JP4298771B2 - Machine translation apparatus and program - Google Patents

Description

  The present invention relates to a machine translation device and a program that automatically translate a sentence in a first language into a sentence in a second language.

  In a machine translation device that automatically translates sentences in the first language (source language) into sentences in the second language (target language) using a computer, the input source sentence is first analyzed by morphological analysis and syntax analysis. Is divided into predetermined translation units. Next, the translation dictionary is searched for each processing unit, a translation rule to be applied is determined, and a corresponding translated phrase is determined. The determined translation phrases are combined according to a predetermined translation generation rule to obtain a translation for the input original sentence.

  Words / sentences arranged according to a certain format, such as the description part of a reference attached to the end of the document, do not have a grammatical sentence structure and have a structure different from that of ordinary sentences. The appropriate translation method varies depending on the element. For this reason, the method of extracting the description part of the reference line by line and performing machine translation on all elements uniformly does not form the description of the reference form in the target language, resulting in deterioration of translation accuracy. It was. For example, “Osamu Hashimoto (2005)“ book of honorific for talking perfectly ”Chikuma Shobo” becomes “Osamu Hashimoto (2005)“ book of honorific for talking perfectly ”.

  In such a case, there is a method in which a description part of a reference document is detected and translated, and the original text is displayed (for example, see Patent Document 1). The original text is displayed when the first language is English, and if the first language is a language that does not use alphabetic characters, it often has to be translated and does not serve the purpose. At least in Japanese-English translation. The Chicago Manual of Style (14th edition, p. 541) explains as follows. `` The titles in languages other than English are treated the same as English.When it is desirable to provide readers with translation of a title, the translation follows the title and is enclosed in parenthesis (sometimes in square brackets.) It's not just a simple translation. For example, it is generally said that book titles, which are one of the elements that make up the description part of a bibliography, are generally not translated into meanings so that readers can refer to the source.

On the other hand, in normal mode machine translation, syntax analysis is performed based on the use of punctuation marks premised on sentence translation, so if this is applied to the description part of a reference, the output order of words may be broken. is there. Therefore, the elements such as author name, book title, publisher, publication / publishing year, page, and place of publication, which are required as items in the bibliography, are manually classified, and based on this classification item, translation or transliteration by meaning (Roman conversion) ), Or a combination of both. As a translation of machine translation, there is one that converts Japanese readings into Romaji and presents them as translations (see, for example, Patent Document 2 and Patent Document 3). These are all limited to proper nouns.
Japanese Patent No. 3394856 JP-A-7-121542 JP 2000-163421 A

  However, in the case of book titles, there are no restrictions such as phrases and sentences as well as proper nouns. In this case, there is no point of how to appropriately write the character string after Romaji conversion. In addition, if it is a proper noun, it is only necessary to capitalize the beginning of the word uniformly, but after converting the book name into Roman letters and dividing it, there was no consideration of where to capitalize and where to keep it lowercase. . Also, assuming that the description part of the bibliography of the first language can be correctly replaced with the format of the target language, and if a translated book of the original source has been published, knowledge of the first language can be obtained if bibliographic information can also be presented. Useful for readers who have nothing at all.

  Further, as another problem, the format of the reference document in the first language does not correspond to the format of the reference document in the second language. In machine translation that does not assume this, for example, when translating Japanese into English, various symbols indicating the delimiter of each element such as punctuation marks are the corresponding English symbols (for example, double quotation marks, double quotation marks, The point is simply replaced with a comma). In addition, even if the Japanese language is the same, there is no unified view of how to write a bibliography. The same applies to English.

  In English, it is common for book titles and journal names to be in italics in principle. In conventional machine translation, uppercase and lowercase letters were used properly, but it appears that there is nothing to convert the translated text to italic or bold format unless the original is italic or bold. Therefore, it was necessary to manually convert the titles of books and magazines to italics. Therefore, if the number of references described is large, labor is required and the range of titles may be wrong.

  Thus, conventionally, translation of the description part of a reference is not subject to translation, or it is uniformly translated without considering factors such as author name, book title, publisher name, and publication year / issue year. Therefore, there is a problem that machine translation cannot obtain a highly accurate translation result. In addition, the necessary adjustments were not made to the second language document, so it did not meet the requirements of the second language bibliography. There is a problem that it is difficult to determine which part corresponds to a book title.

  The object of the present invention is to perform translation suitable for each of the elements constituting the description part of the reference document, and to convert the translated sentence into a format prescribed by the second language and to output the machine translation It is to provide an apparatus and a program.

The machine translation apparatus according to the present invention describes how to use a machine translation program, a translation dictionary that stores knowledge and rules necessary for translation, and parts of speech, parentheses, and punctuation marks for identifying a reference part. Reference data database that stores selected knowledge data, knowledge data for determining which category the elements constituting the reference part of the reference belong to by referring to the symbols and positional relationships before and after the reference part of the reference A storage device that stores a reference form knowledge base that stores information, an input device that inputs a source language in a first language to be translated and information necessary for operation, and a source language in a first language to be translated and a post-translation a display device for displaying a translated sentence of the second language, the machine translation apparatus and a CPU for execution of the machine translation program, stored in the reference identification database With knowledge data that describes how used a part of speech and parentheses, punctuation for identifying that a written portion of the reference, detects the description portion references the textual data input from said input device Stored in the bibliography detector , the bibliography analyzer that divides the character string of the description part of the bibliography detected by the bibliography detector and the separator that divides the character string, and the bibliography format knowledge base The reference analysis unit uses the knowledge data for determining which category the elements constituting the reference part of the reference belong to by referring to the symbols and positional relationships before and after the reference part of the reference. A category determination unit for determining a category of an element constituting a description part of a reference for the obtained character string and assigning the category to the character string, and description of the reference document of the original text to be translated The part is characterized in that a translation unit for translating using said translation dictionary unit strings description part references based on the category that has been granted by the category determining unit.

  According to the present invention, the elements composing the description part of the reference can be translated appropriately, and can be converted into a format prescribed by the second language and output in constructing the translation.

  FIG. 1 is a configuration diagram of a machine translation apparatus according to an embodiment of the present invention. The machine translation apparatus 11 translates and outputs a first language into a second language. For example, a software program such as a machine translation program is installed in a general computer, and the software program is executed by the CPU 12. It is realized by. In the machine translation apparatus 11, a CPU 12, a ROM (Read Only Memory) 13, and a RAM (Random Access Memory) 14 are connected via a bus 15. An input device 16, a display device 17, and a storage device 19 are connected to the bus 15.

  The storage device 19 stores a machine translation program 20, and also stores a translation dictionary unit 21, a reference document identification database 22, a reference document data storage unit 23, a reference document format knowledge base 24, and an interactive screen storage unit 30. .

  The machine translation program 20 includes a control unit 25, a translation unit 26, a reference document detection unit 27, a reference document analysis unit 28, and a category determination unit 29. The translation dictionary unit 21 includes a utilization change dictionary 21a, an analysis grammar dictionary 21b, a word / idiom dictionary 21c, a conversion grammar dictionary 21d, a generation grammar dictionary 21e, a morpheme generation synthesis grammar dictionary 21f, a reading dictionary 21g, a romaji conversion table 21h, and a reference document. The format conversion rule 21i, the division writing rule 21j, and the like are stored.

  The CPU 12 reads out and executes a boot program for starting the machine translation device 11 from the ROM 13 based on an input signal from the input device 16, and further reads an operating system (not shown) stored in the storage device 19. The CPU 12 controls each device based on the input signal of the input device 16, reads out the data of the machine translation program 20 and the translation dictionary unit 21 stored in the storage device 19 and the like and loads the data into the RAM 14. Based on the read program command, the following processing is realized.

  The input device 16 is an input means for inputting data and commands such as character data and file data of the original text to be translated in the first language. Usually, the input device 16 is a pointing device such as a keyboard or a mouse / touch panel, a voice recognition or character recognition function, Alternatively, it is realized by an external storage medium reading device such as a CD drive, a network input device, or the like. That is, the input device 16 inputs a command to be translated, such as a sentence to be translated, and various commands such as a translation instruction and a dictionary registration instruction. In addition, for input of translation target sentence, input such as reading from OCR, flexible disk, magnetic tape, magnetic disk, the Internet, or reading the sentence sentence taken with a microphone and converted into a natural language character string by a dictation device, etc. Is also included. Voice input using a microphone can also be used to input various commands.

  The display device 17 is a means for outputting data input from the input device 16 and translation results. When displaying the screen, translation results, etc. on a display device such as a CRT or a liquid crystal display, the translation results, etc. are stored in a file. and so on. That is, the display device 17 controls various commands to the control unit 25, such as an instruction command that outputs a translation result that is an output of the translation unit 26 or searches and displays a translation example for the input sentence from the example dictionary unit 22. The response from the unit 25 is displayed and output. The display device 17 is usually a display means such as various displays, but the output of the translation result is to a printing means such as a printing machine, a means for directly outputting to a flexible disk, magnetic tape, magnetic disk or other media. Transmission means for transmitting may be used. Moreover, the audio | voice output to a speaker may be sufficient.

  The control unit 25 of the machine translation program 20 controls the entire machine translation apparatus 11 and controls the entire machine translation program 20 and transfers data to and from the input device 16 and the display device 17.

  The translation unit 26 performs a translation process on the translation target original text input by the input device 16 using the translation dictionary unit 21 according to an instruction from the control unit 25, and outputs the translation result to the control unit 25. The reference detection unit 27 detects the description part of the reference from the original text to be translated, and the reference analysis unit 28 detects the character string of the description part of the reference detected by the reference detection unit 27 and its character. The category determination unit 29 divides the column into separators for dividing the column, and the authors, titles, and the like of the elements constituting the description part of the reference for the character string obtained by the reference analysis unit 28. A category such as a publisher name is determined, and the category is assigned to the character string.

  The translation unit 26 also translates the character string of the element of the description part of the reference based on the category assigned by the category determination unit 29 for the description part of the reference detected by the reference detection unit 27 from the original text to be translated. .

The translation dictionary unit 21 is a dictionary for performing translation from the first language into the second language, and stores various dictionary data used when the translation unit 26 performs translation processing. Utilization change dictionary 21a of translation dictionary unit 21 is a dictionary that converts words / idioms whose endings are changed to their original form, and analysis grammar dictionary 21b is a dictionary that stores grammar for analyzing the first language, word / idiom dictionary 21c is a dictionary in which translated words in the second language corresponding to the first language words and idioms are stored together with the part of speech information, and a conversion grammar dictionary 21d is conversion information for converting from the first language grammar to the second language grammar. The generation grammar dictionary 21e is a dictionary that determines sentence structure, the morpheme generation grammar dictionary 21f is a dictionary that completes translated sentences by changing word forms such as endings, and the reading dictionary 21g is a first language heading. dictionary readings reading of stored, conversion table Romanized are stored in correspondence romaji conversion table 21h is having a reading corresponding thereto for each kana character, reference format conversion Law 21i rules that clearly the appropriate translation method in each category, leaving a space between words rules 21j is a rule of leaving a space between words.

  The reference document identification database 22 stores data for identifying the reference document detection unit 27 as a description part of a reference document. In the reference data storage unit 23, elements assigned by the reference analysis unit 28 and their categories are stored in pairs. The reference format knowledge base 24 refers to the symbols and positional relationships before and after the reference description part, and the elements constituting the description part of the reference (author name, title, publisher, publication year / issue year, Data is stored to determine which category the element (page, publication place, etc.) belongs to.

  In the following description, a case where the first language is Japanese and the second language is English will be described. First, a location having a reference format can be detected typically by the use status of parentheses, punctuation marks, and various symbols. For example, in the case of Japanese bibliography in social sciences and humanities, there are usually no punctuation at the end of the line, double parenthesis or single parenthesis is used. For the sake of simplicity, this may be regarded as the contents of the reference from the input line next to the line with the reference or equivalent heading.

  If the reference detecting unit 27 detects such a characteristic part, the reference detecting part 27 transmits the detected part of the document to be translated to the reference analyzing part 28. The reference analysis unit 28 analyzes the internal structure, and the category determination unit 29 assigns categories to elements in the internal structure using the reference format knowledge base 24. The reference data storage unit 23 stores the element assigned by the category determination unit 29 and its category as a set. The stored information is deleted at an appropriate time, for example, when the translation of the “reference document” is completed.

  The dialog screen storage unit 30 stores in advance a dialog screen for the user to interact with the machine translation system via the input device 16 and the display device 17.

  FIG. 2 is an explanatory diagram of an example of a reference document description portion of the original text to be translated, and FIG. 3 is a flowchart showing the processing operation of the machine translation apparatus according to the embodiment of the present invention. Assuming that a document having the reference document description portion shown in FIG. 2 is input to the input device 16 as the original text to be translated, the input device 16 transmits the original text data to the control unit 25. The control unit 25 determines whether the input original sentence is a document composed only of a normal sentence or a document including a portion of “reference document” (S1).

  Whether or not the location of “reference document” is included is determined mainly from the use of parentheses, which are separators for separating character strings, the usage of commas, and the degree of appearance of parts of speech. That is, the reference document detection unit 27 determines whether or not the document is a reference document based on the data for identifying the description part of the reference document stored in the reference document identification database 22. That is, the reference identification database 22 stores knowledge describing how to use characteristic parts of speech, parentheses, and punctuation marks in the reference, and the reference detection unit 27 checks the reference according to this knowledge. Determine whether or not.

  In the case of the example in FIG. 2, the normal sentence in the point that the end of the line is not a punctuation in the third line and after, the point that parentheses are often used in the line, and the line is broken without going to the end of the line. It has a different form. In addition, unless the element is inside the parentheses, it is understood that the particle is not used and the verb is not used at all. Accordingly, the reference document detection unit 27 determines that the third and subsequent lines are the description part of the reference document in light of the reference document identification database 22 in which such knowledge is stored.

  When it is determined that the reference portion has a description part, the reference detection unit 27 further determines whether or not the original text to be translated includes a part having a normal sentence format (S2). That is, it is determined whether or not there is a normal sentence format that does not correspond to the format of the reference document, noting the grammar, the part of speech, etc., but focusing only on the character usage status of the entire document (S2).

  In the example of FIG. 2, it can be seen that the third to fifth lines correspond to references, the first line is a normal sentence, and the second line is a blank line (S2: YES). Therefore, the reference detection unit 27 includes a part corresponding to a normal sentence and a part of “references” (from “country” starting from the first character on the third line to the last character string “(2006)” on the fifth line. (S3). Then, when starting the translation process, the control unit 25 determines whether or not the character string data of the original text to be translated is a normal sentence (S4), and when it is a normal sentence (S4: YES), for example, FIG. The first line and the second line of an example of 2 are sent to the translation unit 26 as they are. Thereby, the translation part 26 performs normal translation about the 1st line and 2nd line of an example of FIG. 2 (S5).

On the other hand, when it is determined that it is not a normal sentence but a description part of a reference (S4: NO), the control unit 25 transmits the data to the reference analysis unit 28. The reference document analysis unit 28 classifies the sent data of the “reference document” portion into a character string and parentheses / symbols (S6). In this case, it is detected that the third line, the fourth line, and the fifth line in the example of FIG. 2 are composed of the following character strings and symbols, respectively.
<Third line>
String 1
・ (Midpoint) One character string 2
・ (Midpoint) One character string 3
"(Open double parenthesis)
String 4
] (Closed in double brackets)
String 5
((Open parenthesis)
String 6
) (Close parentheses)
Line feed <4th line>
String 1
"(Open double parenthesis)
String 2
] (Closed in double brackets)
String 3
((Open parenthesis)
String 4
) (Close parentheses)
Line feed <5th line>
String 1
"(Open single parenthesis)
String 2
"(Close single parenthesis)
"(Open double parenthesis)
String 3
] (Closed in double brackets)
String 4
((Open parenthesis)
String 5
) (Close parentheses)
Line breaks Excluding the character string part, the third line is ",", ",", (,), new line, the fourth line is ",", (,), new line, and the fifth line Are ",", ",", (,), and line feed. Therefore, the reference document analysis unit 28 determines whether or not these three lines have some commonality (S7). For this, a generally known character string matching method is used. Then, although it is a little different, it can be seen that they partially match each other. Specifically, it is common that there is a sequence of “open double parenthesis, close double parenthesis, open parenthesis, close parenthesis, line feed”.

  When the above detection is completed, a set of character strings, that is, “Jiro Kokuryo”, “Jujiro Nonaka”, “Masaka Kataoka”, “Knowledge Management in Network Society”, “NTT Publishing”, “2003”, “Artificial Intelligence Society”, “Artificial Intelligence” The bibliography analysis section 28 is based on the “Science of Intelligence”, “Kyoritsu Publishing”, “2003”, “Junjiro Nonaka”, “Strategy of Knowledge Management”, “Information Processing”, “Volume 47”, “No. 5”, “pp. 547-552”, “2006” Then, the control unit 25 sends these pieces of information to the reference data storage unit 23 (S8). Note that character strings such as “artificial intelligence encyclopedia” and “information processing” are elements constituting the description part of the reference.

  The reference data storage unit 23 stores the data extracted by the reference analysis unit 28. At this time, the information is stored so that information such as what element the element is in the corresponding line, what line the element belongs to, and what parentheses surround it are understood.

  FIG. 4 is an explanatory diagram showing an example of a storage format of the reference data stored in the reference data storage unit 23. The reference data is stored, for example, in a coordinate format as shown in FIG. "Kokuryo Jiro" is stored in the coordinate format, such as (3,1 ,, ...) and "Artificial Intelligence Dictionary" (4, 2, ","). The coordinates (3, 1,...) Of “Kokuryo Jiro” means that “Kokuryo Jiro” is located in the first column of the third row and is separated by a separator “·”. Also, (4, 2, “,”) in the “Artificial Intelligence Encyclopedia” indicates that the “Artificial Intelligence Encyclopedia” is located in the second column of the fourth row and is separated by the separator ““, ””. means.

  Next, the category determination unit 29 sequentially extracts each element of the portion from the third line to the fifth line in FIG. 2 and uses the reference format knowledge base 24 to determine which category each element belongs to. Determine (S9). This reference document format knowledge base 24 is for determining which category an element belongs by referring to the preceding and following symbols and positional relationships. Specifically, the first element is the author, the element is enclosed in double parentheses, and if there are single parentheses on the same line, it is the journal name, if not, the book name, and the front and back are enclosed in parentheses. The number of letters contains the knowledge that it is the year of publication / issue, and that the elements connected with the first element at the midpoint are co-authors.

  There are various forms of “references” (generally known are the science and technology information distribution technical standards and the sociological critique style guide). For example, in the field of information processing, they are listed in FIG. Unlike the example, paper names, book titles, and journal names generally use colons instead of double or single parentheses. Therefore, in this bibliography format knowledge base 24, the general-purpose knowledge applicable to any format (for example, the first element is the author name) is set high in order of application. There is knowledge specific to the field as you go down, and the order of application is lower.

  Each element is information given by the reference document analysis unit 28 (what line the character string such as “artificial intelligence science encyclopedia” “information processing” belongs to, what element the element is in the corresponding line, And the category determined by the category determination unit 29 and the control unit 25 sends the information from the reference data storage unit 23 to the translation unit 26.

  Further, the category determination means 29 uses the reference format knowledge base 24 for each element of the description part of the reference document, using the reference format knowledge base 24, “Jiro Kokuryo” “Yujiro Nonaka” “Masaka Kataoka” "The Society of Artificial Intelligence" "Yujiro Nonaka" is the author name, "Knowledge Management in the Network Society" "Artificial Intelligence Encyclopedia" is the title of the book, "Strategy of Knowledge Management" is the title of the paper, "Information Processing" is the name of the magazine, " “2006”, “2003”, and “2003” determine the category as publication year / issue year.

  In the example in which the category determination unit 29 cannot determine the category, in the example of FIG. 2, “Volume 47”, “No. 5”, “pages 547-552” cannot be determined as a category, and the control unit 25 determines “references”. The fact that it was a part is stored in the reference data storage unit 23 and sent to the translation unit 26 as it is.

  In the translation unit 26, the knowledge (utilization change dictionary 21 a, analysis grammar dictionary 21 b, word / idiom dictionary 21 c, conversion grammar dictionary 21 d, generation) stored in the translation dictionary unit 21 is generated for the original text sent from the control unit 25. Using the grammar dictionary 21e, the morpheme generation grammar dictionary 21f, the reading dictionary 21g, the Romaji conversion table 21h, the reference format conversion rule 21i, the division rule 21j, and the positional relationship information stored in the reference data storage unit 23, A language translation process is performed (S5), and the translation result is output to the output device 17 in a predetermined format (S10).

  Next, the processing when the translation unit 26 translates “knowledge management of the network society”, which is the fourth element in the third row of FIG. 2, will be described. As a premise thereof, the reference document format conversion rule 21i will be described. The bibliographic format conversion rule 21i specifies a translation method suitable for each category.

  In the case of an author name, the [author name] element in the description section of the bibliography should be “first name, comma, first name”. If there are two or more names, the second and subsequent names should be “first name, last name”. Add a comma after the element. After the last author, put “and” behind the comma to become the last author. The editor follows the author, but if it is alone, it will be followed by a string such as “(ed.)” “(Ed.)”. If it is more than one, it will be followed by a character such as “(eds.)” ”(Eds.)” Grant a column.

  The title is converted to italic by converting it into roman characters, and the English translation of the title is enclosed in parentheses following the italic name of the romaji conversion. For publishers, a rule has been accumulated in which a translated word is used as a proper noun in the word / idiom dictionary 21c, and a romaji conversion result is used otherwise. Note that the English translation of the book title is not essential and can be omitted. However, here, the description will be made assuming that the English translation and writing method in which the italicized book name and the English translation of the book title are written together is selected.

  Now, the element of “knowledge management in the network society” that is focused on now belongs to the category “book title” in the category determination unit 29. Therefore, the translation process required for the book title is converted to Roman italic, followed by the English translation enclosed in parentheses.

  The Roman conversion in this case will be described with reference to the flowchart of FIG. FIG. 5 is a flowchart showing an example of processing contents of Roman character conversion in the translation unit 26. First, the target character string “knowledge management of network society” is subjected to morphological analysis and divided into morphemes (S11). In this example, it is divided into “network | society | no knowledge | management”. Next, the division rule 21j is applied to this morpheme (S12).

  For example, the following rules are included. 1. An auxiliary verb or adjunct is combined with the immediately preceding verb or adjective. 2. The prefix and suffix are not separated. 3. Compound words with a length of 6 beats or less are not separated. 4). If the compound word is 7 beats or longer, every part should be 6 beats or less. 5). When one of the particle "ha, no" is followed by one of the particles "ha, no,", or "de", the utilization form of "da" is combined.

  According to this, this example has no place to be combined. However, if the title is “dialogue with archeology, anthropology, and linguistics”, the morphological analysis will be performed as “archaeology | anthropology | linguistics | and | And “no” combine to change to “archaeology | anthropology | linguistics |

  Next, Romaji conversion is performed using the reading information, the reading dictionary 21g, and the Romaji conversion table 21h (S13). In this example, “nettowaku shakai no chishiki keiei”. In this romaji conversion, when the katakana language is recognized as a foreign language, it may be changed to the original language instead of the romaji conversion. In this case, “network” corresponds to this, and according to this, the whole becomes “network shakai no chishiki keiei”.

  Next, uppercase letters are used for the first letters of consecutive independent words separated by spaces (S14). For example, since “network”, “society”, “knowledge”, and “management” do not include any additional words, the initial letters are capitalized. On the other hand, the case particle “no” is an annexed word, and is not processed in upper case. As a result, “Nettowaku Shakai no Chishiki Keiei” is obtained. Then, the whole is italicized (S15), “Nettowaku Shakai no Chishiki Keiei” is obtained, and the essential processing of the book title is completed.

  Subsequently, it is determined whether or not the translation of the title by meaning is desired (S16). When the translation of the meaning by meaning is desired (S16: YES), the title is translated. In the previous translation result, the beginning of the content word is changed to upper case, and the front and back are enclosed in parentheses and added to the end of the Romaji conversion result (S17).

  For example, if the translation result “knowledge management of network society” is obtained as a translation of the book title, the word “knowledge” “management” and “network society” in the previous translation result will be capitalized. Enclose the front and back in parentheses. Then, in this example, content words other than “of” are “Knowledge Management of Network Society”. After this capitalization, the whole is enclosed in parentheses, and this is added to the end of the character string obtained in step S15. Therefore, in this example, the character string “Nettowaku Shakai no Chishiki Keiei (Knowledge Management of Network Society)” is obtained and the processing is terminated.

  In this example, there are “Artificial Intelligence Encyclopedia” and magazine name “Information Processing” as other titles, which are processed in the same manner. In that case, the same processing is performed except that the article title “Knowledge Management Strategy” is not italicized. This is due to the adoption of an English translation and writing system in which the italicized titles for Roman conversion and the English translation of the titles are written together.

  Next, translation of author names will be explained. The author name in the reference document format conversion rule 21i is processed. That is, if the author's name on the 5th line, which is a single work, is machine translated as a human name, “Ikujiro Nonaka” is obtained. Since this is in the order of “first name, last name”, it is reversed like “first name, last name” and separated by commas to obtain “Nonaka, Ikujiro”. On the other hand, the third line consisting of multiple authors such as “Jiro Kokuryo, Kanjiro Nonaka, Masanori Kataoka” is individually translated to obtain “Jiro Kokuryo”, “Ikujiro Nonaka”, and “Masanori Kataoka” respectively. These are all in the order of “first name, last name”. In the case of multiple authors, it is common for only the first author to have “first name, comma, first name”, so “Kokuryo, Jiro”, and connect the remaining authors with a comma and “Kokuryo, Jiro, Ikujiro Nonaka, and Masanori Kataoka ”.

  Next, category assignment is “artificial intelligence society edition” classified as an author, but paying attention to the fact that the end is “hen”, it is re-interpreted as an editor, giving the category editor, and the kanji “ Translate the part except "". In this case, it will be translated as “Japanese Society for Artificial Intelligence”, and will be given a character string such as “(ed.)”, Which is a mark clearly stating that it is an editor, and “Japanese Society for Artificial Intelligence (ed.)” obtain.

  On the other hand, “NTT publishing” and “Kyoritsu publishing” categorized as publishers are translated mainly using the word / idiom dictionary 21c. If the translations of these company names are not registered in these dictionaries, after morphological analysis, they are written in the same way as the titles, and the initials are capitalized. For example, if “NTT Publishing” is not registered in the dictionary, it is converted into a character string “NTT Shuppan”. In addition, with respect to the publication year / issue year, “2003”, “2003”, and “2006” in the example of FIG. 2 are used as they are in English, and no processing is performed. However, there is a method of “2003” and “2006”, and in this case, it is necessary to omit “year”.

  Next, processing for “Volume 47”, “No. 5”, “Pages 547 to 552” whose category is indefinite will be described. Classifiers such as “˜Volume”, “No.”, “˜Page”, “˜Page”, etc. are frequently used in “References”. Classifiers such as “Volume” and “No.” represent various numbers depending on the context, and their meaning is ambiguous alone. For example, the classifier of “No.” is used to count things that are regularly issued, such as magazines, and those that are in order. Other than magazines, as part of the address, such as “X town 1-chome, 2-3 No. 3”, and as the “Article 1, Paragraph 2, Item 3,” indicating the location of the law, treaty or contract document Also used.

  Therefore, a rule that refers to the context of the “reference document” is set in advance in the item of the conversion grammar dictionary 21d. FIG. 6 is an explanatory diagram of an example of a conversion rule of the conversion grammar dictionary 21. In FIG. 6, the conversion rule regarding the headword of "No." is shown. The conversion rule consists of the headword “No.”, the part of speech “classifier”, the rule application condition, and the “conversion rule”. The third case of the rule application condition is “the element of the reference”, As the “conversion rule”, “translate word No. or no. And change word order” is defined. By having such a conversion grammar dictionary 21d, the translation of the corresponding part of the sentence having the description part of the reference document can be converted to “No. 5” in the correct word order in which “No. 5” is a desirable word.

  Similarly, with respect to “pages 547 to 552”, as in the conversion rule shown in FIG. 6, the rule application condition that refers to the context of “reference document” in the heading word of “page” in the conversion grammar dictionary 21 d By setting the conversion rule, the translation of “page” is converted not to “page” but to the abbreviations “p.” And “pp.” Used in “references”, followed by a number. In this case, “pp.” Is selected because it includes a hyphen and covers a plurality of pages. The translation can also be determined in the same manner under the condition that “volume” is also an element of “reference”.

  When the translation of the “reference document” portion is completed in this way, the subsequent portion in which the combination with the translation of the portion other than the “reference document” portion in the same document previously sent to the translation unit 26 is stored is stored. Use information.

In the above description, the typical format knowledge stored in the reference format knowledge base 24 in advance is used for the category determination in step S9 in FIG. 3, but the format can also be defined by the user. This shows some typical formats and allows the user to select a format from among them, or to actually define and define a desired format. For example, a book has the following format.
・ Author Name “Book Name” Publication place, publisher, publication year, author name. Book title. Publication place, publisher, publication year / author name Publication year Book title Publication place: Publisher.

There are also the following formats for papers.
・ Author name “Thesis name”, “Journal name” Volume (issue year), Page ・ Author name Publication year “Paper name” Publisher name “Journal name” Volume: Page ・ Author name: Paper name, Journal name, Vol., No., page, year of publication.

  These formats are specifically shown and displayed on the screen to be selected. If there are no candidates in the options, the user definition is used. In addition, the reference data storage unit 23 may be updated for an element whose category is newly found during translation processing or an element whose interpretation of the category is changed.

  In the above explanation, the case where Japanese is the original language and English is the target language has been described. However, if Japanese is the source language and Chinese is the target language, the Romaji conversion can be read as Pinyin conversion. Similar techniques can be used.

  In the above description, the element of the description part of the reference document is translated into the second language while maintaining the presentation order of the bibliographic information in the first language. It is also possible to translate into For example, in English, there are APA (American Psychological Association) format, CBE (Council of Biology Editors) format, and MLA (Modern Language Association) format as typical formats for describing references in English. For example, regarding the order of presentation for the MLA and APA methods, in the case of books in MLA: 1. author name, 2. book title, 3. place of publication, 4. publisher, 5. order of publication year, papers 1. Author name, 2. Paper title, 3. Journal name, 4. Number of issues, 5. Publication year, 6. Page order.

  On the other hand, in the case of books in the APA format, 1. author name, 2. publication year, 3. book title, 4. place of publication, 5. order of publishing company, and for papers, 1. author name, 2. publication year, 3. Take the order of the article title, 4. journal name, 5. issue number, 6. page. In this way, it can be seen that there is a big difference in the handling of the publication year / issue year whether it is placed near the beginning or near the end.

FIG. 7 is a flowchart of processing contents for outputting the description part of the reference document in the order of presentation of the format requested in the second language. The translation unit 26 performs the translation processing of steps S1 to S9 in FIG. 3 as a premise. For this reason, in principle, each element of the “reference document” is given a category. Therefore, a set of three elements, category, and translation is prepared (S21). Table 1 shows combinations of elements, categories, and translation words in this example.

  Next, the control unit 25 displays and outputs a format selection screen as shown in FIG. 8 on the display device 17 among the dialogue screens stored in advance in the storage device 19 (S22). In FIG. 8, the user can select a format based on a format name, a typical academic field in which the format is used, and a legend. In addition, here, there are the above-mentioned three formats and an option of “others”, and there is a check column for a case where translation of a book title is desired. The user determines whether there is a desired format among these candidates (S23). If there is a desired format (S23: YES), the user selects the desired format (S24). When a desired format is selected by the user, rearrangement conversion is performed using the three sets of step S1 (S25).

  Now, if the APA method is selected and the translation column of the book title is checked, the author's location is Kokuryo, J., Nonaka, I. & Kataoka, M.

  As described above, the format when there are a plurality of authors is described in the reference document conversion rule 21i, and the user does not need to be aware of this. That is, for example, in the above-mentioned APA method, even if there are a plurality of authors, there is a rule that all names are given in the order of last name and first name, and the last two names are connected by & (ampersand). When rearrangement and conversion are completed, the following results are obtained.

Kokuryo, J., Nonaka, I. & Kataoka, M. Nettowaku Shakai no Chishiki Keiei (Knowledge Management of Network Society). Publisher: NTT Shuppan, 2003.
Here, the field of publication is in Japanese characters. This is because the information is not in the original text of FIG. In such a case, the control unit 25 prompts the user to provide information, or checks and translates the location of the publication place by a general search system (such as a www search or a document DB search). Thus, for example, in the case of “Tokyo”, it is translated as “Tokyo”. The processing is completed by substituting this character string into the place of “publication”.

  If it is determined in step S23 that the format desired by the user is “others” (S23: NO), the user uses keywords such as author name, book title, place of publication, publisher, year of publication, and various symbols. The format is defined (S26). In this case, the outline is similar to the APA, CBE, and MLA methods, and if the details are different, the legend of the similar format is copied and the desired format is easily defined by changing the different parts. be able to. When the user definition is finished, as in the case where the previous APA method is selected, conversion into a desired format is performed by rearrangement or the like (S25), and the process is finished.

  Note that the format defined by the user may be stored in the reference format knowledge database 24 so as to save time for redefinition. The triplet of elements, categories, and translations may be stored in the reference document data storage unit 23. By doing so, translation time can be shortened when switching to various formats for output.

  According to the embodiment of the present invention, the separator in the “reference document” is detected, and the elements such as the author name, the book title, the publisher name, and the publication year / issue year are translated appropriately. In constructing the translation, it is converted into the format specified by the second language and output, so that the description part of the reference can be translated with high accuracy and stability.

  In addition, it detects the separator in the “reference document” to be translated, and based on the usage of the separator and the result of dictionary lookup of the character string, any element such as author name, book name, publisher name, etc. Since the translation method and output format of each element are adjusted based on the determination result, the description part of the reference can be translated with high accuracy and stability.

The block diagram of the machine translation apparatus concerning embodiment of this invention. Explanatory drawing of an example of the reference literature description part of the original text for translation in embodiment of this invention. The flowchart which shows the processing operation of the machine translation apparatus concerning embodiment of this invention. Explanatory drawing of an example of the storage format of the reference data stored in the reference data storage part in embodiment of this invention. The flowchart which shows an example of the processing content of the Romaji conversion in the translation part in embodiment of this invention. Explanatory drawing of an example of the conversion rule of the conversion grammar dictionary of the translation dictionary part in embodiment of this invention. The flowchart of the processing content which outputs the description part of a reference document in the presentation order of the format requested | required in the 2nd language in embodiment of this invention. Explanatory drawing which shows an example of the selection screen which designates the format of the reference document in embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 ... Machine translation apparatus, 12 ... CPU, 13 ... ROM, 14 ... RAM, 15 ... Bus, 16 ... Input device, 17 ... Display device, 19 ... Memory | storage device, 20 ... Machine translation program, 21 ... Translation dictionary part, 22 Reference document identification database 23 Reference document data storage unit 24 Reference document format knowledge base 25 Control unit 26 Translation unit 27 Reference detection unit 28 Reference analysis unit 29 Category determination Part, 30 ... Dialog screen storage part

Claims (5)

Bibliography that stores machine translation program, translation dictionary that stores knowledge and rules necessary for translation, knowledge data describing how parts of speech and parentheses and punctuation marks are used to identify the part of the reference Reference format knowledge base that stores knowledge data to determine which category the elements that make up the description part of the reference belong to by referring to the identification database, the symbols and positional relationships before and after the description part of the reference Storage device, input device for inputting the original text in the first language to be translated and input necessary information for operation, and display the original text in the first language to be translated and the translated text in the second language after translation a display device for, in a machine translation system that includes a CPU for execution of the machine translation programs, Oh in the description part of references stored in said reference identification database It used the knowledge data that describes how used a part of speech and parentheses, punctuation for identifying a reference detector for detecting the descriptive portion of the references from the original data inputted from the input device, the A bibliographic analysis unit that divides the character string of the description part of the reference detected by the reference detection unit and a separator that divides the character string; and a description part of the reference stored in the reference format knowledge base. Using the knowledge data for determining which category the element constituting the description part of the reference belongs by referring to the preceding and following symbols and positional relationships, the character string obtained by the reference analysis unit The category determination unit for determining the category of the elements constituting the description part of the reference document and assigning the category to the character string, and the description part of the reference document of the original text to be translated Machine translation apparatus characterized by comprising a translation unit for translating using said translation dictionary unit strings description part references based on the category granted by re determination unit. When the translation part converts the character string of the description part of the reference into Roman characters, it morphologically analyzes the character string and separates it using the division rule stored in the translation dictionary part. The morpheme character string is converted into a Roman character using the Romaji conversion table stored in the translation dictionary unit with the corresponding Romaji spelling having the corresponding reading, and the capital letters of the first letters of consecutive independent words separated by spaces The machine translation apparatus according to claim 1, wherein the machine translation processing is performed. The control unit displays a selection screen of a translation document formula of a reference document stored in advance in the storage device on the display device and notifies the translation unit of the translation document formula selected from the input device. 2. The machine translation apparatus according to claim 1, wherein a character string of a description part of a reference document is translated in accordance with a translated document expression notified from the control unit, and the control unit displays a translation result on the display device. 4. The machine translation apparatus according to claim 3, wherein the control unit adds a translation document expression of the reference defined by a user via the input device to the reference format knowledge base . Bibliography that stores machine translation program, translation dictionary that stores knowledge and rules necessary for translation, knowledge data describing how parts of speech and parentheses and punctuation marks are used to identify the part of the reference Reference format knowledge base that stores knowledge data to determine which category the elements that make up the description part of the reference belong to by referring to the identification database, the symbols and positional relationships before and after the description part of the reference Storage device, input device for inputting the original text in the first language to be translated and input necessary information for operation, and display the original text in the first language to be translated and the translated text in the second language after translation a display device for, in a computer to function as a machine translation system that includes a CPU for execution of the machine translation programs, the computer, the reference Using knowledge data that describes how parts of speech and parentheses and punctuation marks are used to identify the description part of a reference document stored in another database, the reference document is based on the original text data input from the input device. A function for detecting the description part of the reference, a function for dividing the character string of the description part of the detected reference and a separator for dividing the character string, and the reference format for the character string obtained by the division Use the knowledge data to determine which category the elements that make up the reference part of the bibliography belong to by referring to the symbols and positional relationships before and after the reference part of the reference stored in the knowledge base. The function of determining the category of the elements constituting the description part of the document and assigning the category to the character string, and the description part of the reference document of the original text to be translated to the assigned category Zui by machine translation program for causing a string description part references and a function to translate using the translation dictionary unit.

Images