JP4134983B2 - Machine translation device - Google Patents

Description

The present invention relates to a machine translation system for translating the original text by the first language entered in the translation by the predetermined second language.

The machine translation device automatically translates the input original text of the first language (for example, Japanese) into the translated text of the second language (for example, English). In many cases, the translation result is used as it is. Instead, it is necessary for the user to check the translation result and manually correct it if it is not an appropriate translation. Specifically, an appropriate translation is obtained by correcting the translation directly or by re-translating the original text in the first language.
Hereinafter, in the present specification, the first language sentence input earlier is referred to as an original sentence, the translated sentence as a translated sentence, a first language sentence obtained by correcting the original sentence as a corrected original sentence, and the translated sentence as a corrected translated sentence.

The retranslation work to obtain a corrected translation from the corrected original text allows the user to understand what the so-called habit of the machine translation device is, what kind of translation can be obtained by inputting the original text. This is often done because there is a merit that it becomes easy to obtain an appropriate translated sentence from the beginning when the same original sentence is translated next time by learning and saving in the apparatus. However, when the original text to be compared and the corrected original text are long but there are few differences, the difference between the original text and the corrected original text becomes difficult to understand, and the work time required for retranslation check becomes long.
Therefore, in Patent Document 1, the difference between the original text and the corrected original text is extracted to generate difference information, and if the difference is larger than the threshold based on the content of the displayed difference information, the corrected original text is displayed. The invention of a translation support apparatus that performs machine translation efficiently by displaying the original text, the corrected original text, its difference information, and the translated text and accepting the input of the corrected translated text when the machine translation is smaller than the threshold Is disclosed.

JP-A-8-297675

  However, in the translation support device of Patent Document 1, the difference between the original text and the corrected original text can be easily recognized, but when the difference is small, the user himself / herself inputs the corrected translated text corresponding to the corrected original text with reference to the translated text. It will be necessary to redo. Moreover, even if the corrected original was machine-translated, it was still difficult to understand how the corrected translation obtained by reworking the original changed from the translated text, and it was not suitable for translation check work using retranslation.

Therefore, the present invention provides a machine translation device that can easily recognize a difference between a translated sentence and a corrected translated sentence and improve the efficiency of the translation work when performing a translation check work using such retranslation. It is for the purpose.

In order to achieve the above object, the invention according to claim 1 is characterized in that an input means for inputting an original sentence in at least a first language, and an original sentence input by the input means is translated into a translated sentence in a second language using a dictionary. And a display device for displaying at least the original text input by the input means and the translated text translated by the translation means, wherein the original text and the translated text are displayed on the display means. When the corrected original text in which the original text is corrected by the input means is input, the corrected original text and the original text are compared, and at least one of the following (1) to (5) Determining means for determining whether or not there is a relationship, and determining that the corrected original text corresponds to the reworking of the original text if the relationship is present;
(1) The number of characters that differ between the original text and the corrected original text is equal to or less than a predetermined number.
(2) The only difference between the original text and the corrected original text is the notation.
(3) When the original text and the corrected original text are divided into words, only the order is different.
(4) When the original text and the corrected original text are divided into words, different words are stored in the dictionary as synonyms.
(5) When the original text and the corrected original text are divided into words, the number of different words is equal to or less than a predetermined number.
When the discriminating means determines that the original text has been revised, the corrected translated text obtained by translating the corrected original text by the translating means is compared with the translated text obtained by translating the original text. An identification display means for identifying and displaying is provided.

In addition to the object of claim 1, the invention described in claim 2 includes a reverse translation unit that back-translates the translated sentence in the second language translated by the translation unit into the first language. to improve the efficiency of work, said determining means, when the correction translation of the translation is corrected by said input means is input, by comparing the translation and correct translation, the following (1) to Determining whether or not there is at least one relationship in (5), and determining that the corrected translation corresponds to the revision of the translation when the relationship is present;
(1) The number of characters that differ between the translated text and the corrected translated text is equal to or less than a predetermined number.
(2) The only difference between the translated text and the corrected translation is the notation.
(3) When the translated sentence and the corrected translated sentence are divided into words, only the order is different.
(4) When the translated sentence and the corrected translated sentence are divided into words, the different words are stored in the dictionary as synonyms.
(5) When the translated sentence and the corrected translated sentence are divided into words, the number of different words is equal to or less than a predetermined number.
When the discriminating means determines that the identification display means corresponds to the revision of the translated text, the re-corrected original text in which the corrected translated text is reverse-translated by the reverse-translating means and the re-original text in which the translated text is reverse translated And the difference between the two is identified and displayed.

In addition to the object of claim 1 or 2 , the invention described in claim 3 is characterized in that the identification display means includes a sentence pair of the original sentence and the translated sentence in order to securely erase the identification display after the completion of the translation check. When either one of the corrected original text and the corrected translated text pair is selected and the other is deleted, or either the translated text and the rewritten original text pair and the corrected translated text and the recorrected original text pair When the other is deleted and the other is deleted, the difference identification display is canceled.

According to the first aspect of the present invention, when the corrected original text obtained by reworking the original text is translated, the difference is identified and displayed in the translated text and the corrected translated text, so the user can see how the translation result changes due to the rework. This makes it possible to quickly check appropriate translations and improve translation work efficiency. In addition, it is possible to appropriately determine the rework.
In particular, the ability to compare the translated text with the corrected translation at the same time makes it easier for users to understand the characteristics (翻 訳) of how the original text is translated and how it is translated. Improvement in work quality can also be expected.
According to the invention of claim 2, in addition to the above effect, even when performing a translation check by performing reverse translation, the user can know at a glance how the reverse translation result changes by reworking, Similarly, it leads to improvement of translation work efficiency.

According to the invention described in claim 3, in addition to the above effect, when the translation check is finished and an appropriate sentence pair is selected, the identification display can be surely erased and the identification display is forgotten to be erased. Disappears.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[Configuration of machine translation device]
FIG. 1 is a functional block diagram showing an example of a machine translation apparatus according to the present invention. The machine translation apparatus 1 includes a CPU, ROM, RAM, HDD, etc., a data processing unit 2 that performs processing such as translation, a keyboard, An input unit 3 that includes a mouse and the like, inputs an original text to be translated and various operation commands to the data processing unit, a display unit 4 that includes a CRT and the like, displays the input original text and translated text, and the like, a laser A printing unit 5 that includes a printer or the like and prints out text or the like displayed on the display unit;

In the data processing unit 2, an editing unit 6 that checks typographical errors and omissions of input sentences and translation result sentences and edits them as necessary, and a translation unit 7 that translates input sentences edited by the editing unit 6 are ROM. A dictionary (translation dictionary, user dictionary, etc.) 8 used for translation processing is stored and provided in the HDD. The translation unit 7 includes a morpheme analysis unit 9 that divides an input sentence into morphemes, a syntax analysis unit 10 that clarifies the syntactic structure of the input sentence, a conversion unit 11 that converts the input sentence into a target language sentence, and a target language sentence A syntax generation unit 12 that generates a structure and a morpheme generation unit 13 that morphologically arranges a target language sentence are configured.
In the data processing unit 2, the translation check unit 14 capable of performing a translation check process on the translation result obtained by the translation unit 7 is configured by a program stored in the ROM.

  Here, the input unit 3 constitutes input means. Further, the display unit 4 constitutes a display unit, and operates in accordance with the execution of the display procedure in the translation process described later. Further, the translation unit 7 in the data processing unit 2 constitutes a translation unit and a reverse translation unit, and executes a translation procedure and a reverse translation procedure in the translation processing described later. And the translation check part 14 comprises a discrimination | determination means and an identification display means, and performs the discrimination | determination procedure, the identification display procedure, and also the identification display cancellation | release procedure in the translation check process mentioned later.

[Description of normal translation processing]
In the machine translation device 1 configured as described above, for example, when “sunflower is blooming toward the sun” is input from the input unit, the display unit 4 is set in advance as shown in FIG. The Japanese sentence (original text A1) is displayed in the display area. Here, when a translation command is input by clicking a translation button (not shown) displayed on the display unit 4, the translation unit 7 translates the original sentence A 1, and the translated sentence display area set corresponding to the original sentence A 1 is displayed. “A sunflower goes to the sun, and it is blooming.” And the translation (translation B1) are displayed.

  This translation is performed according to the procedure shown in FIG. To explain this procedure, the original text A1 is first input in step 1 (simply abbreviated as “S1”, and the other steps are also the same), and when a translation instruction is issued in S2, the morpheme analysis is started in the morpheme analysis unit 9. (S3). In this method, a character string of a predetermined number of characters that can be registered as a headword is cut out from the head of the input original text A1, and the dictionary 8 is searched for it as a search keyword to check whether there is a corresponding word. . Here, the user dictionary is first searched in S4, and if there is no corresponding word in the determination in S5, the translation dictionary is searched in S6. If it is determined in S7 that there is no corresponding word in the translation dictionary, the word is processed as an unknown word in S8, and the processes after S11 are performed.

  FIG. 4 is a schematic diagram in a state where information used in this morphological analysis is stored. (A) is dictionary information of each entry word, (B) is information of utilization, and other connection information (for example, Information related to a connection rule of words such as a case particle comes immediately after a noun and a connection particle comes immediately after a useful word is also stored. For example, “sunflower” has “sunflower” as a headword (representative word), and the variant words “sunflower” and “sunflower” are stored. Are considered the same word. Similarly, “is” and “reap” are also compared with headwords. In the morphological analysis, these pieces of information are used to divide the information into words (morphemes) associated with attributes such as parts of speech.

If there is a word corresponding to either the user dictionary or the basic dictionary (YES in any of S5 and S7), a connection check with the preceding and following words is performed in S9. If the connection check is OK in the determination in S10, the input sentence is divided into words in S11, the dictionary information of the searched word is given to each word, and the morphological analysis is ended (S12). If the connection check is not possible in S10 (for example, particles are continuous), the process after S11 is performed after processing as an unknown word in S8.
Next, in S13, the syntax analysis unit 10 analyzes the sequence of words obtained by the morphological analysis by an analysis program, collects the words into phrases, decides on the subject or object, and so on. Create a tree structure. After that, in S14, the conversion unit 11 performs an English translation on each word while referring to the user dictionary and the translation dictionary based on the analysis result. In S15, the syntax generation unit 12 creates a tree structure of the translated sentence based on the tree structure created in S13 while referring to the dictionary, and generates a correct syntax. Next, in S16, the morpheme generation unit 13 adjusts the ending change or the like to finish the translation, and the translated sentence B1 is displayed on the display unit 4 (S17).

[Description of translation check processing]
After completing the translation in this way, the user checks whether or not the obtained translation B1 is appropriate. If the translation B1 is not appropriate, the user corrects the translation B1 directly, re-translates the revised text (corrected text), and compares the obtained new translation (corrected text) with the translation B1. There is a case where an operation of adopting a desired translation is performed. In this case, the translation check unit 14 compares the original text with the corrected original text to determine whether the corrected original text corresponds to a revision of the original text, and the corrected original text determined to fall under the rework is retranslated. At this time, a translation check process is performed in which the translated sentence and the corrected translated sentence are compared and the difference between the two is identified and displayed. Hereinafter, the procedure of the translation check process will be described with reference to the flowchart of FIG.

First, in S20, as shown in FIG. 2B, it is assumed that a corrected original A2 “sunflower is blooming toward the sun” is input following the original A1. When a translation instruction is issued for the corrected original text A2 in S21, the translation section 7 translates the corrected original text A2 in the same procedure as in FIG. 3, and in the display section 4, "A sunflower is blooming toward the sun." Is displayed.
In S23, it is determined whether or not there is an input sentence (original sentence A1) before the corrected corrected original sentence A2. Here, if there is no original sentence A1, the translation check ends. However, since there is the original sentence A1, it is determined in S24 whether or not there is a translation result of the original sentence A1 (translation sentence B1). If there is no translation B1, the translation check ends. However, since there is a translation B1, it is determined in S25 whether or not the corrected original A2 is a modification of the original A1 (determination procedure). A rework flag is assigned to the original sentence A1 and the corrected original sentence A2, and the difference between the translated sentence B1 and the corrected translated sentence B2 is identified and displayed in S27 (identification display procedure). As shown in FIG. 2 (B), this identification display includes the difference “goes to the sun, and it is blooming” in the translation B1 and the difference “is blooming toward the sun” in the correction translation B2 as other parts. This is a reverse display S that is displayed in a square frame of a different color (for example, white).

The determination as to whether or not the corrected original text A2 is a modification of the original text A1 in S25 is performed according to the procedure shown in the flowchart of FIG.
First, in S30, the original text A1 and the corrected original text A2 are compared, and the number of characters that the corrected original text A2 is different from the original text A1 is counted and stored in the variable C. The number of characters may be counted by a known difference extraction technique using a difference program such as diff. In addition to simply counting the number of different characters here, it is possible to avoid counting by recognizing a difference in notation such as hiragana and katakana with a conversion matrix.
Next, in S31, it is determined whether or not the variable after the comparison is less than 10. Here, if the number is less than 10, it is determined to be reworked in S32. If it is determined in S31 that the variable is 10 or more, the original text A1 and the corrected original text A2 are morphologically analyzed in S33, and existing words (morphemes) are compared. If the existing word is the same (YES in S34), it is determined that the word is corrected in S32, and even if the existing word is not the same, the number of different words is counted in the morphological analysis in S35 and counted in the determination in S36. If the variable W is less than 3, it is determined in S32 that reworking is performed. If the variable W is 3 or more, it is determined that the original text A1 and the corrected original text A2 are dissimilar (not reworked).

The comparison by the morphological analysis of the original text A1 and the corrected original text A2 in S33 to S35 is performed according to the procedure shown in the flowchart of FIG.
First, the original text A1 is morphologically analyzed in S40, and is cut out and stored for each word from the result of the morphological analysis in S41. Next, morphological analysis is performed on the corrected original text A2 in S42, and the number of different words is reset to 0 in S43. In S44, words are extracted one by one from the result of the morphological analysis of the corrected original text A2, and compared with the word of the morphological analysis result of the original text A1.
Therefore, if there is a matching word in the determination in S45, it is determined in S46 whether or not there is still a word. If there is no matching word in the determination in S45, it is counted as a different word in S47. The processes from S44 to S47 are performed for all the words (the process ends when there are no more words in the determination in S46), and the comparison result is determined in S34 and S36 in FIG. Therefore, according to the morphological analysis, all the words match if there is only a difference in word order. In this case, a variant word or a difference in notation that can be absorbed by the morphological analysis dictionary (for example, “sunflower” and “sunflower”, “reaping” and “reaping” shown in FIG. 4) are also determined as the same word.

  After the difference between the translated text B1 and the corrected translated text B2 that are the translation results of the original text A1 and the corrected original text A2 that have been determined to be corrected in this way is identified and displayed in S27 of FIG. 5, the user displays the displayed translated text. B1 and the corrected translation B2 are compared, a sentence pair of one of the original sentences and the translation is selected, and the other sentence pair is deleted. Therefore, when a sentence pair to be deleted is selected in S28, it is determined in S29 whether or not the original sentence to be deleted has a rework flag. If there is a rework flag in the original text, it is determined in next S30 whether or not there is an original text with a rework flag before and after the original text to be deleted. That is, it is confirmed that there are two pairs of sentence pairs in which a retouched sentence is input and a difference is identified and displayed. After this confirmation, in S31, the identification display of the translated text B1 and the corrected translated text B2 for the original text A1 and the corrected original text A2 with the rework flag is returned to the normal display, and then the text pair for which the deletion command is issued in S32 is deleted. . FIG. 2 (C) shows the sentence pair of the corrected original text A2 and the corrected translated text B2, but after that, when the new corrected original text A3 “Yellow big flower” is input, the correction is performed in the same manner. It is determined whether or not it is reworked with the original text A2 (this time this will be the original text), and if it is not determined that it is reworked, the corrected translation B3 "It is a big yellow flower." Translation B2 “A sunflower is blooming toward the sun” is normally displayed.

[Effect of machine translation device]
Thus, according to the machine translation device 1 of the above form, when the original text and the translated text are displayed on the display unit 4, when the corrected original text is input by the input unit 3, the original text and the corrected original text Comparing the corrected text with the correct text, and if the discriminating means determines that the text has been corrected, the translated text is compared with the corrected text. By providing an identification display means for identifying and displaying the difference between the two, the user can know at a glance how the translation result changes by reworking, and can quickly perform an appropriate translation check. This improves translation work efficiency.
In particular, the user can easily understand the characteristics (癖) of the machine translation program, such as how the original text is changed and how it is translated, since the translated text and the corrected text can be compared at the same time. The improvement of the quality of translation work can also be expected.
In addition, the identification display means is configured to cancel the identification display of the difference when either the original sentence / translated sentence pair and the corrected original sentence / corrected translated sentence pair are selected and the other is deleted. Therefore, when selecting a sentence pair, the identification display can be surely erased, and there is no need to forget to erase the identification display.

[Explanation of changes]
In addition, this invention is not limited to the said embodiment, In the range which does not deviate from the meaning of this invention, implementation with a various form is possible. Hereinafter, main modification examples will be described.

First, in the above embodiment, when the translated text is obtained by translating the original text, the corrected original text is translated and the corrected translated text is compared with the translated text to check the appropriateness of the translation. For example, there may be a case where there is a reverse translation means for translating an original Japanese sentence into an English sentence and then back-translating the English sentence into a Japanese sentence again.
For example, as shown in FIG. 8A, a translation B1 of “I knew that she went out yesterday.” Is output in response to the original sentence A1 “I knew that she went yesterday.” Then, for the purpose of checking this translated sentence B1, this is reverse translated to obtain a re-original sentence A2 "I found out that she went yesterday" ((B) in the figure). In this translation B1, even if the original text A1 was intended to be "what I knew yesterday" regarding the relationship of "Yesterday", there is a possibility that it will be taken as "that she went out yesterday" as in the original text A2. Yes, the translation B1 is not appropriate.
So, I revised translation B1 and entered “Yesterday I knew that she went out.” And corrected translation B2, and back-translated it. “I knew that she went out yesterday.” Original text A3 is obtained ((C) in the figure). Therefore, it is confirmed that the sentence has the same purpose as the original sentence A1, and the corrected translation sentence B2 is selected.

The present invention can also be applied to the execution of such reverse translation means. That is, as shown in FIG. 9, when the corrected translation B2 is input in S50, the reverse translation instruction is issued in S51, and the re-corrected original A3 is displayed in S52, the translated sentence B1 is displayed before the corrected translation B2. And the re-translated sentence A2 that is the reversely translated sentence (YES in S53, 54), it is determined in S55 whether or not the corrected translation B2 is a modification of the translation B1. In S56, a rework flag is given to both translated sentences, and then the difference between the re-corrected original text A3 and the re-original text A2 obtained by reverse-translating the corrected translated text B2 is identified and displayed (inverted display S in FIG. 8C). It is. The process for selecting and deleting a sentence pair after S58 is the same as the process after S28 in FIG. 5 except for checking the retouching flag of the translated sentence. It is possible to judge by comparing different numbers of characters and morphemes.
Therefore, even when such reverse translation is used, it is possible to easily determine whether or not the translation is appropriate by comparing the re-corrected original text A3 in which the difference is identified and displayed with the re-original text A2, and to quickly perform the translation check. Can do. In particular, the user can easily understand how the translated sentence is changed and how it is back-translated, which improves usability.

Further, in the machine translation apparatus of the above-described form, it is possible to automatically determine whether or not there is a correction and to display an identification in accordance with the input of the corrected original text and the translation thereof (in the case of reverse translation, the input of the corrected translation text and the reverse translation thereof). However, it is possible to select the translation check process by determining whether or not it is correct at a timing desired by the user and performing an identification display by a predetermined input operation with a keyboard or a mouse. good.
Further, the identification display is not limited to the reverse display as in the above-described form, but may be a form that can be visually observed, such as blinking display, underline, change of character size or font, surrounding with a frame or parenthesis, adding a mark, etc. For example, various selections can be made, and a plurality of types of identification displays can be used in combination, depending on the length of the difference.

On the other hand, in the determination of retouching between the original text and the corrected original text (the translated text and the corrected translated text in the case of reverse translation), the threshold values for the number of different characters and the number of words are not limited to the above-described forms, and can be increased or decreased as appropriate. The specific revising judgment method is not limited to the above-mentioned form, and the relationship between the original text and the corrected original text, or the translated text and the corrected translated text is different from each other in the number of characters different from each other, or the different characters are related only in the notation. , A relationship in which only the order is different when divided by word, a relationship in which words that differ when divided by word are stored in the dictionary as synonyms, and the number of words that are different when divided by word are predetermined If at least one of the relationships that are less than or equal to the number is selected and adopted, it is possible to make an appropriate rework determination.
However, since the difference extraction by comparing the number of different characters is shorter than the previous sentence, there is a possibility that a completely different sentence may be corrected, so at least the number of morphemes divided by different words It is desirable to employ comparison.

  In addition, in the judgment of rework, in addition to counting the number of different characters and morphemes, the ratio of the number of different characters and morphemes in the number of characters and morphemes in the whole sentence is calculated, and the difference rate is compared with a threshold value. It is also possible to use a method for judging whether or not the work is correct. However, as in the case where only the count of different characters is used, if the sentence is short, the difference rate becomes large and it is difficult to determine that the text has been reworked. It is preferable to combine the comparison by the real number in combination with at least one of the count by the number.

In the above embodiment, the translation program related to the translation check process is stored in the ROM and read. However, the translation program may be stored in an external storage device such as an HDD and read into the RAM and used when necessary. In addition, it can be recorded on a computer-readable recording medium such as an optical disk (CD-ROM, CD-R / RW, DVD-ROM, etc.) or a magnetic disk (MO, FD, etc.), and read and used by a corresponding driver. it can. In addition, if the machine translation apparatus includes a communication unit with an external network, it is possible to perform the same translation check process by downloading a machine translation program from a network such as the Internet. Of course, the effect obtained is the same as that of the said form.
In addition, the machine translation apparatus of the present invention is not limited to the dedicated product as in the above-described form. For example, even in a general-purpose personal computer system, the machine translation program according to the present invention is similarly obtained from the recording medium or the network. It is feasible.

It is a functional block diagram of a machine translation apparatus. It is explanatory drawing which shows the display state in a display part. It is a flowchart of a normal translation process. It is a schematic diagram of dictionary information. It is a flowchart of a translation check process. It is a flowchart of the retouching judgment of the original text. It is a flowchart of the procedure which compares a word using a morphological analysis. It is explanatory drawing which shows the display state in the display part in the case of reverse translation. It is a flowchart of the translation check process in the case of reverse translation.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Machine translation apparatus, 2 ... Data processing part, 3 ... Input part, 4 ... Display part, 6 ... Editing part, 7 ... Translation part, 8 ... Dictionary, 9 ... Morphological analysis part, 10 ... Syntax analysis part, 11 ... Conversion unit, 12 ... syntax generation unit, 13 ... morpheme generation unit, 14 ... translation check unit.

Claims (3)

An input means for inputting at least the original text in the first language;
A translation means for translating the original sentence input by the input means into a translation in a second language using a dictionary;
A machine translation device comprising at least display means for displaying an original text input by the input means and a translation translated by the translation means,
When the original text and the translated text are displayed on the display means, when the corrected original text corrected by the input text is input by the input means, the corrected original text and the original text are compared, and the following (1 ) To (5) for determining whether or not the relationship has at least one relationship, and determining that the corrected original corresponds to the reworking of the original when the relationship is present;
(1) The number of characters that differ between the original text and the corrected original text is equal to or less than a predetermined number.
(2) The only difference between the original text and the corrected original text is the notation.
(3) When the original text and the corrected original text are divided into words, only the order is different.
(4) When the original text and the corrected original text are divided into words, different words are stored in the dictionary as synonyms.
(5) When the original text and the corrected original text are divided into words, the number of different words is equal to or less than a predetermined number.
When the discriminating means determines that the original text has been revised, the corrected translated text obtained by translating the corrected original text by the translating means is compared with the translated text obtained by translating the original text. A machine translation apparatus comprising: identification display means for performing identification display. Comprising reverse translation means for reversely translating the translated text in the second language translated into the first language by the translation means;
The discriminating means compares the translated sentence with the corrected translated sentence when the corrected translated sentence in which the translated sentence is corrected by the input means, and compares at least one of the following (1) to (5) If it has the relationship, it is determined that the corrected translation corresponds to the revision of the translation,
(1) The number of characters that differ between the translated text and the corrected translated text is equal to or less than a predetermined number.
(2) The only difference between the translated text and the corrected translation is the notation.
(3) When the translated sentence and the corrected translated sentence are divided into words, only the order is different.
(4) When the translated sentence and the corrected translated sentence are divided into words, the different words are stored in the dictionary as synonyms.
(5) When the translated sentence and the corrected translated sentence are divided into words, the number of different words is equal to or less than a predetermined number.
When the discriminating means determines that the identification display means corresponds to the revision of the translated text, the re-corrected original text in which the corrected translated text is reverse-translated by the reverse-translating means and the re-original text in which the translated text is reverse translated The machine translation device according to claim 1, wherein the difference between the two is identified and displayed.   The identification display means is selected when either the original text / translated text pair and the corrected original text / corrected text pair are selected and the other is deleted, or the translated text / re-original text pair 3. The machine translation device according to claim 1, wherein when one of the corrected translated sentence and the re-corrected original sentence pair is selected and the other is deleted, the identification display of the difference is canceled.

Images