JPH06295311A - Machine translation system - Google Patents

Abstract

PURPOSE:To improve the translation accuracy of a sentence including an unregistered composite word by referring to a dictionary for the meaning of the unregistered composite word in each its constitutional element and applying proper translation information in accordance with the internal structure of the composite word obtained from the reference of the dictionary. CONSTITUTION:A composite word detecting part 15 detects a composite word combined through a specific character such as a hyphen and a composite word processing part 16 divides the detected composite word into individual constitutional elements by using the specific character as a delimiter and applies translation information such as a part of speech and a translated word to the whole composite word in accordance with the internal structure obtained from the reference of the dictionary. When the composite word detected by the detecting part 15 has no entry as a whole composite word and morphem processing such as affix processing and numeral processing for the composite word is not executed by a translation processing part 17, the processing part 16 gives proper translation information in accordance with the number of constitutional words of the composite word, speech part information obtained by the reference of the dictionary for the meaning of respective constitutional elements and the combination of the number and the information to utilize the information for the translation processing of the original sentence.

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, and more specifically, it extracts a dictionary of unregistered compound words for each constituent element and adds appropriate translation information according to the internal structure of the compound word obtained from the result. By doing so, the present invention relates to a machine translation device that improves the translation accuracy of a sentence including an unregistered compound word. For example, it is applied to a machine translation device having a function of processing unregistered compound words.

[0002]

2. Description of the Related Art In natural language processing such as machine translation, processing of unknown words not registered in a system dictionary is an important issue. In particular, since compound words are rich in productivity, it is practically impossible to add all of them to the system dictionary, because the capacity of the dictionary is enormous.
Even if the entire compound word is not registered, if there is something that has a particularly strong new creativity and is registered in the affix dictionary,
Morphological processing adds translations and parts of speech. Also, those with strong morphological characteristics such as numbers can be put together by morphological processing. As a processing method of a compound word that is not registered in the dictionary and does not correspond to a pattern that is subjected to morphological processing, there is a method of processing the entire compound word as an unknown word, giving some part-of-speech candidates, and translating the original text as it is. . However, compound words joined by special characters such as hyphens are
There is a certain degree of rule in the connection pattern and the semantic relationship between the constituents, and even if it is not registered in the dictionary, it is desirable to perform the process of estimating the translated word and the part of speech according to the rule.

There are the following three conventional techniques for solving such a problem. First, JP-A-3-48
Japanese Patent No. 366 discloses an attempt to parse the inside of a compound word in the word order as it is input, or replace hyphens with spaces to parse and process the entire original sentence as it is. However, the word order of compound words is not always the same as the word order of ordinary sentences. For example, "money-saving" is the reverse of the normal word order, the object comes before the verb, and even if this is analyzed as "money saving", the correct result cannot be obtained. Also, if you replace the hyphen with a space and perform syntactic analysis without distinguishing it from other parts, the constituent elements of the compound word will be confused with other parts, so there is a high possibility that incorrect parsing will occur, and The intelligibility of the translated text is very low.

In Japanese Patent Laid-Open No. 4-262459 and Japanese Patent Laid-Open No. 2-281374, a translation pattern is uniquely determined for a limited pattern such as "noun-participle", and individual configurations are provided. It is intended to generate the translated word of the entire compound word from the result of the dictionary lookup of the elements. However, since only a unique translation pattern is set for the input pattern, whichever method is used, the following problem occurs. Example) college-trained people (a) "People trained at college" (b) "People trained at college" (c) "People trained at college" (a) is a correct interpretation, (b) is JP-A An example of translation by the method disclosed in Japanese Patent Laid-Open No. 4-262459, (c) is Japanese Patent Laid-Open No. 2813/1990.
This is an example of translation by the method disclosed in Japanese Patent Publication No. 74. In the correct answer, the noun on the left side of the hyphen must be given the case label "de" of the place case, but in (b) the noun is treated as the subject of the verb on the right side, and in (c) it is treated as the uniform object. Therefore, the wrong case label is given to each. The case relation of each component in the "noun-past participle" is not uniform, and there may be various relations such as cause, content, actor, etc. in addition to the place, but the conventional technology can handle only limited cases. .

Further, the above-mentioned conventional technique does not support a small number of input patterns, and cannot handle cases where the right side is a participle and the left side is an adjective, and the case where the constituent elements are three words or more is not supported. . In particular, Japanese Unexamined Patent Publication No. 2-281374 only deals with "noun-participle", and corresponds to patterns with relatively high frequency of appearance, where the left side is an adverb or the right side is a noun. I haven't.

[0006]

[Purpose] The present invention has been made in view of the above-mentioned circumstances, and searches an unregistered compound word into a dictionary for each constituent element, and obtains appropriate translation information according to the internal structure of the compound word obtained from the result. By adding it, the translation accuracy of sentences including unregistered compound words can be improved, and in particular, the processing can be divided according to the number of constituent elements of the compound word, the part of speech of the key word in the constituent elements, and the presence or absence of the The purpose was to improve and to provide a machine translation device that treats compound words that do not have a key word as one unknown word in order to avoid parsing failures and erroneous analysis. It is a thing.

[0007]

In order to achieve the above object, the present invention provides (1)
At least an input unit for inputting an original sentence, an original sentence storage unit for storing the original sentence input by the input unit, a translation dictionary unit for storing knowledge information used for translation processing of the original sentence, and the knowledge information are used. A translation processing unit that translates the original sentence stored in the original sentence storage unit, a translation storage unit that stores the translated sentence obtained by the translation processing by the translation processing unit in association with the original sentence, and at least the original sentence or The display unit and the display control unit that display the translated text, the edit control unit that controls the editing process for the original sentence and the translated text displayed on the display unit, and the input source sentence are connected by a special character such as a hyphen. A compound word detection unit that detects a compound word, and a compound word detected by the compound word detection unit is divided into individual constituent elements using special characters as delimiters, and a dictionary is created. A compound word processing unit that gives translation information such as a part of speech or a translated word to the entire compound word according to the structure, and the compound word detected by the compound word detection unit is registered in the translation processing unit as a whole. If it does not have it and it has not undergone morpheme processing such as affix processing or number processing in the translation processing unit, the compound word processing unit obtains the number of constituent elements of the compound word and a dictionary lookup for each constituent element. Appropriate translation information is provided according to the part-of-speech information and the combination thereof, which is used for the translation process of the original sentence. Furthermore, (2) the compound word processing unit has two compound elements, and When the part of speech is an adverb in the left component of the hyphen, the inside of the compound word is translated using the right component as the adjunct of the adverb, a translated word is generated from the result, and it is given to the entire compound word together with the part of speech candidate. , Original text translation process And (3) the compound word processing unit has two compound elements, does not correspond to the first translation process, and is on the right side of the hyphen. When the part-of-speech of a constituent element is the present participle of a verb, the word order of the left and right elements is exchanged, and the inside of the compound word is translated, and a translated word is generated from the result, which is given to the entire compound word together with the part-of-speech candidate. Second translation processing means used for the translation processing of, and (4) in the compound word processing unit, the number of constituent elements of the compound word is two, which does not correspond to the first translation processing,
And when the part of speech of the right component of the hyphen is a past participle of the verb, based on the information of the translation dictionary part, the case frame of the verb and the feature of the left component are matched,
Having a third translation processing means for translating the inside of the compound word by supplementing an appropriate preposition, generating a translated word from the result, adding it to the entire compound word together with a part-of-speech candidate, and utilizing it for the translation processing of the original sentence. , And (5) In the compound word processing section,
When the number of constituent elements of the compound word is two, does not correspond to the first translation process, and the right-side constituent element of the hyphen is a noun,
Translate the inside of the compound word, generate a translated word from the result,
It has a fourth translation processing means that is given to the entire compound word together with the part-of-speech candidate and is used for the translation processing of the original sentence. Furthermore, (6) the compound word processing section has two constituent elements, and If the first translation process to the fourth translation process do not apply, a first unknown word processing unit that processes the entire compound word as an unknown word is further provided, and further (7) the above In the compound word processing unit, when the number of constituent elements is three or more and one of the constituent elements is a coordinate conjunction, translation processing is performed inside the compound word and a translated word is generated from the result. And a fifth translation processing unit which is added to the entire compound word together with the part-of-speech candidate and is used for the translation processing of the original sentence. Furthermore, (8) the compound word processing unit has three or more constituent elements. , And does not match the fifth translation process In the case, two words from the left of the constituent element are taken out, the processing from the first translation processing to the fourth translation processing is performed, and if any of them is matched, the portion is combined with one word on the further right side. Then, the processes from the first translation process to the fourth translation process are performed again, and this process is repeated up to the last component, and a translated word is generated from the result, and is added to the entire compound word together with the part-of-speech candidate. Having a sixth translation processing means used for the translation processing, and further (9) the compound word processing section having three or more constituent elements, whichever of the fifth translation processing or the sixth translation processing. If there is no match, the second unknown word processing means for processing the unknown word as a unit of the entire compound word is featured. Hereinafter, description will be given based on examples of the present invention.

FIG. 1 is a block diagram for explaining an embodiment of a machine translation apparatus according to the present invention, in which 11 is an input unit,
Reference numeral 12 is an original sentence storage unit, 13 is a translated sentence storage unit, 14 is an edit control unit, 15 is a compound word detection unit, 16 is a compound word processing unit, 17 is a translation unit (translation processing unit), 18 is a display control unit, and 19 is It is a display unit. The original text input from the input unit 11 is sent to the original text storage unit 12. The translation unit 17 translates the original sentence in the storage unit 12 under the control of the editing control unit 14.

That is, the original sentence input by the input unit 11 is stored in the original sentence storage unit 12. Knowledge information used for the translation process of the original sentence is stored in the translation dictionary unit. The translation processing unit 17 uses the knowledge information of the translation dictionary unit to translate the original sentence stored in the original sentence storage unit 12. The translated text storage unit 13 stores the translated text obtained by the translation processing in association with the original text. The display control unit 18 and the display unit 9 display at least the original sentence or the translated sentence.
The editing control unit 14 controls the editing process for the displayed original sentence and translated sentence. The compound word detection unit 15 detects a compound word combined with a special character such as a hyphen from the input original sentence. The compound word processing unit 16 divides the compound word detected by the compound word detecting unit 15 into individual constituent elements by using special characters as delimiters and looks up the dictionary, and according to the internal structure obtained from the result, the compound word is extracted. Give translation information such as parts of speech and translations to the whole. With such a configuration, the compound word detected by the compound word detecting unit 15 does not have an entry in the translation dictionary unit as a whole, and the translation processing unit 17 is not subjected to morpheme processing such as affix processing or number processing. In such a case, the compound word processing unit 16 provides appropriate translation information according to the number of constituent elements of the compound word, the part-of-speech information obtained by looking up each constituent element in the dictionary, and the combination thereof, and Used for translation processing.

FIG. 2 is a block diagram showing the processing of the translation unit in FIG. 1 by taking English-Japanese machine translation as an example. In the figure, 21 is a morphological analysis unit, 22 is a syntax analysis unit, 23 is a conversion unit, and 24 is a conversion unit. Is a generation unit, 25 is a dictionary, and 26 is a grammar rule. First, the English sentence which is the input sentence is recognized as an individual word by the morphological analysis unit 21 based on the information in the dictionary 25. Further, by referring to an affix dictionary or the like, information such as a part of speech or a translated word is given to some of the words that are not registered in the system dictionary. Next, the parser 2
In step 2, the grammar rule 26 is applied to determine the sentence structure, and a tree structure that defines the dependency relations is created. The tree structure is converted by the conversion unit 23 into a tree conforming to the structure of the output language Japanese, and the generation unit 24 translates the tree for each node.

The compound word detecting unit 15 cuts out compound words combined by special characters such as hyphens from the original sentence. The compound word processing unit 16 receives the processing result of the morpheme analysis unit 21, and when the cut out compound word does not have the dictionary lookup information or the affix processing information, the special character is used as a delimiter in the dictionary to construct the dictionary. According to the result, translation information such as a part-of-speech and a translated word is given and passed to the processing of the syntax analysis unit 22 and thereafter.
The obtained translated texts are sequentially stored in the translated text storage unit 13. The edit control unit 14 drives the display control unit 18, and the original sentence storage unit 12
The English sentence stored in the and the Japanese sentence stored in the translated sentence storage unit 13 are displayed on the display unit 19 in association with each other. The user performs post-editing processing while looking at this display. The post-editing process is executed by the edit control unit 14 according to the control information input from the input unit 11.

FIG. 3 is a flowchart of the compound word processing in the compound word processing section in FIG. 1 up to the distribution processing according to the number of constituent elements. Hereinafter, each step will be described in order. A compound word delimited by hyphens or the like is detected (S31), and morphological analysis is performed (S32). If it does not have a dictionary entry and is not subjected to affix processing or the like (S33 to S34), the individual words are looked up in the dictionary (S3).
5) The number of components is counted (S36). If the number of constituent elements is 2 words, it is sent to the processing of (a), and if it is 3 words or more, it is sent to the processing of (b) (S37).

FIG. 4 shows an example of the result obtained by looking up the individual constituent elements of the unregistered compound word in the dictionary. In FIG. 5, the number of components is two.
9 is a flowchart of a process of allocating a part of speech candidate of a component in the case of a word. Hereinafter, each step will be described in order. Search for a part-of-speech candidate for the left component (S5
1) If there is an adverb in it, it is sent to the process of (c) in the figure (S52). If there is no adverb, the part-of-speech candidate of the right component is searched (S53), and if there is a current participle, it is sent to the process of (d) in the figure (S54). If there is no present participle, it is checked if there is a past participle, and if there is a past participle, it is sent to the process of (e) in the figure (S55). If not, next, it is checked whether or not there is a noun, and if there is a noun, it is sent to the process of (f) in the figure (S56). If there is no noun, unknown word processing is performed and sent to the syntactic analysis (S57). It should be noted that in this flowchart, if the process is performed in accordance with the flow, it is possible to deal with the case where the component is a multi-part-of-speech word, but it is also possible to separately provide priority information of the multi-part-of-speech and perform the process accordingly. is there. (Below margin)

FIG. 6 is a processing flow chart when the left component is an adverb. Hereinafter, each step will be described in order. Translation processing is performed using the adverb's destination as the right component (S61), and if the analysis is successful (S62),
The entire compound word is used as an adjective, the obtained translated word is given and sent to the syntactic analysis (S63). If the analysis is unsuccessful (S62), the process returns to FIG. In the example of FIG. 4, W1 corresponds to this. Left side component "fastes"
Since "t" includes an adverb in the part-of-speech candidate, the translation process is performed with "growing" on the right as the target, and the translated word "growing fastest" is obtained. An example of translation using a part of speech as an adjective is S1 in FIG.

FIG. 7 is a processing flowchart in the case where the right side component is the present participle. Hereinafter, each step will be described in order. The left and right of the delimiter are replaced and the translation processing is performed (S71). If the analysis is successful (S72),
The entire compound word is used as an adjective, the obtained translated word is given and sent to the syntactic analysis (S73). If the analysis is unsuccessful (S72), the process returns to FIG. In the example of FIG. 4, this corresponds to W2 and W3. The right component "
The threatening “” looking ”both include the present participle in the part-of-speech candidate, so swap the left and right to“ threatening li
The translation process is performed as fe “” looking wistful. ”The former is analyzed as a transitive verb, and the latter is analyzed as an intransitive verb to obtain the translated words“ life threatening ”and“ looks greedy ”.
An example of translation using a part of speech as an adjective is S2 and S in FIG.
It is 3.

FIG. 8 is a processing flowchart when the right component is a past participle. Hereinafter, each step will be described in order. Search the case frame of the right element (S
81), if there is a case frame (S82), the feature of the left element is matched (S83). If there is no case frame (S82), the process returns to FIG. There is an element of the case frame that matches the feature of the left component (S84), and if that element is other than OBJ (S85), in the case of SUBJ, "by" is supplemented (S8).
6, 87), in other cases, a preposition written in the case frame is supplemented to generate a translated word (S88), and the part of speech of the entire compound word is sent as an adjective to the syntactic analysis (S89). There is no match (S84), or the matched element is O
If it is BJ (S85), the process returns to FIG.
The processing after 56 is performed.

In the example of FIG. 4, W4 and W5 correspond to this. "Owned" and "loaded", which are the right components, both include the past participle as a part-of-speech candidate. Examples of "own" and "load" case frames are shown in FIGS. 9 (a) and 9 (b). "Government"
Has the feature ORG that matches the SUBJ of "own", so "b"
The translation process of "owned by government" is supplemented with "y" to obtain the translated word "owned by the government". In addition, since "propane" has the same feature MAT as the "load" WITH, "with" is supplemented by "loaded with propan".
The translation process is performed by "e", and the translated word "loaded with propane" is obtained. Figure 1 shows an example of translation using part-of-speech as an adjective.
2 are S4 and S5.

FIG. 10 shows a processing flowchart when the right component is a noun. Hereinafter, each step will be described in order. Translation processing is performed using the left-side component as a prefix modifier (S101), and if the analysis is successful (S10).
2), add adjectives and nouns as part-of-speech of all compound words,
In the case of an adjective, "no" is added to the obtained translated word, and in the case of a noun, the obtained translated word is given as it is and sent to the syntactic analysis (S
83). If the analysis is unsuccessful (S102), the process returns to FIG. In the example of FIG. 4, W6 corresponds to this. The left component "product" contains a noun in the part-of-speech candidate, so perform translation processing with "new product".
Get the translation "new product". An example of translation using a part of speech as an adjective or noun is S6 in FIG. In this example, the adjective candidates are taken and translated as "new product".

FIG. 11 is a processing flowchart when the number of constituent elements is three or more. Hereinafter, each step will be described in order. The part-of-speech candidate of the component is searched (S11
1), if any of them contains coordinate conjunctions (S11)
2) The inside of the compound word is translated (S113), and if the analysis is successful (S114), the translated word and the part of speech (adjective, noun) are given and passed to the syntactic analysis (S115). If the analysis fails (S
114), unknown word processing is performed, and the result is passed to the syntax analysis (S11).
6). In the example of FIG. 4, W7 corresponds to this. Since "and" in the constituent elements includes coordinate conjunctions in the part-of-speech candidates, "husba"
"nd and wife" is translated and the translated word "husband and wife (no)" is obtained. An example of translation using a part of speech as an adjective or noun is S7 in FIG. In this example, adjective candidates are taken and translated as "husband and wife's".

If there is no coordinate conjunction in the part-of-speech candidate (S
112), two words are taken from the left end (S117), and the processing (S51 to 56) when the constituent element is two words is performed (S11).
8). If it is determined that the processed word is an unknown word (S119), the entire compound word is sent to the syntactic analysis as an unknown word (S116).
If it is not an unknown word but some part-of-speech and translation are added (S119), the word on the right side is included (S121), the process of S118 is performed as two more words, and this is repeated up to the rightmost component ( S120), the final result is sent to the syntactic analysis as the part of speech and translation of the entire compound word (S12).
2).

In the example of FIG. 4, W8 corresponds to this. First, the processing of S51 to S56 is performed with "flexible metal".
Since this example satisfies the condition of S56, the process of FIG. 10 is performed, and the translation of “flexible metal ()” is given.
Part of speech can be an adjective or a noun. Furthermore, including the "pipe" on the right side, S5 as two words "flexible-metal" and "pipe"
The processes 1 to 56 are performed. In this case as well, the condition of S56 is satisfied, so the process of FIG. 10 is performed and "flexible-metal" is performed.
Is parsed as an adjective and gets the translation "flexible metal pipe". An example of translation using a part of speech as an adjective or noun is S8 in FIG. In this example, the adjective candidate is taken and translated as "flexible metal pipe plant". In addition, in the above process, when performing the translation process inside the compound word, even if a normal parsing grammar is used, a grammar that matches the syntactic features inside the compound word (no prefix is attached to the noun phrase, the prefix modifier is Usually only one, etc.) may be used. FIG. 12 is an example of translation of a noun phrase including the unregistered compound words shown in FIG. 4 by the unregistered compound word processing according to the present invention.

[0022]

As is apparent from the above description, the present invention has the following effects. (1) Effect corresponding to claim 1: An unregistered compound word is dictionary-divided for each constituent element, and appropriate translation information is added according to the internal structure of the compound word obtained from the result. It is possible to improve the translation accuracy of sentences including words. (2) Effects corresponding to claims 2 to 5, 7, and 8: Particularly, the processing is divided according to the number of constituent elements of the compound word, the part of speech of the key word in the constituent elements, and the presence or absence thereof, thereby improving accuracy. It will be possible. (3) Effects corresponding to claims 6 and 9: A compound word having no key word can be treated as one unknown word to avoid erroneous analysis and analysis failure.

[Brief description of drawings]

FIG. 1 is a configuration diagram for explaining an embodiment of a machine translation device according to the present invention.

FIG. 2 is a configuration diagram showing processing of a translation unit in FIG. 1 by taking English-Japanese machine translation as an example.

FIG. 3 is a flowchart of the compound word processing in the compound word processing unit in FIG. 1 up to the distribution processing according to the number of constituent elements.

FIG. 4 is a diagram showing an example of a result obtained by looking up individual constituent elements of an unregistered compound word according to the present invention in a dictionary.

FIG. 5 is a flow chart of a process of allocating components by part-of-speech candidates when the number of components is 2 in the present invention.

FIG. 6 is a flowchart of processing when the left-side component is the present participle in the present invention.

FIG. 7 is a flowchart of a process when the right-side component is the present participle in the present invention.

FIG. 8 is a flowchart of a process when the right component is a past participle in the present invention.

FIG. 9 is a diagram showing an example of a case frame according to the present invention.

FIG. 10 is a flowchart of a process when the right component is a noun according to the present invention.

FIG. 11 is a flowchart of a process in the case where the constituent elements of the present invention are three or more words.

FIG. 12 is a diagram showing an example of translation by unregistered compound word processing according to the present invention.

[Explanation of symbols]

11 ... Input unit, 12 ... Original sentence storage unit, 13 ... Translated sentence storage unit,
Reference numeral 14 ... Edit control unit, 15 ... Compound word detection unit, 16 ... Compound word processing unit, 17 ... Translation unit (translation processing unit), 18 ... Display control unit, 19 ... Display unit.

Claims (9)

[Claims] 1. At least an input unit for inputting an original sentence, an original sentence storage unit for storing the original sentence input by the input unit, and a translation dictionary unit for storing knowledge information used for translation processing of the original sentence. A translation processing unit that translates the original sentence stored in the original sentence storage unit using the knowledge information; and a translation sentence storage unit that stores the translated sentence obtained by the translation process by the translation processing unit in association with the original sentence. , At least the display unit and the display control unit for displaying the original sentence or the translated sentence, the edit control unit for controlling the editing process for the original sentence and the translated sentence displayed on the display unit, and the special characters such as hyphens from the input original sentence. The compound word detection unit for detecting the compound word combined by the above, and the compound word detected by the compound word detection unit,
It consists of a compound word processing unit that divides the special characters into individual components as delimiters and looks up the dictionary, and gives translation information such as parts of speech and translation words to the entire compound word according to the internal structure obtained from the result. The compound word detected by the compound word detection unit does not have an entry in the translation processing unit as a whole,
When the translation processing unit does not undergo morpheme processing such as affix processing and number processing, the compound word processing unit obtains the number of constituent elements of the compound word and part-of-speech information obtained by looking up each constituent element in a dictionary. A machine translation device characterized in that appropriate translation information is given according to the combination and the like, and is used for translation processing of the original sentence. 2. In the compound word processing unit, when the number of constituent elements of the compound word is two and the left-side component of the hyphen is a part-of-speech as an adverb, the right-side constituent is used as the adverb's target 2. The first translation processing means for translating a sentence, generating a translated word from the result, assigning it to the entire compound word together with a part-of-speech candidate, and utilizing it for the translation processing of the original sentence. Machine translation device. 3. In the compound word processing unit, when the number of constituent elements of the compound word is two, which does not correspond to the first translation processing, and the part of speech of the component on the right side of the hyphen is the present participle of the verb, Second translation processing means for translating the inside of the compound word by changing the word order of the left and right elements, generating a translated word from the result, assigning it to the entire compound word together with the part-of-speech candidate, and utilizing it for the translation processing of the original sentence. The machine translation device according to claim 1, further comprising: 4. In the compound word processing unit, when the number of constituent elements of the compound word is two, does not correspond to the first translation processing, and the part-of-speech of the component on the right side of the hyphen is the past participle of the verb. , Based on the information of the translation dictionary unit, the case frame of the verb and the feature of the left component are matched, the inside of the compound word is translated with an appropriate preposition, and a translated word is generated from the result, 2. The machine translation device according to claim 1, further comprising a third translation processing unit that is provided to the entire compound word together with the part-of-speech candidate and is used for translation processing of the original sentence. 5. In the compound word processing unit, when the number of constituent elements of the compound word is two, does not correspond to the first translation process, and the right component of the hyphen is a noun, the inside of the compound word is 4. A fourth translation processing unit that performs translation processing, generates a translated word from the result, assigns it to the entire compound word together with a part-of-speech candidate, and uses it for translation processing of the original sentence.
The machine translation device described. 6. In the compound word processing unit, when the number of constituent elements is two and does not correspond to any of the first translation processing to the fourth translation processing, the entire compound word is regarded as one group. 2. The machine translation device according to claim 1, further comprising first unknown word processing means for processing an unknown word. 7. In the compound word processing unit, when the number of constituent elements is three or more and one of the constituent elements is a coordinate conjunction, translation processing is performed inside the compound word as a parallel phrase. The machine translation apparatus according to claim 1, further comprising a fifth translation processing unit that generates a translated word from the result, assigns it to the entire compound word together with the part-of-speech candidate, and uses it for translation processing of the original sentence. 8. In the compound word processing unit, when the number of constituent elements is 3 words or more and does not match the fifth translation processing, two words from the left of the constituent element are extracted and the first word is extracted.
When the processing of the first translation processing to the fourth translation processing is performed and any one of them is matched, the processing of the first translation processing to the fourth translation processing is performed again by a combination of the portion and one word on the right side. It has a sixth translation processing means for performing the processing, repeating the processing up to the last component, generating a translated word from the result, assigning it to the entire compound word together with the part-of-speech candidate, and utilizing it for the translation processing of the original sentence. The machine translation device according to claim 1. 9. In the compound word processing section, when the number of constituent elements is three or more and does not match any of the fifth translation processing and the sixth translation processing, the entire compound word is grouped into one group. And a second unknown word processing means for processing an unknown word as.