JPH07249044A - Machine translation device - Google Patents

Abstract

PURPOSE:To provide a machine translation device which can efficiently and accurately bring cooccurrence relation between words at the time of the translation of a sentence into consideration. CONSTITUTION:An input sentence 12 (a) is analyzed (b) into morphemes and the syntax structure is analyzed to extract words and phrases in cooccurrence relation; and the meanings of the constituent words of the words and phrases, English words corresponding to the semantics the parts of speech and the semantics of the English words, equivalents to the words and phrases are specified (d), equivalents to the words other than the constituent words are determined, and the syntax structure is converted (e) into syntax structure of English to generate a translation (f).

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 a source language sentence into a target language sentence.

[0002]

2. Description of the Related Art Conventionally, when a computer translates a natural language, it is widely known that the ambiguity due to the natural language is a major cause of degrading the accuracy of the translation result.

For example, in Japanese, the meaning of "buy" is basically to "pay and make it your own", but to "highly appreciate" or "receive a grudge or a loss". It is also used when expressing things. The ambiguity of translation due to such polysemy of words increases exponentially as the number of words in the sentence increases. For example, if the number of meanings of one word in the source language (translation source language) is 10,
If two such words are lined up, simply speaking, 10 squared gives 100 ambiguities.

However, in reality, a plurality of words in the source language may correspond to one word in the target language (translation target language) and vice versa, so that the ambiguity of translation is further increased. become. Further, for example, when there is a large difference in structure between languages as well as vocabulary such as Japanese and English, the ambiguity of the translation becomes even greater.

In order to counter the increase in ambiguity due to the combination of various factors as described above, a conventional machine translation apparatus uses a method of determining a bilingual translation in consideration of a co-occurrence relation between words. There is.

[0006]

However, in the conventional machine translation device described above, the translation process includes (1) analysis of a sentence in the source language, (2) conversion from the source language to the target language, and
(3) Generation of a sentence in the target language is modularized into three stages. Usually, the analysis of the source language sentence is performed using an analysis dictionary, and the conversion from the source language to the target language is translation. Since it is done using a special dictionary, we had to deal with co-occurrence between words in different dictionaries. For example, in the case of translating a Japanese sentence "achieve results" into English, after determining the meaning of each word "result" and "raise" using the analysis dictionary, use the translation dictionary (Japanese-English dictionary). It was necessary to select a matching English word from the English parallel translation meanings corresponding to the determined Japanese meanings.

Therefore, the above-mentioned ambiguity cannot be resolved at once, and since all the possibilities are taken into consideration, processing becomes complicated and inefficient, and co-occurrence is handled by different dictionaries. Therefore, the co-occurrence may not be consistent.

The present invention has been made to solve the above problems, and it is an object of the present invention to provide a machine translation device capable of efficiently and accurately considering the co-occurrence relationship between words when translating a sentence. To aim.

[0009]

In order to achieve the above object, the present invention has a parsing means for parsing a syntax of a source language sentence, and a machine for translating a source language sentence into a target language sentence. In the translation device, a co-occurrence condition dictionary that stores information about co-occurrence conditions, and a co-occurrence condition dictionary that is searched from the syntactic analysis result in the syntactic analysis unit using the co-occurrence condition dictionary There is provided a machine translation device characterized by being provided with a co-occurrence condition search means for determining parallel translation of a phrase.

Further, the co-occurrence condition dictionary may hold word senses of words having a co-occurrence relationship, target language words corresponding to the respective words, and parts of speech and word senses of the target language words. preferable.

Further, it is preferable that the co-occurrence condition searching means determines a word meaning of a word in a source language having a co-occurrence relationship and a part of speech and a word meaning of a target language corresponding to the word.

In order to achieve the same object, the present invention has an original sentence holding means for holding a sentence in a source language, a morpheme analyzing means for analyzing a morpheme of the sentence in the source language, and a sentence in the source language. A syntactic analysis unit that analyzes a syntactic structure, a co-occurrence condition dictionary that stores information about co-occurrence conditions, a co-occurrence condition that is searched from the syntactic analysis result in the syntactic analysis unit using the co-occurrence condition dictionary, and a co-occurrence relationship. A co-occurrence condition search means for determining a parallel translation of a phrase in the source language, a semantic analysis means for analyzing the meaning of the sentence in the source language, a translation dictionary for storing a parallel translation dictionary for the source language and the target language, A translation dictionary search means for extracting a word in the target language corresponding to a word in the source language sentence from the translation dictionary, and a structure conversion means for converting the syntactic structure of the source language sentence into the syntactic structure of the target language, In the structure conversion means A translation generating means assembling the sentence of the target language from the conversion results that, to provide a machine translation apparatus is characterized in that a translation holding means for holding the generated sentence by the translation generation means.

[0013]

According to the machine translation device of claim 1, the source language sentence is analyzed for syntactic structure, the co-occurrence condition is searched using the co-occurrence condition dictionary, and the source language phrase in the co-occurrence relation is searched. Will be translated.

According to the machine translation apparatus of claim 2, the co-occurrence condition dictionary stores the meaning of each word having a co-occurrence relationship, the target language word corresponding to each word, and the target language word. Part of speech and meaning are retained.

According to the machine translation apparatus of the third aspect, the word meaning of a word in the source language having a co-occurrence relationship and the part of speech and word meaning of the target translation corresponding to the word are determined.

According to the machine translation apparatus of claim 4, the source language sentence is analyzed for morphemes, for syntactic structure, for co-occurrence conditions are searched using the co-occurrence condition dictionary, and for co-occurrence relations. A parallel translation of a certain source language phrase is determined, the meaning is analyzed, the target language word corresponding to the sentence word is extracted from the translation dictionary, converted into the target language syntactic structure, and assembled into the target language sentence. .

[0017]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a schematic configuration of a machine translation device according to an embodiment of the present invention.

In the figure, 1 is an input device for inputting an original sentence to be translated, which is composed of, for example, a keyboard and a mouse, 2 is a ROM (read only memory), a RAM (random access memory), or A storage device constituted by a combination thereof, or a disk device, 3 is a CPU (central control unit) for controlling the operation of the apparatus of this embodiment, and 4 is a display of the original text input by the input device 1 or its translated text. It is an output device that displays the above or prints the translated text and outputs it. A disc reader or an optical reader may be used as the input device 1.

Further, as shown in FIG. 2, the storage device 2 stores a source sentence (a sentence in the source language which is the source language) input from the input device 1 in the source sentence storage portion 2a, and information on co-occurrence conditions. A co-occurrence condition dictionary 2b, a translation dictionary 2c that stores a bilingual dictionary that is pair information of a source language and a target language,
Also, a translation sentence holding unit 2d that holds the translation sentence generated by the translation sentence generating unit 3g is stored, and the CPU 3 analyzes the original sentence held in the original sentence holding unit 2a into morphemes and the morpheme analyzing unit 3a. From the analysis results in
A co-occurrence condition search unit 3c that searches the syntactic analysis unit 3b that analyzes the syntactic structure of the original sentence and the co-occurrence condition dictionary 2b, and extracts words that have a co-occurrence relationship from the analysis result of the syntactic analysis unit 3. A translation analysis unit 3d that clarifies the semantic relationship of the original sentence, and a translation dictionary search unit 3e that searches the translation dictionary 2c and parallel translates the analysis result in the semantic analysis unit 3d. The structure conversion unit 3f that converts the syntactical structure of the translation conversion unit 3f, and the translation result generation unit 3g that generates a translation sentence from the conversion result in the structure conversion unit 3f.

Although not shown in FIG. 2, the storage device 2 stores an analysis dictionary used in the morphological analysis unit 3b, the syntax analysis unit 3c, and the semantic analysis unit 3d, and the meaning of English words. It stores an English dictionary and a Japanese dictionary that stores the meaning of Japanese words.

Next, the operation of the apparatus according to this embodiment will be described with reference to FIGS.

FIG. 3 is a flow chart showing a translation processing procedure performed by the apparatus of this embodiment, and FIG. 4 is a diagram showing a display example displayed on the display (output device 4) during the translation processing performed by the apparatus. is there.

First, in step S1, the input device 1
When input from, the input sentence 12 "He made a result." Held in the original sentence holding unit 2a is displayed on the display (FIG. 4A). Next, the input sentence 12 is morphologically analyzed by the morphological analysis unit 3a (step S
2), the morphological analysis result 13 is displayed on the display as shown in FIG. Next, the syntax analysis unit 3b analyzes the syntax of the input sentence 12 using the analysis result 13 (step S3), and a syntax analysis result 14 as shown in FIG. 4C is displayed on the display. To be done. In this analysis result 14, "he is" and "results" are
Each is shown to be associated with "raising."

Next, the co-occurrence condition search unit 3c determines whether or not the syntactic analysis result 14 includes components having a co-occurrence relationship.
The co-occurrence condition dictionary 2b is searched to determine (step S
4). FIG. 5 is a diagram showing an example of the storage contents of the co-occurrence condition dictionary 2b. In this storage contents example 15, the storage contents 1, 2, and 3 are the heading, the Japanese word meaning part, and the English word meaning part, respectively. It consists of three items. The co-occurrence condition search unit 3c searches such a co-occurrence condition dictionary 2b, and in the input sentence 12, "he (ga)" and "raise".
Since there is no co-occurrence relationship between and, and “result (wa)” and “raise” match the memory content 2 in FIG. 3, it is determined that they have a co-occurrence relationship (YES). When there is no co-occurrence relationship, the result 15a shown in FIG. 4 (d) is displayed on the display.

On the other hand, when there is a co-occurrence relation, the co-occurrence condition search unit 3c determines that the "result" of the input sentence 12 is a noun from the data stored in advance in the Japanese word meaning part of the storage content 2. Among the headwords “results” stored in the Japanese dictionary shown in FIG. 6, it is specified that the meaning is “finished (good) result” with the meaning number “004”, and “increase”. ”Is a verb, and the headword“ raise ”
Among the above, it is specified that it has the meaning of “getting good results” with the meaning number “018” (step S5). Furthermore, the co-occurrence condition search unit 3c determines that the storage contents 2
"Increase" from the data stored in English
It is specified that the translation word “produce” of is a transitive verb (vt.), And that it has the meaning of the meaning number “005” in the entry word “produce” stored in the English dictionary. The word "result" is a noun (n.), And the meaning number "00" is included in the headword "result" stored in the English dictionary.
3), and the above "result"
Identifies that the adjective "adj.""Good" is required for adnominal modification, and that it is used in the plural form, no article, and that the parallel translation of "achieve results" is "produce good results" Yes (step S6). Then, steps S4 to S
The processing result 15b of No. 6 is displayed on the display as shown in FIG.

After the processing in the co-occurrence condition search unit 3c as described above is completed, the semantic analysis unit 3d analyzes the meaning of the original sentence (step S7), and the translation dictionary search unit 3e.
The words other than the words having the co-occurrence relation are translated (step S8), and the structural conversion unit 3f converts the Japanese syntactic structure into English meaning expressions 16 as shown in FIG.
Converted to and displayed on the display (step S
9), the translated sentence generator 3g generates a translated sentence 17 as shown in FIG. 4F from the semantic representation 16 and displays it on the display (step S10). The generated translated sentence 17 is translated. It is held in the sentence holding unit 2d.

In the conventional translation process, in the semantic analysis process of step S7, for example, the meaning of "raise" is
A plurality of word meaning sentences of the Japanese dictionary shown in FIG. 6 are specified, and further, in the undecided parallel translation determination process of step S8, the parallel translation of the specified word meaning is made into a plurality of English parallel translation meanings of the Japanese-English dictionary shown in FIG. Must be selected from among the candidates.

On the other hand, according to the present embodiment, in step S4, the co-occurrence of words in the source language is taken into consideration immediately after the completion of the syntax analysis, so that the analysis process of the source sentence can be simplified. Further, in steps S5 and S6, a co-occurrence condition dictionary that stores the meaning of each word having a co-occurrence relationship, the word of the target language corresponding to each word, and the part of speech and word meaning of the word of the target language is stored. By using it, it is possible to determine the source word meaning and the target language bilingual translation, so the translation extraction process and translation selection process can be simplified, and the co-occurrence consistency between the analysis dictionary and the translation dictionary cannot be maintained. The problem can be solved.

In the above-described embodiment, an example of translation from Japanese to English has been described. However, since co-occurrence of words is considered to be a phenomenon common to all languages, the present invention is applicable to all languages. It can be applied to translation into a language.

Further, in the above embodiment, when the words having the co-occurrence relation are extracted, the translation is automatically performed using the co-occurrence condition dictionary. It may be possible to interactively control, by using a menu or the like, whether to perform or to normally translate literally.

In the above embodiment, the co-occurrence in which the noun and the verb are connected via the case particle was dealt with, but the invention is not limited to this. The noun and the noun, the adjective and the noun, or the adverb and the verb. Etc. may be applied to co-occurrence of binary relations between all parts of speech, and further, co-occurrence relations of three or more words may be considered as well as between two words.

In the above embodiment, the Japanese 1
Although co-occurrence was treated as a word corresponding to one English word, for example, when replacing two Japanese words "drawing a heart" with one English word "imagine" Alternatively, it may be applied to co-occurrence in which multiple words are translated into a different number of words.

Further, in the above embodiment, the co-occurrence relation in an idiomatic expression was dealt with, but the present invention is not limited to this.
If a frequently used expression such as "buy (book)" is registered in advance, translation accuracy and translation speed can be improved.

[0035]

As described above in detail, according to the machine translation apparatus of the first or fourth aspect, the sentence structure of the source language is analyzed for syntactic structure, and the co-occurrence condition dictionary is searched for using the co-occurrence condition dictionary. In addition, since the bilingual translation of the original language phrase having the co-occurrence relation is determined, the co-occurrence relation can be considered in the shallow analysis stage, and the analysis process of the original sentence can be simplified. Further, the co-occurrence relation can be handled by a single dictionary, and the problem that the co-occurrence consistency cannot be obtained can be solved.

According to the machine translation device of the second aspect, the co-occurrence condition dictionary stores the meaning of each word having a co-occurrence relationship, the target language word corresponding to each word, and the target language word. Since the part of speech and the word meaning are held, the translation word extraction processing and the translation word selection processing can be simplified.

According to the machine translation device of the third aspect, the meaning of a word in the source language having a co-occurrence relationship and the part of speech and the meaning of the target language corresponding to the word are determined. With this, it is possible to select an accurate translated word.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a machine translation device according to an embodiment of the present invention.

FIG. 2 is a diagram showing details of a storage device 2 and a CPU 3 of the device shown in FIG.

FIG. 3 is a flowchart showing a translation processing procedure by the apparatus of FIG.

FIG. 4 is a diagram showing an example of translation processing by the apparatus of FIG.

5 is a diagram showing an example of stored contents of a co-occurrence condition dictionary 4 of the apparatus of FIG.

FIG. 6 is a diagram showing an example of stored contents of a Japanese dictionary of the apparatus of FIG.

FIG. 7 is a diagram showing an example of stored contents of a translation dictionary 7 of the apparatus shown in FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 input device 2 storage device 2a original sentence holding unit (original sentence holding unit) 2b co-occurrence condition dictionary 2c translation dictionary 2d translation sentence holding unit (translation sentence holding unit) 3 CPU 3a morphological analysis unit (morphological analysis unit) 3b syntax analysis unit ( Syntactic analysis means) 3c co-occurrence condition search section (co-occurrence condition search means) 3d semantic analysis section (semantic analysis means) 3e translation dictionary search section (translation dictionary search means) 3f structure conversion section (structure conversion means) 3g translated sentence generation Department (Translated sentence generation means)

 ─────────────────────────────────────────────────── ─── Continued front page (72) Inventor Makoto Hirota 3-30-2 Shimomaruko, Ota-ku, Tokyo Canon Inc. (72) Minor Fujita 3-30-2 Shimomaruko, Ota-ku, Tokyo Canon Within the corporation

Claims (4)

[Claims] 1. A co-occurrence condition dictionary that stores information about co-occurrence conditions in a machine translation device that has a syntax analysis unit that analyzes a syntax of a source language sentence and that translates a source language sentence into a target language sentence. And a co-occurrence condition retrieving unit for retrieving a co-occurrence condition from the syntactic analysis result in the syntactic analysis unit using the co-occurrence condition dictionary and determining a bilingual translation of a source language phrase having a co-occurrence relationship. Machine translation device characterized by. 2. The co-occurrence condition dictionary holds word senses of words having a co-occurrence relationship, target language words corresponding to the respective words, and parts of speech and word senses of the target language words. The machine translation device according to claim 1, which is characterized in that. 3. The co-occurrence condition search means determines a word meaning of a source language word having a co-occurrence relationship and a part of speech and a word meaning of a target language corresponding to the word. Machine translation device. 4. A source sentence holding means for holding a source language sentence,
Morpheme analysis means for analyzing morphemes of the sentence in the source language,
A syntactic analysis unit that analyzes the syntax of the sentence of the source language, a co-occurrence condition dictionary that stores information about co-occurrence conditions, and a co-occurrence condition dictionary that uses the co-occurrence condition dictionary from the syntactic analysis result in the syntactic analysis unit. A co-occurrence condition search means for searching and determining a parallel translation of a source language phrase having a co-occurrence relationship, a semantic analysis means for analyzing the meaning of the source language sentence, and a bilingual dictionary for the source language and the target language are stored. A translation dictionary, a translation dictionary search means for extracting a word in the target language corresponding to a word in the source language sentence from the translation dictionary, and converting the syntactic structure of the source language sentence into the syntactic structure of the target language. And a translated sentence holding unit for holding the sentence generated by the translated sentence generating unit. When That machine translation apparatus.