JPH08185405A - Machine translating device - Google Patents

Abstract

PURPOSE: To prevent the sequence of a mark-up symbol to provide no effect from being generated in a text by detecting the sequences of mark-up symbols to be canceled each other inside the text to be generated and generating the text in a second language while excluding the detected sequences of mark-up symbols. CONSTITUTION: It is discriminated whether a translated word is equipped with a tag in a buffer B or not. When there is any tag, it is discriminated whether that tag is equal with a tag provided for the translated word at present or not. When both the tags are equal, a tag flag is set to '1' but a tag symbol is not stored in a buffer E at such a time. When both the tags are not equal, an end tag is stored in the buffer E. Since the text in the second language is generated like this, the sequences of mark-up symbols to be canceled each other and to provide no effect can be prevented from being generated inside the text in the second language.

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,
Especially RTF (Rich Text Format), SGML (Standa
rd Generalized Markup Langage), a machine translation device capable of translating a document including a markup symbol such as TeX into another language.

[0002]

2. Description of the Related Art Conventionally, the following machine translation devices have been put into practical use. A conventional machine translation device inputs a source language document input from a keyboard or the like into a translation module under the control of a CPU (central processing unit). The translation module analyzes the input source language text using a basic dictionary stored in a memory, a dictionary group including a user dictionary prepared by user registration, and grammatical rules, and a structure analysis tree. To generate. Similarly, by using the tree structure conversion rules from the source language to the target language, which are stored in advance in the memory, the structure analysis tree of the text of the source language is converted into the structure analysis tree of the target language, and each word is converted into each word. Appropriate translations are assigned and the necessary parts are supplemented to generate the target language text.

On the other hand, recently, a DTP (Desk Top
Document editing systems with Puplishing) functions are becoming popular. Therefore, many documents including figures and tables in addition to texts are formed. Further, like TeX, the text may include accompanying information for printing, such as designation for typesetting such as typeface and font size. Such information includes, for example, designation of a title and other portions, designation of a typeface to be used, font size, etc., designation information of a word to be adopted as an index, and the like.

These pieces of information, as markup symbols, are mixed with the text body in the text of the document to be processed. By including such a markup symbol in the document, the document can be automatically printed using the format, typeface, font size, etc. according to the markup information. Further, when the index term is specified, the index can be easily created by searching for a word or a word group to which such a markup symbol is added.

There is a markup language as a system of such markup symbols, and SGML established by ISO (International Organization for Standardization) is one example. SGML is for designating a logical structure of a document, for example, a logical structure such as chapters, paragraphs, and bullets. When actually printing a document created by SGML, a markup language for more specifically determining a format is often used. An example of such a markup language is called TeX.

As a result of the increase in the number of documents designated to be printed using the markup language, there is an increasing demand for a technique for translating these documents into other languages.

Corresponding to this, RTF, SGML, Te
Even in the case of a special source text that includes markup symbols in the sentence such as X, a machine translation device that can perform the same translation as a normal source text that does not include markup symbols and restore the tag symbol to the translation result is also realized. It's starting.

[0008]

However, when attempting to translate a document containing a markup symbol as described above by a conventional machine translation device, a markup symbol different from the source language is included in the document. Therefore, the translation is not performed properly, and the translation becomes impossible, or an incorrect translation occurs. Therefore, conventionally, before inputting text to a machine translation device, it is manually checked by an editor or the like whether or not the input text includes a markup symbol, and if the markup symbol is included, each is deleted one by one. Then, it was necessary to input to the machine translation device after that. Therefore, the efficiency of translating a document including a markup symbol with a conventional machine translation device has been very poor.

In order to solve such a problem, a document processing system for performing non-linguistic data such as format information contained in a document without being aware of it is disclosed in Japanese Patent Laid-Open No. 259057/1992.
No. 6,086,045. In the system disclosed in this publication, only the language data is extracted from the document data in which the non-language data and the language data are mixed, and a predetermined editing process is performed on the extracted language data. Further, the language data edited by this editing process and the language data in the document data originally input are compared and compared to determine corresponding portions. Then, the language data of the input document data is replaced with the corresponding one of the edited language data. By doing so, it is possible to edit document data in which linguistic data and format information are mixed without being aware of the presence of non-language data.
It has been.

However, in the translation of a sentence including a markup symbol, it is not necessary to simply add the corresponding markup symbol to each of the translated words of the translation result corresponding to the word having the markup symbol in the original sentence and output it. , There is a problem that the translated text becomes difficult to read because there are too many markup symbols in the translated text. At the same time, there is a problem in that the system is inefficient because extra memory and display space are used as a result of increasing such useless information.

Further, in most current translation systems, translation is performed in units of one sentence. Therefore, the information of the previous sentence is not carried over to the next sentence. When the markup symbol to be paired spans multiple sentences, the information of those markup symbols is missing from the translated sentence, and there is a problem that the translation that accurately reflects the markup information fails. . For example, in a sequence of multiple sentences, the first sentence has only the mark that indicates the start of markup information and does not have the symbol that indicates the end, and the last sentence does not have the mark that indicates the start of markup information and only the end symbol. , There is a problem that the beginning and end of these plural sentences cannot be correctly translated and the translation fails.

Further, the part to which the markup information is added is considered to be a part having a certain unity in the structure of the text original sentence. Since such information is considered to be useful for structural analysis of the original sentence as one step of translation, it is desirable to perform translation processing in a form that also includes markup information.

Therefore, it is an object of the present invention, even if the source language input text contains markup symbols,
It is an object of the present invention to provide a machine translation device that can perform translation more efficiently than ever before.

An object of the present invention as set forth in claim 2 is to translate a text original text including a markup symbol more efficiently and more accurately as a normal text original sentence by reflecting the markup symbol in the original text. A machine translation device is provided.

The object of the present invention as defined in claim 3 is to convert a text original text including a markup symbol into a text original text in a more efficient, faster and more accurate manner as in a normal text original text. It is to provide a machine translation device capable of reflecting and translating.

[0016]

A machine translation apparatus according to the present invention is a machine translation apparatus for translating a text original text of a first language including a markup symbol into a text of a second language. This machine translation device associates each markup symbol with the word to which each markup symbol is attached, and a separating means for separating the original text into a markup symbol and a text body not containing the markup symbol. Storage means for storing, means for performing a predetermined grammatical process on the text body separated from the markup symbol by the separation means, and creating a structural analysis tree of the first language corresponding to the text body; Based on the means for converting the structure analysis tree of the first language into the structure analysis tree of the second language by using a prepared conversion rule, and the stored contents of the second structure analysis tree and the storage means. And a second word in which a markup symbol equivalent to the markup symbol attached to the translated word is attached to the translated word corresponding to the word to which the markup symbol is attached in the original text. And a text generation means for generating the text. The text generation means detects a sequence of markup symbols that cancel each other in the generated text, and excludes the detected sequence of markup symbols to generate the text in the second language. And means.

A machine translation device according to a second aspect of the present invention is the device according to the first aspect, wherein the pair of markup symbols are arranged in different sentences in the original text of the text. In response to the determination means and the determination that the pair of markup symbols is arranged in a different sentence, the sentence of the sentence including the first markup symbol of the pair of markup symbols is detected. A second markup symbol of a pair of markup symbols is supplemented at a predetermined position, and a first markup symbol is supplemented at a predetermined position of a sentence including the second markup symbol and stored. A markup symbol replenishing means for storing in the means, and a text generated by the text generating means, wherein the markup symbol replenished by the markup symbol replenishing means is generated. And further including means for generating further replenished position determined by the position of the mark-up symbols supplemented.

A machine translation device according to a third aspect of the present invention is the device according to the first or second aspect, wherein the means for creating the structural analysis tree is stored in the storage means of the text original text. It is characterized in that the structure analysis tree is created on the assumption that a part sandwiched between paired markup symbols is one phrase.

[0019]

In the machine translation device according to the present invention, the separating means separates the text original sentence into a markup symbol and a text body not containing the markup symbol. The storage means stores each markup symbol. Each markup symbol is associated with the word to which it was attached. The means for creating a structural analysis tree performs a predetermined grammar process on the text body to create a structural analysis tree for the first language. The conversion means converts this structural analysis tree into a structural analysis tree of the second language. Then, the generation unit generates the text of the second language based on the second structural analysis tree and the storage content of the storage unit. In the generated text, the translated word corresponding to the word to which the markup symbol is attached in the original text of text is attached by the text producing means with a markup symbol equivalent to the markup symbol attached to the translated word. At this time, the text generation unit detects a sequence of markup symbols that cancel each other out in the generated text and excludes the detected sequence of markup symbols to generate the text of the second language. Sequences of markup symbols that offset each other and have no effect are not generated in the text of the second language.

In the machine translation apparatus according to the second aspect of the invention, when it is determined that the markup symbols to be paired are arranged in different sentences in the original text of the text, the markup symbol to be paired is determined. A second markup symbol of the pair of markup symbols is supplemented at a predetermined position of the sentence including the first markup symbol of
The first markup symbol is supplemented at a predetermined position of the sentence including the second markup symbol. In the generation of the text of the second language, in the generated text, at a position determined by the position where the markup symbol is supplemented,
The corresponding markup symbols are supplemented. Since the required markup symbols are placed in different sentences in the original sentence, they will not be restored in the translated sentence.

In the machine translation device according to the third aspect of the present invention, the structure analysis tree is created on the assumption that the portion of the original text that is sandwiched between the pair of markup symbols is one phrase. Therefore, the text original text can be analyzed more accurately and at higher speed.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A machine translation apparatus according to an embodiment of the present invention will be described below with reference to the drawings. The machine translation device of the following embodiment is shown as an example of an English-Japanese machine translation device from English to Japanese. However, the present invention is not limited to this, and can be applied to a machine translation device between any other languages. Further, in the following description, the present invention is referred to as a markup symbol TeX
The markup symbol used in the markup language (called "tag symbol") will be described as an example.
The present invention is not limited to this language and is generally applicable to languages that use markup symbols.

Before explaining the embodiments, the concept of machine translation will be briefly described. Referring to FIG. 1, analysis processing performed in machine translation has various analysis levels.
The machine translation is a process for inputting the source language shown in the upper left of FIG. 1 to sequentially perform the processing of each level to finally obtain the target language shown on the right side of FIG. For example, when a text in the source language is input, first, the level L1 dictionary lookup process, the level L2 morphological analysis process, the level L3 syntactic analysis process, ... Done to generate text in the target language.

Machine translation is roughly divided into the following two types, depending on what level of analysis processing is performed. First
Is a pivot method that analyzes up to the intermediate language shown at level L6 and generates text in the target language from there. Secondly, the internal structure of the source language is obtained by performing analysis on any of the levels L2 to L5 described above.
Next, it is a transfer method in which after converting the obtained internal structure of the source language into the internal structure of the target language at the same level, the text of the target language is generated.

The intermediate language used in the pivot system is a concept that does not depend on either the source language or the target language. Therefore, if the source language is analyzed once to obtain an intermediate language, a plurality of languages can be generated from this intermediate language, which is advantageous for multilingual translation. However, in such a pivot method, it is still unsolved whether the intermediate language which is the basic concept can be really set.

The transfer system is a compromise solution to such a problem of the pivot system, and many current systems adopt this transfer system. The following description is for the transfer method, and the machine translation device of the embodiment described later also uses the transfer method.

The contents of each analysis process shown in FIG. 1 will be described below. (1) Dictionary lookup, morpheme analysis In this process, the process of dividing the input sentence into morpheme strings (word strings) using the longest matching method or the like while referring to the dictionary in which the morphemes are stored First done. Then, a process of obtaining grammatical information such as a part of speech and a translated word for each of the obtained words and further analyzing tense, personal name, and number are performed.

(2) Syntactic Analysis In this process, a process of assembling and determining a structure of a sentence (structure analysis tree) such as a crossing between words based on the part-of-speech / variant and the like is performed. In this process,
No judgment is made as to whether the structure of the obtained sentence represents the correct meaning.

(3) Semantic Analysis From a plurality of structural analysis trees obtained as a result of the syntactic analysis process, semantically correct and improper ones are discriminated and only the correct ones are adopted.

(4) Context Analysis In the context analysis process, a process of understanding the topic of the input sentence and removing abbreviated parts and ambiguities contained in the input sentence is performed.

In the translation module of the machine translation device according to the embodiment of the present invention described below, it is assumed that analysis processing up to level L3 is performed. That is, the translation module of the machine translation device of this embodiment has a structure as shown in FIG. With reference to FIG. 2, this translation module separates the source language text source text into a tag symbol and the text itself, and includes a start tag symbol and an end tag symbol in different sentences in the text source text as described later. If the end tag symbol is included, the end tag symbol is supplemented to the end of the sentence that contains the start tag symbol, and the start tag symbol is supplemented to the beginning part of the sentence that contains the end tag symbol. A replenishing unit 101, a storage unit 17 for storing the tag symbol in association with the word to which the tag symbol is attached,
A dictionary lookup / morpheme analysis unit 11 for performing a dictionary lookup / morpheme analysis process on the text body, a syntax analysis unit 12 for performing a syntax analysis on the input sentence subjected to the morpheme analysis, and a result of the syntax analysis. A conversion unit 13 for converting and generating a structure analysis tree of the target language;
And a translation sentence generation unit 14 for generating a translation sentence of the target language in which the tag symbol is inserted, based on the structure analysis tree of the target language generated by and referring to the contents of the storage unit 17.

Furthermore, this translation module uses a dictionary lookup
A tag symbol extraction unit 102 that extracts a tag symbol from the result of analysis by the morpheme analysis unit 11 and stores it in the storage unit 17, and similarly applies tag information to the analysis result of the dictionary lookup / morpheme analysis unit 11 for translation. A tag for applying necessary information in the translated sentence in the translated sentence in the translated sentence generating unit 14 by referring to the tag information application unit 103 for storing useful information in the memory unit 17 and the stored content in the memory unit 17 The symbol restoration unit 104 is included. The processing performed by each unit will be described in detail in the embodiments.

FIG. 3 is a block diagram of a machine translation apparatus according to an embodiment of the present invention. Referring to FIG. 3, this machine translation device includes a main CPU (central processing unit) 1 and a main C.
A bus 7 to which the PU 1 is connected, a main memory 2 connected to the bus 7, and a CRT (cathode ray tube) connected to the bus 7.
A display device 3 including an LCD or a liquid crystal display device (LCD),
Keyboard 4 and translation module 5 connected to bus 7
And a memory 6 that is connected to the translation module 5 and stores a knowledge base such as a dictionary / grammar rule for translation and a tree conversion structure rule for translation.

When the source language text is input, the translation module 5 translates the text in a predetermined procedure and outputs the target language text. FIG.
And the translation module 5 is connected to the bus 7,
The text in the source language (English in this embodiment) input via the bus 7 is translated according to a predetermined translation program to be used as the text in the target language (Japanese in this embodiment) as the bus 7. Translation CP for output to
U15, a translation program memory 16 connected to the bus 7 for storing a translation program executed by the translation CPU 15, a buffer A for storing an input source language text original text for each word, and a buffer For each word stored in A, a buffer B for storing information such as the part of speech and translation of each word obtained by referring to the dictionary included in the memory 6, and information regarding the structure analysis tree of the source language are stored. Buffer C for storing, a buffer D for storing the structure analysis tree of the target language in which the structure analysis tree of the source language stored in buffer C is converted, and a Japanese structure analysis tree stored in buffer D. A buffer E for storing sentences prepared in Japanese by supplementing appropriate verbs with auxiliary verbs, etc., and used in the tag removal processing described later. Tags and includes a buffer F for storing the number of words such as the tag is attached, and a stack 19 for use in the tag symbol determination replenishment section 101 shown in FIG. These buffers correspond to the storage unit 17 (FIG. 2).

The operation of English-Japanese translation by the machine translation apparatus of this embodiment will be described below with reference to FIGS. In addition,
Details of the translation program executed by the translation CPU 15 will be described later using a specific example.

First, the read text original sentence is decomposed into morphemes by morphological analysis, and the tag symbol discrimination supplementing unit 10
1 (FIG. 2) separates the tag symbol and the other text body. The text body is stored in buffer A (see FIG. 4) as shown in FIG. The tag symbol is stored in the buffer F. At this time, the tag symbol discrimination supplementing unit 101 discriminates whether or not the text original sentence includes a pair of tag symbols in different sentences. Then, when it is determined that the paired tag symbols are included in the different sentences, the markup symbol is supplemented to each of those sentences. How to supplement the markup symbol will be described later with reference to FIGS.

Subsequently, a translation C based on the translation program 16
Under the control of the PU 15, the dictionary lookup / morpheme analysis unit 11 shown in FIG. 2 refers to the dictionary stored in the memory 6 for each word of the original sentence stored in the buffer A, and translates each word. And get information such as part-of-speech information. This information is stored in buffer B shown in FIG. As part of this information, part-of-speech information for each word is included,
These parts of speech information are stored as shown in FIG. That is, "this" is a multi-part-of-speech word and has two parts of speech, a pronoun and a referential adjective. The part of speech of "is" is a verb. Similarly, the parts of speech for "a" and "pen" are also stored in the buffer B. Although “this” is a multi-part-of-speech word, the part-of-speech in a sentence is shown in
Is uniquely determined by the process corresponding to the syntax analysis unit 12 shown in FIG.

In the process corresponding to the syntax analysis unit 12 shown in FIG. 2 of the translation program, the modification / interval between each word according to the dictionary and the grammatical rules stored in the memory 6.
The structural analysis tree indicating the modified relationship is determined as shown in FIG. 7, for example. The result of this parsing is stored in the buffer C in FIG.

The structure analysis tree is determined as follows. Among the grammatical rules stored in the memory 6, the grammatical rules for English are made up of the following.

Sentence → main part, predicate main part → noun phrase predicate → verb, noun phrase noun phrase → pronoun noun phrase → article, noun For example, the first rule of these rules is that "sentence is main part" It is made up of predicates. " The same applies to other rules. The structural analysis tree is determined according to these rules. It should be noted that such grammatical rules are similarly prepared for Japanese, and correspondence is made between English grammatical rules and Japanese grammatical rules. Note that tag information is used in this structural analysis, but details will be described later.

In the process corresponding to the conversion unit 13 shown in FIG. 2 in the translation program, the structure of the input structural analysis tree (see FIG. 7) of the English sentence is parsed for the Japanese sentence shown in FIG. Converted to a tree structure. In this conversion, the “tree structure conversion rule” stored in the memory 6 is used in the same manner as that used by the syntax analysis unit 12 described above. This conversion corresponds to the conversion from the level L3 to the target language level L9 in FIG. The obtained result is stored in the buffer D shown in FIG. The example sentence "This is a" used in this description
"pen" is a Japanese character string "kore
It has been converted to a pen.

The portion of the translation program that performs the processing corresponding to the translated sentence generator 14 of FIG. 2 is created by adding an appropriate particle "ha" or auxiliary verb to the obtained Japanese character string "this is a pen". , And stores it in the buffer E shown in FIG. 4 in the grammatical Japanese form as shown in FIG. This process corresponds to the conversion from level L9 to level L10 shown in FIG. The obtained Japanese sentence “This is a pen.” Is output from the translation module 5 shown in FIG. 4, stored in the main memory 2, and displayed on the display device 3.

Next, the details of the process for translating the original sentence including the tag symbol will be described with reference to FIGS. This process is characterized in that the tag symbol included in the original sentence is removed prior to translation and the original sentence from which the tag symbol is removed is input to the translation module. The removed tag symbol is the original
The removed tag symbol is added to the word corresponding to the original word in the translated sentence in association with the word to which the tag symbol was added.

FIG. 11 shows an example of tag symbols used in an SGML document. Each tag symbol is composed of a set of a start tag and an end tag, and marks up a group of words sandwiched between the start tag and the end tag.

The tag symbol <term><shown in FIG.
\ Term> indicates that the word group sandwiched between the tag symbols is a word group used for a specific purpose, for example, a word group adopted as an index. Also, tag symbol <emp
h><\emph> indicates that the word group sandwiched between the tag symbols is emphasized and printed. For example, the word group sandwiched between the tag symbols is printed in bold type when printing.

The tag symbol differs depending on the document format. For example, the RTF described above is a format for exchanging document data, but in a document in the RTF format, the start tag is a character string starting with "{\" and the end tag is "}". The tag symbols are the same in the TeX document.

FIG. 12 is an example of an English text original sentence to be translated in the following description. This original text is stored in the buffer A shown in FIGS. In the original text shown in FIG. 12, tag symbols <term><\ term
The word or group of words sandwiched between> includes “Tag” and “table chart”. These two words or groups of words serve as headwords when creating an index for a document including this sentence, for example. In addition, the part that appears by being sandwiched between the tag symbols is recorded as the appearance page of each word group in the index. Also, tag symbol <em
Two words “prod” sandwiched between ph><\emph>
Both "use" and "specifying" are printed in a font different from the other portions at the time of printing, for example, the bold font described above.

FIG. 10 is a detailed flowchart of the tag symbol processing performed by the separation unit 10 and the dictionary lookup morpheme analysis unit 11 (FIG. 2) of the translation module 5 shown in FIG. Referring to FIG. 10, first, step S0
1, the input sentence shown in FIG. 12 is buffer A (see FIGS.
Stored).

In step S02, the word position pointer indicating the word position to be processed in the original sentence stored in the buffer A is set to 0. As a result, the original word pointed by the word position pointer is "One".

In step S03, it is determined whether or not the beginning of the word currently pointed to by the word position pointer is the start tag while referring to the tag symbol table as shown in FIG. If it is determined that the tag is not the start tag, the process proceeds to step S04, and if it is determined to be the start tag, the process proceeds to step S09.

When it is determined in step S03 that the word is not the start tag, it is determined in step S04 whether the word pointed by the word position pointer is the end tag. If it is an end tag, the process proceeds to step S08. If it is not the end tag, the process is step S08.
Go to 05.

If the word pointed by the word position pointer is neither a start tag nor an end tag, a process of storing the word currently pointed by the word position pointer in the buffer B (see FIG. 4) in step S05. After that, the process proceeds to step S06. When it is determined that the tag is the end tag, in step S08, the process of deleting the end tag and storing the word pointed by the word position pointer in the buffer B is performed, and the process proceeds to step S06.

On the other hand, when it is determined in step S03 that the beginning of the word pointed by the word position pointer is the start tag, the start tag is stored in the buffer F (see FIG. 4) in step S09. Is performed.

Then, in step S10, the word position (that is, the value of the word position pointer) in the sentence in which the start tag is detected is stored in the buffer F in association with the start tag stored in step S09.

In step S11, it is determined whether or not there is an end tag. If the end tag exists, the process proceeds to step S12, and if not, the process proceeds to step S14.

In step S12, the word number "1" is stored in the buffer F in association with the start tag stored in step S09.

Further, in step S13,
The start tag and end tag attached to the word designated by the word position pointer are deleted, and the buffer B is deleted.
Processing is performed, and the processing proceeds to step S06.

On the other hand, when the process proceeds to step S14, the number of words up to the end tag is first counted. The counted number of words is similarly stored in the buffer F in association with the start tag stored in the buffer F in step S09.

Further, in step S15, a process of deleting the start tag added to the word pointed by the word position pointer and storing this word in the buffer B is performed. The process proceeds to step S06.

In step S06, it is determined whether or not the next word is present in buffer A. If it exists, the process proceeds to step S07, and if not, the process proceeds to step S16.

In step S07, a process of incrementing the word position pointer is performed, and the process proceeds to step S07.
Return to 03. Hereinafter, the above-mentioned processing is repeated.
When the word to be processed does not exist in the buffer A, the process proceeds to step S16.

In step S16, a dictionary lookup process is performed on each entry word stored in buffer B. As a result, the information on the part of speech and the number of words for each headword is stored in the buffer B.

Further, in step S17, a process of further adding information about the tag symbol stored in the buffer F to the result of the buffer B subjected to the dictionary lookup process is performed. The addition of the tag symbol is a process of adding information indicating the added tag symbol to the entry word to which the tag symbol is added in the original sentence, among the entry words.

The original text shown in FIG. 12 is shown in FIG.
The progress of processing when the processing indicated by 0 is performed will be described. First, the word "One" is pointed to by the word position pointer by setting the word position pointer to zero. Since the results of the determinations in step S03 and step S04 are both "NO", step S05
Then, the word "One" is stored in the buffer B. In step S06, the determination result is YES because the next word exists, and the process proceeds to step S07. Since the value of the word position pointer is incremented by 1, the word pointed by the word position pointer is the next word "of".
Becomes

Hereinafter, from the words "of" to "this",
The same processing as described above is repeated, and these words are sequentially stored in buffer B together with their word positions.

If the word position pointer is 8, the answer is YES or NO in step S03, and the process proceeds to step S09. In step S09, the process of storing the detected start tag (<term> in this case) in buffer F is performed, and in step S10, the word position is also stored in buffer F. Step S1
Since the determination of 1 is YES, the word number “1” is stored in the buffer F by the processing of step S12. As a result, the tag symbol information regarding the word "Tag" stored in the buffer F becomes the information shown in the first line of FIG. Further, in step S13, the processing of deleting the start tag and the end tag and storing the word "Tag" in the buffer B is performed.

The presence or absence of the start tag and end tag for each word is determined while incrementing the word position pointer in this way, and each word is stored in the buffer B while removing the tag symbol. As a result, the buffer B shown in FIG.
The information shown in is obtained. Further, in the buffer F, as shown in FIG.
Information about the tag symbol as shown in 3 is obtained.

Referring to FIGS. 13 and 14, the tag symbol <t is added to the word at word position 8, that is, the word “Tag”.
erm> is added, and it is understood that this word is the only word sandwiched by the tag symbols. Similarly, the tag symbol <emph> is added to the word "produce". In addition, the word group "table char
t "is the tag symbol <term><\ t before and after these two words.
It is sandwiched by erm>. Furthermore, the word "specifying" at the word position 25 has the tag symbol <e before and after it.
It is sandwiched between mph><\emph>.

Further, as a result of the dictionary lookup processing performed in step S16, the contents stored in the buffer B are as shown in FIG. That is, the information about the part of speech and the number of words obtained by the dictionary lookup process is added to each headword. However, in FIG. 15, specific contents of the part of speech and the number of words added to each headword are omitted for simplification of the drawing.

In step S17, the process of adding the contents of buffer F shown in FIG. 13 to each entry word in buffer B is performed. This process is performed by the buffer F in FIG.
This is done by adding the tag symbol stored in buffer F to the word in buffer B indicated by the word position of. That is, the 8th word "Tag"
Is added with a tag symbol <term>. Similarly, for the word "produce" at word position 14, the tag symbol <emp
h> is added. Also, 2 starting from the 15th word position
The tag symbol <term> is also added to each of the two words “table chart”. This is because the information of the buffer F shown in FIG. 13 indicates that the tag symbol <term> is added to two words starting from the word position 15. In addition, the word "speci
The tag symbol <emph> is also added to "fying".

Thereafter, based on each headword stored in the buffer B, its part of speech, the number of words, and the like, the parsing unit 12 and the conversion unit 13 in FIG. 3 perform the usual translation processing. Finally, the generation unit 14 generates a translated sentence.

At this time, when the word in the input section corresponding to the word to be generated has tag information, the generating section 14 inserts the buffer B shown in FIG. 16 before and after the translated word.
Generates the start and end tags indicated by the contents of. If the corresponding word has no tag information, only the translated word is generated as usual.

This generation processing will be described with reference to FIGS. 17 and 18.
Will be described below. The processing of steps S21 to S27 shown in FIG. 17 is performed for each word in the “leaf” portion of the structure analysis tree. First, in step S21, for a certain word, the column of "tag information" (see FIG. 16) corresponding to the word in the buffer B is checked to determine whether or not the tag information exists. If so, the control proceeds to step S22, and if not, the control proceeds to step S26.

In step S22, the start tag stored in buffer B is first stored in buffer E.

Subsequently, in step S23, a process of storing the translated word corresponding to the word being processed in the buffer B to the buffer E is performed.

Subsequently, in step S24, a process of storing the end tag in the buffer B to the buffer E is performed.

In step S25 after step S24, an appropriate particle to be added to this translated word is selected, and the buffer E
Stored in. After step S25, the control is step S
Proceed to 27.

On the other hand, if it is determined in step S21 that the tag information does not exist in the buffer B, the control is performed in step S
Proceed to 26. In step S26, no processing regarding tag information is performed. That is, the process corresponding to steps S22 and S24 is not performed, but the process of simply storing the translated word from buffer B to buffer E is performed. Step S2
After 6, the control proceeds to step S25.

In step S27, it is determined whether or not the last word of the structure analysis tree to be processed has been reached. If it has not reached, control returns to step S21, and the processing from step S21 onward is repeated. If it is finished, this generation process is also finished.

The contents of the buffer E thus generated are shown in FIG. The translated sentence shown in FIG. 18 corresponds to the original sentence shown in FIG. When the same tag information is attached to a plurality of consecutive words, a start tag is added before the translated word corresponding to the first word and an end tag is added after the last word. FIG. 18 is thus obtained.

FIG. 9 is a block diagram showing the configuration of the machine translation apparatus of the present invention, and the outline is as already described with reference to FIG. The tag symbol discrimination supplementing unit 101 discriminates whether or not there is a pair of tag symbols in different sentences in the input text original sentence.
a and a tag symbol supplementing unit 101b for supplementing a necessary tag symbol when the tag symbol discriminating unit 101a discriminates that there is a pair of tag symbols in different sentences of the text original sentence. including.

The storage unit 17 includes a tag symbol storage unit 171 for storing tag symbols and a tag information storage unit 1 for storing tag information.
72 and 72.

The source text original text is given to the tag symbol discrimination supplementing section 101 by the input section 31. In addition, the translation sentence of the target language is output from the translation module 5 to the output unit 32.
Given to.

20 and 21 are flowcharts of the generation process according to the embodiment of the present invention, which is a modification of the generation process of FIG. By performing the generation process according to this flowchart, when restoring the tag symbol for each translated word of the translation result in the translation of the sentence including the tag symbol, if the same tag symbol continues, the intermediate tag symbol is omitted. Can be output. In this case, the fact that the same tag symbol is continuous means, for example, that the tag end symbol and the tag start symbol of the same type of tag symbol are consecutively included, and the effects thereof can be offset each other.

First, in step S201, the tag flag is set to 0. The tag flag in this case is a flag for later determining whether to restore or omit the tag symbol.

Then, in step S202, the translated word pointer is set at the head of the tree conversion buffer D.

Further, in step S203, it is determined whether or not the translated word pointed by the translated word pointer has the corresponding tag information in the buffer B. That is, whether there is a tag in the tag portion of the buffer B corresponding to the translated word pointed by the translated word pointer,
Or it is determined whether it is blank.

If there is no tag information, the translated word is read from the buffer B and stored in the buffer E in step S213 as usual. After this, the control is step S214 in FIG.
, Which will be described later.

On the other hand, when the tag information exists, it is determined in step S204 whether the tag flag is 1 or not. As will be described later, the tag flag is 1 when the current translation has a tag, the next translation also has a tag, and both tags are the same. This will be described later with reference to step S211. If the tag flag is 1, the tag symbol may be omitted, and if it is not 1, the corresponding tag code may be restored.

If the tag flag is not 1, step S20
In step 5, the tag is fetched from buffer B and stored in buffer E. When this process is performed first, the tag flag is 0, so the tag is always restored.

On the other hand, if the tag flag is 1 in step S204, the tag is read from the buffer B and stored in the buffer E, and the tag flag is simply set to 0. The control after step S205 and S206 is step S20.
Proceed to 7.

In step S207, the translated word is stored in the buffer B.
To the buffer E.

Subsequently, with reference to step S208 in FIG. 21, it is determined whether or not the next translated word has a tag in the buffer B. If there is a tag, it is determined in step S209 whether the tag possessed by the current translation and the tag possessed by the next translation are equal. If they are not equal, control proceeds to step S210. If the two are equal, control proceeds to step S211.

In step S210, if the next translated word has no tag, or if the next translated word has a tag but is not equal to the tag held by the current translated word, control proceeds. In this case, the end tag corresponding to the tag of the current translated word is obtained from the correspondence table (FIG. 11).

Subsequently, in step S212, the obtained end tag is stored in the buffer E, and control is performed in step S214.
Proceed to.

On the other hand, in step S211, the tag flag is set to 1
Set. That is, the following translation has a tag,
The tag flag is set to 1 when the tag having the current translation and the tag having the next translation are equal. At this time, the tag symbol is not stored in the buffer E.
After step S211, control proceeds to step S214.
In step S214, an appropriate particle is stored in the buffer E.

Subsequently, in step S215, the translation word pointer is advanced by one. Further, in step S216, it is determined whether or not the data (translated word) exists at the position pointed to by the translated word pointer that is advanced by one. If there is data, the control is returned to step S203 of FIG. 20 again, and the above processing is repeated. If there is no translated word, this process ends.

FIG. 22 shows the contents of the translation result buffer E thus obtained. As shown in FIG.
In the tag symbol restoration process before improvement, the same tag symbol repeatedly appears in the portion of “<term> table <\ term><term> chart <\ term>” as shown in FIG. However, in the result of the present invention shown in FIG. 22, the line feed portion is "<term> table chart <\ t
erm> ”, it can be seen that the tag symbol is omitted.

As described above, in the translation of a sentence including a tag symbol, when the tag symbol is restored for each translated word of the translation result, when the same tag symbol continues, the intermediate tag symbol is omitted and output. You can Thus, even if the intermediate tag symbol is omitted and output, it is considered that the obtained translation result correctly reflects the original tag symbol. Moreover, since the extra tag symbols are omitted, the storage area can be saved and the processing speed can be increased.

Next, the operation of the tag symbol discrimination replenishing unit 101 will be described with reference to FIGS. 23 to 25
Indicates the contents of the processing performed by the tag symbol discrimination supplementation unit 101.

First, in step S221, the sentence pointer is set to 1. A sentence pointer is a pointer that points to each sentence in a document.

FIG. 26 shows an example of the input document. In this input document, the start tag 1 and the end tag 1 correspond to each other, and the start tag 2 and the end tag 2 correspond to each other. The statement pointer is set to the first statement first.

Referring again to FIG. 23, step S222
Then, the word pointer is set to 1. The word pointer is a pointer that sequentially points the words in one sentence from the beginning to the end of the sentence.

In step S223, the stack pointer is set to 1
Set to. This stack is the stack 19 shown in FIG. Tag symbols are stacked on the stack 19, and it is determined by referring to them whether or not the tag spans a plurality of sentences.

In step S224, it is determined whether the one at the word position pointed to by the word pointer is the start tag. If it is a start tag, the tag is put on the stack in step S225, the stack pointer is incremented by 1 in step S226, and the process proceeds to step S233.

FIG. 27 schematically shows an example of the stack. When it is determined in step S224 that the one existing at the position pointed by the word pointer is not the start tag, the control is performed as shown in FIG.
4 proceeds to step S227. In step S227,
A determination is made as to whether there is an end tag at the position pointed to by the word pointer. If it is not the end tag, the control proceeds to step S233 in FIG.

If it is the end tag, step S22.
At 8 it is determined whether there is a start tag in the stack. If the stack has a start tag, the start tag is taken out in step S229, and the stack pointer is decremented by 1 in step S230. After step S230, the control proceeds to step S233 (FIG. 23).

On the other hand, if it is determined in step S228 that there is no start tag in the stack, control proceeds to step S231.
Go to, copy the end tag to the stack, step S2
At 32, the stack pointer is incremented by 1, and the process proceeds to step S233 in FIG.

In step S233, the word pointer is set to 1
to add. Subsequently, in step S234, if there is a word at the location pointed to by the word pointer, the control is returned to step S224, and the processes of steps S224 to S234 are repeated.

When the word pointer advances to the end of the sentence, the process proceeds to step S235 in FIG. Step S235
Now, determine if there is a tag on the stack. If there is no tag, control proceeds to step S244. If there is no tag, control proceeds to step S236.

In step S236, it is determined whether the tag is a start tag. If the tag is a start tag, control proceeds to step S237.

In step S237, a process of obtaining the end tag corresponding to the start tag from the correspondence table of FIG. 11 is performed.

Subsequently, in step S238, processing for adding the end tag to the end of the sentence is performed. For example, in the example shown in FIG. 26, since the second sentence includes only start tag 2, the processes of steps S237 and S238 are performed. Here, the result of adding the end tag to the end of the sentence in step S238 is the second sentence in FIG.

Subsequently, in step S239, the start tag is extracted from the stack, and this start tag is stored in the tag storage buffer in step S240. This state is shown in FIG.

Further, in step S241, the stack pointer is decremented by 1, and the control proceeds to step S244.

On the other hand, if it is determined in step S236 that the tag on the stack is not the start tag, control proceeds to step S236.
Proceed to 242. In this case, there is an end tag in the stack. In step S242, the start tag stored in the tag storage buffer is taken out and the start tag is added to the beginning of the sentence (step S243). The result is shown in the third sentence of FIG. After step S243, control proceeds to step S244.

In step S244, the sentence pointer is incremented by 1, and it is determined in step S245 whether or not there is a sentence to be further processed. If there is a sentence, control is shown in FIG.
Returning to step S222, the above processing is repeated. If there is no sentence to be processed, this process ends.

FIG. 29 shows the result of processing the input document shown in FIG. 26 as described above. When the tags span a plurality of sentences in the input document, the tags are supplemented by the above processing so that the start tag and the end tag can be associated with each sentence. The translation processing is performed on the sentence in which the tag is supplemented in this way. Therefore, even when the tag spans a plurality of sentences, the tag information can be restored to the translated sentence to generate the translated sentence without losing the tag information. FIG. 30 shows the dictionary lookup morpheme analysis unit 11 of FIG.
In the process performed in step S1, the syntactic analysis performed by the syntactic analysis unit 12 is performed more accurately and at high speed by extracting and using the tag information. FIG. 30 is a flowchart showing the processing performed by the tag information application unit 103.

First, in step S271, a process of reading the tag symbol storage buffer is performed. In this case, the buffer F shown in FIG. 13 corresponds to the tag symbol storage buffer.

In step S272, the buffer pointer is set to 1
Set to. The buffer pointer is a pointer that sequentially designates data from the beginning of the buffer F.

In step S273, it is determined whether the number of words in the data pointed to by the buffer pointer is greater than one.
If it is greater than 1, control proceeds to step S274, otherwise control proceeds to step S275.

In step S274, the number of words and the word position are stored in the tag information storage buffer. Then step S
At 275, the buffer pointer is incremented by 1. When the number of words is 1, nothing is done except the process of adding 1 to the buffer pointer.

Then, in step S276, it is determined whether or not there is data at the location pointed to by the buffer pointer. If there is data, control is returned to step S273, and the above processing is repeated until there is no data. If there is no data, this process ends.

As a result of processing the tag symbol storage buffer shown in FIG. 13 in this way, a tag information storage buffer as shown in FIG. 31 is obtained.

After the dictionary lookup morpheme analysis processing is completed, the syntax analysis unit 12 shown in FIG. 2 analyzes the syntax using the dictionary and the grammatical rules while referring to the tag information storage buffer shown in FIG. . When applying grammar rules,
A grammatical rule in which this portion is regarded as one "phrase" (phrase) is preferentially applied to "word number" words counted from the "word position" in the tag information storage buffer. Since things such as a group of compound words are often sandwiched between tag symbols, this increases the possibility that a correct parsing result can be obtained quickly.

As described above, according to the present invention, even if a markup symbol such as a tag symbol is inserted in the original sentence, the tag symbol can be removed once and automatically translated. Furthermore, by adding an equivalent tag symbol to the translated sentence as in the present embodiment, the original sentence in which the tag symbol is inserted can be directly translated. As a result, it is not necessary for a human to manually remove the tag symbol prior to translation, and translation can be performed efficiently. Moreover, since unnecessary tag symbols are not restored in the translated text, the storage area can be saved and the processing speed can be increased. Even if the tag symbols are used across different sentences, the corresponding tag symbols can be correctly restored in the translated sentence. Furthermore, by using the tag symbol included in the original sentence for parsing, the parsing can be performed more correctly and at high speed.

[0127]

As described above, in the machine translation apparatus according to the present invention, the text body is automatically separated from the markup symbol, and based on the separated text body, the structure of the first language is the same as the conventional one. A parse tree is created. The first
The structure analysis tree of the language is converted into the structure analysis tree of the second language as in the conventional case. On the other hand, each markup symbol is stored in association with the word to which it is attached. When the text of the second language is generated from the structure analysis tree of the second language, the translated word corresponding to the word to which the markup symbol is attached in the original text of the text is the markup symbol attached to the word. A markup symbol equivalent to is added. Therefore, the markup symbol is attached at an appropriate position. It is not necessary to manually delete the markup symbol from the original text of the text or manually insert the markup symbol in the translated text, and you can obtain the translation result that includes the markup symbol corresponding to the input text. . Moreover, sequences of markup symbols that offset each other and have no effect are not generated in the text of the second language. As a result, it is possible to provide a machine translation device capable of efficiently translating a text original text including a markup symbol in the same manner as a normal text original text.

According to the machine translation apparatus of the second aspect of the invention, in addition to the effect of the first aspect of the invention, a pair of markup symbols are arranged in different sentences in a text original sentence. Even in the case where the text of the second language is generated, the corresponding markup symbol is inserted at the position determined by the position where the markup symbol is supplemented in each sentence in the generated text. . Therefore, even if the necessary markup symbols are arranged in different sentences in the original sentence, they are surely restored in the translated sentence. As a result, it is possible to provide a machine translation device that can translate a text original sentence including a markup symbol more efficiently and more accurately and reflecting the markup symbol in the original sentence, like a normal text original sentence.

According to the machine translation apparatus of the third aspect of the invention, in addition to the effect of the first or second aspect of the invention, the portion of the original text that is sandwiched between the pair of markup symbols is one. Since the structural analysis tree is created assuming that the phrase is one phrase, the text original text can be analyzed more accurately and at high speed. As a result, a machine translation device that can translate a text original text including a markup symbol in a more efficient, faster, and more accurate manner by reflecting the markup symbol in the original text as in a normal text original sentence. Can be provided.

[Brief description of drawings]

FIG. 1 is a diagram schematically showing the concept of machine translation.

FIG. 2 is a schematic diagram showing a configuration of machine translation by a transfer method.

FIG. 3 is a block diagram of a machine translation device according to an embodiment of the present invention.

FIG. 4 is a detailed block diagram of the translation module 5 shown in FIG.

5 is a diagram schematically showing the stored contents of a buffer A. FIG.

FIG. 6 is a diagram schematically showing stored contents of a buffer B.

FIG. 7 is a diagram schematically showing stored contents of a buffer C.

FIG. 8 is a diagram schematically showing stored contents of a buffer D.

FIG. 9 is a diagram schematically showing stored contents of a buffer E.

FIG. 10 is a flowchart of tag symbol processing.

FIG. 11 is a schematic diagram showing an example of tag symbols in the SGML language.

12 is a schematic diagram showing an input text original sentence stored in a buffer A. FIG.

FIG. 13 is a schematic diagram showing storage contents of a buffer F.

FIG. 14 is a schematic diagram showing stored contents of a buffer B.

FIG. 15 is a schematic diagram showing stored contents of a buffer B after dictionary lookup processing.

FIG. 16 is a schematic diagram showing stored contents of a buffer B after adding tag symbol information.

FIG. 17 is a flowchart of a subroutine program of generation processing.

FIG. 18 is a schematic diagram showing stored contents of a buffer E after generation processing.

FIG. 19 is a block diagram of a machine translation device according to an embodiment of the present invention.

FIG. 20 is a flowchart of tag symbol restoration processing according to an embodiment of the present invention.

FIG. 21 is a flowchart of a tag symbol restoration process in the embodiment of the present invention.

FIG. 22 is a diagram showing the contents of a translation result buffer.

FIG. 23 is a flowchart showing a process performed by a tag symbol discrimination supplementing unit according to an embodiment of the present invention.

FIG. 24 is a flowchart showing a process performed by a tag symbol discrimination supplementing unit according to an embodiment of the present invention.

FIG. 25 is a flowchart showing a process performed by a tag symbol discrimination supplementing unit according to an embodiment of the present invention.

FIG. 26 is a schematic diagram showing an example of an input document.

FIG. 27 is a diagram schematically showing a stack used in the tag symbol discrimination supplement unit.

FIG. 28 is a schematic diagram of a tag storage buffer used in the tag symbol determination supplement unit.

FIG. 29 is a diagram schematically showing an output document after processing the input document shown in FIG. 26.

FIG. 30 is a flowchart showing processing performed by a tag information application unit according to an embodiment of the present invention.

31 is a diagram schematically showing a tag information storage buffer as a result of processing the tag symbol storage buffer of FIG.

[Explanation of symbols]

 1 main CPU 2 main memory 3 display device 4 keyboard 5 translation module 6 memory 11 dictionary lookup / morphological analysis unit 12 syntactic analysis unit 13 conversion unit 14 translation sentence generation unit 15 translation CPU 16 translation program memory 17 storage unit 101 tag symbol discrimination supplement Part 102 Tag symbol extraction unit 103 Tag information application unit 104 Tag symbol restoration unit

Claims (3)

[Claims] 1. A machine translation device for translating a text source text of a first language containing a markup symbol into a text of a second language, the text source text not including a markup symbol and a markup symbol. Separation means for separating the markup symbol into a text body, a storage means for storing each markup symbol in association with a word to which each markup symbol is attached, and a separation means for separating each markup symbol from the markup symbol Means for performing a predetermined grammatical process on the text body to create a structure analysis tree of the first language corresponding to the text body, and a structure analysis tree of the first language are prepared in advance. Means for converting into a structural analysis tree of the second language using a conversion rule; and the text based on the stored content of the second structural analysis tree and the storage means. To generate a text in the second language by attaching a markup symbol equivalent to the markup symbol attached to the translated word to the translated word corresponding to the word to which the markup symbol is attached in the sentence Text generating means, wherein the text generating means detects a sequence of markup symbols that cancel each other in the generated text, and excludes the detected sequence of markup symbols. A machine translation device including means for generating text in the language of. 2. A means for determining that a pair of markup symbols are arranged in different sentences in a text original sentence, and a means of determining that a pair of markup symbols is arranged in different sentences. In response to this, the second markup symbol of the pair of markup symbols is supplemented at a predetermined position of the sentence including the first markup symbol of the pair of markup symbols. Of the pair of markup symbols, the markup for supplementing the first markup symbol at a predetermined position of the sentence including the second markup symbol and storing it in the storage means. In the text generation by the text supplementing means and the text generating means,
The means according to claim 1, further comprising means for further supplementing the markup symbol supplemented by the markup symbol supplementing means to a position in the generated text which is determined by the position where the markup symbol is supplemented. Machine translation device. 3. The structure analysis tree is created on the assumption that a portion of the text original sentence sandwiched between paired markup symbols stored in the storage means is one phrase. The machine translation device according to claim 1, wherein the machine translation device is created.