JP2001175655A - Machine translation equipment - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To provide a machine translation apparatus for automatically translating an input document between ideographic languages using a computer. A source language sentence is decomposed into a source language part-of-speech decomposition tree by a source language analysis unit, a syntactic / semantic difference adjustment unit adjusts semantic and syntactic structure differences between source object languages, and a translation language The selection unit sequentially detects each single language in the part-of-speech decomposition tree. Next, the unknown vocabulary translation language determination unit converts them into English character strings, and refers to the translation language reference dictionary of another language to detect the corresponding language and related information. After the object language generating unit generates one object language sentence, a translated sentence is output.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation apparatus for translating an input document using a computer to improve the quality of translation, in order to solve a problem relating to translation of an unknown vocabulary.

[0002]

2. Description of the Related Art The twentieth century is the age of knowledge expansion. Nobody continues to accumulate a great deal of knowledge so that it is not left behind by society. However, these sources of knowledge are not always domestic. Most of the new knowledge comes from abroad. Therefore, it is easier and more efficient to read documents in your native language than in foreign languages,
Document translation is important. To improve the quality and efficiency of document translation, the recent trend is to use computers rather than people to do the translation work. Such a translator is generally known as a machine translator. In such a machine translator, the input language to be translated is known as the source language, while the output language translated from the input language is known as the object language. For example, the source language of the Sino-Japanese machine translation device is Japanese, and the object language is Chinese.

Generally speaking, as shown in FIG. 12, a machine translation apparatus essentially includes the following three parts. (1) Source language analysis unit: forms a part-of-speech decomposition tree such as a dependency structure / part-of-speech decomposition tree from an input language by referring to an analysis dictionary through morphological analysis and syntactic and semantic analysis. Relationship is represented by a positional relationship in the tree). (2) Intermediate structure conversion unit: referring to the difference adjustment conversion dictionary, the part-of-speech decomposition tree of the source language (also known as the source language intermediate structure)
Is adjusted and converted into an object language intermediate structure. Then, the translation language for each node is sequentially determined by referring to the translation language selection conversion dictionary, and related information is searched.
(3) Object language generation unit: Referring to the object language syntax, from the object language intermediate structure,
Information for the nodes is sequentially detected to generate an object language sentence. After morphological processing of the object language, such as singular and plural processing and punctuation specification,
An object statement is output.

[0004] An example of the prior art is disclosed in Japanese Patent No. 2814634. The block diagram is shown in FIG. As shown in FIG. 13, reference numeral 1 denotes an analysis dictionary that stores source language vocabulary and information related thereto.
2 represents a difference adjustment translation dictionary that stores information on syntactic, semantic and conceptual differences between the source language and the object language. Reference numeral 3 denotes a translation language selection conversion dictionary that stores information related to the source language vocabulary and the translation language of the object language. Reference numeral 10 denotes an input unit for inputting a source language sentence.

Reference numeral 15 denotes a source language analysis / intermediate structure generation unit for decomposing the source language sentence input from the input unit 10 into a part of speech with reference to the syntactic and semantic information of the analysis dictionary 1. Reference numeral 20 denotes a syntactic / semantic difference adjustment unit for converting the source language intermediate structure input from the source language analysis / intermediate structure generation unit 15 into an object language intermediate structure with reference to the difference adjustment conversion dictionary 2. Reference numeral 25 denotes a semantic dominating code that governs each node and a semantic code that determines the semantic code that governs each node with reference to the association between the translation language selection conversion dictionary 3 and the object language intermediate structure input from the syntactic and semantic difference adjusting unit 20. Represents a semantic code / semantic dominant code selector.

Reference numeral 30 refers to the translation language selection conversion dictionary 3 and uses the semantic dominant code and semantic code of each node of the object language intermediate structure input from the semantic code / semantic dominant code selecting unit 25 as search keys. , A translation language selection unit for determining an object language translation language for each node.
35 is the object language intermediate structure whose translation language selection unit 3
After input from 0, the object language generating unit for generating an object language sentence by referring to the information for each node and the object language structure.

Reference numeral 40 denotes an output unit for outputting the object language sentence generated by the object language generation unit 35 to an associated output device. Reference numeral 45 denotes a buffer area for storing intermediate processing results.

If a vocabulary that is not defined in the analysis dictionary (generally known as an unknown vocabulary) is found in the source language sentence during the part-of-speech decomposition of the source sentence, the source language analysis / intermediate structure generation unit performs processing on these unknown vocabulary. Predicts the part of speech of these unknown vocabularies or uses those specified by humans to maintain the normal operation of the linguistic analyzer in accordance with the above example of the prior art. However, when selecting a translation language, there is no vocabulary heading corresponding to itself in the translation language selection dictionary, so a translation language for it cannot be found. Therefore, the vocabulary item of the source language is used as it is in the translated language. As an example, if we take a Japanese-Chinese machine translation, if the input Japanese sentence is "This area is called storage" and if "Storage" is an unknown vocabulary, the translation result will be "General Storage ". In addition, in “Machine Translation” by Makio Takenori, edited by the Japan Electronic Information and Telecommunications Association and published by Ohmsha, the unknown vocabulary is processed and described as described above.

[0009]

The prior art machine translation apparatus has the following problems. 1. Generally speaking, users of machine translators are not very proficient in the source language, so if the translated text is intermingled with the source language, there is no use for the user.

[0010] 2. Since the source language and the object language may use the same character code such as BIG5 code for Chinese and Shift JIS code for Japanese, the same output device cannot distinguish the same code. It may output wrong characters. This has a significant negative effect on the quality of machine translation. 3. If the user has to register the necessary information during translation, the efficiency of automatic machine translation is dramatically reduced.

Accordingly, it is a primary object of the present invention to provide a machine translation apparatus which can overcome the above-mentioned problems associated with the prior art.

[0012]

According to the present invention, a machine translator according to the present invention comprises the following components: each vocabulary item in a source language, and corresponding syntactic markers, frequency of use, and vocabulary. Analysis dictionary for storing item phonetic symbols, main verb before conversion, class of auxiliary verb before conversion, other item set before conversion, main verb after conversion, class of auxiliary verb after conversion, and after conversion Difference adjustment conversion dictionary for storing other itemsets of the source language vocabulary items,
A translation language selection conversion dictionary for storing syntactic markers, semantic codes, and corresponding object language translation languages, source language marker sequences, possible corresponding English character strings, and source language marker sequences and corresponding corresponding English characters An English character conversion table for storing weights corresponding to the respective character strings, translation language reference dictionaries of other languages for storing other language character strings, syntactic markers, and corresponding object language translation languages, input According to the source language sentence input from the part, the segmentation process is executed by referring to the analysis dictionary to detect related information, and further, the source language syntactic rule is referred to generate the source language part-of-speech decomposition tree. Source language analysis unit, based on the main verb of the source language part-of-speech decomposition tree, refers to the difference adjustment conversion dictionary and converts After the main verb, the class of the auxiliary verb before the conversion, the other item set before the conversion, and the main verb after the conversion, the class of the auxiliary verb after the conversion, and the other item set after the conversion are found, the matching means Using replacement means, search the source language part-of-speech decomposition tree, syntactic semantic difference adjustment unit for converting to a tree-like intermediate structure of the object language, sequentially, each node and its syntactic and semantic information, A translation language selection unit for detecting as a search item from the object language intermediate structure, referring to the translation language selection conversion dictionary, and detecting a corresponding translation language, searching for an object language intermediate structure to find a vocabulary item of an unknown vocabulary After detecting each and referring to the English character conversion table to find all possible English character strings and their combined weights, By referring to the reference dictionary, using each English character string as a search item, an unknown vocabulary translation language determination unit for detecting a translation language having the largest merged weight, and each node information of the object language intermediate structure After sequentially generating source language sentence structures with reference to the source language syntax, using a linear approach,
An object language generator for generating an object language sentence.

According to the machine translation apparatus of the present invention, when a source language sentence is input from the input unit, the source language analysis unit refers to the analysis dictionary and parses the sentence into a source language part-of-speech decomposition tree. The syntactic / semantic difference adjustment unit refers to the difference adjustment conversion dictionary, adjusts the difference related to the source language of the source language part-of-speech decomposition tree, and forms a tree-like intermediate structure of the object language. Uses the source language vocabulary item of each node in the intermediate structure of the object language and related information as a search item and refers to a translation language selection conversion dictionary to detect a translation language.

Next, the translation language determining unit sequentially detects the unknown vocabulary items of the source language intermediate structure, and refers to the alphabetical character conversion table to convert this source language intermediate structure into possible English character strings and merged. Convert to weight. Thereafter, the object language generating unit generates an object language syntactic tree with reference to the object language intermediate structure and the object language syntax. After the object language sentence is obtained using the linear approach, the object language sentence is output from the output unit.

[0015]

FIG. 1 is a system block diagram of a preferred embodiment of Japanese-to-Japanese machine translation according to the present invention. 100 denotes an input unit for inputting a source language sentence. Reference numeral 260 represents an analysis dictionary for storing each vocabulary item of the source language, and corresponding syntactic markers, usage frequencies, and kana phonetic symbols. A schematic structural conceptual diagram of the analysis dictionary 260 is shown in FIG. 200 refers to the analysis dictionary 260, executes segmentation processing of the source language sentence input from the input unit 100, detects related information, further refers to the source language syntactic rule, and executes the source language part of speech. Represents a source language analyzer for generating a decomposition tree. This process
The flowchart is shown in FIG.

Reference numeral 350 denotes a main verb before conversion, a class of auxiliary verb before conversion, another item set before conversion, a main verb after conversion, a class of auxiliary verb after conversion, and another item set after conversion. Indicates a difference adjustment conversion dictionary to be stored. The schematic structural concept of the difference adjustment conversion dictionary 350 is shown in FIG. Reference numeral 300 denotes a main verb before conversion, a class of auxiliary verb before conversion, another item set before conversion, and
After the main verb after the conversion, the class of the auxiliary verb after the conversion, and the other item set after the conversion are found, the collating unit and the replacing unit are used to search the source language part-of-speech decomposition tree to find the object language tree. FIG. 9 shows a syntactic and semantic difference adjuster for converting to a symmetric intermediate structure. FIG.

Reference numeral 450 denotes a translation language selection conversion dictionary for storing source language vocabulary items, syntactic markers, semantic codes, and corresponding object language translation languages. The schematic structural concept of the translation language selection conversion dictionary 450 is shown in FIG. 400 sequentially detects each node and its syntactic and semantic information from the object language intermediate structure as a search item, and refers to the translation language selection conversion dictionary 450 to detect a corresponding translation language. Represents a translation language selection unit. This process flowchart is shown in FIG.

Reference numeral 540 represents an English character conversion table for storing a source language marker string, a possible corresponding English character string, and a weight corresponding to each of the source language marker string and a possible corresponding English character string. The schematic structural concept of the English character conversion table 540 is shown in FIG. 550 represents a translation language reference dictionary of another language for storing other language strings, syntactic markers, and corresponding object language translation languages. Translation language reference dictionary 5 for other languages
The schematic structural concept of 50 is shown in FIG.

The search unit 500 searches the object language intermediate structure to detect vocabulary items of unknown vocabulary, and refers to the English character conversion table 540 to find all possible English character strings and their combined weights. After the search, referring to the translation language reference dictionary 550 of another language, using each English character string as a search item, an unknown vocabulary translation language determination unit for detecting a translation language having the maximum merged weight Represents The process flowchart is shown in FIG.

Reference numeral 600 denotes a method for sequentially generating source language sentence structures with reference to each node information of the object language intermediate structure and the source language syntax, and then generating an object language sentence using a linear approach. Represents an object language generation unit. The process flow diagram is shown in FIG. An output unit 700 includes, for example, a monitor. Reference numeral 800 denotes a buffer area for storing intermediate processing results.

FIG. 2 is a process flowchart of the source language analysis unit. In step S201, after a source language sentence is input from the input unit 100,
At 205, the morphological analysis is performed with reference to the analysis dictionary 260. In step S210, the semantic and syntactic analysis of the source language is performed using the result of the morphological analysis, and the source language phrase structure is generated. In step S215, the source language phrase structure includes, for example, the dependency structure It is converted into a source language intermediate structure, known as a source language part-of-speech tree, such as a part-of-speech tree. The process ends after step S215 has been performed.

FIG. 3 is a process flowchart of the syntactic semantic difference adjusting unit. In step S301, after the source language intermediate structure is input from the source language analysis unit 200, in step S305, the main verb of the part-of-speech decomposition tree is used as a search item, and the difference adjustment conversion dictionary 350 is referred to. The converted information is detected, and the difference in the phrase structure is adjusted by using the collating and replacing means.

Thereafter, in step S310, the main verb of the part-of-speech decomposition tree is used as a search item, the difference adjustment conversion dictionary 350 is referred to, the information before and after conversion is detected, and the collating and replacing means is used. Conceptual differences are adjusted. Next, in step S315, the main verb of the part-of-speech decomposition tree is used as a search item, the difference adjustment conversion dictionary 350 is referred to, information before and after conversion is detected, and the collating and replacing unit is used to check the source language. Is translated into the corresponding object language structure. The process ends after step S315.

FIG. 4 is a process flowchart of the translation language selection unit. After the source language intermediate structure is input from the syntactic and semantic difference adjusting unit 300 in step S401, in step S405, each node is sequentially detected from the upper end to the lower end and from the left end to the right end. . Then, in step S410, it is determined whether an unprocessed node exists. If the answer is no, the source language is converted to an object language intermediate structure and the process ends.

If the answer is yes in step S410, then in step S415, the translation language selection conversion dictionary 450 searches for a corresponding translation language based on the source language vocabulary item, vocabulary class, and semantic code. Is done. Then, step S4
At 20, it is determined whether a translated language is present. If the answer is yes, the process returns to step S405. Otherwise, the process proceeds to step S425, where the vocabulary item of the source language is used as the translation language. Thereafter, the process proceeds to step S405
Return to

FIG. 5 is a process flowchart of the unknown vocabulary translation language determination unit. In step S501, the object language intermediate structure is
After input from 0, in step S505, unprocessed nodes (that is, the translation language is not an object language) are sequentially detected from the upper end to the lower end, from the left end to the right end. Then, the process proceeds to step S510, where it is determined whether there is an unprocessed node. If the answer is no, the process ends.

Otherwise, in step S515, it is determined whether this node is an unknown vocabulary to be processed. If the answer is no, the process returns to step S505. If the answer is yes, the process proceeds to step S520, and it is determined whether or not this node is a katakana unknown vocabulary. If no, the process returns to step S505.
If yes, the process proceeds to step S525, in which the katakana is set as a search item and the English character conversion table 54 is used.
0 is referenced to find all possible English string candidates and their merged weights.

Thereafter, in step S530, processing of each candidate English character string (eg, for repeated Japanese consonants, the corresponding English character must be repeated only once) is performed. Then, in step S535, the translation language reference dictionary 550 of another language is referred to using each candidate English character string and syntactic marker as search items, and the corresponding translation language is detected. Step S
After 535, it is determined in step S540 whether or not the candidate character string having the translation language is “0”.
If yes, in step S550, the comparing means determines the English character string A having the largest weight. Then, in step S555, it is determined whether or not the character string A has a translation language. If the character string A does not have a translation language, in step S560, the character string A is stored in the corresponding node as a translation language, and the process returns to step S505 for subsequent processing.

If the character string A has a translation language in step S555, this translation language is stored in the corresponding node in step S565, and the process proceeds to step S505 for subsequent processing.
If the candidate character string having the translation language is not “0” in step S540, it is determined in step S545 whether the candidate English character string is “1”. If the answer is no, the process proceeds to step S
Proceed to 550. Otherwise, the process proceeds to step S565.

FIG. 6 is a process flowchart of the object language generator. In step S601, after the object language intermediate structure is input from the unknown vocabulary translation language determining unit 500, in step S605, various object language clauses such as noun clauses, verb clauses, and participle clauses are converted based on the object language intermediate structure. Generated. Then, in step S610, one simple sentence is generated by referring to the object language sentence.

Next, in step S615, this single sentence is referred to to generate a compound sentence. Step S61
After step 5, in step S620, the object language and the related properties of the node are referred to, and special object language clauses such as a command clause and a passive clause are generated. Then, in step S625, after the standard symbols are generated in the object language sentence, the object language is detected from the object language phrase structure by the linear approach, and the process ends.

The operation of the preferred embodiment of the Japanese-Chinese translation of the present invention is further described in the following example. One Japanese sentence “This area is called storage” is input unit 1
Input from 00. Then, the source language analysis unit 20
At 0, after the parsing dictionary is referenced and the part-of-speech decomposition process is performed, the dependency structure yields the source language intermediate results shown in [Table 1]. Relevant information for each node, such as vocabulary class, semantic code, auxiliary verb, particle, case marker, etc., is shown in parentheses in this dependency structure.

[0033]

[Table 1]

Thereafter, the adjustment of the difference structure is executed by the syntactic / semantic difference adjustment unit 300. The difference adjustment conversion dictionary 350 is referred to, and it is determined that the structure adjustment is unnecessary. Using the Japanese vocabulary item and syntactic and semantic information of each node of this intermediate structure as search items, the translation language selection conversion dictionary 450 is referred to, and the corresponding translation language is sequentially detected from the translation language selection unit 400. Is done. Accordingly, the object language intermediate structure shown in [Table 2] is obtained. .

[0035]

[Table 2]

Thereafter, in the unknown vocabulary translation language determining section 500, the unknown vocabulary node (storage) is found by searching the intermediate language structure. The possible English character strings and weights are obtained as follows by referring to the English character conversion table 540 using this vocabulary item as a search item.

S → s2 → t2, to1 → la2, le, 1, ra1, ra
i0 di → g2, ge1, di0

The candidate strings and the merged weights are obtained from combining the English strings shown above and are shown below. stag 8, stag 7, stladi
6, stage 7, stage 6, stl
edi 5, storage 7, storage 6,
strati 5, straig 6, straig
e 5, straidi 4, stolag 7,
storage 6, storage 5, stol
eg 6, storage 5, dtoladi
5, storage 6, storage 5, st
oradi 4, storage5, storage
ge 4, storaidi 3.

Next, using the English character string shown above as a search item, a translation language reference dictionary 550 of another language is referred to, and only "storage" is translated into "memory" designated as a translation language. Discover what you can do. Therefore, the object language intermediate language changes as shown in [Table 3].

[0040]

[Table 3]

Subsequently, with reference to the object language intermediate structure and the Chinese syntax, the object language generation unit 600 generates a Chinese phrase structure.
Then, the quantifier "individual" of the demonstrative pronoun "craw" and the invariants "in" and "in" of the case markers "loc" and "opponent" are generated and translated into the translated Chinese sentence "in crawling individual area name storage memory". Is obtained using the linear approach (detecting the Chinese translation languages of leaf nodes in order from left to right). Finally, the Chinese sentence is output from the output unit 700.

[0042]

As described above, the machine translation apparatus of the present invention can overcome the problems associated with the prior art. That is, the advantages of the present invention are as follows: (1) A translation language of an unknown vocabulary can be translated using an object language or an English character string, so that a user unfamiliar with the source text can easily understand the source text. Can be. This dramatically improves the quality of automatic machine translation. (2) Since the translation language of an unknown vocabulary is translated using an object language or an English character string, a wrong character is not output due to a collision in a code area during output. Thus, the output quality is greatly improved. (3) The efficiency of automatic machine translation is dramatically increased because no manual work is required for processing unknown languages.

While the present invention has been described in terms of what is considered to be the most practical and preferred embodiment, the present invention is not limited to the disclosed embodiment, but may be within the broadest spirit and scope of interpretation. It is intended to cover various modifications and equivalent modifications. For example, a dynamic calculation of weights can be used instead of a static calculation of weights.

[Brief description of the drawings]

FIG. 1 shows a system according to a preferred embodiment of the present invention;
FIG.

FIG. 2 shows a process flowchart of a source language analysis unit according to a preferred embodiment of the present invention.

FIG. 3 shows a process flow chart of a syntactic semantic difference adjuster according to a preferred embodiment of the present invention.

FIG. 4 shows a process flowchart of a translation language selection unit according to a preferred embodiment of the present invention.

FIG. 5 shows a process flowchart of an unknown vocabulary translation language determination unit according to a preferred embodiment of the present invention.

FIG. 6 shows a process flowchart of an object language generator according to a preferred embodiment of the present invention.

FIG. 7 is a schematic diagram showing a structure of an analysis dictionary according to a preferred embodiment of the present invention.

FIG. 8 is a schematic diagram showing a structure of a difference adjustment conversion dictionary according to a preferred embodiment of the present invention.

FIG. 9 is a schematic diagram showing a structure of a translation language selection conversion dictionary according to a preferred embodiment of the present invention.

FIG. 10 is a schematic diagram showing a structure of an English character conversion table according to a preferred embodiment of the present invention.

FIG. 11 is a schematic diagram showing a structure of a translation language reference dictionary of another language according to the preferred embodiment of the present invention.

FIG. 12 is a block diagram showing the structure of a conventional machine translation device.

FIG. 13 shows a system block diagram of a conventional machine translation apparatus.

[Explanation of symbols]

REFERENCE SIGNS LIST 100 Input unit 260 Analysis dictionary 200 Source language analysis unit 350 Difference adjustment conversion dictionary 300 Syntactic / semantic difference adjustment unit 450 Translation language selection conversion dictionary 400 Translation language selection unit 540 English character conversion table 550 Translation language reference dictionary for other languages 500 Unknown vocabulary translation language determination unit 600 Object language generation unit 700 Output unit 800 Buffer area

Claims (1)

[Claims] 1. A machine translation apparatus having an input unit, a storage unit, and an output unit, and stores each vocabulary item of a source language, a corresponding syntactic marker, a use frequency, and a vocabulary item phonetic symbol. For analysis, main verb before conversion, class of auxiliary verb before conversion, other item set before conversion, main verb after conversion, class of auxiliary verb after conversion, and other item set after conversion Translation dictionary, source language vocabulary item, syntactic marker, semantic code, and translation language selection translation dictionary for storing the corresponding object language translation language, source language marker sequence, possible corresponding English characters Conversion table for storing the weights corresponding to the columns and the corresponding corresponding English character strings with the source language marker strings, other language character strings, A translation language reference dictionary of another language for storing a tick marker and a corresponding object language translation language; and performing a segmentation process by referring to the analysis dictionary according to a source language sentence input from the input unit. A source language analysis unit for detecting relevant information, further referring to source language syntactic rules, and generating a source language part-of-speech decomposition tree, based on the main verb of the source language part-of-speech decomposition tree, To find the main verb before conversion, the class of auxiliary verb before conversion, other itemsets before conversion, the main verb after conversion, the class of auxiliary verbs after conversion, and other itemsets after conversion. After that, a syntactic and a syntactic method for searching the source language part-of-speech decomposition tree using a matching unit and a replacing unit and converting it into a tree-like intermediate structure of an object language are used. A mantic difference adjustment unit, sequentially detecting each node and its syntactic and semantic information as a search item from the tree-like intermediate structure of the object language, and referring to the translation language selection conversion dictionary, A translation language selection unit for detecting a translation language, searching the tree-like intermediate structure of the object language to detect a vocabulary item of an unknown vocabulary, and referring to the English character conversion table to find all possible English characters After searching the columns and the merged weights of these, the translation language having the maximum merged weight is determined by using each English character string as a search item by referring to the translation language reference dictionary of the other language. An unknown vocabulary translation language determining unit for detecting, referring to each node information of the tree-like intermediate structure of the object language and a source language syntax, sequentially converting a source language sentence structure; After produced, using a linear approach, consisting consist object language generation unit for generating an object language sentence translation apparatus.