JP2002132764A - Machine translation preprocessor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To mechanically translate contents of a dictionary integrated inside a machine translation system with accuracy similar to the case of practically increasing the number of words registered on the dictionary without changing the contents. SOLUTION: In the machine translation preprocessor for preprocessing an input text for improving the accuracy of machine translation before inputting an input text in a source language to a machine translation device, this device is provided with an input part 102 for inputting the text in the source language, a table memory 104 for storing a preprocessing dictionary table 1041 having an entry containing pairs of index words in the source language and correspondent words in a predetermined language, a dictionary reference part 1051 for retrieving an entry having a correspondent index word from the preprocessing dictionary table 1041 concerning a word contained in the inputted text, and a replacing part 1052 for replacing one part of the text with the correspondent word, which is contained in the retrieved entry, in the predetermined language.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation system for translating a text in a source language into a text in a target language, and more particularly to improving the accuracy of machine translation and the speed of machine translation prior to machine translation. Device for pre-processing text in a source language.

[0002]

2. Description of the Related Art A machine translation system translates text written in a natural language into text in another language mainly using computer processing. In machine translation, it is desired that all sentences to be translated in a source language be correctly analyzed and appropriately translated into a target language.

In analyzing a sentence, a dictionary in a source language is referred to. When a word that is not in the dictionary appears in a sentence to be translated in the source language, the analysis of that part fails, and the translation of that part often fails.

Therefore, in order to properly perform machine translation, it is most important to register as many vocabularies as possible in the dictionary of the source language of the system and to enhance the contents of the dictionary.

[0005] However, an attempt to increase the number of words registered in the dictionary requires a great deal of trouble, which is a problem when a practical machine translation system is to be commercially supplied to the market. In particular, as the size of the machine translation system increases and the number of words registered in the dictionary increases, the number of words registered in the dictionary must be increased while maintaining consistency with the dictionary and system that have been built up to then. Is a difficult task, requires a lot of human resources, and increases the cost of the dictionary and thus the cost of the machine translation system.

Further, when using a machine translation system purchased from another source, it is sometimes impossible to increase the number of words registered in the dictionary because the inside of the system cannot be referenced or changed. is there. Therefore, despite such restrictions, there is a need for a machine translation system that can perform machine translation with high accuracy as in the case where the number of registered words in the dictionary is substantially increased. Since the contents of the dictionary built into the machine translation system itself cannot be changed, it is necessary to achieve such an effect by pre-processing the text of the source language in some form as the pre-processing of the machine translation.

[0007]

As a prior art for improving the accuracy of machine translation by preprocessing text in a source language, Japanese Patent Application Laid-Open No. 5-225232 discloses a technique for performing preprocessing of machine translation. There is something. However, this prior art publication attempts to indirectly increase the accuracy of machine translation by automating the pre-editing of text for machine translation, which is usually done manually. It does not directly improve the accuracy of machine translation by providing the same effect as increasing the number of words registered in the dictionary.

Therefore, an object of the present invention is to provide machine translation with the same accuracy as that when the number of words registered in a dictionary is substantially increased without changing the contents of a dictionary incorporated in a machine translation system. To provide a pre-processing device for machine translation that enables

[0009]

SUMMARY OF THE INVENTION A machine translation preprocessing apparatus according to an aspect of the present invention pre-processes an input text in order to improve the accuracy of machine translation before inputting the input text in a source language to the machine translation apparatus. A machine translation pre-processing apparatus for processing, comprising: input means for inputting a text in a source language; an entry including a pair of a headword in a source language and a corresponding word in a predetermined language. Means for storing a processing dictionary table, searching means for searching for an entry having a corresponding headword from the preprocessing dictionary table for words included in the text input by the input means, and searching by the searching means And a replacement unit for replacing a part of the text with a corresponding word of a predetermined language included in the entered entry.

[0010] Preferably, the machine translation preprocessing device further comprises:
Including means for storing a morphological information table including morphological information on words in the source language, and morphological analyzing means for analyzing text with reference to the morphological information table, dividing the text into morphemes, and providing the morphemes to a search means .

Preferably, the predetermined language is a language different from the source language, and each of the entries in the preprocessing dictionary table includes a pair of a headword of the source language and a corresponding word of the predetermined language. In addition, the replacement means further includes semantic information of the headword of the source language, and the replacement means replaces a part of the text with a corresponding word of a predetermined language included in the entry searched by the search means, and further replaces the text. Means for adding semantic information of the corresponding headword to the corresponding word of the determined language, wherein the machine translation preprocessing device is configured to pre-determine the text partially substituted by the providing device by the machine translation device. The apparatus further includes means for deleting the semantic information given by the means for giving from the result of the machine translation into the sentence of the language.

[0012] Preferably, the semantic information corresponding to the headword of the source language in each entry of the preprocessing dictionary table is one of the morphemes constituting the corresponding headword.

Preferably, the headword of the entry in the preprocessing dictionary table includes a combination of ideographic characters, and the semantic information corresponding to the headword may include only one character at the end of the headword.

According to one aspect of the invention, the source language is a language that uses kanji as ideographic characters, and the semantic information includes only one kanji at the end of the corresponding heading.

According to still another aspect, each of the entries in the pre-processing dictionary table includes, in addition to a pair of a headword of a source language and a corresponding word of a predetermined language, a correspondence of a predetermined language. Word replacement information, the replacement unit replaces a part of the text with a corresponding word of a predetermined language included in the entry searched by the search unit, and further replaces a corresponding word of the predetermined language. The machine translation preprocessing device includes a means for providing the word form information, and the machine translation preprocessing device is configured to provide, from the result of machine translation of the text partially replaced by the providing device into a sentence in a predetermined language by the machine translation device, Means for correcting the word form of a corresponding word of a predetermined language included in the result of machine translation according to the word form information given by the means.

[0016] The predetermined language may be the same language as the source language.

[0017]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, machine translation preprocessing devices according to first to seventh embodiments of the present invention will be described. In the following description, the machine translation is a translation from Japanese to English (Japanese-English translation system), but the present invention is not limited to the Japanese-English translation system. [First Embodiment] Referring to FIG. 1, a first embodiment of the present invention will be described.
The machine translation preprocessing device according to the embodiment is configured by a CPU (Central Processing Unit) of a computer,
A control unit 101 for controlling each unit of the machine translation pre-processing device according to the control program; input of a sentence input in a natural language; input of an instruction by an operator for pre-editing processing; Input unit 1 for performing data communication, installation of a control program, and the like
02 and an output unit 103 for displaying or printing an input result by the input unit 102, a conversion result by preprocessing of the source language text under the control of the control unit 101, and the like.
A table memory 104 for storing a dictionary for pre-editing including an entry consisting of a pair of a headword for pre-processing a text in a source language and a translation thereof, and a control program executed by the control unit 101 , A buffer memory 106 used as a work area in the pre-edit processing of the source language, and these units 101, 102, 103, and 10.
4, 105 and 106, and a bus 108 for transferring control programs and address data therebetween.

The control unit 101 includes a program memory 105
From the control program, and by executing the control program, controls each unit via the bus 108;
A machine translation preprocessing device according to the present embodiment is realized.

The input unit 102 includes input devices such as a keyboard, a mouse, a pen, a tablet, and a scanner, a character recognition device, and a CD-ROM (Compact Disc-Re).
adOnly Memory), FD (Flexible)
e Disk), DVD (Digital Video)
Disc) for reading information from a storage medium, a communication device connected to a communication line, and the like.

The output unit 103 includes a display device such as a CRT (cathode ray tube) display, an LCD (liquid crystal display device), a PD (plasma display), and a printing device such as a thermal printer or a laser printer.

The table memory 104, the program memory 105, and the buffer memory 106 include, for example, a mask ROM (Read-Only Memory),
PROM (Erasable Programmable)
e ROM), EEPROM (Electrical
y Erasable ROM), a semiconductor memory such as a flash ROM, a tape-based storage medium such as a magnetic tape or a cassette tape, a magnetic disk such as an FD or a hard disk, or a CD
-ROM, MO (Magneto-Optical),
It includes any combination of an optical disk or a magneto-optical disk such as a DVD, or a card-based storage medium such as an IC card or an optical card including a memory card.

The table memory 104 stores a preedit dictionary table 1041 in which Japanese headwords necessary for preprocessing and their English translations are linked and stored.

Referring to FIG. 2, pre-edit dictionary table 10
Reference numeral 41 stores a plurality of headings in which the vocabulary of the source language and the vocabulary corresponding to the translation of the target language in the machine translation system are paired.

The program memory 105 stores a pre-processing dictionary table 10
Dictionary lookup unit 105 that performs dictionary lookup processing with reference to 41
1 and a program that functions as a replacement unit 1052 for replacing the vocabulary in which the corresponding headword is found in the preprocessing dictionary table 1041 with an appropriate vocabulary according to the contents of the headword.

A buffer memory 106 stores an input sentence buffer 1061 for storing an input source sentence to be translated, and a preprocessing dictionary table 104 for the input sentence.
1. A dictionary lookup result buffer 1062 for storing the result of the dictionary lookup process performed with reference to No. 1, and a replacement for storing the result of replacing the vocabulary of the input sentence with an appropriate vocabulary according to the dictionary lookup result. And a result buffer 1063.

The storage medium 107 includes, for example, a mask RO
M (Read-Only Memory), EPROM
(Erasable Programmable RO
M), EEPROM (Electrically Er
assable ROM), a semiconductor memory such as a flash ROM, a tape-based storage medium such as a magnetic tape or a cassette tape, a magnetic disk such as an FD or a hard disk, or a CD-RO
M, MO (Magneto-Optical), DVD
And any combination of a card-based storage medium such as an IC card or an optical card including a memory card. Storage medium 1
Reference numeral 07 denotes a unit which is detachably attached to the input unit 102 of the machine translation preprocessing apparatus and fixedly carries a machine translation preprocessing program for implementing the machine translation preprocessing of the present embodiment. .

The machine translation pre-processing program is input to the input unit 1
By installing the program in the area of the program 105 via the storage medium reading device 02, the machine translation preprocessing device of the present embodiment can be realized. Or, the input unit 102
The program may be received from any computer on the external network connected via the inside communication device and installed in the program memory 105. When a program is received from a communication network, a communication program for that program is stored in the program memory 105 in advance.
And so on.

Referring to FIG. 3, a control program for realizing the machine translation preprocessing device of the first embodiment has the following control structure. The control unit 101 converts the Japanese sentence input to the input unit 102 into an input buffer 106.
1 (step 201).

The dictionary lookup unit 1051 checks whether there is a corresponding entry in the pre-processing dictionary table 1041 for each word of the Japanese sentence stored in the input buffer 1061 (202). If there is a corresponding entry, the headword and the translation of the entry are stored in the dictionary lookup result buffer 1062, and the process proceeds to step 204. If there is no corresponding entry, the input sentence buffer 10
The contents of 61 are copied to the replacement result buffer 1063, and the flow advances to step 205.

In step 204, the replacement unit 1052
The headword stored in the dictionary lookup result buffer 1062 in the Japanese sentence stored in the input sentence buffer 1061 is converted into the corresponding translation stored in the dictionary lookup result buffer 1062 and stored in the replacement result buffer 1063.

Subsequently, in step 205, the Japanese sentence stored in the replacement result buffer 1063 is provided to a machine translation system (not shown) via the output unit 103 to perform translation. Thus, a translation result is obtained. Even if the obtained translation result is a word that is not in the dictionary of the machine translation system, it is replaced by an appropriate translation before machine translation by performing preprocessing by the machine translation preprocessing device. The obtained translation result is more appropriate as compared with the case where the preprocessing is not performed.

For example, a case where a sentence "He has entered Princeton University" is translated will be described as an example. It is assumed that the word "Princeton University" was not found in the machine translation system dictionary. If the preprocessing is not performed, the word "Princeton" will not be properly translated, and a desired result cannot be obtained. On the other hand, if the contents of the pre-processing dictionary table 1041 are as shown in FIG.
With the function 051, the word "Princeton University" in the input sentence buffer (see the upper part of FIG. 4) is changed to "Pr
The word "inceton University" is stored in the dictionary lookup result buffer as a translation. Further, the function of the substitution unit 1052 replaces the word “Princeton University” in the sentence buffer with “Princeton University” and stores “He is Princeton Universi” in the substitution result buffer 1063.
I entered ty. And stored. By giving this sentence to the machine translation system and translating it, "He entered
Princeton University ".

On the other hand, when such preprocessing is not performed and the contents of the input sentence buffer 1061 are directly provided to the same machine translation system and translated, the result is "He entered Princeton university". Comparing the two, when the machine translation pre-processing system of the present embodiment was performed, the portion of "Princeton University" in the original text was properly translated, and a more appropriate English text was obtained as a translation result.

In the system according to the present embodiment, "Princeton University" and "Princeton
An appropriate translation can be obtained without registering an entry consisting of a pair with the word "University". Even if it is impossible to change the contents of the dictionary of the machine translation system, it is possible to obtain substantially the same effect as increasing the number of words registered in the dictionary of the machine translation system.
Moreover, in the machine translation preprocessing device of the present embodiment, the preprocessing dictionary table 1041 may be created independently of the dictionary format of the machine translation system, and the number of registered words can be freely set. Therefore, without changing the inside of the machine translation system at all, it is possible to obtain the effect that the accuracy of translation can be improved as well as substantially increasing the number of registered words in the dictionary included in the machine translation system. Can be. [Second Embodiment] The machine translation preprocessing device according to the second embodiment is different from the device of the first embodiment in that morphological analysis is performed on an input sentence. Referring to FIG. 5, the device according to the second embodiment has the same structure as the first device.
The difference from the apparatus of the embodiment is that the table memory 10
4 includes a morphological information table 1042 for morphological analysis in addition to the preprocessing dictionary table 1041, and the program memory 105 is stored in the input sentence buffer 1061 in addition to the dictionary lookup unit 1051 and the replacement unit 1052. A morphological analysis unit 1054 for performing a morphological analysis on an input sentence of a source language is included, and a buffer memory 106 includes an input sentence 1061, a dictionary lookup result buffer 10
62, in addition to the replacement result buffer 1063, a morphological analysis result buffer 1064 for storing a morphological analysis result by the morphological analysis unit 1054 is further included. 5, the same components as those in FIG. 1 are given the same reference numerals and the same names. Their functions are the same. Therefore, detailed description thereof will not be repeated here. Note that morphological analysis is a well-known matter in this field, and its realizing method is also well-known. Therefore, the details of the contents will not be described here.

Referring to FIG. 6, the program for realizing the machine translation preprocessing device of the second embodiment is the same as the control program of the device of the first embodiment shown in FIG. Instead, the difference is that steps 3021 and 3022 corresponding to the morphological analysis unit 1054 are included. 6, the same steps as those in FIG. 3 are denoted by the same reference numerals as those in FIG.

In the apparatus according to the second embodiment, a morphological analysis is performed on the input sentence stored in the input sentence buffer 1061 in step 3021 with reference to the morphological information table 1042, and the result is obtained by the morphological analysis. It is stored in the result buffer 1064. Based on the result of the morphological analysis, the dictionary lookup unit 1051
At 22, it is checked whether a corresponding entry exists in the preprocessing dictionary table 1041. If the input sentence is the result of morphological analysis,
Since it is appropriately divided into morphemes, the dictionary lookup unit 105
1 only needs to search a corresponding entry for only the divided morphemes and their combination. Therefore, the processing speed can be increased compared with the case of performing dictionary lookup including a result obtained by dividing a sentence in an inappropriate form, and the dictionary lookup accuracy,
Therefore, the accuracy of the final machine translation can be improved.

For example, consider the above sentence "He has entered Princeton University." This sentence is the result of the morphological analysis "he / ha / Princeton / college /
/. Is divided into morphemes. At the time of dictionary lookup, corresponding entries may be searched for only these words and their combinations. As a result, there is no need to search for an entry with a character string that should not exist as a dictionary entry, such as "He is a pu" or "Just enrolled in a school". It turns out that it improves.

As described above, in the apparatus according to the second embodiment, in addition to the effects of the first embodiment, the speed of the preprocessing can be increased, and the accuracy thereof can be improved. As a result, the accuracy of translation by the machine translation system can be improved. Third Embodiment FIG. 7 is a block diagram showing a machine translation preprocessing system according to a third embodiment of the present invention. This apparatus differs from that shown in FIG. 1 in that each entry of the preprocessing dictionary table 1041 includes a headword, a translation word, and semantic information, and that the program memory 105
Deletes the part corresponding to the semantic information from the result of machine translation using the replaced word including the semantic information of the word obtained as a result of the dictionary lookup, in addition to the dictionary lookup unit 1051 and the replacement unit 1052 Including a post-processing unit 1053 for performing post-processing,
However, a translation result buffer 1065 for storing a result translated by a machine translation system (not shown) is included.

FIG. 8 shows an example of a heading of the preprocessing dictionary table 1041 according to the third embodiment. For example, the heading "Major bleeding" is assigned the translation "hemorrhage", but the semantic information is [bleeding]
Is assigned. The same applies to other headings.

FIG. 9 shows a flowchart of a control program in the device according to the third embodiment. Referring to FIG. 9, control unit 101 stores a Japanese sentence input to input unit 102 in input sentence buffer 1061 (40).
1).

The dictionary lookup unit 1051 lookups the entries of the pre-processing dictionary table 1041 for Japanese sentences stored in the input sentence buffer 1061. Subsequently, in step 403, it is determined whether or not the dictionary lookup is successful. If the dictionary lookup is successful, the headword, translation, and semantic information of the result are stored in the dictionary lookup result buffer 1062, and control proceeds to step 4.
04, and if unsuccessful, the input statement is copied to the replacement result buffer 1063, and control proceeds to step 405.

In step 404, the replacement unit 1052 converts the headword stored in the dictionary lookup result buffer 1062 in the Japanese sentence stored in the input sentence buffer 1061 into a translated word and stores it in the replacement result buffer 1063. At this time, the semantic information stored in the dictionary lookup result buffer 1062 is stored in the replacement result buffer 1063 in a format that can be correctly interpreted by the machine translation system. For example, if a machine translation system translates the relevant part of the word before the symbol “◎” when the semantic information is added after the symbol “◎” to a certain word, If the user has a function of using a word according to the subsequent semantic information, in step 404, the semantic information is added after the translated word with a symbol “◎” therebetween.

Subsequently, the Japanese sentence stored in the replacement result buffer 1063 is translated via the output unit 103 by a machine translation system (not shown). The result is stored in the translation result buffer 1065.

At step 406, it is determined whether or not the sentence has the semantic information added at step 404. If the sentence has been added, control is passed to step 407; If so, control goes to step 4.
Go to 08 respectively.

In step 407, the translation result buffer 1
Post-processing for deleting the portion corresponding to the semantic information given in step 404 from the translation result sentence stored in 065 is performed, and the process of re-storing in the translation result buffer 1065 is performed. This processing corresponds to the processing performed by the post-processing unit 1053 in FIG. In the case of the above description, the word following the symbol “◎” is deleted. The resulting sentence is output in step 408.

The operation of the machine translation preprocessing apparatus according to the third embodiment will be described with a more specific sentence as an example.
As an example, it is assumed that a sentence "pregnant woman has caused massive bleeding." As a result of dictionary lookup for this sentence,
Suppose that "hemorrhage" was dictionary searched for the word "massive bleeding". At this time, the dictionary lookup result buffer 10
In the column 62, semantic information “disease” is stored in addition to a pair of a headword and a translated word such as “mass bleeding” and “hemorrhage”. As shown in FIG. 8, "massive bleeding" and "hemorrhage"
This is because semantic information “disease” is added to the pair.

As a result of step 404, the content of the replacement result buffer 1062 is "pregnant woman has had <hemorrhage ◎ sickness." If you give this to a machine translator,
As described above, this machine translation system has a function of translating a related part of the word before the symbol “◎” in consideration of the semantic information added after the symbol “◎”.
"A pregnant woman caused hemorrhage ◎ illness." Is obtained as a translation result. Since “◎ illness” is deleted in step 407 by the post-processing unit 1053, the translated text finally obtained is “A pregnant woman caused hemorr
hage. " This result is obtained by machine-translating the sentence "A pregnant woman who has caused hemorrhage." set up large-scal
e bleeding. "shows that more appropriate English sentences are obtained. In other words, the word "raised" has been replaced with a more appropriate word reflecting the meaning of "massive bleeding" from "set up" to "caused".

As described above, according to the device of the third embodiment, the preprocessing dictionary table is provided with the semantic information, and the machine translation system is provided with the semantic information in addition to the English translation. Therefore, the machine translation system can perform more appropriate translation using this semantic information,
The translation accuracy of the entire translation system can be improved. [Fourth Embodiment] A machine translation pre-processing apparatus according to a fourth embodiment has substantially the same configuration as the apparatus according to the third embodiment. The difference is that a partial morpheme of the headword is used.
That is, the headword is subjected to morphological analysis in advance, and one of the obtained morphemes is used as semantic information.

FIG. 10 shows an example of the contents of the preprocessing dictionary table 1041 in this case. Referring to FIG. 10, for example, a headword “massive bleeding” is morphologically analyzed as “massive / bleeding”. Therefore, of the two morphemes "mass" and "bleeding", the word "bleeding" which is considered more appropriate to represent the meaning is used as semantic information.

When the preprocessing dictionary table 1041 is created in this way, the hardware configuration, functional block configuration, and software configuration are exactly the same as those of the apparatus of the third embodiment, and the same as in the third embodiment, Pre-editing processing for realizing high-precision machine translation using semantic information becomes possible.

For example, in the device of the third embodiment,
The contents of the replacement result buffer 1063 for the input sentence “Pregnant woman has caused massive bleeding.”
age ◎ I was ill. It was. On the other hand, in the apparatus of the present embodiment, the content of the replacement result buffer 1063 is “pregnant woman has experienced <hemorrhage bleeding>”. As a result of subjecting this sentence to machine translation, an appropriate translation was selected according to the semantic information "bleeding", and "A pregnant w
oman caused hemorrhage ◎ bleeding. ” As a result of deleting the word next to “◎” by the post-processing by the post-processing unit 1053, “A pregnant woman c
aused hemorrhage. ", which is the same translation as that obtained by the apparatus of the third embodiment.

In order to add semantic information to the headword in the pre-processing dictionary table, it is necessary to determine the semantic information for the headword using a large-scale thesaurus or to manually check the semantic information. Therefore, the cost of the pre-processing dictionary table may increase. However, in the apparatus according to the fourth embodiment, a headword is subjected to morphological analysis in advance, and one of the divided morphemes is used as semantic information. In particular, in the case of a compound word composed of a combination of Japanese kanji, one of the divided morphemes, for example, the last one, often appropriately represents the semantic information of the compound word. Therefore, it is not necessary to use a large-scale thesaurus, and semantic information can be added to each headword.
In addition, an operation of performing morphological analysis and assigning one of the words to the headword as semantic information can be automatically performed. Therefore, an increase in the cost of the pre-processing dictionary table 1041 can be suppressed. [Fifth Embodiment] A machine translation preprocessing apparatus according to a fifth embodiment has the same hardware configuration, functional block configuration, and the same configuration as those of the machine translation preprocessing apparatus according to the third embodiment shown in FIG. Only the method of adding the meaning information to each headword in the preprocessing dictionary table differs depending on the software configuration. Therefore, the configuration and functions of the apparatus according to the fifth embodiment are similar to those of the machine translation preprocessing apparatus according to the fourth embodiment.

In the pre-processing dictionary table of the apparatus according to the fifth embodiment, one character at the end of the headword is used as the meaning information added to each headword. For example, FIG.
As shown in, "blood" for "massive bleeding", "person" for "hemophilia" and "primary" for "hemochrome"
Like.

In general, in a compound word composed of a combination of Japanese kanji, the last word best represents the meaning of the compound word. Kanji is an ideographic character and can represent one character. In compound words, the combination of "adjective kanji" + "kanji representing the concept to be modified" is common. That is, if a word consists of two kanji, for example,
Two letters so that "bleeding", "hemostatic" and "hematuration" all belong to the group of words related to "blood", and "white cloud", "black cloud" and "blue cloud" all belong to the group of words related to "cloud". Of the characters, the last kanji character often indicates a comprehensive word group to which the word belongs. Therefore, one character at the end of the headword is adopted as the meaning information of the headword as in the fifth embodiment.

In the case of the fifth embodiment, the contents of the replacement result buffer 1063 for the above-mentioned input sentence “Pregnant woman has caused massive bleeding.”
◎ Blood was raised. ". As a result of subjecting this sentence to machine translation, an appropriate translation was selected according to the semantic information of "blood" and "A pregnant woman caused hemorrhage ◎ bloo
d. "is obtained. As a result of deleting the word next to “◎” by the post-processing by the post-processing unit 1053, “A
"Pregnant woman caused hemorrhage.", which is the same translation as that obtained by the devices of the third and fourth embodiments.

As described above, when the last character of the headword is adopted as its semantic information, the work of assigning the semantic information is further simplified as compared with the case of the fourth embodiment. Therefore, it is possible to reduce the labor required when creating the pre-processing dictionary table 1041, and suppress an increase in cost. Further, when such a method is adopted, it is possible to easily add the semantic information at the time of the work by the replacing unit 1052 at the time of the preprocessing, without adding the semantic information to the preprocessing dictionary table 1041. You. [Sixth Embodiment] The configuration of hardware and functional blocks of a machine translation preprocessing device according to a sixth embodiment is the same as that of the third embodiment shown in FIG. The device of the sixth embodiment differs from the third embodiment in that each entry in the preprocessing dictionary table 1041 is made up of a headword, an English translation, and English word form information. It is. FIG.
An example of the contents of the preprocessing dictionary table 1041 according to the embodiment will be described.

In the example shown in FIG. 12, word form information on a plurality of English words is given as word form information.

FIG. 13 shows a flowchart of a control program of the machine translation preprocessing device according to the sixth embodiment. The flowchart shown in FIG. 13 is similar to the flowchart shown in FIG.
In the flowchart of the apparatus according to the embodiment, steps 403 and 404 are replaced by steps 503 and 504, respectively, and steps 406 and 407 are replaced by steps 506 and 507, respectively.
Steps 401, 402, 405, and 408 are shown in FIG.
Is the same as that shown in FIG. Therefore, detailed description of these steps will not be repeated here.

In step 503, it is determined whether or not the dictionary lookup has been successful as a result of the dictionary lookup process in step 402. If the dictionary lookup is successful, the resulting headword, translated word, and word form information are stored in the dictionary lookup result buffer 1062, and control proceeds to step 504. If dictionary lookup fails, the input sentence is copied to the replacement result buffer 1063, and control proceeds to step 405.

At step 504, the replacement unit 1052 replaces the headword stored in the dictionary lookup result buffer 1062 in the Japanese sentence stored in the input sentence buffer 1061 with a translated word, and stores it in the replacement result buffer 1063. At this time, the contents of the replacement result buffer 1063 are corrected to a format in which the plural form information can be correctly interpreted by a machine translation system to be used later, in order to make the translated word into the plural form by the word form information corresponding to the headword later. . Here, in order to be able to correctly interpret and translate plural information in a machine translation system to be used later, a heading “◎” is inserted as an example after the English translation obtained by dictionary lookup. It is assumed that a word that represents the meaning of the word and that can be easily pluralized by the machine translation system is added.

At step 405, the replacement result buffer 10
The content of 63 is translated by a machine translation system.

In step 506, step 504
Then, it is determined whether or not the word form information has been added. If the word form information has been added, the control proceeds to step 507, and if not, the control proceeds to step 408.

In step 507, the word form of the word corresponding to the headword is corrected from the translation result of the information given to the input sentence in step 504 and the word form information obtained as a result of the dictionary lookup in step 402. Here, as a result of the machine translation, immediately after the English translation corresponding to the headword, the symbol "◎"
The plural form of the word assigned in step 504 is output and stored in the translation result buffer 1065. If the word immediately after the symbol “◎” is translated as a plural form by the machine translation system, it is understood that the word immediately before the symbol must be plural. Therefore, based on the plural form obtained as a result of the dictionary lookup, the English translation word in the translation result is corrected to the plural form, and the result is stored in the translation result buffer 1065 again.

At step 408, the translation result is output, and the process ends. A specific example will be described. It is assumed that "three pregnant women have been called" has been given as an input sentence. By dictionary lookup for the pre-processing dictionary table shown in FIG.
urient "and the plural form information" -s ".

In step 504, the word “three pregnant women have been called” is converted into a sentence “three <parturient ◎ girls> have been called” by the processing of the replacement unit 1052. By giving this sentence, which is information added to the original sentence for correcting the inflected form after translation, to the machine translation system, the "girl" immediately after "◎"
e parturient ◎, girls, were called. " Since “girl” is translated as “girls”, it is understood that “parturient” immediately before “◎” must be pluralized. On the other hand, step 40
As a result of dictionary lookup processing in 2 and 503, "parturien
I know that the plural of "t" is "parturients". Therefore, in step 507, as post-processing, “par” of “Three parturient ◎, girls, were called.”
turient ”is corrected to its plural form“ parturients ”, and the symbol“ ◎ ”assigned in step 504 and the part“ ◎, girs ”corresponding to the word immediately after it are deleted from the translation result. As a result, "Three parturients were calle
d. "is obtained.

As described above, according to the machine translation preprocessing apparatus of the sixth embodiment, it is possible to correct word forms such as plural forms of translated words correctly, and to improve translation accuracy. [Seventh Embodiment] A machine translation preprocessing apparatus according to a seventh embodiment has the same hardware configuration, functional block configuration, and software configuration as the apparatus of the first embodiment shown in FIG. Have. In the apparatus according to the seventh embodiment, each entry of the preprocessing dictionary table 1041 is
As shown in FIG. 4, the third embodiment differs from the first embodiment in that a pair of a headword in Japanese as a source language and a word in which the headword is paraphrased in the same Japanese. Otherwise they are the same.

Here, "paraphrasing" means replacing words that are not included in the dictionary of the machine translation system with words having substantially the same meaning as those registered in the dictionary of the machine translation system. Thus, in the input sentence,
Improve the accuracy of machine translation by applying the process of replacing words that are not included in the dictionary of the machine translation system with words with similar meanings included in the machine translation system to the input sentence as preprocessing in advance Those skilled in the art will readily understand what will be done.

For example, consider a case where a dictionary of a certain machine translation system does not have “pregnancy” but has “pregnancy”. If the input sentence is "The pregnancy of a princess has been conveyed."
"Pregnancy" could not be translated well, for example, "Princess' bo
som pregnant was transmitted. " However, by rephrasing this as "pregnancy of the princess was conveyed" by preprocessing, when translated by the same machine translation system, "Princess' pregnancy was transm
itted. ", and a more appropriate translation result can be obtained as compared with the case where no paraphrase is made.

In the apparatus of this embodiment, not only synonyms such as “pregnancy” and “pregnancy” are registered as the pre-processing dictionary table 1041, but also “estimation”, “estimation”
Alternatively, such a misspelled word may be registered as a paraphrase. Further, an entry may be registered in which a compound word such as “collection of material” is paraphrased into a combination of “collection of material” and a basic word. The point is that words not registered in the dictionary of the machine translation system are replaced with words registered in the dictionary of the machine translation system or a combination thereof by preprocessing.

According to the apparatus of this embodiment, each entry of the preprocessing dictionary is a pair of words in the source language, so that the entries are composed of pairs of the source language and the translated words. There is an effect that creation of the processing dictionary table is easy.

The embodiments disclosed this time are to be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a machine translation pre-processing device according to a first embodiment of the present invention.

FIG. 2 is a diagram illustrating an example of the contents of a preprocessing dictionary table of the device according to the first embodiment.

FIG. 3 is a flowchart of a control program of the device according to the first embodiment.

FIG. 4 is a diagram showing the contents of each buffer in the device according to the first embodiment.

FIG. 5 is a functional block diagram of a machine translation pre-processing device according to a second embodiment of the present invention.

FIG. 6 is a flowchart of a control program of the device according to the second embodiment.

FIG. 7 is a functional block diagram of a machine translation preprocessing device according to a third embodiment of the present invention.

FIG. 8 is a diagram illustrating an example of contents of a pre-processing dictionary table of the device according to the third embodiment.

FIG. 9 is a flowchart of a control program of the device according to the third embodiment.

FIG. 10 is a diagram illustrating an example of contents of a pre-processing dictionary table of the device according to the fourth embodiment.

FIG. 11 is a diagram illustrating an example of contents of a pre-processing dictionary table of the device according to the fifth embodiment.

FIG. 12 is a diagram illustrating an example of contents of a preprocessing dictionary table of the device according to the sixth embodiment.

FIG. 13 is a flowchart of a control program of the device according to the sixth embodiment.

FIG. 14 is a diagram illustrating an example of contents of a pre-processing dictionary table of the device according to the seventh embodiment.

[Explanation of symbols]

101 control unit, 102 input unit, 103 output unit, 1
04 table memory, 105 program memory, 1
06 buffer memory, 1041 pre-processing dictionary table, 1042 morphological information table, 1051 dictionary lookup unit, 1052 replacement unit, 1053 post-processing unit, 1054
Morphological analysis unit, 1061 input sentence buffer, 1062
Dictionary lookup result buffer, 1063 replacement result buffer, 1064 morphological analysis result buffer, 1065 translation result buffer.

Claims (8)

[Claims] 1. A machine translation pre-processing device for pre-processing input text in order to improve the accuracy of machine translation prior to inputting the input text of the source language into the machine translation device, comprising: Input means for inputting text, means for storing a pre-processing dictionary table having an entry including a pair of a headword of the source language and a corresponding word of a predetermined language, and the input means Searching means for searching the pre-processing dictionary table for an entry having a corresponding headword for a word included in the input text; and the predetermined means included in the entry searched by the searching means. Machine translation pre-processing device, comprising: a replacement unit for replacing a part of the text with a language equivalent. 2. A means for storing a morphological information table including morphological information on words in the source language; analyzing the text with reference to the morphological information table, dividing the text into morphemes, and performing the search 2. The machine translation pre-processing apparatus according to claim 1, further comprising: a morphological analysis unit for giving to the unit. 3. The predetermined language is a language different from the source language, and each of the entries in the pre-processing dictionary table has a correspondence between a headword of the source language and the predetermined language. In addition to the pair with the word, further includes semantic information of the headword of the source language, the replacement means, the corresponding word of the predetermined language included in the entry searched by the search means, the text And further comprising means for adding semantic information of a headword corresponding to the corresponding word of the predetermined language, wherein the machine translation preprocessing device is partially replaced by the providing means. From the result of machine translation of the text that has been translated into a sentence of the predetermined language by the machine translation device,
3. The machine translation pre-processing apparatus according to claim 1, further comprising: means for deleting the semantic information given by the giving means. 4. The method according to claim 2, wherein the semantic information corresponding to the headword of the source language in each of the entries of the preprocessing dictionary table is one of morphemes constituting the corresponding headword. Machine translation preprocessor. 5. The headword of an entry of the preprocessing dictionary table includes a combination of ideographic characters, and the semantic information corresponding to the headword includes only one character at the end of the headword. Machine translation preprocessor. 6. The machine translation pre-processing device according to claim 5, wherein the source language is a language using kanji as an ideographic character, and the semantic information includes only one kanji at the end of a corresponding heading. 7. Each of the entries of the preprocessing dictionary table includes a pair of a headword of the source language and a corresponding word of the predetermined language, and a corresponding word of the predetermined language. Further including word form information, wherein the replacement unit replaces a part of the text with a corresponding word of the predetermined language included in the entry searched by the search unit, and further replaces the predetermined language. The machine translation preprocessing device includes means for adding word form information of a corresponding word, wherein the machine translation preprocessing device converts the text partially substituted by the providing means into a sentence of the predetermined language by the machine translation device. From the translated result,
2. The apparatus according to claim 1, further comprising: a unit configured to correct a form of the corresponding word of the predetermined language included in the result of the machine translation according to the form information provided by the providing unit. 3. 3. The machine translation preprocessing device according to 2. 8. The machine translation preprocessing apparatus according to claim 1, wherein the predetermined language is the same language as the source language.