), and the extracted translated word, that is,
Find the one that matches "paper" (step 15)
08). If there is a match with the translated word corresponding to the word in the line of the same sentence number in the second word division table 42, it can be said that the word in the first word division table 41 has a correspondence relationship. If there is no "paper" in the line of the same sentence number in the second word division table 42 and there is no unprocessed translated word in the word, the correspondence relation determination table 43 as shown in FIG. Meaning "??"
? Is stored together with “” (step 1509). The "paper" and the "paper" for which the correspondence has been obtained are also stored in the correspondence determination table 43 as shown in FIG. 8 (step 1510). In FIG. 8, 431 stores the numbers of the combinations of the extracted all correspondences, 432 stores the words of the original text, and 433 stores the words of the translated text. A word in the original text with "???" in 433 indicates that there is no corresponding translation. The above processing operation is repeated for all the words in the row of sentence number 1 of the first word division table 41, and when the search for one word is completed, the process proceeds to the next word (step 150).
5) When one sentence ends (step 1511), move to the next line, that is, the next sentence (step 1512),
When the processing for all the words in the first word division table 41 is completed (step 1501), the processing operation for creating the correspondence relationship determination table 43 is completed.

Next, from the correspondence relation determination table 43, correspondence relations in which the translated words corresponding to the original words are not unified are extracted (step 16). FIG. 9 is a flowchart showing a processing operation for extracting, from the correspondence relationship determination table 43, a correspondence relationship that is a target for issuing a warning that the translated words corresponding to the original word are not unified. The processing operation will be described in detail with reference to FIG. The processing program in the main memory 2 of FIG. 1 then extracts the word “paper” of the original text of the first set “paper-paper” from the correspondence determination table 43 of FIG. 9 (step 1602), which The group number 1 is assigned to the variable COUNT in order to store whether to extract the group of the correspondence relationship (step 16).
03). Then, the extracted word "paper" of the first set of original texts is stored in the first word variable (step 160).
4) The word "paper" of the same set of translated texts is extracted and assigned to the second word variable (step 1605).
While proceeding to the second, third, and fourth groups (step 161
0), the word matching the original text in the first word variable is searched for in the correspondence determination table 43 (step 1607). If the matching word is not found even after the search in the correspondence relationship determination table 43 is completed (step 1606), there is only one original word in the original text, and therefore it is necessary to check whether or not the translated words are unified. Since there is no variable, the variable COUNT is incremented by 1, and in this case, the correspondence relation of the second set "print-print" of the correspondence relation determination table 43 is moved to (step 160).
8). If a matching word is found (step 160)
In 9), the corresponding translated word and the character string of the word of the translated text in the second word variable are compared (step 161).
1). Here, a character string that matches the word “paper” of the original text in the first word variable exists in the fourth set of the correspondence determination table 43. Therefore, the word "form" of the fourth set of translated text is extracted. If the word and the word of the translated text in the second word variable match, the translated words have been unified, so the process proceeds to the next group (step 1610), and steps 1607 and 1609 are repeated. here,
The word of the translated text in the second word variable is "paper"
So the two do not match. That is, it can be seen that the word "paper" in the original text is assigned with two different translations "paper" and "form". Therefore, the character string "paper" of the first word variable and the character strings "paper" and "form" of the second word variable are stored in the warning information table 44 of FIG. 10 (step 161).
2). In the warning information table 44 of FIG. 10, 441
Is the number of sets of correspondences displaying warning information, 442 is the word of the original text, 443 is the number of translated words assigned to the word in the translated text, 444 is assigned to the word in the translated text Stores translated words. Of the correspondences stored in the warning information table 44, the word “key” of the second set of original texts corresponds to “k”.
"ey" and "???", but because one of the translated words of "key" is misspelled with "ky", there is no corresponding relationship and it has a special corresponding relationship. It is shown that. Thus, the correspondence determination table 43
When the above-mentioned search processing is performed for all the groups in the inside and the correspondence relationship determination table 43 ends (step 1601),
The processing operation ends.

Then, the correspondence relationship stored in the warning information table 44 is highlighted or summarized on the display device 7 as shown in FIG. 11 (step 17). As the highlighting method, the method disclosed in Japanese Patent Laid-Open No. 58-101365 can be used. Regarding the display of the above warning information, for example, if the operator intentionally selects different translated words such as Japanese "i" that need to be translated in English, it is possible that Since the warning display is rather annoying, it is assumed that the operator can specify a word that does not need to be detected.

As another embodiment of the present invention, a text that has been machine translated is used as the first text instead of the original text, and the machine translation result is post-edited as the second text instead of the translated text. With translated text,
Even if there is no original text, the same effect as the above-described embodiment of the present invention can be obtained. In the case of the present embodiment, in the process of creating the correspondence relationship determination table, in the above-described embodiment of the present invention, the translated word of the word in the first word division table is obtained by referring to the bilingual dictionary, and then the translated word is set to the second word. While the word division table is searched, the words in the first word division table may be directly searched from the second word division table, so that the process is simplified.

[0013]

As described above in detail, according to the present invention, a translated word having a non-uniform or a misspelled word is extracted from a text translated by a machine translation processor and post-edited manually. It can be automatically detected, and the translations can be unified or spelling errors can be corrected efficiently. Therefore, the translated text is divided, and after multiple workers perform post-editing work, those texts are merged and post-editing work is performed by multiple workers. It becomes possible to unify the translated words, and as a result, regarding the post-editing work on the text translated by the machine translation processing device, the time required for the whole post-editing work is shortened and the labor is remarkably reduced. The effect occurs.

[Brief description of drawings]

FIG. 1 is a block diagram of an apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a processing operation of an embodiment of the present invention.

FIG. 3 is a flowchart showing a processing operation for creating a word division table in an embodiment of the present invention.

FIG. 4 is a first word division table showing a result of word division of an original text according to an embodiment of the present invention.

FIG. 5 is a second word division table showing a result of word division of the translated text according to the embodiment of the present invention.

FIG. 6 is a flowchart showing a processing operation for creating a correspondence relationship determination table in an embodiment of the present invention.

FIG. 7 is a diagram showing the structure and contents of a bilingual dictionary used in an embodiment of the present invention.

FIG. 8 is a correspondence relationship determination table in which a word of an original text and a word of a translated text are associated with each other according to an embodiment of the present invention.

FIG. 9 is a flowchart showing a processing operation for creating a warning information table in the embodiment of the present invention.

FIG. 10 is a warning information table in which correspondences that need to be displayed as warnings on a screen according to an embodiment of the present invention are extracted.

FIG. 11 is a diagram showing an example of a display screen associated with processing in an embodiment of the present invention.

[Explanation of Codes] 1 Analysis Processor 2 Main Memory 3 Dictionary Memory 4 Translated Word Check Memory 41 First Word Division Table 42 Second Word Division Table 43 Correspondence Determination Table 44 Warning Information Table 5 Document File 6 Keyboard 7 Display Device

Claims (6)

[Claims] 1. A method of translating a first text described in a first natural language into a second text described in a second natural language, performing post-editing work, and then converting the minimum of the first text. A translated word inconsistency detection method characterized in that when the minimum syntactic element unit of the second text corresponding to a syntactic element unit becomes inconsistent, a warning is displayed about the inconsistent correspondence. 2. The method for detecting inconsistency of translations according to claim 1, wherein the first text is divided into minimum syntax element units and stored in a first table, and the second text is divided into minimum syntax element units. And storing in the second table by using the bilingual dictionary and associating the minimum syntax element unit of the first text with the minimum syntax element unit of the second text in the third table The step of extracting the minimum syntactical element unit of the first text from the third table in which the correspondence with the minimum syntactical element unit of the second text is inconsistent, the extracted correspondence relation being inconsistent Is displayed on the display screen. 3. A first text after machine-translating a text described in a first natural language into a second natural language as a first text and post-editing the first text as a second text. The second corresponding to the smallest syntax element unit of
If the minimum syntactic element unit of the text of
A method for detecting inconsistency in translated words, which is characterized by displaying a warning about inconsistent correspondence. 4. The method for detecting inconsistency of translations according to claim 3, wherein the first text is divided into minimum syntax element units and stored in the first table, and the second text is divided into minimum syntax element units. And storing in the second table, and storing in the third table in association with the minimum syntax element unit of the first text and the minimum syntax element unit of the second text. From the step of extracting from the third table the minimum syntax element unit of the first text in which the correspondence relationship with the minimum syntax element unit does not match, and the step of displaying on the display screen that the extracted correspondence relationship does not match A method for detecting inconsistency in translated words, which is characterized by being formed. 5. An analysis processor, a main memory, a dictionary memory, a word translation memory, a document file, a keyboard and a display device, wherein the first text described in the first natural language is converted into the second natural language. Means for translating and post-editing into a second text described by, a means for dividing the first text into minimum syntax element units, a means for dividing the second text into minimum syntax element units, and a bilingual dictionary A means for associating the minimum syntactical element unit of the first text with the minimum syntactical element unit of the second text by utilizing, and the correspondence relation between the minimum syntactical element unit of the second text are inconsistent. An apparatus for detecting inconsistency in translated words, comprising means for extracting the minimum syntactical element unit of the text and means for displaying that the extracted correspondence is inconsistent. 6. An analysis processor, a main memory, a dictionary memory, a translation check memory, a document file, a keyboard, and a display device, wherein the text described in the first natural language is converted into the second natural language in the first natural language. Means for machine-translating the text as a text, a means for post-editing the first text as a second text, a means for dividing the first text into minimum syntax element units, and a second text as a minimum syntax element unit. A first means in which the correspondence between the dividing means, the means for associating the minimum syntax element unit of the first text with the minimum syntax element unit of the second text, and the minimum syntax element unit of the second text do not match. A united word inconsistency detection device characterized by having means for extracting the minimum syntactical element unit of the text of the text and means for displaying the fact that the extracted correspondences do not match. Place