JPH07295985A - Machine translation device - Google Patents

Abstract

PURPOSE:To generate partial symbols of kinds different from the inside and the outside of parentheses by inserting delimiter symbols of kinds different for parts satisfying a prescribed condition from parts not satisfying a prescribed condition to borders of partial structures in a translation sentence. CONSTITUTION:A target language generating section (translation sentence generating section) 15d generates a translation sentence based on a syntax structure and a translated word strings of a target language. A delimiter symbol generating section 15f inserts delimiter symbols of kinds different for a position in the translated word strings corresponding to a border in a part satisfying a prescribed condition from a part not satisfying a prescribed condition to the border of partial structures in the translated sentence generated based on the result of discrimination by an in-sentence condition discrimination section 15e. Thus, plural kinds of delimiter symbols are generated and partial symbols of kinds different from the inside and the outside of parentheses are generated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device for outputting a translated sentence divided into smaller parts according to some conditions when translating an original sentence into sentence units.

[0002]

2. Description of the Related Art Most current machine translation devices perform translation using a "sentence", which is a unit of delimiters such as punctuation marks or periods, as one processing unit. Such a machine translation device analyzes the structure of a sentence within one "sentence" and translates it by constructing one independent structure (that is, syntax) from one sentence based on the analysis result. Carry out.

Generally, the machine translation processing has levels as shown in FIG. The left side of the figure is the source language (source language)
Is the analysis level, and the right side is the generation level of the translation destination language (target language). Now, when a sentence in the source language is input, analysis is proceeded with dictionary lookup at the analysis level L ₁ , morphological analysis at the analysis level L ₂ , syntactic analysis at the analysis level L ₃ , and so on.

Machine translation is roughly divided into the following two types according to the above analysis level. One is analysis level L ₆
Source language or in any of the target language analyzed to concepts that are independent (referred to as the intermediate language), the context generation in generation level L ₇ therefrom, generating means at production level L _8, syntax of the generated level L ₉ of This is a pivot method for generating a sentence in a target language by proceeding with morpheme generation and generation at the generation and generation level L ₁₀ . The other is the above analysis level L ₂
Obtaining an internal structure of the source language by performing morphological analysis, parsing the analysis level L _3, semantic analysis of the analysis level L _4, the analysis until either contextual analysis at the analysis level L ₅ in. Next, it is a transfer method in which the obtained internal structure of the source language is converted into the internal structure of the target language at the same level as that, and then a sentence of the target language is generated.

Here, each analysis at the above analysis level will be described in detail.・ Dictionary lookup Morphological analysis Dividing the original sentence into morphemes (words: simply called words) using the information of the dictionary stored in the memory etc., and obtaining grammatical information such as the part of speech of each word and translated words, and Analyze the tense, person, number, etc. -Syntactic analysis Determine the structural analysis tree that represents the structure of the sentence such as the dependency relation between each word.・ Semantic analysis Discriminate semantically correct and improper ones from multiple syntactic analysis results.・ Context analysis Understand the topic and estimate the omitted ones and resolve the ambiguity. The machine translation device described below is based on the transfer method for performing at least parsing up to the parsing level L ₃ .

In the above machine translation, there are cases where one structure (syntax) cannot be determined from one sentence. In such a case, in many conventional machine translation devices, one sentence can be divided into a plurality of parts and translated in some way. For example, JP-A-63-886
In the translation device disclosed in Japanese Patent Laid-Open No. 4 publication, when it is unsuccessful to construct one structure from one input sentence using the grammatical rules registered in the table, the sentence is divided into several parts. The structure consisting of the substructures is created.

The machine translation device having a division translation processing function for outputting a translation sentence in the form of a partial structure obtained by dividing the input sentence into several pieces as described above will be described in more detail below. FIG. 7 is a block diagram of a machine translation device having the divisional translation processing function.

The input unit 2 is composed of an input device such as a keyboard, a mouse, a pen, a tablet, a scanner or a character recognition device and a communication device connected to a communication line. Is entered. The output unit 3 is composed of a display device such as a CRT (cathode ray tube) or LCD (liquid crystal display device), a printing device such as a printer, or a communication device connected to a communication line, and outputs a translation result or the like. The translation main memory 7 includes various buffers such as an original sentence buffer 7a, a dictionary lookup buffer 7b, a pre-conversion syntax buffer 7c, a post-conversion syntax buffer 7d, and a translation buffer 7e for temporarily storing information obtained during translation processing. have.

The translation module 5, as shown in FIG.
The dictionary lookup / morpheme analysis unit 5a, the syntax analysis unit 5b, the syntax conversion unit 5c, and the target language generation unit 5d are provided, and the target language generation unit 5d further includes a division symbol generation unit 5f that generates one type of division symbol. Then, the analysis level L ₃ shown in FIG.
Performs parsing up to the parsing of. When a source language sentence is input, the source language input source sentence is converted using various rules used in translation processing such as a dictionary, grammar rules, and tree structure conversion rules registered in the storage unit 6. Convert to target language. Here, the syntactic analysis unit 5b and the syntactic conversion unit 5c in the translation module 5 are adapted to perform syntactic analysis process and syntactic conversion process suitable for divided translation.

The translation control unit 1 is a CPU (central processing unit)
The input unit 2, the output unit 3, the translation module 5, the storage unit 6 and the translation main memory 7 are controlled to perform translation processing. Incidentally, 4 is a bus line.

The machine translation device having the above configuration operates as follows to convert an input source sentence in a source language (English) into a translated sentence in a target language (Japanese). Translation control unit 1
Under the control of the original sentence "The demand (as measured by the aggregate, busines
s investment) increased all through the year. "is stored in the original text buffer 7a in the main translation memory 7, as shown in FIG. In this way, the original sentence stored in the original sentence buffer 7a is displayed by the output unit 3 as shown in FIG. 9 under the control of the translation control unit 1. In this case, the translated sentence column on the right side of the drawing is blank.

Under the control of the translation control unit 1, the dictionary lookup / morphological analysis unit 5a in the translation module 5
The original sentence stored in the original sentence buffer 7a is read and divided into each morpheme using the information of the dictionary stored in the storage unit 6. Then, information such as a part of speech, a translated word, a tense, a personal name, and a number is obtained for each word, and stored in the dictionary lookup buffer 7b in the translation main memory 7. The part-of-speech information at that time is stored as shown in FIG. Where the word "deman
"d" is a multi-part word having two parts of speech, "noun" and "verb", but is uniquely determined by the syntactic analysis unit 5b.

Next, the syntactic analysis unit 5b, based on the information obtained by the dictionary lookup / morpheme analysis unit 5a and the grammatical rules stored in the storage unit 6, expresses the dependency relation between each word. 11 is determined as shown in FIG. Then, the obtained English structural analysis tree is stored in the pre-conversion syntax buffer 7c in the translation main memory 7. At that time, the structural analysis tree is constructed by using a special element called "splitting element" together.

Next, the syntax conversion unit 5c converts the English structural analysis tree into a Japanese structural analysis tree based on the tree structure conversion rules stored in the storage unit 6. further,
Using the translation word stored in the dictionary lookup buffer 7b in the translation main memory 7, the translation word of each element in the Japanese structural analysis tree is determined. The Japanese structural analysis tree thus obtained and the translated words of each element thereof are stored in the translated syntax buffer 7d in the translation main memory 7, as shown in FIG. Next, the target language generation unit 5d refers to the Japanese structural analysis tree stored in the syntax buffer 7d and the information stored in the dictionary lookup buffer 7b, and adds them to the Japanese structural analysis tree. Translation column “Demand (corporate investment as measured by aggregate)”
With the appropriate particles and auxiliary verbs added to "all increasing during the year", the sentence is transformed into an optimal sentence pattern, and translated into Japanese "Demand (as measured by total business investment").
Increased ‖ all over the year. Is generated. The translated sentence thus generated is stored in the translated sentence buffer 7e in the main translation memory 7, as shown in FIG. The translated sentence stored in the translated sentence buffer 7e is output by the output unit 3 under the control of the translation control unit 1 in association with the original sentence as shown in FIG.

As described above, when the translated sentence is generated by the target language generating unit 5d, one kind of dividing symbol (in this case, "‖") is generated by the dividing symbol generating unit 5f. . The division symbols used in this case are: (1) There is a subtree having the "partitioning element" node as a vertex in the structure analysis tree of the target language. (2) The target word is the end node of the subtree with the "partitioning element" node as the apex, and the target word of the node located at the right end (if the target word has an attached word, the attached word also It is a translation word including). Furthermore, it is not the last translated word in the translated word string. According to the generation criterion, the translated word satisfying the condition (2) is stored in the translated text buffer 7e and then generated.

[0016]

However, the above-described conventional machine translation device having a division translation function has the following problems. First, when the target language generation unit 5d of the translation module 5 generates a translated sentence in the target language as described above, the division symbol generation unit 5f causes only one type of division symbol (in the case of the above example, "‖") Is generated. Therefore, the impression of the role of the shape of the generated division symbol strongly controls the entire translated sentence, which may make it difficult to read the translated sentence.

For example, in the case of the division symbol “‖”,
The impression of the shape's role as a "break" strongly controls the entire translation. Therefore, the above-mentioned Japanese sentence "Demand (as measured by aggregate ‖ corporate investment)
Increased ‖ all over the year. Is essentially large "Demand (as measured by aggregate ‖ corporate investment)
Is divided into two: “increased ‖” and “everything during the year”, while “demand (as measured by aggregate ‖” and “corporate investment” has increased ‖ It seems to be divided into three parts, "and all of that year." It is difficult to read.

Secondly, there is the following problem. That is, a specific pair of character symbols is included in the translated sentence in the target language generated by the target language generation unit 5d.
There may be a part sandwiched by (for example, ““, ””, etc.). The use of such a character symbol is for the purpose of emphasizing that the part sandwiched by the specific pair of character symbols is syntactically or semantically different from the others. At that time, if a partition symbol with a strong impression of “break” (for example, “‖”) is used as the partition symbol for indicating the boundary of the subtree having the “partition element” node as a vertex obtained by the syntactic analysis, The degree of emphasis made by a specific pair of character symbols inserted to emphasize a certain portion becomes weak. However, even if this is the case, if a partition symbol that does not give a strong impression as a "delimiter" is used, the partition symbol is not conspicuous in the case of a translated sentence that has no syntactically or semantically emphasized part, and the partition symbol The effect of will not be fully utilized.

For example, in the case of the translated sentence illustrated in FIG. 13, there is a part sandwiched between parentheses as the above-mentioned specific pair of character symbols, and the part sandwiched between the parentheses and the other part. There is a split sign in each. Originally parenthesis
In a translated sentence that includes (for example, "(,)" or "[,]"), parentheses serve as a significant delimiter.
However, when the translated text is displayed as shown in FIG. 9, the impression of the “delimiter” in the division symbol is too strong, and the impression of the unity between the brackets is weakened. However, even if this is the case, if you use a delimiter that does not give the impression of being a "delimiter," the delimiter will not be conspicuous in the part that is not enclosed by parentheses or in a translation that does not have parentheses. The effect will not be fully utilized.

Therefore, an object of the present invention is to be able to generate a plurality of types of division symbols, and to generate different types of division symbols between a portion sandwiched by a specific pair of symbols including parentheses and other portions. To provide a machine translation device.

[0021]

In order to achieve the above object, the invention according to claim 1 is such that a morpheme analysis unit performs a morpheme analysis process on an input sentence from an input unit, and a syntactic analysis unit performs the morpheme analysis. The parsing process is performed to obtain the syntactic structure of the source language based on the result of the parsing process, and the syntactic conversion unit converts the syntactic structure of the source language into the syntactic structure of the target language and the translated word corresponding to the syntactic structure of the target language. Perform the syntax conversion process to get the sequence,
A translation sentence generation process is performed by the translation sentence generation unit to generate a translation sentence based on the syntactic structure of the target language and a translated word string, and when the input sentence is converted into a translation sentence, the syntax analysis unit and the syntax conversion unit In the machine translation device that executes a divided translation process in which the input sentence is divided into a plurality of partial structures and processed when the input sentence matches a predetermined translation condition, the division translation When processing,
Based on the determination result by the in-sentence condition determination unit that determines whether or not a portion satisfying a predetermined condition exists in the translated word string obtained by the syntax conversion unit, and the determination result by the in-sentence condition determination unit,
At the boundary position of the partial structure in the translated sentence generated based on the translated word string by the translated sentence generation unit, the position in the translated word string corresponding to the boundary position is in a portion that satisfies the predetermined condition. It is characterized in that a division symbol generation unit for inserting a division symbol of a different type from that in the case where the predetermined condition is not satisfied is provided.

The invention according to claim 2 is the machine translation device according to claim 1, wherein the in-sentence condition determination part is a pair of specific character symbols in the translated word string obtained by the syntax conversion part. The division symbol generation unit has a character / symbol determination means for determining whether or not there is a portion sandwiched between, and the division symbol generation unit includes the portion sandwiched by the specific pair of character symbols in the translation and the specific It is characterized in that different types of division symbols are inserted in a portion that is not sandwiched by a pair of character symbols.

The invention according to claim 3 is the machine translation device according to claim 2, wherein the character symbol determination means includes a portion between parentheses in the translated word string obtained by the syntax conversion section. Parenthesis determination means for determining whether or not they exist, wherein the division symbol generation unit inserts different types of division symbols between the portion enclosed by the parentheses and the portion not enclosed by the parentheses in the translated sentence. It is characterized by that.

[0024]

In the invention according to claim 1, when a sentence in the source language is input from the input unit, the morpheme analysis process is executed by the morpheme analysis unit. Then, by the syntactic analysis process by the syntactic analysis unit, based on the result of the morphological analysis process, when the input sentence matches a predetermined translation condition, it is divided into a plurality of partial structures to obtain the syntactic structure of the source language. . Further, by the syntax conversion processing by the syntax conversion unit, the syntax structure of the source language is converted into the syntax structure of the target language, and a translated word string corresponding to the syntax structure of the target language is obtained. Then, the in-sentence condition determination unit determines whether or not there is a portion satisfying a predetermined condition in the translated word string obtained by the syntax conversion unit. Then, when the translated sentence is generated by the translated sentence generation unit based on the syntactic structure and the translated word string of the target language, determination is made by the in-sentence condition determination unit at the boundary position of the partial structure in the generated translated sentence. Based on the result, by the division symbol generation unit, different types of division symbols depending on whether the position in the translation string corresponding to the boundary position is in a portion that satisfies the predetermined condition or not. Is inserted.

In this way, different kinds of division symbols are inserted at the boundary position of the partial structure in the translated text output from the translated text generation section, for the part that satisfies the predetermined condition and the part that does not.

In the invention according to claim 2, there is a portion sandwiched by a specific pair of character symbols in the translated word string obtained by the syntax conversion unit by the character symbol determination means of the sentence condition determination unit. It is determined whether or not to do it. Then, based on the determination result by the in-sentence condition determination unit, the division symbol generation unit inserts the specific pair of character symbols at the boundary position of the partial structure in the translation sentence generated by the translation sentence generation unit. Different types of division symbols are inserted between the part that is sandwiched and the part that is not sandwiched.

Further, in the invention according to claim 3, the parenthesis determining means in the sentence condition determining section determines whether or not there is a portion enclosed by parentheses in the translated word string obtained by the syntax converting section. To be done. Then, based on the determination result by the in-sentence condition determination unit, the division symbol generation unit inserts a portion sandwiched between parentheses at the boundary position of the partial structure in the translated sentence generated by the translated sentence generation unit. Different types of division symbols are inserted depending on the part that is not filled.

[0028]

The present invention will be described in detail below with reference to the embodiments shown in the drawings. FIG. 1 is a block diagram of the machine translation apparatus of this embodiment. The input unit 12, the output unit 13, the bus line 14, and the storage unit 16 in this machine translation device have the same configuration as the input unit 2, the output unit 3, the bus line 4, and the storage unit 6 in the conventional machine translation device shown in FIG. The detailed description is omitted.

The translation main memory 17 is similar to the translation main memory 7 in the conventional machine translation device shown in FIG. 7, and includes an original sentence buffer 17a, a dictionary lookup buffer 17b, a pre-conversion syntax buffer 17c, and a post-conversion syntax buffer 17d. And a translation buffer 17e for temporarily storing information obtained during translation processing. Further, it has a translation word pointer TW for pointing a translation word of each element in the Japanese structural analysis tree, and a parenthesis flag K indicating that an opening parenthesis has appeared. Further, the translation module 15 is the same as the translation module 5 in the conventional machine translation device shown in FIG. It has a language generation unit 15d and has an analysis level L ₃ shown in FIG.
The input sentence in the source language is converted into the translated sentence in the target language by performing the parsing up to the parsing.

Further, the target language generation unit 15d of the translation module 15 in this embodiment has a sentence condition determination unit 15e, which is a feature of this embodiment, and a division symbol generation unit 15f that generates a plurality of types of division symbols. There is.
Here, the in-sentence condition determination unit 15e is provided so that it can be determined whether or not a specific pair of character symbols is present in the translated word string in the target language determined by the syntax conversion unit 15c.
With this configuration, the division symbol generation unit 15f can generate different division symbols depending on whether or not the portion is sandwiched by a specific pair of character symbols in the translated sentence.

In particular, the in-sentence condition determining unit 15 is provided so that it can be determined whether or not a pair of parentheses, which is one of the specific pair of character symbols, exists in the translated word string in the target language.
By configuring e, the division symbol generation unit 15f can generate different division symbols depending on whether or not the portion is sandwiched by a pair of parentheses in the translated sentence. The in-sentence condition determination unit 15e in the machine translation device of the present embodiment is configured to determine whether or not a pair of parentheses are present in the translated word string. In the following description, the source language is English and the target language is Japanese.

The machine translation device shown in FIG.
Under the control of No. 1, the input unit 12, the output unit 13, the translation module 15, and the storage unit 16 perform the divided translation process as follows. 2 and 3 are flowcharts of the divided translation processing operation. The divided translation processing operation will be described in detail below with reference to FIGS. 2 and 3.

In step S1, the original text in English read by the input unit 12 is converted into the main memory 1 for translation.
7 is stored in the original text buffer 17a. The original sentence in the present embodiment is the same as the original sentence used when explaining the operation of the conventional machine translation device having the divided translation processing function, `` The demand (as measured by the aggregate, busines
s investment) increased all through the year. "

In step S2, the translation module 15
The original sentence stored in the original sentence buffer 17a is read by the dictionary lookup / morpheme analysis unit 15a, and the same morpheme analysis as that of the dictionary lookup / morpheme analysis unit 5a in the translation module 5 shown in FIG. 7 is executed. Then, the obtained information is stored in the dictionary lookup buffer 17b in the translation main memory 17. In step S3, the parser 1
5b executes the same syntactic analysis as in the syntactic analysis unit 5b shown in FIG. 7 to determine the English structural parse tree as shown in FIG. Then, the obtained English structural analysis tree is stored in the pre-conversion syntax buffer 17c in the translation main memory 17. At that time, since the original sentence matches the translation condition that "the structural analysis tree cannot be determined using the registered grammatical rule", the structural analysis tree uses a special element called "dividing element" together. It is assembled. In step S4, the syntactic conversion unit 15c executes syntactic conversion similar to that of the syntactic conversion unit 5c shown in FIG. 7 based on the tree structure conversion rule stored in the storage unit 16 to analyze the English structure. The tree is converted into a Japanese structural analysis tree as shown in FIG. Then, the obtained Japanese structural analysis tree is stored in the translated syntax buffer 17d in the translation main memory 17.

Thereafter, the translated sentence generation process by the target language generation unit 15d is executed (steps S5 to S23). However, among them, steps S7 to S
Reference numeral 10 denotes a sentence condition determination process by the sentence condition determination unit 15e, and steps S19 to S21 are division symbol generation unit 15
This is a division symbol generation process by f. In step S5, the translation word pointer TW provided in the translation main memory 17
Is set to "1". In step S6, the parenthesis flag K in the translation main memory 17 is set to "0".

In step S7, the contents of the translated word string added to the Japanese structural analysis tree and stored in the converted syntax buffer 17d is referred to, and the "TW" -th point designated by the translated word pointer TW is displayed. It is determined whether or not the translated word is an open parenthesis "(". As a result, if it is an open parenthesis, the process proceeds to step S8, and if not, the process proceeds to step S9.
In step S8, the parenthesis flag K is set to "1". In step S9, it is similarly determined whether or not the "TW" th translated word is a closing bracket ")". As a result, if it is a closing parenthesis, the process proceeds to step S10, and if not, the process proceeds to step S11. In step S10, the parenthesis flag K
Is set to "0". That is, the parenthesis determination means is constituted by the steps S7 and S9.

In step S11, the dictionary lookup buffer 1
By referring to the information in 7b, it is judged whether or not the "TW" th translation word is a translation word that requires a pattern change. As a result, if the translated word requires a pattern change, the process proceeds to step S12, and if not, the process proceeds to step S13. In step S12, the above "T
The W "-th translated word is changed in type. In step S13, the" TW "-th translated word (if the translated word requires a type change, the translated word after the type is changed in step S12).
Is a translation buffer 17 in the main memory 17 for translation.
Stored in e. At that time, if the translated word string is already stored in the translated text buffer 17e, it is stored at the end of the translated word string.

In step S14, the contents of the converted syntax buffer 17d and the information in the dictionary lookup buffer 17b are referred to, and it is determined whether or not the "TW" -th translation word requires an adjunct word. . As a result, if necessary, the process proceeds to step S15, and if not, the process proceeds to step S16. In step S15, the adjunct word to be added to the "TW" th translated word is the step S13 in the translated text buffer 17e.
It is stored in the position next to the "TW" -th translated word stored in.

In step S16, when the Japanese structural analysis tree stored in the converted syntax buffer 17d is traced upward from the "TW" th translated word, it is determined whether the "partition element" node is reached. That is, it is determined whether or not there is a “partition element” node above the “TW” -th translated word in the Japanese structural analysis tree. As a result, if there is, step S
If not, go to step S22. Step S17
Whether the "TW" th translated word is the node located at the right end of the subtree whose vertex is the "splitting element" node encountered first when the Japanese structural analysis tree is descended Is determined. As a result, if the node is the rightmost node in the subtree, the process proceeds to step S18, and if not, the process proceeds to step S22. In step S18, it is determined whether or not the "TW" th translation word is the final translation word in the translation word string stored in the converted syntax buffer 17d.
As a result, if it is the final translated word, the process proceeds to step S22, and if not, the process proceeds to step S19.

In step S19, it is determined whether or not the content of the parenthesis flag K in the translation main memory 17 is "0". If the result is "0", the process proceeds to step S20, and if not, the process proceeds to step S21. Step S
At 20, the first partition symbol “‖” to be inserted at the boundary of the subtree whose vertex is the “partition element” node other than the part between the opening parenthesis “(” and the closing parenthesis “)” is the translation buffer. It is stored in the position next to the "TW" -th translated word stored in step S13 in step 17e or the attached word stored in step S15. Here, since the first division symbol is inserted at a delimiter portion other than the portion sandwiched between the opening parenthesis and the closing parenthesis, it is more preferable than the impression that the shape of the parentheses “(,)” is the delimiter. The symbol “‖” having a strong impression is used as the first division symbol.

In step S21, the second division symbol "..." to be inserted into the boundary of the subtree having the "partitioning element" node in the portion sandwiched between the opening parenthesis "(" and the closing parenthesis ")" is , Is stored in the translated text buffer 17e at a position next to the "TW" th translated word or the adjunct word. Here, since the second division symbol is inserted at the delimiter in the part sandwiched between the opening parenthesis and the closing parenthesis, it is weaker than the impression that the shape of the parentheses "(,)" is as a delimiter. The symbol "..." Having an impression is used as the second division symbol.

In step S22, it is determined whether or not the "TW" th translation word is the final translation word in the translation word string stored in the converted syntax buffer 17d. As a result, if it is the final translated word, the process proceeds to step S24, and if not, the process proceeds to step S23. In step S23, the content of the translated word pointer TW is incremented and the above step S7 is executed.
The process returns to the process for the next translated word in the translated word sequence. In step S24, the output unit 13 causes
As shown in FIG. 4, the Japanese translated sentence stored in the translated sentence buffer 17e in the translation main memory 17 is read out and is output in association with the original sentence as shown in FIG. After that, the divided translation processing operation ends.

Next, regarding the translated sentence generation processing executed by the target language generation unit 15d, the original sentence "The demand (as measured by the aggregate, busines
s investment) increased all through the year. " Here, the morphological analysis by the dictionary lookup / morphological analysis unit 15a, the syntactic analysis by the syntactic analysis unit 15b, and the syntactic conversion by the syntactic conversion unit 15c of the translation module 15 have already been performed as described above, and the obtained Japanese is obtained. The structural analysis tree of is the sequence of translated words of each element in this Japanese structural analysis tree "Demand (Corporate investment as measured by total)".
It is assumed that it is stored in the post-conversion syntax buffer 17d in the translation main memory 17 together with "all the increasing year."

First, of the translated word strings stored in the converted syntax buffer 17d, the processing for the (TW = 1) th translated word "demand" is executed. Since the translated word "demand" is neither an open parenthesis nor a closed parenthesis and is a noun and does not need to be changed in shape, the translated word "demand" is stored in the translated text buffer 17e as it is. ... Step S5 to Step S7,
Steps S9, S11, S13 Furthermore, the translated word "demand" does not require an adjunct word, and there is a "partitioning element" node at the upper level in the Japanese structural analysis tree, and the "partitioning element" node is the vertex. Since it is not the node at the right end of the subtree and is not the final translated word in the translated word string, the process relating to the first translated word "demand" ends. ... Steps S14, S16, S17, S22

Next, since the (TW = 2) th translated word "(" is an open parenthesis, "1" is set in the parenthesis flag K. Moreover, since it is not a closing parenthesis and there is no need to change the type, The translated word “(” is stored in the translated text buffer 17e. As a result, the content of the translated text buffer 17e becomes “demand (). Furthermore, the translated word“ (”does not require an adjunct word,
Since there is a "partitioning element" node in the upper part of the Japanese structural analysis tree, and it is not the node at the right end of the subtree with the "partitioning element" node as the vertex, and is not the final translation word in the above translation string, the second The process for the translated word "(" is finished .... Step S23, Step S7 to Step S9, Steps S11, S13, S14, S16, S17, S22.

Next, since the (TW = 3) th translated word "statistics" is neither an open parenthesis nor a closed parenthesis, and there is no need to change the type, the translated word "statistics" is stored in the translated text buffer 17e. Further, since the translated word "statistical" requires the adjunct word "ni", the adjunct word "ni" is stored in the translated sentence buffer 17e. As a result, the content of the translated text buffer 17e becomes "demand (statistically). Furthermore, the translated word" statistical "has a" partitioning element "node at the upper level in the Japanese structural analysis tree, and the" partitioning element " Since it is not the node at the right end of the subtree with the node as the vertex and it is not the final translation in the above translation string,
The process for the third translated word "statistics" is ended. ... Steps S23, S7, S9, S11, Steps S13 to S17, S
22

Next, the processing for the (TW = 4) th translated word "by" is performed in the same manner as the case of the first translated word "demand", and the translated word "by" is stored in the translated text buffer 17e. As a result, the content of the translated text buffer 17e becomes "demand (according to statistics). Next, the type of the (TW = 5) th translated word" measure "needs to be changed in the above step S11 due to its syntax condition. Is determined, and in step S14 the attached word "ta"
Is determined to be necessary, and the translated word "measured" after the type change
And the adjunct “ta” are stored in the translated text buffer 17e. As a result, the content of the translated text buffer 17e becomes "demand (measured by statistics").

Next, the (TW = 6) th translated word "youyo" is
Since it is neither an opening parenthesis nor a closing parenthesis, and there is no need to change the type, the translated word "Yoyo" is stored in the translated text buffer 17e. In addition, the translated word "yoyo" does not require an adjunct word, and there is a "partition element" node at the upper level in the Japanese structural analysis tree, and the right end of the subtree whose vertex is the "partition element" node Node, which is not the final translated word in the translated word string, and the content of the parenthesis flag K is not “0” (“1” is set in the process for the second translated word “(”). The division symbol "..." of 2 is stored in the translated text buffer 17e, and as a result, the content of the translated text buffer 17e becomes "demand (as measured by statistics ...) .... Steps S23, S7, S9, S11 ,. S13, S14, Step S16 to Step S19, Step S21, S22

Next, the processing relating to the (TW = 7) th translation word "corporate investment" is performed by the first translation word "demand" and the fourth translation word "by".
The translation word "corporate investment" is stored in the translation sentence buffer 17e in the same manner as in the above case. As a result, the content of the translated text buffer 17e becomes "demand (as measured by statistics ... corporate investment". Next, the (TW = 8) th translated word ")" is a closing parenthesis in step S9. Is determined. Then, in step S10, the parenthesis flag K is set to "0". In step S14, it is determined that the attached word "ha" is necessary, and the translated word ")" and the attached word "ha" are stored in the translated text buffer 17e. As a result, the content of the translation buffer 17e is "demand (as measured by statistics ... corporate investment).
Will be ".

Next, the (TW = 9) th translated word "increase"
Is neither an open parenthesis nor a closing parenthesis, and it requires a type change, and the translated word "increase" after the type change requires the adjunct "ta". The adjunct word "ta" is stored in the translated text buffer 17e. In addition, the above-mentioned translated word "increase" means "partitioning element" at the top of the Japanese structural analysis tree.
There is a node, which is the node at the right end of the subtree whose vertex is the "splitting element" node, and is not the final translated word in the translated word string, but the content of the parenthesis flag K is "0" (the eighth translated word " ) ”Is set to“ 0 ”during the process), the first division symbol“ ‖ ”is displayed in the translation buffer 17e.
Stored in. As a result, the content of the translation buffer 17e is "demand (as measured by statistics ... corporate investment).
Is increased ‖. " ... Steps S23, S7, S9, Steps S11 to S20, S22

Next, the process for the (TW = 10) th translated word "that" is executed in the same manner as the case of the first translated word "demand", etc., and the translated word "that" is stored in the translated sentence buffer 17e. . As a result, the content of the translation buffer 17e is "demand (as measured by statistics ... corporate investment).
Has increased ‖ that. " Next, the processing for the (TW = 11) th translated word “year” is performed in the same manner as the case of the third translated word “statistical”,
The translated word "year" and the attached word "no" are stored in the translated text buffer 17e. As a result, the content of the translation buffer 17e is "demand (as measured by statistics ... corporate investment).
Will increase that year. ” Next, the (TW = 12) th translated word "ma" and
The process for the (TW = 13) th translated word "all" is performed in the same manner as the case of the first translated word "demand", and the translated words "between" and the translated word "all." Are stored in the translated text buffer 17e. It At that time, punctuation marks are added to the final translated word "all". As a result, the content of the translation buffer 17e is "demand (as measured by statistics ... corporate investment).
Increased ‖ all over the year. Will be Then, in step S22, the 13th translated word is determined to be the final translated word, and the translated sentence generation process ends.

As described above, in the machine translation apparatus of this embodiment, the target language generation unit 15d in the translation module 15 has the sentence condition determination unit 15e and the division symbol generation unit 15f, while the translation main memory 17 has A parenthesis flag K is provided. Then, the sentence condition determination unit 1
If the "open parenthesis" is present in the translated word of each element in the Japanese structural analysis tree obtained by the syntax conversion unit 15c in the translation module 15 by 5e, the parenthesis flag K
Is set to "1", while "closed parenthesis" is present, the parenthesis flag K is set to "0". Then, the split symbol generation unit 15f generates the split symbol to be inserted into the boundary of the subtree having the “split element” node in the Japanese structural analysis tree obtained by the syntax conversion unit 15c as the vertex. Is "1", the second division symbol "...", which has a weaker impression than the shape of the parenthesis, is generated, while the content of the parenthesis flag K is "0". In some cases, a first split symbol "|" is generated that has a stronger impression than the impression as a delimiter that the shape of the brackets has.

That is, according to the present embodiment, the second division symbol "...", which is weaker than the parentheses, is inserted into the parentheses in the translated sentence generated by the target language generating unit 15d, while the parentheses are used. The first division symbol “‖”, which is stronger than the parentheses, will be inserted outside. Therefore, the output section 13 outputs the translated sentence "Demand (as measured by statistics ... Corporate investment) output as shown in FIG.
Increased ‖ all over the year. ", The translation is largely" demand (as measured by statistics ... corporate investment).
Is divided into two: “increased ‖” and “everything during the year.” The first half of them, “Demand (as measured by statistics… corporate investment).
"Increased ‖" has the introductory phrase "as measured by statistics ... corporate investment", and this introductory phrase "as measured by statistics ... corporate investment" is "measured by statistical It can be seen at a glance that it has a structure that is divided into two parts, such as “” and “corporate investment”. As a result, the translated text is easy to read, the impression of the unity inside the parentheses is not weakened by the division symbols inserted in the parentheses in the translation, and the effect of the division symbols outside the parentheses in the translation is sufficiently utilized. It is done.

The types of the first division symbol and the second division symbol generated by the division symbol generation unit 15f are not limited to those in the above embodiment. The point is that the first division symbol has a shape with a stronger impression than the parenthesis, while the second division symbol has a shape with a lower impression than the parentheses.

In the in-sentence condition determining unit 15e in the above embodiment, the translated word string added to the structure analysis tree of the target language obtained by the syntax converting unit 15c has a specific pair of character symbols other than parentheses. If it is configured such that it can be discriminated, it is possible to change the type of division symbol between inside and outside of the specific pair of character symbols in the translated sentence. By doing so, it is easy to read, and the degree of emphasis by the pair of character symbols is not weakened by the division symbol inserted in the specific pair of character symbols in the translated sentence, and
The effect of the division symbol in a translated sentence that does not have a portion to be emphasized is fully utilized. At that time, in step S7 in the flowchart of the divided translation processing operation described above, it is determined whether or not the "TW" th translated word is a forward character symbol ("""or the like) forming a specific pair of character symbols. In step S9 above, the "TW" th translated word is the backward character symbol.
It suffices to determine whether it is ("""or the like).

[0056]

As is apparent from the above, the machine translation device according to the first aspect of the present invention uses the morphological analysis unit, the syntactic analysis unit, the syntactic conversion unit and the translation sentence generation unit for the input sentence from the input unit. When performing the divided translation process, the sentence condition determination unit determines whether or not there is a portion satisfying a predetermined condition in the translated word string obtained by the syntax conversion unit, and based on this determination result, the At the boundary position of the partial structure in the translated sentence generated based on the translated word sequence by the translated sentence generation unit by the symbol generation unit, the position in the translated word sequence corresponding to the boundary position satisfies the predetermined condition. Since different kinds of division symbols are inserted depending on whether it is in a part or in a part that does not satisfy it, at the boundary position of the partial structure in the translation obtained by the machine translation device, the predetermined Different types of split symbols are inserted in the portion not filled with the part that meets the matter. Therefore, according to the present invention, when the portion of the translated sentence that satisfies the predetermined condition is a portion that distinguishes it from other portions, the divided signal inserted in the portion that satisfies the predetermined condition is set to the other portion. By using a divided signal of a kind having a weaker impression than the inserted divided signal, the effect of distinguishing the portion satisfying the predetermined condition in the translated sentence from other portions is not deteriorated by the insertion of the divided signal. , It is possible to generate a translation that is easy to understand.

Further, in the sentence condition determination unit in the machine translation device according to the second aspect of the present invention, does the translated word sequence obtained by the syntax conversion unit include a portion sandwiched by a specific pair of character symbols? A character / symbol determination unit for determining whether or not the divided symbol generation unit determines different types of divided symbols in a portion sandwiched by the specific pair of character symbols and a portion not sandwiched in the translated sentence. Since it is inserted, at the boundary position of the partial structure in the translated text obtained by the machine translation device, different kinds of division symbols are inserted between the part sandwiched by the specific pair of character symbols and the part not sandwiched. To be done. Therefore, according to the present invention, if the division symbol inserted in a specific pair of character symbols in a translated sentence is made to be a division symbol having a weaker impression than the division signal inserted in another part, The division symbol inserted in the pair of character symbols does not weaken the degree of emphasis by the pair of character symbols, and the division symbol in the translated sentence in which there is no portion emphasized by the specific pair of character symbols The effect of is fully utilized and a legible translated sentence is generated.

Further, in the character symbol determination means of the sentence condition determination unit in the machine translation device according to the third aspect of the present invention, is there a portion between parentheses in the translated word string obtained by the syntax conversion unit? Parenthesis determination means for determining whether or not the division symbol generation unit inserts different types of division symbols in the portion inside the parentheses and the portion outside the parentheses in the translated sentence, and thus is obtained by the machine translation device. At the boundary position of the partial structure in the translated sentence, different kinds of division symbols are inserted between the portion inside the parentheses and the portion outside the parentheses. Therefore, according to the present invention, if the division symbol inserted in the parentheses in the translated sentence is made to have a weaker impression than the division signal inserted in the portion outside the parentheses, the division symbol inserted in the parentheses in the translated sentence is inserted. The division symbols do not weaken the impression of the unity inside the parentheses, and the effects of the division symbols outside the parentheses are sufficiently utilized to generate an easy-to-read translated sentence.

[Brief description of drawings]

FIG. 1 is a block diagram of a machine translation device of the present invention.

FIG. 2 is a flowchart of a division translation processing operation performed under the control of a translation control unit in FIG.

FIG. 3 is a flowchart of a divided translation processing operation subsequent to FIG.

4 is a diagram showing an example of translated texts stored in a translated text buffer in FIG.

FIG. 5 is a diagram showing an example of output contents by an output unit in FIG.

FIG. 6 is an explanatory diagram of a translation level.

FIG. 7 is a block diagram of a conventional machine translation device having a divisional translation processing function.

FIG. 8 is a diagram showing a concept of an original text buffer in FIGS. 1 and 7.

9 is a diagram showing an example of output contents by an output unit in FIG.

10 is a diagram showing the concept of a dictionary lookup buffer in FIGS. 1 and 7. FIG.

11 is a diagram showing the concept of a structure analysis tree in the source language stored in the syntax buffer before conversion in FIGS. 1 and 7. FIG.

12 is a diagram showing the concept of a structure analysis tree in the target language stored in the syntax buffer after conversion in FIGS. 1 and 7. FIG.

FIG. 13 is a diagram showing an example of translated sentences stored in a translated sentence buffer in FIG. 7.

[Explanation of symbols]

11 ... Translation control unit, 12 ... Input unit, 1
3 ... Output unit, 15 ... Translation module, 15a ... Dictionary lookup / morpheme analysis unit, 15b ... Syntax analysis unit, 15c ... Syntax conversion unit, 15d ...
Target language generation unit, 15e ... Sentence condition determination unit,
15f ... division symbol generation unit, 16 ... storage unit,
17 ... Main memory for translation, 17a ... Original text buffer, 17b ... Dictionary lookup buffer,
17c ... Syntax buffer before conversion, 17d ... Syntax buffer after conversion, 17e ... Translation buffer, T
W ... translation pointer, K ... parenthesis flag.

Claims (3)

[Claims] 1. A morphological analysis process is performed on an input sentence from an input unit by a morphological analysis unit, and a syntactic analysis process is performed by a syntactic analysis unit to obtain a syntactic structure of a source language based on the result of the morphological analysis process. , Performing a syntax conversion process of converting the syntax structure of the source language into the syntax structure of the target language by the syntax conversion unit and obtaining a translation string corresponding to the syntax structure of the target language, and the translation sentence generation unit of the target language. When performing a translation sentence generation process for generating a translation sentence based on a syntactic structure and a translated word string and converting the input sentence into a translation sentence, the syntax analysis unit, the syntax conversion unit, and the translation sentence generation unit are In a machine translation device that performs a divided translation process that divides an input sentence into a plurality of partial structures and processes the input sentence when the input sentence meets a predetermined translation condition, In the divided translation process, based on the determination result by the in-sentence condition determination unit that determines whether or not a portion satisfying a predetermined condition exists in the translated word string obtained by the syntax conversion unit, and the determination result by the in-sentence condition determination unit, At the boundary position of the partial structure in the translated sentence generated based on the translated word string by the translated sentence generation unit, a position in the translated word string corresponding to the boundary position is in a portion that satisfies the predetermined condition. A machine translation device comprising: a division symbol generation unit that inserts a division symbol of a different type from that in a portion that does not satisfy the predetermined condition. 2. The machine translation device according to claim 1, wherein the in-sentence condition determination section includes a portion sandwiched by a specific pair of character symbols in the translated word string obtained by the syntax conversion section. The divided symbol generation unit has a character / symbol determination unit for determining whether or not the portion sandwiched by the specific pair of character symbols in the translated sentence and the specific pair of character symbols in the translated sentence are not sandwiched. A machine translation device, characterized in that different types of division symbols are inserted between parts. 3. The machine translation device according to claim 2, wherein the character / symbol determination means determines whether or not there is a portion sandwiched between parentheses in the translated word string obtained by the syntax conversion unit. A machine translation device, which is parentheses determination means, wherein the division symbol generation unit inserts different types of division symbols in a portion sandwiched between parentheses and a portion not sandwiched between parentheses in the translated sentence. .