JPH08329091A - Machine translation device - Google Patents

Abstract

PURPOSE: To guarantee a translation result of high quality even if an original text has unknown words, etc., mixed. CONSTITUTION: A term freezing means 2 substitute a frozen term which is not translated for an arbitrary term in the original text l and outputs it. A collation table 7 contains information on the frozen term and unfrozen term corresponding to the term frozen by the term freezing means 2. A machine translating means 3 translates the original text containing the frozen term outputted by the term freezing means 2 and outputs the frozen term as a term which is not translated. A term unfreezing means 4 refers to the collation table 7 and converts the frozen term into the corresponding unfrozen term and outputs when the translation outputted by the machine translating means 3 includes the frozen term.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device for translating an original document described in natural language into a translated document in a target language.

[0002]

2. Description of the Related Art In a machine translation device that translates an arbitrary language into another arbitrary language, the handling of unknown words becomes a problem. This unknown word is a character string that is not registered in the dictionary of the machine translation device, and because the machine translation device cannot obtain information for translation related to such a character string, it fails in the analysis of the original sentence. , It adversely affects the translation generation. Therefore, it is necessary to take some measures for unknown words.

As a countermeasure for this, for example, Japanese Patent Laid-Open No.
As described in Japanese Patent Laid-Open No. 306766, if an unknown English character string or the like can be recognized as a Roman character string, the character string is replaced with Kana. That is, this is to prevent words that are not registered in the dictionary in the process of performing machine translation of the entered English character string, etc., for the purpose of preventing the appearance of English character strings in the translated text and improving the quality of the translated text. When it appears, it is determined whether or not the word can be recognized as a Roman character, and if it is recognized as a Roman character, it is converted to Kana using the Roman character Kana table and the translation result is output.

[0004]

However, in the above-mentioned conventional apparatus, it is necessary to provide a storage unit for unknown words and a Romaji-kana table for unknown words. This is because the unknown words are stored in the dictionary of the machine translation device. It is equivalent to registering. Considering the time and effort required for this registration, it is certainly effective when converting an English character string to Kana through Roman letters, but there is no doubt as to whether it can be applied to other unknown words. There was a problem that From this point of view, it has been desired to realize a machine translation device that can guarantee a high-quality translation result without adversely affecting the generation of a translated sentence of an original sentence in which unknown words are mixed.

[0005]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention is to change the viewpoint, and the behavior which is originally a problem of the machine translation device with respect to the input unknown word, that is, the unknown word On the other hand, the machine translation device is constructed so as to construct a useful translation device by utilizing the property that the translation process is not actively performed. Therefore, in order to prevent the machine translation device from actively translating the conventional unknown words of the input sentence (original sentence) and the portion that is not desired to be translated, the present invention uses them. A means to make a fake unknown word was provided.

That is, the machine translation apparatus of the present invention replaces an arbitrary term in the input original sentence with a frozen term that is not a translation target and outputs the term, and a frozen term corresponding to the arbitrary term. If the original text contains a frozen term, it will not be translated. And a term decompressing means for converting and outputting the corresponding decompressed term when the frozen term matches, with reference to the collation table for the translated sentence output from this machine translating means. It is characterized by having.

[0007]

In the machine translation device of the present invention, the term freezing means replaces an arbitrary term with a frozen term and outputs the original sentence. Further, the collation table stores information on frozen terms and defrosted terms corresponding to arbitrary terms. When the original text from the term freezing means includes a frozen term, the machine translation means outputs the frozen term as a non-translation target. The term decompressing unit determines whether or not there is a frozen term conforming to the matching table in the translated text output from the machine translating unit, and if there is a matching frozen term, converts it to the corresponding defrosting term. Output.

[0008]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a machine translation device of the present invention. The device shown in the figure is a device that pre-edits the original sentence 1 and machine-translates it to output a translated sentence 5. As a configuration therefor, the term freezing means 2, machine translation means 3, term decompressing means 4, dictionary means 6 are provided. , A collation table 7 is provided.

Original sentence 1 is a sentence written in natural language that is input to the machine translation apparatus. The term freezing unit 2 has a function of replacing an arbitrary term in the input original sentence 1 with a frozen term that is not a translation target and outputting it to the machine translation unit 3. The freezing means an operation in which the machine translation unit 3 corresponds to an unknown word that is not a translation target. In addition, the term freezing means 2 has a function of storing information of the term, the frozen term, and the defrosted term in the collation table 7 when the freezing process is performed.

The machine translation unit 3 is a known machine translation unit that translates an original sentence described in a natural language and generates a translated sentence of a target language. Further, the machine translation unit 3 identifies the original sentence as a frozen state. If a possible part is included, the part is configured to be output as a non-translation target.
The term decompressing unit 4 selects, from the terms in the translated text output by the machine translation unit 3, a term frozen by the term freezing unit 2.
It has a function of referring to a collation table 7 which will be described later and restoring it to a term of an original sentence or converting it into a term of a target language as needed. The operation of restoring / converting will be referred to as decompression hereinafter. The translated sentence 5 is a sentence resulting from the decompression performed by the term decompressing means 4.

The dictionary means 6 is a known dictionary means used by the machine translation means 3 at the time of translation, and used by the term freezing means 2 and the term decompressing means 4 as necessary, and the character type for an arbitrary word. It has information, code system and morpheme information. The morpheme means, for example, in English or Japanese, a group of characters such as words having a certain meaning. The collation table 7 is created by the term freezing means 2, stores the freezing method and the defrosting method of the term frozen by the term freezing means 2, and the term defrosting means 4 refers to the defrosting operation to perform the defrosting operation. It is for.

FIG. 2 is a diagram showing a configuration example of the collation table 7. This collation table 7 is created by the term freezing means 2 at the time of freezing or created in advance. In the figure, 10 indicates the general format of the information stored in the collation table 7, which is composed of three description fields (from the left, "term", "frozen term", and "decompressed term") separated by slash "/". Has been done. Here, in the “term” field, a portion of the original sentence 1 that is not a target of translation processing, that is, a translation non-target term is described. Since this translation non-target term may be composed of a plurality of words, it is necessary to show it in the original sentence 1 as one translation non-target term part. In that case, an unknown word equivalent word (hereinafter, referred to as a frozen term) corresponding to the translation non-target term described in the “term” field is described in the “frozen term” field. In the “decompression term” field, the term of the target language that should be replaced when the translated sentence 5 is generated with respect to the translation non-target term described in the “term” field (hereinafter, referred to as decompression term). ) Is described.

The frozen and thawed terms may be omitted if they are equivalent to the terms listed in the term field. In the figure, 11 indicates this state, which is information for a typical unknown word "ODA" in English.
In the English sentence, "ODA" may be an English expression of "Oda" which is a proper noun or "ODA" which is an abbreviation of "government development assistance". In this case, both the frozen term and the thawed term are the original terms. The description in these fields is omitted because there is no problem with the notation of the term.

In the figure, 12 is information for the term "Machine Translation" when the original text and the translated text are different. In this case, the frozen term is described as “μτ” which does not exist in the dictionary means 6 on the assumption that it is treated as an unknown word, and the decompressed term has the replacement notation “MT” in the translated sentence.

In the figure, 13 is a Japanese verb "cop".
This is information when the “copy” part of “y” is treated as a translation non-target term. Usually, most of the translation non-target terms are nouns, but this example shows a case where a verb is treated as a translation non-target term. As can be seen from this example, a term such as a verb that does not uniquely correspond between the source language and the target language is a redundant term (“co
It is also possible to include a description including the "do" portion of "py do". The notation of frozen terms in the collation table 7 is not particularly limited, but in the present embodiment, it is realized by using a symbol string as in the example of 12. That is, normally,
Since there is no term such as "τ", it is assumed to be an unknown word.

The above-mentioned collation table 7 is temporarily created during the operation of the machine translation apparatus. However, an equivalent collation table is prepared in advance and is used when the machine translation apparatus is executed. Good. Further, after the processing of the machine translation device is executed, the contents of the used collation table 7 can be saved and accumulated to be used as learning knowledge. That is, when the original sentence 1 is given to the term freezing means 2, the term freezing means 2
May refer to the collation table 7 to determine whether the original sentence 1 does not include a non-translated term.

Next, the operation of the machine translation device will be described. FIG. 3 shows the term freezing means 2 in this machine translation device.
3 is a flowchart showing the operation of FIG. When the original sentence 1 is input, the term freezing means 2 first divides the original sentence 1 into words (step S1). The division into words is performed by the technique of morphological analysis. That is, the term freezing unit 2 first obtains information for morphological analysis using the dictionary unit 6. Then, using the morphological analysis information obtained from the dictionary means 6, it is checked whether or not the divided words of the original sentence 1 are to be translated (step S2). Here, that the word is not a translation target, that is, is not a translation target, includes the following requirements.

(A) The word is an unknown word. (B) The word is in a language different from the original sentence. (C) Do not intentionally set the word as a translation target.

The information as to whether or not it is an unknown word in the above (a) can be judged by the information obtained from the dictionary means 6, that is, whether or not the term is registered as a morpheme in the dictionary. As for (b), it is possible to make judgments based on information from the character type (Kanji / Hiragana / Katakana / Alphabet / Symbol, etc.) of the word in the original sentence, and the code system (JIS / Shift JIS, etc.). Also, these pieces of information can be obtained from the dictionary means 6.

Further, with respect to (c), it can be determined when the user of the apparatus has designated in advance that a particular word or word string is not to be translated. To achieve this, there is a method to mark non-translated terms in the original sentence. As a method of this mark, for example, a symbol such as a parenthesis generally called a preedit symbol is used, and the translation non-target term is parenthesized. Further, when the user of this apparatus designates, the input / output interface such as a keyboard or a pointing device (not shown) is used.

In this way, when it is determined that an arbitrary word in the original sentence 1 is not to be translated, that is, to be frozen (step S3), it is frozen corresponding to the requirements (a) to (c). The method is determined (step S4). For example,
For the unknown word “ODA” corresponding to (a), a method of fixing the term as it is is adopted, as indicated by 11 in FIG. In addition, the term “copy” that corresponds to (b)
2), the freezing term is set to “★” and the decompressing term is set to “copy”, and the English “copy” that appears in the original Japanese sentence is identified and frozen. Adopt the method. Further, the term "Machine Translation" corresponding to (c) should be able to obtain a translated word that is not a problem even if it is normally translated, but in this case, the intention of the user of this device is high. Due to
The method of translating as an abbreviation "MT" as shown by 12 in FIG. 2 is adopted.

When the freezing method is determined in this way, finally, the target term in the original text is frozen by the determined freezing method, and if necessary, the information, that is,
Information including terms, frozen terms, and defrosted terms as shown in FIG. 2 is stored (step S5). On the other hand, the above step S
In the process of determining whether or not the object is the freezing target in step 3, if the term is not the freezing target, steps S4 and S5 are omitted. Finally, it is checked whether there is an unprocessed word (step S6), and if there are still words left, steps S2 to S6 are repeated. And the original 1
After processing all the words in
Passes the frozen original text to the machine translation means 3 and ends the operation.

Next, the term decompression operation in this machine translation device will be described. FIG. 4 is a flowchart showing the operation of the term decompression means 4. First, the translated text is input from the machine translation means 3 (step S11). Next, the term of the input translated sentence is searched by the information obtained from the collation table 7 (step S1.
2) Determine whether the term is subject to decompression (step S13). According to this judgment, if the term is to be decompressed, the decompression method is determined according to the decompression term of the term obtained from the collation table 7 (step S14), the term is decompressed, and the result is reflected in the translated sentence 5. (Step S15). On the other hand, if it is determined in step S13 that decompression is not necessary, steps S14 and S15.
Is omitted. Finally, it is checked whether or not there is an unprocessed term (step S16), and if the term still remains, steps S12 to S16 are repeated. When all the terms of the translated sentence have been processed, the term decompression means 4 outputs the decompressed translated sentence and ends the operation.

As described above, according to the above-mentioned embodiment, the term is frozen in order to prevent the known machine translation means from translating unknown words in the original sentence or portions which are not desired to be translated, and Since the frozen terms are decompressed after machine translation, it is possible to generate a translated sentence from an original sentence containing unknown words and different languages even if the machine translation means does not actively translate unknown words. High-quality translation results can be guaranteed and translated without adverse effects.

Further, in this embodiment, the collation table 7 is composed of information having a description field consisting of a term to be frozen, a frozen term to replace the term, and a defrosting term corresponding to the frozen term. Therefore, it is possible to easily create and change any term. Furthermore, in the present embodiment, the term freezing unit 2 stores the term to be frozen, the frozen term, and the defrosted term in the collation table 7 when any term in the original sentence 1 is put into a frozen state. Therefore, it is not necessary to separately prepare the collation table 7, and it is possible to contribute to improvement of operability as a machine translation device.

Further, as described in the embodiment, the term freezing means 2 refers to the matching table 7 when the original sentence 1 is given, and the original sentence 1 contains the term to be frozen. If the term is included, it may be configured to perform the freezing process based on the collation table 7. In such a configuration, the freezing process is performed each time. There is no need, and the operability can be improved.

In the above embodiment, the freeze / thaw process is
In the case of translating an unknown word or a non-translated term, that is, an example applied to the translation of the term itself in the original sentence 1, the document format or document structure (for example, font information or document) attached to the original sentence 1 is shown. It is also possible to apply to things such as (tag information of). If you do this,
The collation table 7 is configured to show frozen terms and decompression methods for frozen document formats and document structures, and the term decompression means 4 converts the translated sentence based on the collation table 7 and outputs it.

[0028]

As described above, according to the machine translation apparatus of the present invention, a machine translation is performed by using a frozen term for a portion which is not desired to be translated, and the translated sentence is decompressed corresponding to the frozen term. Since this is performed, it is possible to guarantee a high-quality translation result even for an original sentence in which unknown words and the like are mixed.

[Brief description of drawings]

FIG. 1 is a block diagram showing a machine translation device of the present invention.

FIG. 2 is a diagram showing a configuration example of a matching table in the machine translation device of the present invention.

FIG. 3 is a flowchart showing an operation of a term freezing unit in the machine translation device of the present invention.

FIG. 4 is a flowchart showing an operation of a term decompressing unit in the machine translation device of the present invention.

[Explanation of symbols]

 1 original sentence 2 term freezing means 3 machine translation means 4 term decompressing means 5 translation 6 dictionary means 7 collation table

Claims (5)

[Claims] 1. A term freezing means for replacing an arbitrary term in an input original sentence with a frozen term that is not translated and outputting the term, and a collation indicating a frozen term and a defrosting term corresponding to the arbitrary term. A machine translation unit that translates a table and an original sentence described in natural language to generate a translated sentence of a target language, and when the original sentence includes the frozen term, outputs the frozen term as a non-translation target. And a term decompressing means for converting the translated text output from the machine translation means into the corresponding decompressed term and outputting the defrosted term when the frozen term is matched. Characteristic machine translation device. 2. The machine translation device according to claim 1, wherein the collation table has a description field including a term to be frozen, a frozen term to replace the term, and a decompression term corresponding to the frozen term. A machine translation device comprising: 3. The machine translation device according to claim 2, wherein the term freezing means includes a term to be frozen, a frozen term, and a thawed term when an arbitrary term in the original sentence is frozen. A machine translation device characterized in that is stored in a collation table. 4. The machine translation device according to any one of claims 1 to 3, wherein the term freezing unit refers to a matching table when an original sentence is given, and a term to be frozen is included in the original sentence. A machine translation device that determines whether or not a term is included, and if the term is included, performs a freezing process based on the matching table. 5. A term freezing unit which replaces and outputs arbitrary document information other than the terms in the original text, which are not included in the original text, into a frozen term that is not to be translated, and a frozen term corresponding to the arbitrary document information. And a decompression method, and a source text written in natural language is translated to generate a translated sentence in the target language. If the original text contains the frozen term, the frozen term is translated. For the machine translation means output as outside and the translated text output from the machine translation means, refer to the collation table, and if there is the frozen term, convert the translated text based on the corresponding decompression method. A machine translation device, comprising: