JP2007052700A - Machine translation apparatus and machine translation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce the generation of unknown words or non-sentences in translating an original into another language and to improve translation accuracy and operation efficiency by correcting words including lost characters or misinputted characters when an input original to be translated is acquired from an original on the basis of the characteristics of an acquiring source of the input original text. <P>SOLUTION: A translation original estimation apparatus is provided with: a morpheme analysis part 14 for determining whether the head character of a head word out of a plurality of words obtained by dividing character strings acquired from a web page in each word unit is a capital letter or a small letter, and when the head character is a capital letter, generating a plurality of head word substituting words by adding a capital letter of alphabet to the head of a head word; a syntax analysis part 15 for performing the syntax analysis of one or more sentences in each of which a head word substitution candidate existing in a dictionary part 10 which is selected on the basis of dictionary data in the dictionary part 10 is substituted for the head word of an original on the basis of the dictionary data of the dictionary part 10 about the plurality of generated head word substituting words; and a syntax analysis control part 16 for selecting a grammatically correct original candidate from the candidates and outputting the selected candidate to a translation part 18. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a machine translation device and a machine translation program that convert a sentence in a first language such as English into a sentence in a second language such as Japanese.

  For example, text data (characters) obtained by scanning a novel or newspaper article written in a language such as English with a scanner, inputting the image data to a computer, and performing character recognition processing with character recognition software installed on the computer There is a machine translation system in which a translation software further translates a character string consisting of a code) into a sentence in another language such as Japanese.

  On the cover of a document such as a novel or newspaper article, the first character of the sentence is different from the following characters, and it may be written in large characters extending over several lines, or decorative characters (image data) may be pasted. Many.

  The contents of such a document can be read as a single sentence by humans, but in the case where a sentence is read by character recognition software such as a computer, the size of the first character is the size of the sentence in the ground. Therefore, it may not be possible to determine the connection with the local sentence, or it may be analyzed as a sentence including unread words with only the decoration characters excluded, and the correct character string as the original sentence may be passed to the translation software. You may not be able to obtain the desired translation results.

In addition, a machine translation system is often provided with a function of translating an English Web page displayed on a computer screen into a Japanese page, and the first character (first character) of the English Web page is text. There are cases in which the characters are created by other characters, such as images (image data), or the first character is written in a font that is significantly different from other characters.
Also in this case, as in the case of reading a newspaper article with the scanner, the character recognition software determines that the first character of the first character string is one word in which the first character of the first word and the subsequent characters are continuous. There are cases in which words that cannot be interpreted and output as words that do not exist in the translation dictionary, that is, unknown words or unread words, fail in syntax analysis, or output incorrect translation results.

Hereinafter, with reference to FIG. 20, it demonstrates with four specific cases (the following (1)-(4) cases).
Case (1) is a case where the first letter “W” of the original text is a decorative character. The local sentence in this sentence will be “orld is too big.”, And the character recognition software will output the sentence (character string) “orld is too big.” To the translation software. When translation software translates into Japanese for (string), "orld" is extracted as an unknown word, and the translation result is "orld is too big."Can't translate correctly.

  Case (2) is a case where the first letter “T” in the original text is a decorative character. The local text in this sentence is "his dish is very delicious." If the translation is done as it is, the translation result will be “His dish is very delicious.” Although it is correctly analyzed, it is different from “This dish is very delicious.” It is a pattern. In such a case, since the meaning is understood for the time being, there is a problem that it is not known that an incorrect translation result is obtained unless the user refers to the original text.

  The case (3) is a case where the first letter “T” in the original text is a decorative character. The local sentence in this sentence is "he problems of the last few days have been fixed." If the translation is performed as it is, the translation result will be “He, the problem of the past several years has been resolved.” The syntax analysis failed and the translation cannot be performed accurately.

  Case (4) is a case where the first character “T” in the original text is an extra large font. In this case, the translation software cannot determine whether "T" is the first letter of "he problems of the last" or "few days have been fixed." Even if is executed, accurate translation results are not obtained.

  As in the cases (1) to (4) above, if the translation software translates the character recognition result of the image of the scanner or the text character string obtained by copying the web page as it is, the user originally expects In some cases, the translation result cannot be output.

  As a prior art of this type of technology, for example, in a machine translation device that detects a word (unknown word) that does not exist in a dictionary, an unknown word is detected during machine translation, and the display format of the detection result (unknown word list) is A technique that can improve work efficiency and effectiveness of processing by allowing the user to arbitrarily specify the information has been disclosed (for example, see Patent Document 1).

On the other hand, there is a method in which a person reads a content of an article or a Web page and inputs a key directly from a keyboard without relying on an automatic character recognition process such as a scanner for a document such as a novel or a newspaper article or a Web page sentence.
In this case, because people read and input the contents of articles and web pages, you can correctly enter and input extremely large initial characters and decorative characters. However, because people enter text, input such as incorrect key operations The possibility of mistakes increases.
JP-A-7-31878

  Thus, in the case of the above prior art, although the user can arbitrarily specify the display format of the unknown word list, unknown words themselves are not reduced because unknown words are detected during translation. Since character strings including unknown words are to be translated, there is a problem that further improvement in translation accuracy cannot be expected.

  In addition, if there is an input mistake when a person reads the original text and inputs a key, unknown words will increase in the character string to be translated, resulting in a backward work for correcting the character, resulting in reduced work efficiency. There is a problem.

  The present invention has been made to solve such a problem, and a word is detected as an unknown word or non-sentence when machine translation is performed on a character string obtained by inputting an original sentence with a key input or a Web page or a scanner. An object of the present invention is to provide a machine translation apparatus and a machine translation program capable of reducing the above-described problem and improving translation accuracy and work efficiency.

  In order to achieve the above object, the machine translation apparatus of the present invention provides a second language based on dictionary data registered in advance for translating an original sentence created in a first language into a sentence in another language. In the machine translation apparatus provided with the translation unit for translating the sentence into the text, the data receiving means for receiving the character string created in the first language, and the character string received by the data receiving means based on the dictionary data Dividing means for dividing into word units, unknown word extracting means for extracting words that are not registered in the dictionary data among the words divided by the dividing means as unknown words, and extracted by the unknown word extracting means A replacement candidate character generating means for generating a replacement candidate character by adding a character to the unknown word or replacing the character according to a predetermined rule; and the replacement candidate character generating means It has and a character string output means for outputting the generated replacement text is replaced with candidate characters to replace the original unknown word as a translated string to the translation unit, characterized in.

  The machine translation apparatus according to the present invention is based on a dictionary unit that stores dictionary data including characters, words, dictionaries, translations, and parts of speech in advance in order to translate an original sentence created in a first language into a sentence in another language. In a machine translation apparatus including a translation unit that translates text into a second language, data accepting means for accepting the original text in the first language as a character string that can be processed by a computer, and characters accepted by the data accepting means Dividing means for dividing a character string into word units by performing morphological analysis based on the dictionary data stored in the dictionary section for the column; and a head of the first word among a plurality of words divided by the dividing means A word generating means for determining whether a lowercase letter is an uppercase letter and generating a plurality of new words by adding an uppercase alphabet to the beginning of the first word when the first letter is lowercase; A replacement candidate word extracting unit that extracts a word existing as the dictionary data among a plurality of words generated by the word generating unit, and a replacement candidate word extracted by the replacement candidate word extracting unit Character string output means for outputting a sentence replaced with a word to the translation unit as a character string to be translated is provided.

  The machine translation apparatus of the present invention is a machine translation system for translating an original text created in a first language into a text in a second language, a data receiving means for receiving a character string of the original text input by the key, and the data By performing morphological analysis on the character string input by the accepting unit based on dictionary data set in advance, a dividing unit that divides the character string into word units, and each word divided by the dividing unit An unknown word extraction means for extracting a word that is not registered as dictionary data as an unknown word, and a new word generated by replacing characters of the unknown word extracted by the unknown word extraction means based on a predetermined rule Replacement candidate word generation means for extracting a replacement candidate word that becomes a grammatically correct sentence when the word is replaced with the original word, and the replacement candidate word generation means The candidate words replacement which is more produced, characterized by comprising a character string output means for outputting to said translation unit replaces the word corresponding position of the character string of the original as a string to be translated.

  A machine translation program according to the present invention includes a translation unit that translates an original sentence created in a first language into a sentence in a second language based on dictionary data stored in advance to translate the sentence in another language. In the machine translation program for causing the machine translation device to execute processing, the machine translation device is configured to receive a character string created in the first language and a character string accepted by the data acceptance unit. Dividing means for dividing into word units based on dictionary data, unknown word extracting means for extracting words that are not registered in the dictionary data among the words divided by the dividing means, and the unknown words A replacement candidate character generating means for generating a replacement candidate character by adding a character or replacing a character according to a predetermined rule with respect to the unknown word extracted by the extracting means; The replacement candidate character replacement candidate characters generated by the generation means, characterized in that to function as a character string output means for outputting the sentence by replacing the replacement source unknown word to the translation unit as a translated string.

  The machine translation program of the present invention is based on a dictionary unit that stores dictionary data including characters, words, dictionaries, translations, and parts of speech in advance to translate an original sentence created in a first language into a sentence in another language. In a machine translation program that causes a machine translation device including a translation unit to translate a sentence in a second language to execute processing, the machine translation device receives the original text in the first language as a character string that can be processed by a computer. A data receiving unit; a dividing unit that divides the character string into word units by performing morphological analysis on the character string received by the data receiving unit based on dictionary data stored in the dictionary unit; and Among the plurality of words divided by the means, it is determined whether the first letter of the first word is a lowercase letter. If the first letter is a lowercase letter, an uppercase alphabet is added at the beginning of the first word. A word generation unit that generates a plurality of new words with a bet; a replacement candidate word extraction unit that extracts words existing as the dictionary data among the plurality of words generated by the word generation unit; and the replacement The replacement candidate word extracted by the candidate word extraction unit is made to function as a character string output unit that outputs a sentence in which the first word of the original sentence is replaced to the translation unit as a character string to be translated.

  The machine translation program of the present invention is a machine translation program that causes a machine translation device to execute processing on a machine translation device that translates an original sentence created in a first language into a sentence in a second language. Data receiving means for receiving the original text string, and dividing the character string into word units by performing morphological analysis on the character string input by the data receiving means based on preset dictionary data A dividing unit; an unknown word extracting unit that extracts a word that is not registered as the dictionary data among the words divided by the dividing unit; and an unknown word character extracted by the unknown word extracting unit A replacement candidate word that becomes a grammatically correct sentence when a new word generated by replacing a word with a predetermined rule is replaced with the original word A replacement candidate word generation unit, and a character string output that outputs the replacement candidate word generated by the replacement candidate word generation unit to the translation unit as a translation target character string by replacing the replacement candidate word with a word at a corresponding position in the original text string It is made to function as a means.

  In the present invention, in order to translate a sentence in a first language into a sentence in a second language by performing machine translation after correctly supplementing characters that are missing or erroneously input at the time of inputting the original text before performing machine translation, It is possible to reduce the number of unknown words and non-sentences in the sentence of the first language to be translated as much as possible.

  In other words, by compiling words and phrases damaged when inputting the original text as known words existing in the dictionary, machine translation is performed with fewer unknown words and non-sentences. It is possible to reduce the backtracking work that occurs due to an error in inputting.

  As described above, according to the present invention, it is possible to reduce the detection of a word as an unknown word or non-sentence when machine translation of a character string obtained by inputting an original sentence with a key input or a Web page or a scanner is performed. The accuracy can be improved and the working efficiency can be improved.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing a configuration of a translation original text estimation apparatus according to the first embodiment of the machine translation apparatus and machine translation program of the present invention. FIG. 2 shows an uppercase alphabet at the beginning of the first word in the translation text estimation apparatus of FIG. It is a figure which shows the example which attached | subjected the flag whether the word attached | subjected in order and produced | generated exists in a dictionary.

  As shown in FIG. 1, the translation original text estimation apparatus according to the first embodiment includes a dictionary unit 10, an input unit 11, an original text estimation unit 12, an original text estimation determination unit 13, a statistical data storage unit 17, a translation unit 18, and an output unit. 19 is provided. The original sentence estimation unit 12 refers to the input original sentence by referring to the dictionary unit 10 and performs a morpheme analysis unit 14, and refers to the dictionary unit 10 for a morpheme analysis result analyzed by the morpheme analysis unit 14. A syntax analysis unit 15 that performs syntax analysis, a syntax analysis control unit 16 that determines whether the syntax analysis result analyzed by the syntax analysis unit 15 is appropriate with reference to the dictionary unit 10, and the like.

  In other words, this translation original sentence estimation device is a dictionary unit that stores in advance dictionary data including characters, words, dictionaries, translated words, and parts of speech in order to translate an original sentence created in a first language (English) into a sentence in another language. 10 is provided with a translation unit 18 for translating text into a second language (Japanese) based on 10.

  The input unit 11 includes, for example, communication means for a LAN or the Internet, software such as a web browser, a keyboard and character input software that accepts commands input by key operations from the keyboard, and converts the character into a CCD line. This is realized by an image reading device (hereinafter referred to as a scanner) that scans a sensor or the like to form an image of a sentence described in a form, and software that recognizes characters of image data of a document obtained by the scanner.

  The input unit 11 receives a sentence or sentence in a first language to be translated, which is input from the Internet, a keyboard, or the like. A sentence in a first language shown on a form or a web page, for example, English It functions as a data receiving means for receiving an input of a text data string (hereinafter referred to as a character string) such as a sentence (hereinafter referred to as an original text).

  The input unit 11 receives input of a character string recognized from image data obtained by reading an original text with a scanner. The input unit 11 accepts input of a character string obtained by text copying a Web page. That is, the input unit 11 accepts a computer-processable character string obtained by character recognition after the original text is read, or a computer-processable character string obtained by key-inputting the original text.

  The dictionary unit 10 is realized by a memory or a hard disk device. The dictionary unit 10 stores (stores) dictionary data such as translated words, parts of speech, and semantic analysis dictionaries associated with headwords (words and characters). The parts of speech are, for example, nouns, verbs, adjectives, adverbs, and the like. The dictionary data includes a character code for each character constituting the word. The translation of the dictionary data is data for converting between other languages in one direction or in two directions, such as Japanese → English, English → Japanese. The dictionary data includes, for example, character tables of national languages such as English alphabet tables, Japanese kanji tables, hiragana tables, and katakana tables.

  The original sentence estimation unit 12 performs a morphological analysis on a sentence (text data) input from the input unit 11 or a sentence in which image data from a scanner is character-recognized, and divides the sentence into a plurality of character strings. It is determined whether or not the first character of the character string is a lower case letter, and upper case letters are added in order from A to Z at the beginning of the lower case character string, so that 26 new character strings are obtained. Is generated.

  The original sentence estimation unit 12 refers to the dictionary unit 10 for the new 26 character strings, determines whether or not the word is an unknown word, and adds a flag to each character string. If it is not an unknown word, a flag such as ◯ is added, and if it is an unknown word, a flag such as x is added.

  The original sentence estimation unit 12 replaces the character string to which the flag of ○ is added as a replacement candidate with the first character string of the input original sentence, performs a syntax analysis to determine whether the sentence passes the meaning, and passes the meaning. Is selected as the original text candidate.

  The original sentence estimation determination unit 13 uses one or more original sentence candidates selected (extracted) by the syntax analysis unit 15 of the original sentence estimation unit 12 based on statistical data set (stored) in the statistical data storage unit 17 in advance. An original sentence candidate using an appropriate word as the first word in the sentence is selected and output to the translation unit 18 as a character string to be translated.

  In other words, the original sentence estimation determination unit 13 functions as a character string output unit that outputs a sentence obtained by replacing the replacement candidate word extracted by the original sentence estimation unit 12 with the first word of the original sentence to the translation unit 18 as a character string to be translated. .

  The original sentence guessing / determining unit 13 selects first sentence candidate selection means for selecting a grammatically correct sentence candidate based on dictionary data for each sentence replaced with the first word of the original sentence when a plurality of replacement candidate words are extracted. Function as.

  The original sentence estimation determination unit 13 uses an appropriate word as the first word in the sentence based on statistical data set in advance in the statistical data storage unit 17 for one or more selected grammatically correct sentence candidates. It functions as second sentence candidate selection means (original sentence estimation determination unit) that selects a sentence candidate and outputs it to the translation unit 18 as a character string to be translated.

  The morpheme analysis unit 14 of the original text estimation unit 12 functions as a dividing unit that divides the character string into word units by performing morphological analysis on the character string received by the input unit 11 based on the data in the dictionary unit 10. . Further, the morpheme analysis unit 14 determines whether the first character of the first word among the plurality of divided words is an English character table, that is, whether the lowercase letter of the alphabet is uppercase, based on the dictionary data of the dictionary unit 10, and the first character is alphabetic In the case of lowercase letters, it functions as a word generating means for generating a new word for replacing a plurality of head words by adding a capital letter alphabet to the head of the head word.

  The morpheme analysis unit 14 selects (extracts) a first word replacement candidate that selects (extracts) a first word replacement candidate word existing in the dictionary unit 10 based on the dictionary data of the dictionary unit 10 for the plurality of generated first word replacement words. Functions as a means.

  The syntax analysis unit 15 performs syntax analysis on each sentence in which the first word replacement candidate word selected (extracted) by the morpheme analysis unit 14 is replaced with the first word of the original text based on the dictionary data of the dictionary unit 10, and It functions as an original text candidate selection means for selecting correct original text candidates.

  The syntax analysis control unit 16 performs syntactic analysis based on dictionary data for each sentence replaced with the first word of the original sentence, or grammatically correct sentence candidates or semantic sentence candidates by means of semantic analysis. Functions as a sentence analysis means to elect.

The statistical data storage unit 17 stores statistical data in which the order of words having a high appearance frequency, which is previously obtained by taking statistics for each part of speech among the words existing in the dictionary unit 10, is set. As statistical data, for example, at least one of word appearance frequency data, sentence word appearance frequency data, and sentence character appearance frequency data in a sentence is used.
That is, the statistical data storage unit 17 stores, for each part of speech, data having a high appearance frequency as a head word or data having a high appearance frequency as a head character. For example, Sato is the most suitable noun of Sato and Suzuki. Suzuki is set as second.

  The translation unit 18 uses the dictionary data including the corresponding character table of the dictionary unit 10 according to the language translation direction (such as English → Japanese) designated from the input unit 11 to translate an English sentence that is a sentence in the first language. Translate to a second language sentence, for example, a Japanese sentence.

  The dictionary unit 10, the input unit 11, the source sentence estimation unit 12 (the morpheme analysis unit 14, the syntax analysis unit 15, the syntax analysis control unit 16), the source sentence estimation determination unit 13, the statistical data storage unit 17, the translation unit 18, and the output unit 19. Are realized by a cooperative operation of hardware such as a computer keyboard, scanner, CPU, memory, hard disk device, and monitor, and software such as an operating system and a translation original text estimation program installed in the hard disk device.

Hereinafter, the operation of the translation original text estimation apparatus according to the first embodiment will be described with reference to FIGS.
2 is a diagram showing a word list created by dictionary lookup processing performed by the translation original text estimation apparatus of this embodiment, FIG. 3 is a diagram showing dictionary data in the dictionary unit 10 of FIG. 1, and FIG. 4 is this translation text estimation apparatus. FIG. 5 is a flowchart showing dictionary lookup word estimation processing, and FIG. 6 is a flowchart showing syntax analysis original text estimation processing.

  In describing this embodiment, it is assumed that the first language to be input is English and the second language to be output is Japanese. That is, in this example, the original sentence is English, and a case where the original sentence is translated into a Japanese sentence will be described. However, the present invention is not limited to only these languages.

  In the translation original text estimation apparatus according to the first embodiment, in the case of 1 described in the above prior art, “orld is too big.” Is input to the PC 1 from the input unit 11. This is referred to as input original text A.

  In PC1, when the original sentence estimation unit 12 acquires the input original sentence A input from the input unit 11 (S601), the acquired input original sentence A is subjected to morphological analysis with reference to the dictionary unit 10 and divided into words, It is determined whether the language of the input original text A is a language that can be estimated (S602).

  As a result of this determination, when the language of the input original sentence A is English and the English text data exists in the dictionary unit 10 as in this example, the original sentence inference part 12 is the original sentence of the input original sentence A. It is determined whether or not it is necessary to guess the original text (S603).

  Whether or not it is necessary to infer the original text is determined based on whether the first letter is uppercase or lowercase when the language of the input original text A is English as in this example. When the form of the sentence of the input original sentence A extends over two or more lines, the determination is made using character position information.

  For example, when the input language is English, the first character of a sentence is usually written in uppercase letters. For this reason, when the first character of the input data is an uppercase letter, it is determined that the correct original text has been input, and it is not necessary to infer the original text that is the original of the input original text A. The original sentence estimation unit 12 determines that the original sentence needs to be estimated when the first character is a lowercase letter.

  In addition, when the input source sentence A has two or more lines and character position information is used, assuming that the font size of the first character on the first line is twice that of the other fonts, one line is assumed. The position of the second character of the eye and the position of the second character of the second line are greatly different. For this reason, when the position of the second character on the second line and the position of the second character on the first line are greatly different, the original sentence estimation unit 12 determines that it is necessary to estimate the original sentence.

  Accordingly, the original sentence estimation determination unit 13 determines whether or not to estimate the original sentence. 1. In the example, since the first character is a lowercase letter, it is necessary to perform original text estimation, and first, a process of estimating a word (dictionary search word estimation process) in the dictionary reference original text estimation process is performed (S604).

  In the dictionary lookup word estimation process, as shown in FIG. 5, the original sentence estimation determination unit 13 first acquires the character string “orld” by dividing the input original sentence A by the first delimiter position (delimiter character) ( S701). This word “orld” is designated as word B.

  Then, the original text guess determination unit 13 changes the guess character x to be added to the head of the acquired word B from 'A' to 'Z' (S702), and adds the word B after the guess character x, in other words, the word B The guess character x is added to the head of “Aorld” (S703). This “Aorld” is assumed to be a guess word Cn (n changes to A-Z), and as shown in FIG. 2A, 26 guess words Cn for the word B “orld” are generated and temporarily stored in the memory.

  After that, the original text guess determination unit 13 performs dictionary lookup using the translation dictionary of the dictionary unit 10 in order from the first guess word CA stored temporarily in the memory (S704), and the corresponding word (text data) is stored in the dictionary unit 10. Is determined (S705).

As a result of the determination, when the corresponding word exists in the dictionary unit 10, the original text estimation determination unit 13 holds the estimated word Cn (for example, “World”, etc.) in the memory (S706).
Next to the guess word CA, dictionary lookup is performed with the guess word CB with the guess character x set to “B”, the guess character is changed to “Z”, word guess processing is performed, and it is registered in the translation dictionary as described above. The estimated words Cn are temporarily stored in the memory sequentially. Note that words that do not exist in the dictionary unit 10 are deleted from the memory.

  As shown in FIG. 2B, when the word B is “he”, for example, the guess word Cn “Ahe”, “Bhe”... “Zhe” is generated, and “She”, “The”, etc. As described above, a flag in the translation dictionary is marked with a circle, and a flag in the translation dictionary is marked with a cross, and only words registered in the translation dictionary are temporarily stored in the memory.

  As shown in FIG. 2C, when the word B is “at”, for example, the guess word Cn “Aat”, “Bat”. Is marked with a ◯ flag, and with a x flag for those not in the translation dictionary, only the words registered in the translation dictionary are stored in the memory.

  As shown in FIG. 2D, when the word B is “ato” or the like, for example, the guess words Cn “Aato”, “Bato”. Is marked with a ◯ flag, and with a x flag for those not in the translation dictionary, only the words registered in the translation dictionary are stored in the memory.

  The original text guess determination unit 13 holds, in the memory, words registered in the translation dictionary of the dictionary unit 10 among the guess words Cn generated by changing the guess character x to ‘Z’ as described above.

In this example, as shown in FIG. 3, the guess word Cn existing in the translation dictionary is, for example, “The” is “qualifier”, “She” is “pronoun”, “Bat” is “noun”, etc. Are combined in a correspondence relationship between a headword and a headword part-of-speech and stored in the memory.
When one or more guess words Cn held in the memory exist (S605! = 0), the original sentence guess determination unit 13 makes the guess word Cn present in the dictionary part 10 held in the memory. Is replaced with the word B in the input original A to generate a translation original candidate Dm (m is up to AZ). In other words, the original sentence estimation determination unit 13 sequentially outputs the translation original sentence candidates Dm in which the estimated word Cn replaced with the estimated character x is a replacement candidate from the word B to the original sentence estimation unit 12 (the syntax analysis unit 15 and the morpheme analysis unit 14). Let parsing and morphological analysis.

  When there is only one translation original text candidate Dm with the estimated word Cn as a replacement candidate from the word B (S609 = 1), the original text estimation determining unit 13 replaces the word B with the estimated word Cn. The translated original sentence candidate Dm is output to the translation unit 18, and the sentence is translated by the translation unit 18 (S611).

  When the translation source sentence candidate Dm from the source sentence estimation determination unit 13 is input to the source sentence estimation unit 12 (the syntax analysis unit 15 and the morpheme analysis unit 14), the source sentence estimation unit 12 uses the syntax analysis unit 15 and the morpheme analysis unit 14 An analysis original sentence estimation process is performed and determined by the syntax analysis control unit 16 (S606).

  In the syntax analysis source text estimation process of S606 in the source text estimation unit 12, the contents of the translation source text candidate Dm are analyzed using syntax analysis technology and morpheme analysis technology, and whether or not the translation source text candidate Dm is appropriate in the syntax analysis control unit 16 is determined. Determined.

  In the syntax analysis original text estimation process, as shown in FIG. 6, the morpheme analysis unit 14 performs a morpheme analysis process on the input translation source text candidate Dm (S802), and then the syntax analysis unit 15 inputs the input translation source text. Parsing processing of the candidate Dm is performed (S803). These processes are performed for the number of input translation source candidates Dm (S801, S806).

  When the parsing process for the translation source text candidate Dm is successful (Yes in S804), the syntax analysis unit 15 primarily holds the translation source text candidate Dm for which the parsing has been successful in the memory (S805). The process proceeds to the translation original sentence candidate Dm (S806), and the above process is repeated until there is no translation original sentence candidate Dm.

When the processing of all the translation original sentence candidates Dm is completed, the syntax analysis control unit 16 checks the number of translation original sentence candidates Dm held in the memory.
As a result of this check, if there is no translation original text candidate Dm held in the memory (S607), the syntax analysis control unit 16 outputs the input original text A itself input from the input unit 11 to the translation unit 18, and translates it. The unit 18 translates the input original sentence A based on the data in the dictionary unit 10 (S608).

  When the input source text A is “orld is too big.”, For example, only one source text candidate Dm is specified by the parsing source text estimation process.

  However, if the source text is “at eats rat.”, There are a plurality of translation source candidates Dm such as translation source candidates DB, DC, DE, DF, DG, etc. The translation source candidates cannot be narrowed down to one.

  Therefore, as a result of the syntax analysis, when the number of translation original sentence candidates cannot be narrowed down to one (> 1 in S607), a plurality of translation original sentence candidates Dm are output from the original sentence estimation unit 12 to the original sentence estimation determination unit 13.

  The original sentence estimation determination unit 13 to which a plurality of translation original sentence candidates Dm are input performs semantic analysis on the remaining part of the input original sentence A using the semantic analysis dictionary of the dictionary unit 10 (S608), and is expected to be an estimated word. Guess the word from part of speech and meaning.

  For example, in an example such as “at eats rat.”, It is possible to infer from the keywords “rat = mouse” and “eat = eat” that the presumed word contains information such as “noun” and “living organism”. . Based on such information, the original text estimation determination unit 13 selects a translation original text candidate Dm and sends it to the translation unit 18 for translation.

However, there are a plurality of words having an expected meaning (> 1 in S609), and there are cases where the translation source sentence candidates cannot be narrowed down to one.
In this way, when a plurality of translation source text candidates exist and cannot be narrowed down to one, the source text estimation determination unit 13 uses the statistical data stored in the statistical data storage unit 17 and the appearance frequency data of words in the same document. Use to select any one translation source candidate with high priority. In this way, selecting one translation source candidate having a high statistical priority from a plurality of translation source candidates is referred to as estimating a translation source candidate (S610).

For example, in the case of the input original text A “ato is a Japanese name.”, There is a pattern in which a plurality of words having the expected meaning exist. The statistical data storage unit 17 is searched using a search key, and is translated based on statistical data such as data having a high appearance frequency as a head word, data having a high appearance frequency as a head character, or data on the frequency of appearance of a word in the document. Select one candidate by registering the original text candidates.
Then, the original text estimation determination unit 13 outputs the selected one translation original text candidate to the translation unit 18.
The translation unit 18 translates the input translation source text candidate based on the data in the dictionary unit 10 (S611).

  Thus, according to the translation source text estimation apparatus of the first embodiment, it is determined whether the first letter of the input source text A obtained from the Web page is a lower case letter or upper case letter. Morphological analysis is performed, and capitalization is applied to the first word divided into words, and a word that does not become an unknown word is replaced with a word in the input source text as a replacement candidate. Since the leading word, which is often an unknown word in the web page import, is corrected (complemented) and output to the translation unit 18, the accuracy of the translation result by the translation unit 18 can be improved.

  That is, in consideration of the characteristics of the Web page, the word existing in the dictionary is added to the original unknown word with one letter added to the head of the word originally supposed to be an unknown word in the input original sentence A obtained from the Web page. The correct translation result can be obtained by translating it by replacing it with the first word that should be taken.

  In the first embodiment described above, translation from a Web page has been described as an example. In addition, the present invention also uses a missing character portion when, for example, image data read by a scanner is recognized. can do. That is, when trying to perform character recognition using a scanner, there are often characters that are difficult to recognize such as decorative characters in the image data, and character recognition may not be performed accurately. By applying the invention, it is difficult to recognize with the image data from the scanner, so that a more correct recognition result can be output by estimating the missing characters from the original text written in the original form applied to the scanner.

  Next, the translation original text estimation apparatus according to the second embodiment of the present invention will be described with reference to FIGS. FIG. 7 is a diagram showing a configuration of the translation original text estimation apparatus according to the second embodiment of the present invention.

  As shown in FIG. 7, the translation original text estimation apparatus according to the second embodiment includes an input unit 201 as data receiving means for receiving a text string of an original text key-input from a keyboard, and a character string input in a first language. And an input character string storage unit 202 for storing type information of the input means, etc., determination of non-sentence, unknown word, etc. by performing word analysis, presence / absence of unknown word, syntactic analysis on input character string By performing a spell correct process on the unknown words stored in the unknown word list storage unit 204 and the unknown word list storage unit 204 in which the unknown words extracted by the determination unit 203 are stored. A spell correct unit 205 that generates a word and replaces the word with a word of the original sentence, and translates the word generated by the spell correct unit 205 with a word of the original sentence A display unit 206 for displaying a result to be output to 208, and a dictionary unit 207 in which a translation dictionary (hereinafter referred to as dictionary data) including syntax rules and vocabulary for sentences in the first language and the second language is registered. Have.

  The relationship between the hardware and software including the OS is the same as that in the first embodiment. In other words, this system is composed of a PC having a CPU, memory, hard disk device, keyboard, mouse, monitor, etc., and the above components are processed by the software installed in the hard disk device being read into the memory by the CPU. It is realized by executing. Storage units such as the input character string storage unit 202 and the unknown word list storage unit 204, the dictionary unit 207, and the like are realized by a memory or a hard disk device.

  When the user inputs a first language character string such as English from the keyboard, the input unit 201 accepts the character string input by the key and stores it in the input character string storage unit 202 together with the type information of the input means. The type information of the input means is, for example, a keyboard, a scanner, a microphone, or the like. The type information also includes the type of keyboard.

  The dictionary unit 207 stores (stores) dictionary data such as translated words, parts of speech, and semantic analysis dictionaries associated with headwords (words and characters). The parts of speech are, for example, nouns, verbs, adjectives, adverbs, and the like. The dictionary data includes a character code for each character constituting the word. The translation of the dictionary data is data for converting between other languages in one direction or in two directions, such as Japanese → English, English → Japanese. The dictionary data includes, for example, character tables of national languages such as English alphabet tables, Japanese kanji tables, hiragana tables, and katakana tables. As the dictionary data, many words from various countries such as English words corresponding to headwords and Japanese words are registered.

The translation unit 208 uses the dictionary data including the corresponding character table of the dictionary unit 207 according to the language translation direction (such as English → Japanese) designated by the user from the input unit 201, and is English that is a sentence in the first language. Is translated into a second language sentence, for example, a Japanese sentence.
The determination unit 203 functions as a dividing unit that divides the character string into word units by performing morphological analysis on the character string input by the input unit 201 based on preset dictionary data.

  The determination unit 203 functions as an unknown word extraction unit that extracts, as unknown words, words that are not registered as dictionary data in the dictionary unit 207 among the divided words.

The determination unit 203 extracts, from the dictionary data, the part of speech of the word existing before and after the extracted unknown word from the original text string, and associates the part of speech that is established as the part of speech of the unknown word with the unknown word. It functions as a part of speech combination list generating means for generating a part of speech combination list.
The determination unit 203 functions as a syntax analysis unit that performs syntax analysis of the original text string based on the generated part-of-speech combination list and dictionary data.

  The determination unit 203 replaces the replacement candidate word generated by the spell correction unit 205 with the original sentence word and outputs the replacement sentence to the translation unit 208. In other words, the determination unit 203 functions as a character string output unit that replaces a replacement candidate word with a word at a corresponding position in the original character string and outputs it to the translation unit 208 as a character string to be translated. The determination unit 203 selects a grammatically correct sentence candidate by performing syntax analysis based on dictionary data for each sentence replaced with the first word of the original sentence, or a sentence candidate having a correct meaning by performing semantic analysis. Functions as a sentence analysis means.

  The spell corrector 205 performs morphological analysis on the character string input by the key based on the English words in the dictionary unit 207, and determines whether the word obtained by the division is registered in the dictionary unit 207. Is an unknown word that is not registered in the dictionary unit 207, a table that is set according to the characteristics of the input means, for example, characters that are likely to cause typographical errors due to the keyboard layout (see FIGS. 16 to 18). ) To correct spelling to the correct spelling.

The spell corrector 205 functions as an unknown word part-of-speech specifying means for specifying a part of speech that has been successfully parsed as a part of speech of an unknown word.
The determination unit 203 functions as a candidate character string generation unit that generates a candidate character string by replacing characters of an unknown word whose part of speech is specified with a predetermined rule. The spell corrector 205 functions as a replacement candidate word extraction unit that outputs the generated replacement candidate character string that exists in the dictionary data and matches the part of speech to the determination unit 203 as a replacement candidate word.

  When the word obtained by dividing the input character string is an unknown word that is not registered in the dictionary unit 207, the spell corrector 205 corrects the spelling of the word according to the characteristics (type information) set in advance. And generate a replacement candidate word with the wrong word corrected to the correct spelling.

  In other words, the spell corrector 205 replaces the unknown word character extracted by the determination unit 203 based on a predetermined rule and replaces the new word with the original word so that the replacement candidate becomes a grammatically correct sentence. It functions as replacement candidate word generation means for extracting words.

  These determination unit 203 and spell correction unit 205 replace candidate words that become grammatically correct sentences when new words generated by replacing the extracted unknown word characters based on predetermined rules are replaced with original words. It functions as a replacement candidate word generation means for extracting.

Hereinafter, the operation of the translation original text estimation apparatus according to the second embodiment will be described with reference to FIGS.
In explaining the operation of the translation source sentence estimation apparatus according to the second embodiment, in order to make the processing flow easy to understand, the first language input by the user is English, the second language is Japanese, and an English input sentence is taken as an example. This will be described with reference to the flowcharts shown in FIGS.

  When the translator issues a translation request instruction from the input unit 201, the determination unit 203 obtains a character string input sentence by sentence from the input character string storage unit 202 in accordance with the translation request instruction input to the input unit 201 (S100). , S110).

Subsequently, the determination unit 203 performs morphological analysis on the input character string using word data in the dictionary unit 207, thereby dividing the character string into words and assigning numbers to the respective words.
For example, when the input sentence in English is “I yake a bictcle.”, The input sentence is a word unit such as “I”, “yake”, “a”, “bictcle”... (S120), each word is associated with a number and stored in the memory.
In this example, “1” is associated with “I”, “2” is associated with “yake”, “3” is associated with “a”, and “4” is associated with “bictcle”. Held in.
Thereafter, the determination unit 203 initializes the value of the counter for determining the processing order, that is, the value of the variable m indicating the position of the word in the input sentence to 1 (S130).

Subsequently, the determination unit 203 extracts the mth word in numerical order, searches the dictionary unit 207 using the word as a search key, and performs a dictionary lookup process (S140).
If no matching word exists in the dictionary unit 207 as a result of the dictionary lookup processing (No in S150), the determination unit 203 outputs the word as an unknown word to the unknown word list storage unit 204, as shown in FIG. It is stored in the unknown word list 21 (S160). Again, each unknown word is numbered to accommodate the increment of the variable p and is held in memory.

  After storing the unknown word in the unknown word list 21, the determination unit 203 determines whether or not the currently processed word is the last word based on the number assigned to the divided word (S170), and is not the last word. In the case (No in S170), 1 is added to the value of the variable m (S180), and the processes in S140 to S180 are repeated until the process for the last word is completed.

When the processing for all the words is completed (Yes in S170), the determination unit 203 subsequently determines whether or not an unknown word exists in the unknown word list storage unit 204 (S190), and whether or not there is an unknown word Different processing is executed depending on the situation.
For example, when an unknown word exists in the unknown word list storage unit 204 (Yes in S190), the determination unit 203 executes the processing of the flowchart illustrated in FIG. 9 and when the unknown word does not exist in the unknown word list storage unit 204 ( The determination unit 203 executes the processing of the flowchart shown in FIG.

When an unknown word exists (Yes in S190), the determination unit 203 checks how many words are stored in the unknown word list storage unit 204, and according to the number of unknown words, as illustrated in FIG. An unknown word part-of-speech combination list 23 is created (S300 in FIG. 9).
At this time, the determination unit 203 does not extract candidates that are not correct as part-of-speech arrangements using part-of-speech information of words before and after the unknown word.
For example, in the case of the unknown word “bictcle” listed in the unknown word part-of-speech combination list 23, the previous word is the indefinite article “a”, and the following word ends with “.” (Period). In such a case, the part-of-speech candidates for “bictcle” can be limited to only “nouns”.
When creating the unknown word part-of-speech combination list 23 in the process of S300, the determination unit 203 initializes the value of the variable n indicating the position of the created unknown word part-of-speech combination list 23 to 1 (S310), and each unknown word in the sentence Then, the part of speech of the unknown word part of speech combination list 23 is added (S320), and syntax analysis is performed (S330).

When the parsing is successful (Yes in S340), the determination unit 203 identifies the part of speech as the part of speech of each unknown word (S340).
Further, when the syntax analysis fails (No in S340), the determination unit 203 adds 1 to the value of the variable n (S350), and repeats the processes of S320 to S350 until the syntax analysis is successful.

  In the case of the example sentence of this embodiment, the unknown word “bictcle” is limited as “noun” in the unknown word part-of-speech combination list 23, and the unknown word “yake” in the sentence has an unknown word part-of-speech combination. The parsing succeeds when the “verb” in the part-of-speech candidates in the list 23 is applied.

  If the parsing by the determination unit 203 is successful, the spell corrector 205 initializes the value of the variable p indicating the position of the unknown word list to 1 (S360), and sets the value of the variable q indicating the character position of the unknown word p to 1 (S370), the value of the variable r indicating the position of the alphabet (az) is initialized to 1 (S380), and one character indicated by the variable q is replaced with az to create a candidate character string. (S390).

The spell correction unit 205 compares the created candidate character string with the translation dictionary of the dictionary unit 207 (S400), and determines whether the candidate character string exists in the dictionary unit 207 and the parts of speech match (S410). .
If the candidate character string satisfies the condition as a result of the determination, the spell corrector 205 outputs the candidate character string as a replacement word (S430).

  In this way, the spell corrector 205 increments the variable r indicating the alphabet position and the variable q indicating the character position of the unknown word (S420, S460), and the generated character string is compared with the data in the dictionary unit 207. When comparison is made and the above condition is satisfied, the word is extracted as a replacement word (S390 to S460) and registered in the replacement candidate table 30 (see FIG. 14) set in the memory.

The spell corrector 205 registers a replacement candidate for each unknown word in the replacement candidate table 30 by performing a series of processes from S370 to S480 for all unknown words.
Thereafter, the determination unit 203 reads replacement candidates for each unknown word registered in the replacement candidate table 30, generates an original sentence by each combination, and outputs the generated original sentence to the translation unit 208.
The translation unit 208 outputs the translation result (see FIG. 15) obtained by translating the original text input from the determination unit 203 based on the translation dictionary (dictionary data) of the dictionary unit 207 to the display unit 206. That is, the unknown word in the original text is replaced with the spell correct candidate, translated, and displayed on the screen of the display unit 206 (S490).
In this example, the replacement candidate of the unknown word “bictcle” registered in the replacement candidate table 30 is only “bicycle”, and replacement candidates of the unknown word “yake” are “bake”, “make”, “ Since there are three “take”, there are three original sentences generated by combinations of replacement candidates for each unknown word, as shown in FIG. 15, and the translation results are also three sentences.

If no unknown word is detected in the process of S190, the determination unit 203 performs the process as shown in the flowchart of FIG.
That is, when an unknown word is not detected, the determination unit 203 performs syntax analysis on a sentence composed of divided words (S500). When the syntax analysis is successful (Yes in S510), the sentence (English) is selected. The data is output to the translation unit 208 and translated.
If the parsing fails (No in S510), a word with a different part of speech compared to a similar basic sentence pattern is output as an unknown word to the unknown word list storage unit 204 (S520), and the processing of S300 to S490 described above is performed. I do.

Thus, according to the translation original text estimation apparatus of the second embodiment, when translating an original text created in English into a Japanese text, a character string of the original text key-input from the keyboard by a translator or the like is input. When the unit 201 accepts and stores the input character string in the input character string storage unit 202, the determination unit 203 reads the character string from the input character string storage unit 202 and performs morphological analysis based on dictionary data set in the dictionary unit 207 in advance. By doing so, the character string is divided into units of words, and among the divided words, words that are not registered as dictionary data are listed (extracted) as unknown words. After that, the spell corrector 205 converts the character of the unknown word listed (extracted) by the determination unit 203 into a predetermined rule (a grammatical check by a parsing process, an association with a part of speech, a character using an alphabet table). A replacement candidate word that becomes a grammatically correct sentence when a new word generated by replacing the original word with the original word is generated, and the generated replacement candidate word is replaced with the original text string. Since it is replaced with the word at the corresponding position and output to the translation unit 208 as a character string to be translated, there is no word that is an unknown word due to a typo or the like before the character string is input to the translation unit 208, and the translation result Accuracy can be improved.
In other words, when an English sentence is translated into a Japanese sentence by machine translation, the accuracy of the translation result can be improved by translating after improving the accuracy of the English sentence as a translation source.

In the present invention, the example described in the second embodiment, that is, an example of correcting a word with respect to a typo (keyboard mistake) is an example, and there is a method as described below.
[Third Embodiment]
For example, when a translator tries to input a sentence such as “I take a bus.” From the keyboard, he or she mistyped “I yake a bus.” A case where a sentence including an unknown word “yake” is input will be described as an example.

  The character string “I yake a bus.” Key-inputted from the input unit 201 is stored in the input character string storage unit 202 in association with the type information of the input means that the input unit 201 is “keyboard”.

The determination unit 203 determines “yake” in the key-input character string as an unknown word and stores it in the unknown word list storage unit 204.
The spell corrector 205 reads the unknown word “yake” stored in the unknown word list storage unit 204 from the unknown word list storage unit 204, and replaces the characters in the replacement candidate table 31 shown in FIG. A spell correct candidate character string (another word obtained by changing a character included in an unknown word) is generated based on the data.

  The replacement candidate table 31 is not simply divided by the type of input means such as “keyboard”, “microphone”, “scanner”, etc., and among the keyboards, 101/102 English keyboard, 106 English keyboard, 109 English keyboard, thumb shift keyboard For example, the key layout may be finely divided according to the keyboard type.

  In this case, in the character replacement process shown in FIG. 9 (the process of S390), when the target character string input means is designated as “keyboard”, the spell corrector 205, as shown in FIG. The replacement candidate table 31 based on the key arrangement using the key whose key position is close to that of the target character string based on the key arrangement is preferentially used. The replacement candidate table 31 based on the key arrangement may be stored in advance in a memory or a hard disk device, or may be set as a conversion rule in a program.

When using the replacement candidate table 31 based on this key arrangement, for example, in the example of the character string “I yake a bus.”, As replacement candidates for “yake”, first, “uake”, “take”, “gake”, “Jake” and “hake” are generated, and “take”, which is present as a word and matches the part of speech, is adopted.
As a result, the translation result candidate sentences generated for the example sentence case are ranked as shown in FIG. 17, and the correct translation result is displayed as the top candidate. Can be confirmed.

  In addition, when the input means is a keyboard, there are a key pressed with the left hand and a key pressed with the right hand. If there are a key pressed with the finger of the left hand and a key pressed with the finger of the right hand during spelling, for example, “make” If you try to enter "amke" after pressing "a" first.

  For this reason, as shown in FIG. 18, the finger replacement candidate table 32 is stored in the memory in advance, and the spell corrector 205 stores the keys to be pressed with the left and right fingers in the input character string. If there is a sequence of keys to be pressed with the left finger and keys to be pressed with the right hand finger in the input string, the keys to be pressed with the left finger and the right hand in the input string For the part of the key sequence to be pressed with a finger, the correct spelling word “make” is derived by replacing the character “am” of the word “amke” with “ma” using the replacement candidate table 32 with the finger. Good.

  On the other hand, when the input means is character recognition software using a “scanner” or the like, there is a strong tendency that characters having similar shapes are erroneously recognized depending on the state of image data to be recognized.

  Therefore, as shown in FIG. 19, for example, a similar character replacement candidate table 33 in which similar characters are associated with each other between lowercase letters, uppercase letters, half-width characters, full-width characters, and the like is stored in a memory in advance, and the spell correction unit When “scanner” is designated as an input unit 205, the similar character replacement candidate table 33 may be preferentially used to perform character replacement.

As described above, according to the third embodiment, a rule for character replacement based on the replacement candidate table 31 based on the key arrangement is used as a priority enemy, or the replacement candidate table 32 based on the positions of the fingers of the left and right hands. Precisely use the one based on, or preferentially use the one based on the similar character replacement candidate table 33, thereby correcting the unknown word in the character string of the translation source keyed in with higher accuracy. be able to.
That is, by using the replacement candidate tables 31, 32, and 33 according to the type of input means, more appropriate replacement candidates are selected, and the character string corrected to the correct word is output to the translation unit 208. Thus, a highly accurate translation result can be derived.

It is a figure which shows the structure of the translation original text estimation apparatus of 1st Embodiment of this invention. (A) It is a figure which shows the example which added the flag of whether a capital letter alphabet is added to the head of the head word "orld" detected as an unknown word, and it exists in a dictionary. (B) It is a figure which shows the example which added the capital alphabet to the head of the top word "he" detected as an unknown word, and attached | subjected the flag of whether it exists in a dictionary. (C) It is a figure which shows the example which added the capital alphabet to the head of the top word "at" detected as an unknown word, and attached | subjected the flag whether it exists in a dictionary. (D) It is a figure which shows the example which added the capital letter alphabet to the head of the head word "ato" detected as an unknown word, and attached | subjected the flag whether it exists in a dictionary. It is a figure which shows the example memorize | stored in the memory as a word which exists in a dictionary from the thing which added the character to the head of the top word. It is a flowchart which shows operation | movement of this translation original text estimation apparatus. It is a flowchart which shows the dictionary lookup word estimation process of this translation original sentence estimation apparatus. It is a flowchart which shows the syntax analysis original text estimation process of this translation original text estimation apparatus. It is a figure which shows the structure of the translation original text estimation apparatus of 2nd Embodiment which concerns on this invention. It is a flowchart which shows operation | movement of the translation original text estimation apparatus of 2nd Embodiment. It is a flowchart which shows a process when an unknown word is detected in the translation original text estimation apparatus of 2nd Embodiment. It is a flowchart which shows a process when an unknown word is not detected in the translation original text estimation apparatus of 2nd Embodiment. It is a figure which shows the divided | segmented word. It is a figure which shows an unknown word list. It is a figure which shows an unknown word part of speech combination table. It is a figure which shows the replacement candidate table for replacing an unknown word with the word registered into the dictionary part. It is a figure which shows the translation result at the time of using a replacement candidate table. It is a figure which shows the replacement candidate table by key arrangement | positioning. It is a figure which shows the translation result at the time of using the replacement candidate table by key arrangement | positioning. It is a figure which shows the replacement candidate table by a finger | toe. It is a figure which shows a similar character replacement candidate table. It is a figure which shows four cases which may output an incorrect translation result.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Dictionary part, 11 ... Input part, 12 ... Original sentence estimation part, 13 ... Original sentence estimation determination part, 14 ... Morphological analysis part, 15 ... Syntax analysis part, 16 ... Syntax analysis control part, 17 ... Statistical data storage part, 18 ... Translation unit, 19 ... Output unit, 30 ... Replacement candidate table, 31 ... Replacement candidate table by key arrangement, 32 ... Replacement candidate table by finger, 33 ... Similar character replacement candidate table, 201 ... Input unit, 202 ... Input character string Storage unit 203... Determination unit 204 204 Unknown word list storage unit 205 Spell correction unit 206 Display unit 207 Dictionary unit 208 Translation unit

Claims (15)

In a machine translation apparatus including a translation unit that translates a sentence written in a first language into a sentence in a second language based on dictionary data registered in advance to translate the original sentence created in another language into a sentence in another language,
Data receiving means for receiving a character string created in the first language;
A dividing unit that divides the character string received by the data receiving unit into word units based on the dictionary data;
An unknown word extracting means for extracting a word that is not registered in the dictionary data among the words divided by the dividing means, as an unknown word;
Replacement candidate character generation means for generating a replacement candidate character by adding a character or replacing a character according to a predetermined rule with respect to the unknown word extracted by the unknown word extraction means;
And a character string output means for outputting a sentence obtained by replacing the replacement candidate character generated by the replacement candidate character generation means with a replacement-source unknown word to the translation unit as a character string to be translated. Translation device. Translates the original text created in the first language into the text in the second language based on the dictionary part that stores the dictionary data including characters, words, dictionaries, translated words, and parts of speech in advance to translate the text in other languages In a machine translation device equipped with a translation unit,
Data receiving means for receiving the original text of the first language as a computer-processable character string;
Dividing means for dividing the character string into word units by performing morphological analysis on the basis of the dictionary data stored in the dictionary unit for the character string received by the data receiving means;
Of the plurality of words divided by the dividing means, it is determined whether the first letter of the first word is a lowercase letter, and if the first letter is a lowercase letter, a plurality of new letters with an uppercase alphabet added to the beginning of the first word Word generation means for generating a word;
Replacement candidate word extraction means for extracting words existing as the dictionary data among a plurality of words generated by the word generation means;
Character string output means for outputting a sentence obtained by replacing the replacement candidate word extracted by the replacement candidate word extraction means with the first word of the original sentence to the translation unit as a character string to be translated, Machine translation device. The character string output means includes:
A first sentence candidate that selects a grammatically correct sentence candidate based on the dictionary data for each sentence replaced with the first word of the original sentence when a plurality of replacement candidate words are extracted by the replacement candidate word extracting means; 3. The machine translation apparatus according to claim 1, further comprising selection means. The first sentence candidate selection means includes:
Sentence analysis means for selecting a grammatically correct sentence candidate by performing syntax analysis based on the dictionary data for each sentence replaced with the first word of the original sentence, or a sentence sentence having a correct meaning by performing semantic analysis The machine translation apparatus according to claim 3, further comprising:   For one or more grammatically correct sentence candidates selected by the first sentence candidate selecting means, a sentence candidate in which an appropriate word is used as the first word in the sentence based on preset statistical data is selected. 4. The machine translation apparatus according to claim 3, further comprising second sentence candidate selection means for outputting it as a character string to be translated to the translation unit.   6. The machine translation apparatus according to claim 5, wherein the statistical data uses appearance frequency data of words in a sentence.   6. The machine translation apparatus according to claim 5, wherein appearance frequency data of a head word is used for the statistical data.   6. The machine translation apparatus according to claim 5, wherein the appearance data of initial characters is used for the statistical data. In a machine translation system that translates an original text created in a first language into a text in a second language,
Data receiving means for receiving the original text string input by the key;
A dividing unit that divides the character string into word units by performing morphological analysis based on dictionary data set in advance for the character string input by the data receiving unit;
An unknown word extracting means for extracting a word that is not registered as the dictionary data among the words divided by the dividing means, as an unknown word;
Extracting a replacement candidate word that becomes a grammatically correct sentence when a new word generated by replacing the character of the unknown word extracted by the unknown word extraction means based on a predetermined rule is replaced with the original word Replacement candidate word generation means;
A character string output unit that replaces the replacement candidate word generated by the replacement candidate word generation unit with a word at a corresponding position in the original character string and outputs it to the translation unit as a character string to be translated; A machine translation device. The replacement candidate word generation means includes:
The part of speech of the word existing before and after the unknown word extracted by the unknown word extraction means is extracted from the dictionary data, and the part of speech that is established as the part of speech of the unknown word is defined as an unknown word. A part-of-speech combination list generating means for generating an associated part-of-speech combination list;
Syntax analysis means for performing syntax analysis of the text string of the original text based on the part of speech combination list generated by the part of speech combination list generation means and the dictionary data;
An unknown word part-of-speech specifying means for specifying a part of speech that has been successfully parsed by the syntax analysis means,
Candidate character string generating means for generating a candidate character string by replacing the character of the unknown word whose part of speech is specified by the unknown word part of speech specifying means with a predetermined rule;
A replacement candidate word extraction unit that outputs a candidate character string generated by the candidate character string generation unit that is present in the dictionary data and has a matching part of speech as a replacement candidate word to the character string output unit. The machine translation apparatus according to claim 9, wherein the machine translation apparatus is a machine translation device. The predetermined rule is:
10. The machine translation apparatus according to claim 9, wherein the machine translation apparatus is based on a replacement candidate table based on key arrangements or positions of fingers of left and right hands. The predetermined rule is:
The machine translation apparatus according to claim 9, wherein the machine translation apparatus is based on a similar character replacement candidate table in which similar characters are associated with each other. Executes processing in a machine translation device having a translation unit that translates into a second language sentence based on dictionary data stored in advance to translate an original sentence created in the first language into a sentence in another language In the machine translation program to let
The machine translation device;
Data receiving means for receiving a character string created in the first language;
A dividing unit that divides the character string received by the data receiving unit into word units based on the dictionary data;
An unknown word extracting means for extracting a word that is not registered in the dictionary data among the words divided by the dividing means, as an unknown word;
A replacement candidate character generating unit that generates a replacement candidate character by adding a character or replacing a character according to a predetermined rule with respect to the unknown word extracted by the unknown word extracting unit;
A machine that functions as a character string output unit that outputs a sentence obtained by replacing the replacement candidate character generated by the replacement candidate character generation unit with a replacement source unknown word as a character string to be translated to the translation unit. Translation program. Translates the original text created in the first language into the text in the second language based on the dictionary part that stores the dictionary data including characters, words, dictionaries, translated words, and parts of speech in advance to translate the text in other languages In a machine translation program for causing a machine translation device including a translation unit to execute processing,
The machine translation device;
Data receiving means for receiving the original text of the first language as a computer-processable character string;
Dividing means for dividing the character string into word units by performing morphological analysis on the basis of the dictionary data stored in the dictionary unit for the character string received by the data receiving means;
Of the plurality of words divided by the dividing means, it is determined whether the first letter of the first word is a lowercase letter, and if the first letter is a lowercase letter, a plurality of new letters with an uppercase alphabet added to the beginning of the first word Word generation means for generating a word;
Replacement candidate word extraction means for extracting words existing as the dictionary data among a plurality of words generated by the word generation means;
The replacement candidate word extracted by the replacement candidate word extraction unit functions as a character string output unit that outputs a sentence in which the first word of the original sentence is replaced as a character string to be translated to the translation unit. Machine translation program. In a machine translation program that causes a machine translation device that translates an original sentence created in a first language into a sentence in a second language,
The machine translation device;
Data receiving means for receiving the original text string input by the key;
A dividing unit that divides the character string into word units by performing morphological analysis based on dictionary data set in advance for the character string input by the data receiving unit;
An unknown word extracting means for extracting a word that is not registered as the dictionary data among the words divided by the dividing means, as an unknown word;
Extracting a replacement candidate word that becomes a grammatically correct sentence when a new word generated by replacing the character of the unknown word extracted by the unknown word extraction means based on a predetermined rule is replaced with the original word Replacement candidate word generation means;
The replacement candidate word generated by the replacement candidate word generation means is replaced with a word at a corresponding position in the original text character string and functions as a character string output means for outputting to the translation section as a character string to be translated. A machine translation program.

Images