JPH0683861A - Machine translation device - Google Patents

Abstract

PURPOSE:To provide the high-quality machine translation device provided with an operation processing method and a proverb/idiom dictionary to be referred to so as to simultaneously resolve a proverb, idiom, which are mixed among words, and polysemy related to them. CONSTITUTION:A source language analyzing intermediate structure generation part 200 provides intermediate structure depending on a source language by performing the analysis of syntax and meaning to an inputted sentence while referring to an analysis dictionary 250. A source language proverb/idiom processing part 300 reads the intermediate structure of the source language from a buffer 900, refers to a proverb/idiom dictionary 350 and prepares one node by defining the node of the proverb or idiom contained in the intermediate structure as one morpheme. Then, a new meaning code is applied again and after the node related to the relevant proverb or idiom is excluded, the processing result is stored in the buffer 900.

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 capable of automatically processing proverbs and idioms mixed between words and ambiguity associated with each other, which are not fixed phrases.

[0002]

[Prior Art] The 20th century can be said to be an era of knowledge explosion with the emergence of new information. Everyone must constantly absorb their knowledge so that they will not be slaughtered in the times.
However, the knowledge acquired is not only domestic but also foreign. It is needless to say that translation is important because ordinary people read their native language faster than foreign languages.
In order to improve the quality and efficiency of translation, we have come to the time when we have to consider some method by machine instead of manpower, that is, a machine translation system. In a machine translation system, a language that is input and translated is a source language, and a language that is translated and output is a target language. A machine translation device that adopts an intermediate conversion method is generally referred to as (1) a source language analysis unit (2) an intermediate structure conversion unit (3) a target language generation unit (4) a reference dictionary and a dictionary, as shown in FIG. It consists of four parts. The quality of machine translation is
Based on the grammatical rules of the target language generated by the target language generation unit, whether the input phrase was correctly analyzed by the source language analysis unit, and whether the difference between the source language and the target language was resolved by the intermediate structure conversion unit. It often depends on whether the target language is generated correctly. In particular, when the target of translation is a fixed sentence such as a proverb or an idiomatic phrase, each word often does not satisfy the grammatically correct dependency relation, and as a result, it becomes impossible to translate or the formally grammatical dependency Even if the above condition is satisfied, the translated text generated based on the translated text for each word often becomes a meaningless text, and the quality of the translation deteriorates. For example, How do you do? When (ordinary phrase) is subjected to ordinary English sentence analysis while ignoring the parsing process of the phrase, the analysis result is as shown in FIG.

The Chinese translation is "

[0004]

[Outer 1]

How is it? (What do you do?)

[0006]

[Outside 2]

Good! (Hello) "and it can not be properly translated. In order to solve this problem, a conventional technique, for example, a machine translation system of Japanese Patent Laid-Open No. 62-82464, is proposed, which can more appropriately translate a standard sentence such as a proverb and an idiom. Has been done.
This machine translation system is as shown in FIG.
In FIG. 9, the input unit 1 inputs a source language. The editing control unit 4 processes editing in two languages before and after translation processing. The original sentence storage unit 2 and the translated sentence storage unit 3 respectively store the original sentence and the translated sentence that are being processed. The translation unit 5 processes translation. The translation dictionary 6 is, as shown in FIG.
It is a dictionary that is divided into seven parts of g and needs to be searched for translation processing. The display control unit 7 processes the screen display of the display unit 8. The printing unit 9 prints the contents of files and screens.
As shown in FIG. 10, the processing flow of the translation unit 5 refers to the fixed text dictionary 6g of the translation dictionary 6 in s71 to determine whether the input source language is a fixed text. If it is a fixed phrase, s72
That is, the Japanese translation is taken out from the fixed text dictionary 6g, and then the translation is output in s73.
If it is determined by s71 that it is not a fixed phrase, s7
In step 4, general translation processing is performed. The detailed flow is shown in FIG. First, the morpheme is analyzed by referring to the rule / irregular change dictionary 6a by the morpheme analysis s51. Next, by referring to the word / jukugo dictionary 6b by the dictionary search s52, a morpheme that needs to be processed is searched, and then s53, s54, s
55, the parsing grammar 6c and the conversion grammar 6d are referred to, and the syntactic analysis and the structure transformation are repeatedly performed to find an intermediate structure of an appropriate target language. Then, the syntax generation s56 generates the surface structure of the target language while referring to the generation grammar 6e. Finally, the morpheme generation s57 refers to the morpheme generation grammar 6f to select an appropriate translated word. The above processing steps are called general translation processing. While referring to the standard sentence dictionary shown in FIG. 13 according to this conventional example using the following English sentence as an example, the following English sentence is determined to be a standard sentence in s71 of FIG. After performing the translation and extracting the translated word, the translation result is output via s73. The following correct processing result is obtained.

How do you do? (English original) Nice to meet you (Japanese translation)

[0009]

[Problems to be Solved by the Invention] Since the input original sentence has a degree of freedom in word order, when a proverb or an idiom is not a fixed sentence but is mixed between words, for example, it is referred to as "get it". If you register a fixed phrase in the dictionary as a Chinese translation of "Hikage", the following error will occur. First example He puts a coat on himself. Originally dressed as a garment. (Chinese translation of the prior art) Other * Deployment * Large garment. (Appropriate Chinese translation) The second example He acquires skill. Original technology. (Chinese translation of conventional technology) Other * learning * technology. (Appropriate Chinese translation) In the first example, the degree of freedom in word order does not allow the matching of fixed phrases, so an appropriate translation cannot be obtained. The second example is an error caused by the polysemy of idioms and proverbs.

In view of the above-mentioned drawbacks, the present invention provides an arithmetic processing method capable of simultaneously solving proverbs and idioms mixed between words and ambiguities related to them, and a proverb idiom dictionary that can be referred to to improve quality. An object is to provide a high machine translation device.

[0011]

According to the present invention, for each proverb or idiom of a source language, the main morpheme of the proverb or idiom is used as a search key, and related information of the main morpheme or the submorpheme,
A proverb dictionary that stores all proverbs and idioms of the main morpheme and their corresponding meaning codes, and an intermediate structure obtained by analyzing the input source language are stored in the proverb idiom dictionary. The proverb phrase processing part that makes a proverb or phrase phrase that matches a predetermined condition into one node compared to the information that is input, and the sentence that is input by the information such as the number of nodes of the processed intermediate structure is a simple proverb or Equipped with an intermediate structure discriminator that determines whether the phrase is an idiomatic phrase, and a translation selection dictionary that stores the part of speech code, the meaning code, the meaning control code, and the translation of the corresponding target language corresponding to each proverb and idiomatic phrase of the corresponding source language. It is a machine translation device characterized by that.

[0012]

According to the present invention, a proverb idiom processing unit is provided in the source language analysis unit and the intermediate structure conversion unit of the machine translation device to process proverbs and idioms in the intermediate structure (analysis result) of the source language. By doing so, the shortcomings of the matching process of fixed phrases are solved.
Then, wasteful processing is reduced as an intermediate structure determination unit for determining whether other processing is necessary is provided. This machine translation device does not only process idioms and proverbs mixed between words, but also corrects semantic codes during processing,
By removing unnecessary, too long nodes, etc.
Can solve the ambiguity of idioms and proverbs. Furthermore, since the intermediate structure of the source language can be modified more cleanly and appropriately, the quality and efficiency of translation can be improved.

[0013]

1 is a block diagram showing the configuration of a machine translation apparatus according to an embodiment of the present invention. In FIG. 1, 10
Reference numeral 0 is an input unit for inputting the source language to be processed into the system. Reference numeral 200 denotes a source language analysis intermediate structure generation unit that analyzes the syntax and meaning of an input sentence with reference to the analysis dictionary 250 and acquires an intermediate structure depending on the source language. 300 reads the intermediate structure of the source language from the buffer 900, refers to the proverb phrase dictionary 350, and makes the proverb and phrase phrases included in the intermediate structure into one node as one morpheme. Then, it is a proverbial phrase processing section that stores a processing result in the buffer 900 after giving a new meaning code again and deleting the node related to the relevant proverb or phrase. The processing flow of the proverbial phrase processing unit 300 is shown in FIGS. FIG. 5 shows a proverb phrase dictionary 350.
It is explanatory drawing which shows a part of structure. Proverb idiom dictionary 350
Contains information such as main morphemes, main morpheme restriction elements, auxiliary search elements, idioms, and related meaning codes. The intermediate structure discriminating unit 400 takes out the intermediate structure of the source language from the buffer 900, judges whether it is a simple proverb or idiom according to the processing flow shown in FIG. 4, and judges the judgment result (IDOM
The flag value) is stored in the buffer 900. The conceptual structure conversion unit 500 performs the following operation according to the determination result of the intermediate structure determination unit 400. (1) When the judgment result is a simple saying or idiom, it is not necessary to generate the target language, and the translation word selection unit 6
00, the proverb or idiom can be directly converted into the target language. (2) If the judgment result is not a simple saying or idiom , take out the intermediate structure of the source language from the buffer 900,
Referring to the difference adjustment conversion dictionary 550, the meaning of the source language,
After converting the syntax-dependent intermediate structure into the target language-dependent intermediate structure, the conversion result is stored in the buffer 900. The translated word selection unit 600 extracts the intermediate structure of the target language from the buffer 900, refers to the translated word selection dictionary 650 to determine the translated words of each node of the intermediate structure, and then stores the processing result in the buffer 900. Target language generator 7
00 does not need to be processed if the determination result of the intermediate structure determination unit 400 is a simple saying or phrase. However,
If it is not a simple saying or idiom, the intermediate structure is taken out from the buffer 900, the target language is generated by the generation grammar of the target language, and the translation result is stored in the buffer 900. Finally, the translation result of the buffer 900 is output by the output unit 800 that is associated with the print or the like.

2 and 3 show a source language proverb idiom processing unit 30.
It is a processing flow chart of 0. 2 and 3, step 201 performs a matching process on the intermediate structure of the source language, and finds a node whose PROC flag (the PROC flag determines whether a node with an intermediate structure has been processed) is not 1. . If there is no such node, the operation of the source language proverb idiom processing unit 300 is ended, and the intermediate structure determining unit 400 of FIG. 1 is entered. If there is a node whose PROC flag is not 1, then at step 202, it is judged whether the attribute of this node is a modifier. If it is a modifier, the process of step 230 is entered and the PROC flag of this node is set to 1. Then, the process returns to step 201. If it is not a qualifier, it is determined in step 203 that it is a leaf node (node without child node), and the processing in step 204 is performed.
With reference to the proverb phrasebook 350 of FIG. 1, it is determined whether there is a proverb or phrase. If you determine that there are no proverbs or idioms,
The process of step 230 is performed. If there are proverbs and idioms, the process of step 205 is performed, and after all the proverbs and idiom sets S (i) are searched from the proverb idiom dictionary 350, the process of step 206 is started. A node whose child node is a leaf node is taken out from the processed node to form a set Y. Then, the logical product operation of the set Y and the set S (i) is performed, and the set of proverbs and idioms that may be obtained is set to M (i).
And Then, perform the determination process of step 208,
If all M (i) are an empty set, it means that there are no proverbs or idioms. Therefore, the processing of the proverbial idiom processing unit 300 ends, and the process of the intermediate structure determination unit 400 of FIG. 1 starts. If it is not an empty set, the process of step 209 is performed, the main morpheme restriction element L (i) is extracted by referring to the proverb phrasebook 350, and the auxiliary verb information is extracted from the node attribute value of the main morpheme of step 210. After taking out A, L (i) and set A
And a logical product operation is performed to obtain the set J (i) for judgment. In step 201 of FIG. 2, it is determined whether the set J (i) obtained above is an empty set. If it is an empty set, the process of step 230 is performed. If it is not an empty set, the proverb phrase dictionary 350 is referred to, and the phrase phrase X and the new meaning code Y are set using M (i) as a search key. Find out. By the processing of steps 216 and 217, the intermediate structure is corrected or the node that is too long is deleted, and then the processing returns to the above-mentioned step 230. FIG. 4 is a flowchart showing the processing of the intermediate structure discriminating unit 400. In FIG. 4, step 301 is an intermediate structure. In order to determine whether the corresponding node is a simple saying or idiom, it is checked whether it is a single node. The process of step 302 is performed, the IDOM flag value is set to 1, and when the node is not a single node, the process of step 303 is performed and the IDOM flag value is set to zero. The operation of the machine translation apparatus of the present invention will be described by taking Japanese-Chinese translation as an example.

A source language sentence "He will not turn around due to debt" is input by the input unit 100, processed by the source language analysis intermediate structure generation unit 200, and then the source language intermediate structure shown in FIG. Can be earned. Next, the proto-language proverb idiom processing unit 300 processes as in the flow of FIGS. In step 201 of FIG. 2, the PROC flag value of the intermediate structure is searched for from the top to the bottom and from the right to the left in the intermediate structure shown in FIG. Then, the node "turn" is found. Therefore, according to the determination processing of step 202, since this node is not a modifier (for example, an embedded sentence), the processing of step 203 is started. "Turn" by the intermediate structure in Fig. 6
You can often determine that the node is not a leaf node, so
The processing of step 204 is entered. Proverb dictionary 35 of FIG.
Since the proverb idiom information using the morpheme "turn" as a search key is stored in 0, it is determined that the node "turn" has a proverb or idiom related to it, and step 205
Enter the process. In the proverb phrase dictionary shown in FIG. 5, the auxiliary search element set S of all the related proverb phrases using the morpheme "turn" as a search key is as follows.

S (1) = ((head) (in debt)) S (2) = ((hand)) S (3) = ((eye)) S (4) = ((head)) ) S (5) = ((tongue is)) S (6) = ((ki)) Next, the process of step 206 of FIG. 2 is performed, and the leaf node set Y is as follows by the intermediate structure of FIG. Can be earned.

Y = ((head) (debt) (he)) Then, the process of step 207 in FIG. 2 is started. here,
M (i) = S (i) ∩Y, i = 1 to 6. The processing result is as follows.

M (1) = ((head) (in debt)) M (2) = Ф: M (6) = Ф Then, the process of step 206 is started, and M (i) is not an empty set. Then, the process of step 209 starts. According to the proverb dictionary, it looks like this:

L (1) = (no no numazu) Next, the processing of step 210 in FIG. 3 is entered, and since the auxiliary verb attribute set A = (n) of the morpheme “turn”, J (i) = L ( 1). The result of the processing of Kaji is that J (1) is not the only empty set. Next is the determination process of step 211 in FIG. Since J (1) is not an empty set, step 215 is entered. With the proverb dictionary and the auxiliary search key M (1), the processing result of X (idiom) = unsuccessful due to debt Y (new meaning code) = M371 is obtained. Enter step 216,
After substituting the morpheme and the meaning code attribute value of the morpheme “turn” node of the intermediate structure of FIG. 6 with X and Y, step 2
Since the nodes and attribute values included in M (1) and L (1) have already been processed in step 216 by the processing of 17, it is possible to delete unnecessary and too long nodes and attribute values here. it can. According to the example of FIG. 6, as the result of this processing, two nodes “debt” and “neck”, and the auxiliary verb attribute “n” of the morpheme “turn” are deleted as in the explanation example on the right side of FIG. 6. That is why. Next, in step 230 of FIG. 3, when the attribute PROC of the morpheme “turn” node is set to 1, the process returns to step 210, and the node “he” whose PROC is not 1 is found from the intermediate structure, and then step 202 is executed. Enter, step 23 because node "he" is not a modifier.
Enter 0, set PROC of this node to 1, and then return to step 201, where PROC of the intermediate structure is 1
Since there is no other node, the processing of the source language proverb idiom processing unit 300 in FIG. 1 ends. Then, the process of the intermediate structure discriminating unit 400 is started as shown in the process flow of FIG. In FIG. 4, first, the determination process of step 301 is performed, and as shown in the explanatory example on the right side of FIG. 6, since only two nodes remain in the intermediate structure, it is determined that this input sentence is not a fixed sentence. So you can go to step 303 and flag
After setting IDOM to zero, the process of the intermediate structure determination unit 400 is finished. Then, the process of the conceptual structure conversion unit 500 of FIG. 1 starts. Since the flag IDON value of the node shown in FIG. 6 is zero, the difference in the meaning and syntax of Chinese and Japanese is adjusted by referring to the difference adjustment conversion dictionary 550 of FIG. After that, the translated word selection unit 600 of FIG. 1 refers to a translated word selection dictionary 650 as shown in FIG. 7 to determine the morpheme of each node of the intermediate structure. For example, the translation of "he" is "other", and the translation of "I can't turn my head because of debt" is "Bond Tachitaka"
Is. Next, the target language generation unit 700 generates a Chinese surface structure, and generates a translated sentence "other bond Tachitakachi" in this example.
Becomes Finally, the output unit 800 of FIG. 1 outputs the translation result.

[0020]

According to the present invention, the following effects can be obtained. (1) Properly translated words can be obtained because proverbs and idioms that are not fixed phrases can be processed accurately. Therefore, the quality of translation can be improved. (2) The intermediate structure can be simplified and the speed of machine translation can be speeded up by deleting unnecessary and too long structures during processing or correcting improper meaning codes. (3) As the quality of machine translation is improved, the necessity of re-editing can be significantly reduced, so that manual correction can be reduced and automatic machine translation can be achieved.

As described above, the present invention can solve the problem that only the proverbs and idiomatic phrases of conventional fixed sentences can be processed, and the efficiency of execution is high, and the practicality is much greater.

[0022]

[Brief description of drawings]

[0023]

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

[0024]

FIG. 2 is a flowchart showing processing of a source language proverb idiom processing unit in the embodiment.

[0025]

FIG. 3 is a flow chart showing a continuation of FIG. 2 of the source language proverb phrase processing unit in the embodiment.

[0026]

FIG. 4 is a flowchart showing a process of an intermediate structure discriminating unit in the embodiment.

[0027]

FIG. 5 is a diagram showing a partial structure of a proverbial phrase dictionary in the embodiment.

[0028]

FIG. 6 is a diagram showing an intermediate structure of a source language of an example sentence used in the example.

[0029]

FIG. 7 is a diagram showing a partial structure of a translation word selection dictionary in the embodiment.

[0030]

FIG. 8 is a flowchart showing a translation processing process of a machine translation device in a general intermediate structure system.

[0031]

FIG. 9 is a block diagram showing a configuration example of a conventional machine translation device.

[0032]

FIG. 10 is a flowchart showing processing of a conventional machine translation device.

[0033]

FIG. 11 is a flowchart showing a translation process of a conventional example.

[0034]

FIG. 12 is a diagram showing a configuration of a reference dictionary of a conventional example.

[0035]

FIG. 13 is a diagram showing a part of the configuration of a conventional fixed phrase dictionary.

[0036]

FIG. 14 is a diagram showing an example in which conventional node analysis is performed on an idiom.

[0037]

[Explanation of symbols]

 100 input unit 200 source language analysis intermediate structure generation unit 250 analysis dictionary 300 source language proverb idiom processing unit 350 proverb idiom dictionary 400 intermediate structure discrimination unit 500 conceptual structure conversion unit 550 difference adjustment conversion unit 600 translation word selection unit 650 translation word selection dictionary 700 target language generation unit 800 output unit 900 buffer

Claims (1)

[Claims] 1. A main morpheme, related information of a main morpheme and a submorpheme, and all proverbs and idioms of the relevant main morpheme using the main morpheme of the proverb and idiom as a search key for each proverb and idiom of the source language. And a proverb phrase dictionary that stores the corresponding meaning codes, and an intermediate structure obtained by analyzing the input source language are compared with the information stored in the proverb phrase dictionary and meet a predetermined condition. A proverb phrase processing part that makes a sentence of a proverb or an idiom into one node, and an intermediate structure that determines whether the sentence entered is a simple proverb or idiom based on information such as the number of nodes in the processed intermediate structure A machine translation device comprising: a discriminating unit; and a translation word selection dictionary that stores a part-of-speech code, a meaning code, a meaning control code, and a translation of a corresponding target language corresponding to each proverb and idiomatic phrase of the corresponding source language.