JP4054353B2 - Machine translation apparatus and machine translation program - Google Patents

Description

  The present invention relates to a machine translation apparatus and a machine translation program that automatically translate a sentence in a first language (source language) into a sentence in a second language (target language) using a computer.

  Machine translation devices and software that automatically translate sentences written in a first natural language into sentences in a second natural language using a computer have been put into practical use. For example, there are many products in translation software for personal computers, and there are examples in which machine translation services are provided on the Internet.

  For example, consider a case where Japanese as a first natural language is translated into English as a second natural language. First, the original text before translation is divided into several words by morphological analysis. Then, for each divided word, a translation dictionary that is a database between original words and translated words prepared in advance is searched, and an appropriate translated word is selected to translate Japanese into English.

  Here, if the word being searched for is not found in the heading of the translation dictionary between the original word and the translated word, the translated word cannot be obtained, and the word before translation is treated as an unknown word. Then, in the English text output as a translation, some words that have become unknown words may appear as original words, that is, in Japanese.

  In order to avoid this, there is a method in which, for example, a word in katakana notation is converted into a romaji notation as it is as a so-called foreign word and output. In this case, the converted Roman letter notation obtained in this way is often very different from the original spelling of English words. Therefore, it may be impossible to infer the original English word from the converted Roman alphabet.

  Therefore, in order to process an unknown word of Katakana that was not found in the translation dictionary, there is one that divides each Katakana character into one character, considers them as one word, and analyzes it as a chain of one character words (for example, Patent Document 1).

Also, instead of just a single word, it is divided into several katakana character strings consisting of a plurality of characters, and other notations with the same reading (for example, hiragana characters) In such a case, there is a morpheme analysis method in which a word registered in the translation dictionary is used instead of an unknown word by searching for whether or not (notation based on kanji or kanji) is in the translation dictionary (see, for example, Patent Document 2).
JP 62-208169 A JP-A-5-20304

  However, in the prior art, whether or not the translation is successful depends on whether or not the original word is finally registered in the translation dictionary even if the original word is converted into a one-letter word or another notation. That is, if they are not registered in the translation dictionary, they cannot still be properly translated.

  Even if the original words in katakana notation are converted into romaji notation as so-called foreign words, words that are not originally foreign words are extremely inappropriate.

  Accordingly, an object of the present invention is to provide a machine translation device and a machine translation program that can output a translation with high accuracy even if it is an unknown word that is not registered in a translation dictionary.

The machine translation device according to the present invention includes a translation dictionary unit that stores a pair of a first language word and a second language word corresponding thereto, and a morphological analysis of the first language source sentence into several words. An input processing unit to be decomposed, a translation dictionary search unit that searches the translation dictionary unit for words / phrases decomposed by the input processing unit and selects words in the second language as translated words, and one or more characters in the words in the first language And a spelling correspondence table for storing one or more spellings of spelling of characters in the second language corresponding thereto, and spelling of the phrases in the second language, meanings of words, parts of speech, fields, co-occurrence information When the knowledge database of the second language that stores attribute information such as and the phrase decomposed by the input processing unit is an unknown word that does not exist in the translation dictionary unit, the unknown word is further composed of one or more characters To decompose It was lexical retrieved from the spelling correspondence table extracting lexical second language, when synthesized lexical second language extracted this finding the translation of the unknown word, the knowledge database of the second language Referring to and determining the priority of the translation of the unknown word based on the attribute information, the unknown word processing unit for obtaining the translation of the unknown word, and receiving and assembling the translations output from the translation dictionary unit search unit and the unknown word processing unit And an output processing unit for outputting the translated sentence.

  According to the present invention, it is possible to accurately output a translated word for an unknown word that is not registered in the translation dictionary.

  The best mode for carrying out the present invention will be described. Of the words in the original text of the first language, the words that are not registered in the translation dictionary, that is, unknown words, are further divided into words, and the spelling is obtained with reference to the second language spelling database corresponding to the words, A combination of these spellings yields a second language phrase. Thereby, even an original sentence including an unknown word can be translated with high accuracy.

  Hereinafter, in the embodiment of the present invention, a case where the machine translation device of the present invention is translated into English as the second language when the first language is Japanese will be described. FIG. 1 is a block diagram of a machine translation apparatus according to an embodiment of the present invention.

  The machine translation apparatus 101 includes an input processing unit 102, a translation dictionary search unit 103, a translation dictionary unit 104, an unknown word processing unit 105, a spelling correspondence table 106, an output unit processing unit 107, and a vocabulary in a second language. A database 302 and a knowledge database 303 of a second language. The second language knowledge database 303 may be provided in the translation dictionary unit 104.

  The machine translation apparatus 101 receives a translation target document described in a first language (source language), that is, an original text as electronic data, performs translation, and outputs a translation result document described in a second language, that is, a translated text. To do. The input processing unit 102 captures the original text. Then, the phrases constituting the original text are extracted by morphological analysis or the like and output to the translation dictionary search unit 103.

  The translation dictionary unit search unit 103 searches the translation dictionary unit 104 to determine whether or not the word / phrase is an entry word. The translation dictionary unit 104 is a dictionary for performing translation from the first language into the second language, and stores words and phrases of the first language, that is, original words, and phrases and translations of the second language corresponding to the words. For the original language registered as a headword in the translation dictionary unit 104, the translated word is sent to the output processing unit 107.

  The unknown word processing unit 105 treats an original word (for example, a foreign word or a new word) that is not registered as a headword in the translation dictionary unit 104 as an unknown word, and converts this into an appropriate translated word. The spelling correspondence table 106 pre-registers spelling candidates in the second language corresponding to the original language words in the first language, and the vocabulary database 302 in the second language lists the words in the second language that are candidates for unknown words. The knowledge database 303 of the second language stores attribute information such as the meaning of words, parts of speech, fields, and co-occurrence information in addition to the spelling of words in the second language. These are referred to when the unknown word processing unit 105 processes an unknown word.

  The output processing unit 107 performs processing to receive the translations output from the translation dictionary search unit 103 or the unknown word processing unit 105, assemble them, and output them as translations.

  FIG. 2 is a hardware configuration diagram of the machine translation apparatus according to the embodiment of the present invention. The machine translation apparatus 101 according to the embodiment of the present invention is configured by a computer. That is, a central processing control device 201, a ROM (Read Only Memory) 202, and a RAM (Random Access Memory) 203 are connected to the bus 210, while an input device 204, a display device 205, a communication control device 206, a storage device 207, and a removable device. The disk 208 is connected to an input / output interface 209, and the input / output interface 209 is connected to a bus 210.

  The central processing control device 201 reads and executes a boot program from the ROM 202 based on an input signal from the input device 204, and further reads an operating system stored in the storage device 207. Furthermore, the central processing control device 201 controls various devices based on input signals from the input device 204, the communication control device 206, etc., reads out programs and data stored in the RAM 203, the storage device 207, etc., and loads them into the RAM 203. In addition, the processing device implements a series of processes to be described later, such as data calculation or processing, based on a program command read from the RAM 203.

  The input device 204 is configured by an input device such as a keyboard and a mouse through which an operator inputs various operations. The input device 204 creates an input signal based on the operation of the operator, and inputs via the input / output interface 209 and the bus 210. It is transmitted to the central processing control apparatus 201. The display device 205 is a CRT (Cathode Ray Tube) display, a liquid crystal display, or the like, and receives an output signal to be displayed on the display device 205 from the central processing control device 201 via the bus 210 and the input / output interface 209. This is a device that displays the processing result of the control device 201. The communication control device 206 is a device such as a LAN card or a modem, and is a device that connects the machine translation device 101 to a communication network such as the Internet or a LAN. Data transmitted / received to / from the communication network via the communication control device 206 is transmitted / received to / from the central processing control device 201 via the input / output interface and bus 210 as an input signal or an output signal.

  The storage device 207 is a magnetic disk device, and stores programs and data executed by the central processing control device 201. The removable disk 208 is an optical disk or a flexible disk, and signals read / written by the disk drive are transmitted / received to / from the central processing control apparatus 201 via the input / output interface 209 and the bus 210.

  When such a computer functions as the machine translation apparatus 101 according to the embodiment of the present invention, the machine translation program is stored in the storage device 207 and the translation dictionary unit 104 is stored. In addition, the machine translation program is read and executed by the central processing controller 201 of the machine translation apparatus 101, whereby the input processing unit 102, the translation dictionary unit search unit 103, the unknown word processing unit 105, and the output processing unit 107 are realized. The The spelling correspondence table 106, the second language vocabulary database 302, and the second language knowledge database 303 of the machine translation device 101 are stored in the RAM 203. Note that the second language knowledge database 303 is stored in the storage device 207 when provided in the translation dictionary unit 104. The input processing unit 102 of the machine translation apparatus 101 can be implemented by the input device 204 or the communication control device 206, or data can be input from the storage device 207 or the removable disk 208. . The output processing unit 107 of the machine translation device 101 can be implemented by the display device 205 or the communication control device 206, or data can be output to the storage device 207 or the removable disk 208. .

  FIG. 3 is a block diagram of the unknown word conversion unit 105 in the machine translation apparatus 101. The unknown word conversion unit 105 includes a lexical division unit 105 a, a translation word synthesis unit 105 b, and a spelling correspondence table search unit 301, and refers to the spelling correspondence table 106. Further, the second language vocabulary database 302 and the second language knowledge database 303 are referred to as necessary.

  When the unknown word conversion unit 105 receives the aforementioned unknown word in the first language, the unknown word is divided into several lexical words by the lexical dividing unit 105a. For each divided word, the spelling correspondence table search unit 301 searches the spelling correspondence table 106. If the word is found, the spelling registered in the spelling correspondence table 106 is converted. The translation composition unit 105b refers to the vocabulary database 302 of the second language or the knowledge database 303 of the second language as necessary, and synthesizes the spelling of each lexical phrase, thereby outputting a translation for the original source word. .

  Hereinafter, each component of the unknown word conversion unit 105 will be described in detail. The lexical division unit 105a divides the unknown word into several lexical terms. First, the number of characters in each divided lexical phrase is made as large as possible. For this purpose, first, it is divided into two. For example, when “ubiquitous” is an unknown word, it is divided into a lexical phrase “ubi” and a lexical phrase “kitas”. If an appropriate translation of each lexical phrase is not found in the processing of the spelling correspondence table search unit 301 thereafter, the division location is shifted back and forth or the division location is increased. For example, the division part is shifted like “ubiquity” and “tas”. If the appropriate translation for each axis is still not found by the processing of the spelling correspondence table search unit 301, the number of divisions is increased. For example, it is divided into three phrases such as “Yu”, “Biki”, and “TASS”, or divided into three phrases such as “YUBI”, “KI”, and “TASS”.

  By doing this, the unknown word is efficiently divided because it increases the possibility of finding an appropriate translated word at an early stage by roughly dividing the lexical phrase without dividing it into a single character unit from the beginning. Can be translated.

  The spelling correspondence table search unit 301 searches the spelling correspondence table 106 for each word passed from the word division unit 105a. Here, even if the lexical phrase is not found in the spelling correspondence table 106 even by various division modes in the lexical segmentation unit 105a, the lexical phrase is converted into a Roman character notation as a default operation. For example, “ubi” is converted to “yubi”.

  In the spelling correspondence table 106, candidates for spelling of the second original word corresponding to the characters or character strings of the first original word are registered in advance. In particular, if the first source language is Japanese and the second source language is English, the Japanese phonetic characters will correspond to Roman characters, as well as the Japanese phonetic characters will simply correspond to Roman characters. In addition, the handling of foreign words that Japanese tend to do is registered in this spelling correspondence table 106. For example, mouth is often a mouse, so "su" is "th", and meter is often a meter, so "tor" is "ter".

  The spelling correspondence table search unit 301 extracts the spelling of each lexical phrase of the language divided from the spelling correspondence table 106. When the translated word synthesizing unit 105b receives the spelling of each lexical phrase of the language divided from the spelling correspondence table searching unit 301, it synthesizes it and outputs it as a translated word. Here, since a plurality of spelling candidates may be found in the spelling correspondence table 106, there may be a plurality of candidates for a translated word synthesized by combining them. Therefore, the most appropriate translation can be selected by determining the priority for each of a plurality of translations in consideration of the priority of each spelling candidate. Specifically, any one of the following processes (1) to (4) is performed.

(1) Eliminate words that are impossible or rarely appear in the second language as synthesized translations. Alternatively, the priority as a translated word is lowered.

(2) Lower the priority as a translated word for romaji writing.

(3) Search the vocabulary database 302 of the second language, and increase the priority as a translated word for those found in this.

  (4) Referring to the knowledge database 303 of the second language, the tendency of the words in the field is checked from the information on the field related to the original language, and this is used for setting the priority in the synthesis of the translated words. For example, many substance names and plant names used in chemical formulas have characteristic spellings, so if the original language is the name of a substance or plant used in a chemical formula, Increase the priority of certain spellings.

The vocabulary database 302 in the second language is a collection of vocabularies that are candidates for unknown words in the second language. Words that are candidates for unknown words are new words, coined words, foreign words, abbreviations, trade names, and the like. For example, words are collected from general magazines and newspaper articles in the second language. Therefore, words that are not registered or words in the utilization form may be included as headwords in the translation dictionary unit 104 of the machine translation apparatus 101. In this vocabulary database 302, it is not necessary to know the meaning and usage of a word, and it is only necessary that the translation synthesizer 105b can refer to whether or not the spelled word exists. Thereby, the accuracy of the translated word output by the translated word synthesizing unit 105b can be effectively improved.

  Next, the second language knowledge database 303 stores not only spelling of words in the second language but also a huge amount of various types of information such as word meaning, part of speech, field, and co-occurrence information. Is. This is generally provided in a machine translation apparatus. As described in the above process (4), the translated word synthesizing unit 105b refers to this, so that the priority as a translated word can be set for each of the synthesized translated words, and the translated word synthesizing unit 105b The accuracy of the translated word to be output can be improved effectively.

  FIG. 4 is an explanatory diagram of the spelling correspondence table 106. In the spelling correspondence table 106, the first language word 401 is used as a headword, and the spelling 402 in the second language corresponding thereto is registered in advance. For example, as described above, the English word “meter” is handled as a foreign word “meter” in Japanese, and the English word “ter” corresponds to the Japanese word “tor”. For example, the spelling of “th” in English often corresponds to the phrase “su” in Japanese.

  By registering such correspondence in advance in the spelling correspondence table 106, it is possible not only to convert an unknown word into a simple Roman character notation as in the past, but also to translate it into a more realistic and appropriate translation.

  FIG. 5 is a flowchart showing the processing contents of the unknown word conversion unit 105. The processing is started by receiving an original word treated as an unknown word from the translation dictionary search unit 103 (S501). The original word is divided into several words by the word dividing unit 105a (S502).

  The spelling correspondence table search unit 301 searches the spelling correspondence table 106 for each divided word / phrase (S503). It is determined whether or not the lexical phrase exists in the spelling correspondence table 106 (S504). If the lexical phrase exists in the spelling correspondence table 106, the lexical phrase is converted by the spelling registered in the spelling correspondence table (S505). ). On the other hand, if the lexical phrase does not exist in the spelling correspondence table 106, it is converted into a Roman alphabet (S506).

  Then, it is determined whether or not the search of the spelling correspondence table 106 has been completed for all the divided characters (S507). If the search of the spelling correspondence table 106 has not been completed for all of the divided characters, the process returns to step S503. When the search of the spelling correspondence table 106 is completed for all the divided characters, it is determined whether or not the Roman character notation is included (S508). If the Roman character notation is included, all the dividing methods for the unknown word are determined. It is determined whether or not it has been examined (S509), and if there is another division method, the process returns to step S502, and the unknown word is subdivided by the other division method. When all the division methods are examined in the determination in step S509, or when the Roman character notation is not included in the determination in step S508, the translated word composition unit 105b determines the priority of the spelling combination (S510). ), The process is terminated (S511).

  According to the embodiment of the present invention, if an unknown word is found during translation, even if it is an unknown word as a whole, if a spelling is obtained for each decomposed lexical word, a translated word is derived by connecting them together be able to. At that time, it is possible to derive a translation with higher accuracy by taking in not only the spelling conversion of lexical words but also the conversion that Japanese people tend to perform when deciding Japanese readings of foreign words. .

  By using the knowledge database of the second language, it is possible to determine priorities between translated words based on a huge amount of information such as meaning, part of speech, field, and co-occurrence as well as spelling. Similarly, the priority between translated words can be determined using information stored in a vocabulary database of a second language in which unknown word candidates are stored in advance.

It is a block diagram of a machine translation device concerning an embodiment of the invention. It is a hardware block diagram of the machine translation apparatus concerning embodiment of this invention. It is a block diagram of the unknown word conversion part of the machine translation apparatus concerning embodiment of this invention. It is explanatory drawing of the spelling | correspondence table | surface of the machine translation apparatus concerning embodiment of this invention. It is a flowchart of the processing content of the unknown word conversion part of the machine translation apparatus concerning embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 ... Machine translation apparatus, 102 ... Input processing part, 103 ... Translation dictionary search part, 104 ... Translation dictionary part, 105 ... Unknown word processing part, 105a ... Lexical division part, 105b ... Translation word synthesis part, 106 ... Spelling correspondence table, 107: Output processing unit

Claims (3)

A translation dictionary unit that stores a pair of a first language word and a corresponding second language word, an input processing unit that morphologically analyzes the original text of the first language and decomposes it into several words, and the input A translation dictionary search unit that searches the translation dictionary unit for a phrase decomposed by the processing unit and selects a second language phrase as a translation word; a phrase composed of one or more characters in the first language phrase; and a corresponding phrase A spelling correspondence table that stores one or more words spelled in bilingual characters in association with each other and a second language that stores attribute information such as spelling of words in the second language, meaning of words, parts of speech, fields, co-occurrence information, etc. When the language knowledge database and the word / phrase decomposed by the input processing unit are unknown words that do not exist in the translation dictionary unit, the unknown word is further decomposed into a word / phrase composed of one or more characters and the decomposed word / phrase is From spelling correspondence table Search and extract the lexical second language, said when synthesized lexical second language extracted this finding the translation of the unknown word, based on the reference to the attribute information knowledge database of the second language An unknown word processing unit that determines the priority of the translation of the unknown word and obtains the translation of the unknown word, and an output processing unit that receives and assembles the translation output from the translation dictionary unit search unit and the unknown word processing unit and outputs the translation A machine translation apparatus comprising: A vocabulary database of the second language for previously storing the word of the second language as a candidate for the unknown word, the unknown word processor, when determining the translation of an unknown word by combining the lexical second language 2. The machine translation apparatus according to claim 1, wherein when there is a match with reference to the vocabulary database of the second language, the priority of the unknown word as a translated word is increased. The computer is decomposed from a translation dictionary unit that stores a pair of a phrase of the first language and a phrase of the second language corresponding to a means for decomposing the original sentence of the first language into a plurality of phrases. Means for searching for the first language word and selecting the second language word as a translation word, and when the decomposed first language word is an unknown word that does not exist in the translation dictionary, Spelling means that associates and memorizes one or more lexical words spelled with one or more characters in a first language word and a second language character corresponding to the means for decomposing the lexical word consisting of the above characters It means for extracting the lexical second language by searching the token mentioned above degraded from the correspondence table, extracted by synthesizing the lexical second language when obtaining the translation of the unknown word, spelled words in a second language , Word meaning, part of speech, Field, means for determining a translation of the unknown word in the referenced based on the attribute information knowledge database of the second language attribute information accumulated decided priorities of the unknown word translation of such co-occurrence information, the decomposed phrases A machine translation program for functioning as a means for receiving and assembling the translated words and the translated words of the unknown word and outputting them as translated sentences

Images