JP3982726B2 - Translation knowledge learning device and machine translation device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a translation knowledge learning method and apparatus having a function of learning translation knowledge such as a translation pattern and document information from an existing bilingual document, and an example-driven machine translation method and apparatus for performing translation processing using this translation knowledge It is about.
[0002]
[Prior art]
Conventionally, as a method for learning translation knowledge such as a translation pattern and document information using an existing parallel translation document, for example, there is the following method.
[0003]
Literature 1: “Japanese Dictionary / Rule Learning Method for Machine Translation System and Dictionary / Rule Learning Device for Machine Translation System”
Reference 2: “Japanese Laid-Open Patent Publication No. 7-85056 Dictionary Creation Support Method”
[0004]
The above two conventional techniques are machine translation methods or apparatuses that extract translation knowledge from a bilingual document and translate based on the translation knowledge. Translation patterns and grammatical rules correspond to the translation knowledge here. The machine translation apparatus as described above is generally called an example-driven machine translation apparatus. The advantage of this type is that the target language expression in the bilingual document given in advance is reflected in the translation result of the machine translation device.
[0005]
The technique of Document 1 reads a bilingual document to be learned in advance, and compares the translation result with the bilingual sentence (ideal translation) each time the source language input sentence to be translated is translated. The dictionary and grammatical rules are changed using the ideal translation with the smallest difference in the ideal translation. This is not a learning process at the time of reading a bilingual document, but it is typical of the type in which an optimal example is selected from a bilingual sentence each time an original sentence to be translated is input, and a dictionary and grammar rules are learned from that example. It is an example.
[0006]
The advantage of this method is that, since no special processing is applied to the example, it is used for translation processing in units of one sentence, so that information of one sentence is used for translation processing. Also, because the dictionary and grammar rules are added and modified when the original text to be translated is input, it is not necessary to create useless dictionaries and grammar rules, and there are side effects between dictionaries and grammar rules. Hateful.
[0007]
On the other hand, the technique of Document 2 is a typical example of a dictionary learning device that automatically creates a conversion dictionary between a source language and a target language called a translation pattern from a parallel translation document in advance. Such an apparatus extracts a candidate for dictionary to be registered in a user dictionary or the like by extracting a translation pattern from a bilingual document, and at the same time, edits the data of the registration candidate, original document information and translated document Extract information. Then, by translating using this extraction result, the expression in the bilingual document is reflected in the translation result.
[0008]
The advantage of this method is that the example learning unit and the translation processing unit exist separately, and the example is a translation pattern format that is easy for the user to understand, so the user can easily manage (add to) the dictionary created by the learning unit. , Delete, change). Further, the dictionary can be managed for each bilingual document to be learned.
[0009]
[Problems to be solved by the invention]
However, the above two conventional techniques have the following problems.
The technique of Document 1 takes time to translate because a search process for an optimal example, addition of a dictionary, and grammar rules are added and changed each time a sentence is input. Furthermore, since the dictionary and grammar rules are rewritten every time a sentence is input, even if an adverse effect due to the learning effect is found, the dictionary and grammar rules cannot be restored. Furthermore, since the user cannot predict the translation result, it is difficult to improve the translation by adding and correcting the dictionary by the user (in a practical machine translation device, automatic learning by the machine translation device and cooperation of user support are indispensable) Absent).
[0010]
Also, if different types of documents (for example, letters, manuals, etc.) are mixed in the bilingual document, rules and dictionaries with completely different representation formats and dictionaries are mixed in the dictionary and grammar, resulting in poor visibility of the dictionary and rules. Side effects are likely to occur, causing a reduction in translation quality. Furthermore, even though the entire document is stored as an example, only one sentence of information is used during translation.
[0011]
On the other hand, in the technique of Document 2, a translation pattern dictionary is created from a bilingual document regardless of whether or not it is used for translation processing and regardless of the dictionary that the machine translation apparatus has. For this reason, the user must always select an appropriate dictionary for the input document to be translated before performing translation. Even if all the dictionaries are used for translation, it causes a side effect between dictionaries, resulting in a decrease in translation quality.
[0012]
In addition, when automatically creating a dictionary, it refers to which translation a word in a source language corresponds to, and what is the condition when it is translated into that word, even though the entire bilingual document is referenced. Even if the local knowledge is extracted, for example, the knowledge about the entire bilingual document such as how many times the word appears in the bilingual document and what kind of bilingual document is not extracted.
[0013]
In view of the above, it has been desired to realize a translation knowledge learning method and apparatus, and a machine translation method and apparatus that allow a user to easily manage translation knowledge and improve translation quality.
[0014]
[Means for Solving the Problems]
  The present invention employs the following configuration in order to solve the above-described problems..
[0015]
<Constitution1>
  A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in a natural language; and a bilingual document input in the bilingual document input unit, the morpheme of each of the source language and the target language and the number of appearances of the morpheme Based on the lexical knowledge learning unit to be extracted and the grammatical rules and grammatical rules provided in advance for use in translation processing, from the document input in the bilingual document input unitGenerate new grammar rulesAnd the morpheme frequency based on the morpheme extracted by the vocabulary knowledge learning unit and the number of appearances of the morphemedictionaryFind these morpheme frequenciesdictionaryWhenNew grammar rulesA translation knowledge learning apparatus comprising: a document environment learning unit that adds the above to the morpheme dictionary / grammar rule.
[0016]
<Constitution2>
  A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in a natural language; a bilingual document input by the bilingual document input unit; Vocabulary knowledge learning unit that extracts the translation pattern and the number of appearances of the translation pattern, and input from the bilingual document input unit based on the grammar rule in a morpheme dictionary / grammar rule provided in advance for use in translation processing From documentGenerate new grammar rulesAnd the morpheme frequency based on the morpheme extracted by the vocabulary knowledge learning unit, the number of appearances of the morpheme, the translation pattern, and the number of appearances of the translation patterndictionaryAnd translation patternsdictionaryFind these morpheme frequenciesdictionaryWhenNew grammar rulesTo the morpheme dictionary / grammar rules, and the translation patterndictionaryA document environment learning unit for registering an image in a translation pattern learning dictionary, and the morpheme frequency from the document environment learning unitdictionaryWhenNew grammar rulesA machine translation apparatus comprising: a translation engine unit that translates a source language document into a target language document by using the morpheme dictionary / grammar rule to which is added and the translation pattern learning dictionary.
[0017]
<Configuration 3>
  A bilingual document input unit for inputting a bilingual document consisting of a source language document and a target language document in natural language, a translation morpheme dictionary including a large number of morphemes constituting a sentence, and a storage unit storing grammar rules; A translation knowledge learning apparatus comprising a translation pattern dictionary including a translation pattern in which a source language and a target language phrase are associated with each other, wherein the input source language document and the target language document are based on the grammar rules. Each morpheme is analyzed to generate a source language morpheme frequency dictionary and a target language morpheme frequency dictionary in which each morpheme is associated with the number of appearances indicating the number of occurrences of the morpheme in the document. A vocabulary knowledge learning unit that generates a translation pattern dictionary in which the number of appearances indicating the number of appearances of the translation pattern is associated with a translation pattern in which a word of the document is associated, and for each bilingual document A source language morpheme frequency dictionary, another storage unit for storing the target language morpheme frequency dictionary and the translation pattern dictionary, the generated source language morpheme frequency dictionary, and the generated target language morpheme frequency dictionary One or more morpheme frequency dictionaries of other bilingual documents in which the number of morphemes in common among the other storage units is at least a predetermined number or more are determined for each morpheme in each of the generated morpheme frequency dictionaries. Each of the generated morphemes is given a weight number value obtained by adding a weighting coefficient to the number of appearances, and a weight number value obtained by adding a weighting coefficient to each appearance number in the generated translation pattern dictionary. If the weight count value of the morpheme in the frequency dictionary is smaller than the appearance count of the same morpheme in each morpheme frequency dictionary of the other bilingual document, the weight count value is changed to the appearance count Then, the morpheme dictionary for translation is updated with each morpheme in each of the generated morpheme frequency dictionaries and the number of appearances and the number of weights, and the number of times of the weight of the translation pattern in the generated translation pattern dictionary is the other bilingual translation. If it is smaller than the number of appearances of the same translation pattern in the translation pattern dictionary of the document, the weight number value is changed to the number of appearances, and then each translation pattern in the generated translation pattern dictionary and the number of appearances and the weight number value are used. A translation knowledge learning apparatus comprising: a document environment learning unit for updating the translation pattern dictionary.
[0018]
<Configuration 4>
  A bilingual document input unit for inputting a bilingual document consisting of a source language document and a target language document in natural language, a translation morpheme dictionary including a large number of morphemes constituting a sentence, and a storage unit storing grammar rules; A machine translation apparatus comprising a translation pattern dictionary including a translation pattern in which a source language and a target language word are associated with each other, and the input source language document and the target language document are based on the grammar rules. Analyzing each morpheme to classify it, generating a morpheme frequency dictionary in the source language and a morpheme frequency dictionary in the target language in which the number of appearances indicating the number of appearances of the morpheme in the document is generated, and the both documents A vocabulary knowledge learning unit that generates a translation pattern dictionary in which the number of appearances indicating the number of appearances of the translation pattern is associated with the translation pattern corresponding to A morpheme frequency dictionary of the language, another storage unit for storing the morpheme frequency dictionary of the target language and the translation pattern dictionary, at least the generated morpheme frequency dictionary of the source language and the generated morpheme frequency dictionary of the target language On the other hand, one or more morpheme frequency dictionaries of other parallel translation documents having a common number of morphemes from the other storage unit are determined, and the number of appearances of each morpheme in each of the generated morpheme frequency dictionaries Each of the generated morpheme frequency dictionaries is assigned a weight number value obtained by adding a weighting coefficient to each translation pattern in the generated translation pattern dictionary The weight count value is changed to the appearance count when the weight count value of the morpheme is smaller than the appearance count of the same morpheme in each morpheme frequency dictionary of the other bilingual document The morpheme dictionary for translation is updated with each morpheme in each of the generated morpheme frequency dictionaries and the number of appearances and the number of weights, and the weight number value of the translation pattern in the generated translation pattern dictionary is the value of the other bilingual document. If the number of appearances of the same translation pattern in the translation pattern dictionary is smaller than the number of appearances, the weight number value is changed to the number of appearances, and the translation is performed according to each translation pattern, the number of appearances and the weight number value in the generated translation pattern dictionary. A document environment learning unit for updating a pattern dictionary, and when a document to be translated in a source language is input, the translation pattern dictionary, the translation morpheme dictionary, and the document to be translated based on the grammar rules A machine translation apparatus comprising: a translation engine unit for translating into a language document.
[0019]
<Constitution5>
  A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in a natural language, and a morpheme dictionary indicating morpheme analysis information and grammar rules provided in advance for use in translation processing from the source language to the target language・ Grammar rulesPartA morpheme frequency dictionary / translation pattern dictionary for each bilingual document indicating the number of appearances and translation patterns of each morpheme for each bilingual document, and a morpheme, its number of appearances and its translation from the bilingual document using the morpheme dictionary / grammar rule Extract patterns and their number of appearances, and extract theseinformationThe lexical knowledge learning unit for outputting to the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary as dictionary information and the bilingual document-specific morpheme frequency dictionary / translation pattern dictionaryinformationThe optimal dictionary is selected based on theinformationAnd the resulting morpheme frequencydictionaryAnd a grammar extracted from the bilingual document based on the grammatical rules in the morpheme dictionary / grammatical rulesRulesBased on the document environment learning unit that outputs the document environment setting file for each bilingual document, the morpheme frequency dictionary / translation pattern dictionary for each bilingual document, and the document environment setting file for each bilingual documentdictionaryAnd grammarRulesAnd createShapeA translation knowledge learning apparatus comprising: an environment adaptive dictionary construction unit that outputs information as additional information to an elementary dictionary and grammatical rules.
[0020]
<Constitution6>
  Claim5The translation knowledge learning device according to claim 1, further comprising a dictionary / environment setting file editing unit for editing the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary and the bilingual document-specific document environment setting file. .
[0021]
<Constitution7>
  A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in a natural language, and a morpheme dictionary indicating morpheme analysis information and grammar rules provided in advance for use in translation processing from the source language to the target language・ Grammar rulesPartA morpheme frequency dictionary / translation pattern dictionary for each bilingual document indicating the number of appearances and translation patterns of each morpheme for each bilingual document, and the morpheme, its number of appearances, and its translation from the bilingual document Extract patterns and their number of appearances, and extract theseinformationThe lexical knowledge learning unit for outputting to the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary as dictionary information and the bilingual document-specific morpheme frequency dictionary / translation pattern dictionaryinformationThe optimal dictionary is selected based on theinformationAnd the resulting morpheme frequencydictionaryAnd a grammar extracted from the bilingual document based on the grammatical rules in the morpheme dictionary / grammatical rulesRulesBased on the document environment learning unit that outputs the document environment setting file for each bilingual document, the morpheme frequency dictionary / translation pattern dictionary for each bilingual document, and the document environment setting file for each bilingual documentdictionaryAnd grammarRulesAnd createShapeOutput as additional information to the lexicon and grammar rules and translation patternsdictionaryCreate the translation patterndictionaryIs generated as a translation pattern learning dictionary, and the morpheme frequency is generated by the environment adaptive dictionary construction unit.dictionaryAnd grammarRulesAnd a translation engine unit that translates a source language document into a target language document using the translation pattern learning dictionary.
[0022]
<Constitution8>
  Constitution7The machine translation apparatus according to claim 1, further comprising a dictionary / environment setting file editing unit for editing the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary, the bilingual document-specific document environment setting file, and the translation pattern learning dictionary. Machine translation device.
[0023]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
<< Specific Example 1 >>
<Constitution>
FIG. 1 is a flowchart showing a specific example 1 of the machine translation method of the present invention. Prior to this description, the configuration of the machine translation apparatus according to the specific example 1 of the present invention will be described.
[0024]
FIG. 2 is a configuration diagram of the machine translation apparatus according to the first embodiment of the present invention.
The function of the machine translation apparatus of the present invention is realized by a computer, and this apparatus is roughly divided into an input / output unit 1, a learning engine unit 2, a translation engine unit 3, a morpheme dictionary / grammar rule 4, a morpheme frequency for each bilingual document. It comprises a dictionary / translation pattern dictionary 5 and a translation pattern learning dictionary 6.
[0025]
The input / output unit 1 includes a general input / output device such as a keyboard and a display in a computer, an input / output terminal, a network, and the like, and includes a bilingual document input unit 11, a source language sentence input unit 12, and a target language sentence output unit 13. ing.
[0026]
The bilingual document input unit 11 is a functional unit that inputs a bilingual document composed of a source language and a target language in a natural language. The source language sentence input unit 12 is a functional unit for inputting source language sentences in the translation engine unit 3. The target language sentence output unit 13 is a functional unit for outputting the target language sentence translated by the translation engine unit 3.
[0027]
  The learning engine unit 2 is a functional unit that learns translation knowledge, such as a vocabulary knowledge learning unit 21, a document environment learning unit 22, a morpheme frequency dictionary / translation pattern dictionary 23.Storage deviceIt is composed of
[0028]
The vocabulary knowledge learning unit 21 refers to the morpheme dictionary / grammar rule 4 and extracts the morpheme, its number of appearances, its translation pattern, and its number of appearances from the bilingual document input by the bilingual document input unit 11, It has a function of creating a translation pattern dictionary 23.
[0029]
The document environment learning unit 22 creates a translation pattern learning dictionary 6 with reference to the morpheme frequency dictionary / translation pattern dictionary 23 and the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 5, and creates a morpheme dictionary / grammar rule 4. It has a function to add information. That is, the document environment learning unit 22 selects an optimum dictionary from the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 5 based on the number of appearances of the morpheme with the morpheme frequency dictionary / translation pattern dictionary 23, and this optimum dictionary. Are merged with the morpheme frequency dictionary / translation pattern dictionary 23, and the morpheme frequency knowledge obtained as a result is newly added to the morpheme dictionary / grammar rule 4 and the bilingual document based on the grammatical rule in the morpheme dictionary / grammar rule 4 It has a function of newly adding the grammatical knowledge extracted from the morpheme dictionary / grammar rule 4.
[0030]
The translation engine unit 3 translates the source language sentence input from the source language sentence input unit 12 with reference to the morpheme dictionary / grammar rule 4 and the translation pattern learning dictionary 6 and outputs it to the target language sentence output unit 13. It has a function.
[0031]
The morpheme dictionary / grammar rule 4 is a database used in common with the learning engine unit 2 and the translation engine unit 3. Translation knowledge obtained by the learning engine unit 2 is added, and the translation engine unit including the added information is added. 3 is for translation.
[0032]
The bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 5 is a database that stores a plurality of morpheme frequency dictionaries / translation pattern dictionaries for each bilingual document to be learned.
[0033]
The translation pattern learning dictionary 6 is a dictionary created by the document environment learning unit 22 and used by the translation engine unit 3 during translation processing.
[0034]
In the above-described configuration, the vocabulary knowledge learning unit 21 and the document environment learning unit 22 in the learning engine unit 2 and the translation engine unit 3 include a program corresponding to this function and a CPU or memory for executing the program. The function is realized. The morpheme frequency dictionary / translation pattern dictionary 23, the morpheme dictionary / grammar rule 4, the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 5, and the translation pattern learning dictionary 6 are temporarily created on the main memory. A database or a database stored in an auxiliary storage device.
[0035]
<Operation>
Hereinafter, the operation of this example will be described with reference to the flowchart of FIG.
[0036]
The bilingual document in this specific example is a document that already has Japanese and English sentences.
The user can use the English document system. en, Japanese document system. A case where a bilingual document consisting of ja is inputted and an English sentence “This command pushes modules on and / or pops modules of the stream” is translated will be described as an example. In addition, system. en's "system" represents the document name, and system. en's ".en" is an English document, system. “ja” of ja is an identifier indicating a Japanese document.
[0037]
FIG. 13 shows a system. en and system. It is a figure which shows the example of ja.
As shown in the figure, it can be seen that tags indicating correspondence between sentences are written as en = 1, 2,..., S and ja = 1, 2,..., S (s is the number of translated sentences).
[0038]
First, the user inputs a bilingual document system. en, system. When ja is input (step S1), the bilingual document is transferred to the vocabulary knowledge learning unit 21 of the learning engine unit 2.
[0039]
  The vocabulary knowledge learning unit 21 automatically extracts a translation pattern, all morphemes constituting the source language sentence and the target language sentence, and the number of appearances from the bilingual document, and the result is a morpheme frequency dictionary / translation pattern dictionary 23.As storage(Step S2).
[0040]
Further, the morpheme analysis process in this specific example uses a morpheme dictionary / grammar rule 4. The reason for this is to prevent deterioration in translation quality due to a morpheme unit shift between the learning engine unit 2 and the translation engine unit 3.
[0041]
FIG. 14 is a diagram showing an example of the morpheme frequency dictionary 23 after the vocabulary knowledge learning process in this specific example.
FIG. 15 is a diagram illustrating an example of the translation pattern dictionary 23.
[0042]
As shown in both figures, with the field separator as a tab, the first field stores morphemes (FIG. 14) or translation patterns (FIG. 15), and the second field stores the number of appearances. In FIG. 15, * NP means a noun phrase and * VP means a verb phrase.
[0043]
Next, the document environment learning unit 22 creates the translation pattern learning dictionary 6 from the morpheme frequency dictionary / translation pattern dictionary 23 and the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 5, and moves to the morpheme dictionary / grammar rule 4. Is added (step S3).
[0044]
FIG. 3 is a flowchart for explaining the operation of the document environment learning unit 22.
First, an optimal dictionary selection process, which is a process of selecting a dictionary similar to the morpheme frequency dictionary / translation pattern dictionary 23 created by the vocabulary knowledge learning unit 21 from the existing bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 5 (step S31). )I do.
[0045]
FIG. 4 is a flowchart for explaining the operation of the optimum dictionary selection process.
First, morpheme top n words appearing frequently from the morpheme frequency dictionary 23 (which may be either English or Japanese morpheme frequency dictionary, or both may be used. In this example, the English morpheme frequency dictionary is used). Extracted and stored in the work table WT1 (step S41) (n is a natural number of 1 or more. In this specific example, n = 50).
[0046]
Next, if the processing has not been completed for all the bilingual document-specific morpheme frequency dictionaries 5 (step S42), one of the morpheme frequency dictionaries is stored in the work table WT2 (step S43). Similar to the processing in step S41, the top morpheme n words that appear frequently are extracted and stored in the work table WT3 (step S44).
[0047]
Next, the number x of common morphemes stored in WT1 and WT3 is obtained (step S45). If the value obtained by dividing x by n is larger than the threshold value y, the dictionary is recognized as the optimum dictionary, The dictionary name is stored in the table WT0 (step S46). Here, y is a number from 0 to 1. In this specific example, y = 0.8.
[0048]
If it is determined in step S42 that the above-described processing (steps S43 to S46) has been repeated for all bilingual document-specific morpheme frequency dictionaries 5, the present processing is terminated and the processing proceeds to learning dictionary construction processing (step S32).
[0049]
FIG. 16 is a diagram illustrating an example of WT0 in this specific example after the optimum dictionary selection processing.
In this specific example, a dictionary named “os.dic, unix.dic, users_guide.dic” is stored in the bilingual document system. en, system. It is similar to a dictionary created from ja.
[0050]
  Next, the process proceeds to a learning dictionary construction process (step S32).
  5 and 6 are flowcharts for explaining the operation of the learning dictionary construction process.
  First, the work tables WT1 to 6 are initialized (step S501), the English morpheme frequency dictionary 23 is stored in WT1, the Japanese morpheme frequency dictionary 23 is stored in WT2, and the translation pattern dictionary 23 is stored in WT3 (step S502). However, when storing in the work tables WT1 to WT3, the morpheme frequency dictionary 23And the translation pattern dictionary 23The value of the number of occurrences multiplied by the weighting factor αweightTimesvalueStore as. Here, when the existing dictionary (translation document-specific morpheme frequency dictionary / translation pattern dictionary 5) is set to 1, the weighting factor α gives priority to the dictionary (morpheme frequency dictionary / translation pattern dictionary 23) based on the currently input document. Is a numerical value indicating the degree of superiority, and is usually a value of 1 or more (in this specific example, α = 2).
[0051]
  Next, if the processing has not been completed for all the dictionaries stored in WT0 (step S503), the dictionary nameFor example, os. dic to WRetrieved from T0 and its dictionary nameCorresponding toThe English morpheme frequency dictionary 5 is stored in WT4, the Japanese morpheme frequency dictionary 5 is stored in WT5, and the translation pattern dictionary 5 is stored in WT6 (step S504).
[0052]
  Next, it is stored in WT4English morpheme frequency dictionary 5If processing has not been completed for all morphemes (step S505),RawMorphology and the number of appearancesSet ofIs extracted (step S506). If the extracted morpheme is registered in WT1 (step S507), WT1Morpheme weight count value in the English morpheme frequency dictionary 23And the appearance times of each extracted morphemeNumberIf the extracted morpheme is larger, WT1Weight value of English morpheme frequency dictionary 23The appearance times of the extracted morphemesTo numberIf the changed and extracted morpheme is smaller, nothing is done (step S508).
[0053]
  On the other hand, if it is not registered in step S507, the extracted morpheme and its appearance times are displayed.NumberWTlNew English Morphological Frequency Dictionary 23(Step S509
[0054]
  After repeating the processing of step S506 to step S509 for all morphemes (step S505), the Japanese morpheme frequency dictionary23, 5WT2 and WT5 (steps S510 to S514), and a translation pattern dictionary23, 5The same processing is performed for WT3 and WT6 (steps S601 to S605) in which is stored.
[0055]
  In step S601, if the processing is completed for all the translation patterns of WT6, the process returns to step S503 and is stored in WT0.remainingAfter repeating the same process for the dictionary name (steps S504 to S514 and steps S601 to S605), the process proceeds to the next process.
[0056]
  Next, all the morphemes registered in WT1 and their occurrence timesNumber etc.English morpheme dictionaryUpdated as 4Register (step S606), all morphemes registered in WT2 and their appearance timesNumber etc.Japanese morpheme dictionaryUpdated as 4Registration is performed (step S607).
[0057]
  That is, Morpheme dictionary4InalreadyIf morphemes exist,Corresponding to the morphemeAppearance times of WT1 or WT2Register numberTo do. If there is no morpheme,In the morpheme dictionary 4WT1 or WT2New morphemes and correspondingAppearance timesRegister numberTo do.
[0058]
On the other hand, a default value d is given to the appearance count value of a morpheme that does not exist in WT1 or WT2 and exists in the morpheme dictionary (d = 1 in this specific example).
[0059]
  FIG. 17 shows an English morpheme before and after the learning dictionary construction process in this specific example.frequencydictionary23It is a figure which shows the example of.
  Before processing, appearance timesnumber("-" Is registered in (third field with "," as a field separator), but it can be seen that a numerical value of 1 or more is registered after processing.
[0060]
  Next, all the translation patterns registered in WT3 and their occurrence timesNumber etc.It is registered in the translation pattern learning dictionary 6 (step S608).
[0061]
  FIG. 18 is a diagram showing an example of the translation pattern learning dictionary after the learning dictionary construction processing in this specific example.
  As shown, “Translation pattern, appearance timesnumber"It can be seen that a translation pattern learning dictionary is created in the format.
[0062]
Next, grammarknowledgeProceed to the extraction process (step S33).
Figure 7 shows the grammarknowledgeIt is a flowchart explaining the operation | movement of an extraction process.
First, the work tables WT1 to WT5 are initialized (step S71), and the Japanese document and English document input by the bilingual document input unit 11 are converted into morpheme dictionaries / grammar rules for each language.Rule 4Is used to perform morphological analysis (step S72) and syntactic analysis (step S73).
[0063]
Next, from this morphological analysis result and syntactic analysis result, Japanese sentence ending expression, passive tendency, question tendency, punctuation expression, conjunction translation (eg, translation of if, and, of)knowledgeIs extracted (step S74). The processing here is as follows.
[0064]
For example, Japanese sentence ending expressionknowledgeIn the extraction process, “Affirmative form: is, is, is”, “Affirmative form: is, is”, “Negative form is not:” , “Negative Sul: Do not do, do not”, “Instruction affirmative: Do, do, let's do”, “Negative negation: Do, do, do, do not, do” "have.
[0065]
All the prescriptive suffixes are extracted from the result of Japanese morphological analysis, and the sentence end expression (prescriptive suffix) having the highest number of appearances is stored in WT1 from the above candidates. On the other hand, for the passive tendency and question tendency, the ratio of passive form and question sentence to the whole sentence is calculated from the Japanese parsing result. If the value is equal to or greater than the threshold value, it is determined that the tendency is strong. Is stored. In the punctuation mark expression, it is determined from the Japanese morpheme result whether “.,” Or “.,” Is used in the bilingual document, and the result is stored in WT1.
[0066]
Conjunction translation extracts the morphological analysis result of the English sentence in which the conjunction (for example, “if”) that is the target in the English morphological analysis and the bilingual Japanese sentence thereof, extracts the Japanese translation corresponding to “if”, and sends it to WT1. Store. For example, if a Japanese sentence corresponding to an English sentence including “if” contains “if if”, “if if” is stored, and if “to and” are included, “ Store "~". If there are a plurality, all of them are stored.
[0067]
FIG. 19 shows the grammar in this example.knowledgeIt is a figure which shows the example of WT1 after an extraction process.
Here we ’re going to use the field separatortabIn the first field, the attribute is shown, and in the second field, the attribute value (extraction result) is shown. For example, it can be seen that “Shi” is stored in the positive sul form (kout_da) of the Japanese sentence end expression. Note that “passive_tendency high” indicates that the passive tendency is strong, and “interactive low” indicates that the question tendency is low.
[0068]
After completing the grammar information extraction process, the value of WT1 is set to the grammar rule.Rule 4(Step S75). Grammar rulesRule 4Then, depending on the value of WT1, the subsequent translation process (stepS4) Is applied to change the grammar rules.
[0069]
  Next, the morphological frequency dictionary / translation pattern dictionary 23 automatically created by the vocabulary knowledge learning unit 21 is obtained by adding “.dic” to the bilingual document file name identifiers “.en” and “.ja”. As the dictionary name, the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 5As storage(Step S34). In the case of this specific example, the file name of the input parallel translation document is system. en, system. ja, the name of the dictionary to be stored is system. dic. Here, the document environment learning process (step S3) ends.
[0070]
Next, the translation process is started.
FIG. 8 shows a flowchart for explaining the operation of the translation process.
First, an English sentence to be translated “This command pushes modules on and / of pops modules off the stream.” Is input by the source language sentence input unit 12 (step S81).
[0071]
The translation engine unit 3 divides the input document into sentence units (step S82), and performs morpheme analysis processing (step S84), syntax / translation pattern analysis processing (step S85), and morpheme generation processing (step S86) for each sentence. The translation result is output by the target language sentence output unit 13 (step S87). When all the source language sentences are processed (step S83), the translation process is terminated.
[0072]
That is, in the source language sentence morpheme analysis process (step S84) and the target language sentence morpheme generation process (step S86), the value of the number of appearances of the morpheme extracted by the learning engine unit 2 (stored in the morpheme dictionary / grammar rule 4). On the other hand, in the syntax / translation pattern analysis process (step S85), the translation pattern extracted by the learning engine unit 2 and the value of the number of appearances (value of the translation pattern learning dictionary 6) are referred to. A translation result reflecting the expression of the bilingual document is obtained.
[0073]
FIG. 20 is a diagram illustrating an example of an original text input screen and an example of a translation result output screen in this specific example.
In the figure,Example of original text input screen(A)In the example of translation result output screenShown in (c). Also, (b) shows an example of a translation result output screen without learning in this input sentence. The learning effect can be read from the examples (b) and (c).
[0074]
<effect>
As described above, Example 1 has the following effects.
As shown in FIG. 18 and FIG. 17, the learning knowledge is in a format that is easy for the user to understand, such as the translation pattern and the number of morpheme occurrences, so that the user can easily manage the translation knowledge and also predict the translation result. .
● For example, from a learning document, system. When there are a plurality of learning documents such that dic is obtained, learning knowledge can be classified and managed according to each learning document, and can be reused.
● Morphological dictionaries and grammar rules used in the translation engine section 34 isMorphological dictionary and grammar rules used in the learning engine 2Rule 4Since the knowledge acquired by the learning engine unit 2 is added to the learning engine unit 2, and the language knowledge that is the core of each engine is common, side effects are unlikely to occur.
● The translation knowledge acquired by the learning engine 2 is the morpheme dictionary and grammar rules used in the translation process.Rule 4Since it is not rewritten, it can be easily restored to the original dictionary state.
The learning engine unit 2 not only acquires lexical knowledge such as correspondence between words in the source language and the target language from the bilingual document, but also knowledge obtained by referring to the entire bilingual document (for example, appearance frequency, etc.) ), And use the result for translation processing.
[0075]
<< Specific Example 2 >>
<Constitution>
FIG. 9 is a block diagram of a specific example 2 of the machine translation apparatus of the present invention.
[0076]
The apparatus of Example 2 is roughly divided into an input / output unit 100, a learning engine unit 200, an environment adaptive dictionary construction unit 300, a translation engine unit 400, a morpheme dictionary / grammar rule 500, and a bilingual document-specific morpheme frequency dictionary / translation. It consists of a pattern dictionary 600, a bilingual document-specific document environment setting file 700, and a translation pattern learning dictionary 800.
[0077]
The input / output unit 100 includes a bilingual document input unit 101, a source language sentence input unit 102, a target language sentence output unit 103, and a dictionary / environment setting file editing unit 104. Here, the bilingual document input unit 101, the source language sentence input unit 102, and the target language sentence output unit 103 are the same as the input / output unit 11 to the target language sentence output unit 13 in the specific example 1.
[0078]
The dictionary / environment setting file editing unit 104 is a functional unit for editing the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 600, the bilingual document-specific document environment setting file 700, and the translation pattern learning dictionary 800. Specifically, it comprises input / output devices such as a display and a keyboard.
[0079]
The learning engine unit 200 includes a vocabulary knowledge learning unit 201 and a document environment learning unit 202. The vocabulary knowledge learning unit 201 refers to the bilingual document input from the bilingual document input unit 101, refers to the morpheme dictionary / grammar rule 500, extracts the morpheme, the number of appearances thereof, the translation pattern thereof, and the number of appearances thereof, This extraction result is stored in the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 600. Further, the document environment learning unit 202 selects an optimum dictionary from the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 600 based on the knowledge extracted by the vocabulary knowledge learning unit 201, and extracts the optimum dictionary and the extracted knowledge. And the morphological frequency knowledge obtained as a result and the grammatical knowledge extracted from the bilingual document based on the grammatical rules in the morpheme dictionary / grammatical rule 500 are output as a bilingual document-specific document environment setting file 700. ing.
[0080]
The environment adaptive dictionary construction unit 300 creates a translation pattern learning dictionary 800 by referring to the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 600 and the bilingual document-specific environment setting file 700, and morphemesFrequency knowledgeAnd grammarknowledgeAnd create theseknowledgeIs added to the morpheme dictionary / grammar rule 500.
[0081]
The translation engine unit 400 is a functional unit that performs translation processing with reference to the morpheme dictionary / grammar rule 500 and the translation pattern learning dictionary 800.
[0082]
The dictionary / environment setting file editing unit 104 refers to or edits (adds, deletes, modifies) a bilingual document-specific morphological frequency dictionary / translation pattern dictionary 600, a bilingual document-specific document environment setting file 700, and a translation pattern learning dictionary 800. ), Specifically, an input / output device such as a display and a keyboard.
[0083]
<Operation>
There are three major differences between this example and Example 1. First, the document environment learning unit 202 of the learning engine unit 200 outputs the document environment setting file 700, and secondly, the environment adaptive dictionary construction unit 300 newly exists, and the morpheme frequency dictionary / translation pattern dictionary 600 Based on the document environment setting file 700, a morpheme dictionary / grammar rule 500 and a translation pattern learning dictionary 800 are constructed. Third, each dictionary 600, 800 acquired by learning and a document environment setting file 700 for each bilingual document are stored. The dictionary / environment setting file editing unit 104 can refer to and edit the file.
[0084]
FIG. 10 is a flowchart for explaining the operation of the second specific example.
Specific example 2 will be described using the same example as specific example 1.
First, the user translates bilingual document system. en, system. When ja is input (step S1001), the bilingual document is transferred to the vocabulary knowledge learning unit 201 of the learning engine unit 200, and lexical knowledge learning processing is performed (step S1002).
[0085]
The lexical knowledge learning process is based on the specific example 1, but the learning result is not held in the learning engine unit 2 as the morpheme frequency dictionary / translation pattern dictionary as in the specific example 1, but the bilingual document-specific morpheme frequency dictionary / It is stored in the translation pattern dictionary 600. The storage method conforms to the dictionary storage process (step S34) of the document environment learning process (step S3) of the first specific example.
[0086]
Next, a document environment learning process is performed (step S1003).
FIG. 11 is a flowchart for explaining the operation of the document environment learning unit 202 in the second specific example.
[0087]
First, optimal dictionary selection processing (step S1101), grammarknowledgeAn extraction process (step S1102) is performed. These processes also conform to the specific example 1. As a result, the result of the optimal dictionary selection process is WT0,knowledgeThe result of the extraction process is stored in WT1.
[0088]
The result of WT0 is combined into one value and stored in the bilingual document-specific document environment setting file 700 together with the result of WT1 (step S1103). The document environment setting file name is assumed to be obtained by removing the identifier of the file name of the bilingual document to be learned and adding “.env”.
[0089]
FIG. 21 shows an example of a document environment setting file according to this specific example.
As illustrated, in the second specific example, the optimum dictionary item use_dic and its values “os.dic, unix.dic, users_guide.dic” are added to the state illustrated in FIG. 19 of the first specific example.
[0090]
Next, environment adaptive dictionary construction processing is performed (step S1004).
FIG. 12 is a flowchart for explaining the operation of the environment adaptive dictionary construction unit 300 in the second specific example.
[0091]
First, after initialization of the work table WT0 (step S1201), the bilingual morpheme frequency dictionary / translation pattern dictionary file name set in the environment setting file 700 (here, the value of the attribute use_dic of system.env). Is stored in WT0 according to the same format as WT0 of specific example 1 (step S1202), and then a learning dictionary construction process (step S1203)knowledgeA setting process (step S1204) is performed. Learning dictionary construction process, grammarknowledgeThe setting process is in accordance with the first specific example.
[0092]
At this point, the user refers to / edits the bilingual document-specific environment setting file 700 (steps S1005 and S1006), and references / edits the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 600 and translation pattern learning dictionary 800 ( Steps S1007 and S1008) can be performed by the dictionary / environment setting file editing unit 104.
[0093]
After completion of the editing process, the translation process is performed in the same manner as in the specific example 1, and the translation result is output by the target language sentence output unit 103 (step S1009).
[0094]
<effect>
Specific Example 2 has the following effects in addition to Specific Example 1.
By providing a method in which the learning engine unit 200 and the translation engine unit 400 are clearly separated and the translation knowledge acquired by the learning engine unit 200 is disclosed to the user as a bilingual document environment setting file 700. Can see the result and predict the translation result.
[0095]
The user uses the dictionary / environment setting file editing unit 104 to convert the bilingual document-specific document environment setting file 700, the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary 600, and the translation pattern learning dictionary 800 based on the above prediction. You can change it according to your preference. That is, the translation knowledge acquired by learning can be processed to build a higher quality translation environment.
[0096]
In Specific Example 1, it was necessary to execute translation processing after inputting a bilingual document in order to learn the document environment. However, in Specific Example 2, the environment setting file can be stored in the same manner as the dictionary. Once an environment setting file is created, it is not necessary to input a bilingual document at the time of retranslation.
[0097]
<< Usage form >>
Specific examples 1 and 2 described above can be modified (or applied) as follows.
The learning engine unit 2,200 and the environment adaptive dictionary construction unit 300 in the specific example 2 can function separately from the translation engine unit 3,400. Further, only the learning engine unit 2,200, or the learning engine unit 2,200 and the environment adaptive dictionary construction unit 300 can be set as a translation knowledge learning apparatus.
[0098]
● Document Environment Learning Department22,As the grammar knowledge extracted by 202, Japanese sentence ending information and the like are mentioned, but other information such as translation of conjunctions other than if and tendency to stop speaking can also be extracted and used for translation processing.
[0099]
Various parameters described in the first specific example (for example, the weighting factor α and the threshold value y in the optimum dictionary selection process) can be set according to the user's preference by providing a parameter setting function by the user.
[0100]
● When a monolingual document is input instead of a bilingual document, a translation pattern dictionary cannot be constructed, but part of the morpheme frequency dictionary and document environment information can be learned. If the range is limited, a part of the apparatus functions even when inputting a monolingual document.
[0101]
● English-Japanese translation has been described as an example, but this device can be applied to both Japanese-English translation and bilingual languages other than English and Japanese.
[0102]
In specific example 2, if the translation environment setting file of the source language document to be translated exists in the bilingual document-specific document environment setting file 700, the process can be started from the environment adaptive dictionary construction process. In this case, the user does not input the bilingual document but inputs the environment setting file name. Thereby, the file name is passed to the environment adaptive dictionary construction unit 300, and the environment adaptation dictionary construction process is started.
[0103]
The editing of the dictionary and the environment setting file by the user is not limited to the operation shown in the flowchart of the specific example 2, and can be edited at any time.
[0104]
In Specific Example 2, the vocabulary knowledge learning process and the document environment learning process are processed in series, but can be performed in parallel.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a specific example 1 of a machine translation apparatus of the present invention.
FIG. 2 is a flowchart showing the overall operation of a specific example 1 of the machine translation apparatus of the present invention.
FIG. 3 is a flowchart showing an operation of a document environment learning process in the specific example 1 of the machine translation device of the invention.
FIG. 4 is a flowchart showing an operation of optimum dictionary selection processing in the specific example 1 of the machine translation device of the invention.
FIG. 5 is a flowchart (part 1) showing an operation of a learning dictionary construction process in the specific example 1 of the machine translation device of the invention.
FIG. 6 is a flowchart (part 2) showing the operation of the learning dictionary construction process in the specific example 1 of the machine translation device of the invention.
FIG. 7 shows a grammar in the first specific example of the machine translation apparatus of the present invention.knowledgeIt is a flowchart which shows operation | movement of an extraction process.
FIG. 8 is a flowchart showing an operation of translation processing in the specific example 1 of the machine translation apparatus of the present invention.
FIG. 9 is a configuration diagram of a specific example 2 of the machine translation device of the invention.
FIG. 10 is a flowchart showing the operation of a specific example 2 of the machine translation device of the invention.
FIG. 11 is a flowchart showing document environment learning processing in a specific example 2 of the machine translation device of the invention.
FIG. 12 is a flowchart showing environment adaptive dictionary construction processing in a specific example 2 of the machine translation device of the invention.
FIG. 13 is an explanatory diagram showing an example of a bilingual document in the machine translation device of the present invention.
FIG. 14 is an explanatory diagram of a morpheme frequency dictionary after lexical knowledge learning processing in the specific example 1 of the machine translation device of the invention.
FIG. 15 is an explanatory diagram of a translation pattern dictionary after lexical knowledge learning processing in the specific example 1 of the machine translation device of the invention.
FIG. 16 is an explanatory diagram of the work table WT0 after the optimum dictionary is selected in the specific example 1 of the machine translation device of the invention.
FIG. 17 is an explanatory diagram of an English morpheme dictionary after learning dictionary construction processing in the specific example 1 of the machine translation device of the invention.
FIG. 18 is an explanatory diagram of a translation pattern learning dictionary after learning dictionary construction processing in the specific example 1 of the machine translation device of the invention.
FIG. 19 shows a grammar in the first specific example of the machine translation apparatus of the present invention.knowledgeIt is explanatory drawing of the work table WT1 after extraction.
FIG. 20 is an explanatory diagram of a translation processing result in the specific example 1 of the machine translation device of the invention.
FIG. 21 is an explanatory diagram of an environment setting file in the specific example 2 of the machine translation device of the invention.
[Explanation of symbols]
2,200 Learning Engine Department
3,400 translation engine
4,500 Morphological Dictionary / Grammar Rules
5,600 Bilingual Document Morphological Frequency Dictionary / Translation Pattern Dictionary
6,800 translation pattern learning dictionary
11, 101 Bilingual document input section
15,600 Bilingual Document Morphological Frequency Dictionary / Translation Pattern Dictionary
21,201 Vocabulary Knowledge Learning Department
22, 202 Document Environment Learning Department
23 Morphological Frequency Dictionary / Translation Pattern Dictionary
104 Dictionary / environment setting file editor

Claims (8)

A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in natural language;
A vocabulary knowledge learning unit that extracts the morpheme of each of the source language and the target language and the number of appearances of the morpheme from the bilingual document input in the bilingual document input unit;
Based on the grammar rules in prearranged morpheme dictionary and grammar rules for use in the translation process, The rewritable generates a new grammar rules from a document input in the bilingual document input unit, in the lexical knowledge learning unit A document environment learning unit that obtains a morpheme frequency dictionary based on the extracted morpheme and the number of appearances of the morpheme and adds the morpheme frequency dictionary and a new grammar rule to the morpheme dictionary / grammar rule; Translation knowledge learning device. A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in natural language;
A lexical knowledge learning unit that extracts morphemes of each of the source language and the target language, the number of appearances of the morpheme and the number of appearances of the translation pattern, and the number of appearances of the translation pattern,
Based on the grammar rules in prearranged morpheme dictionary and grammar rules for use in the translation process, The rewritable generates a new grammar rules from a document input in the bilingual document input unit, in the lexical knowledge learning unit Based on the extracted morpheme, the number of appearances of the morpheme, the translation pattern, and the number of appearances of the translation pattern, a morpheme frequency dictionary and a translation pattern dictionary are obtained, and the morpheme frequency dictionary and a new grammar rule are used as the morpheme dictionary / grammar rule. And a document environment learning unit for registering the translation pattern dictionary in the translation pattern learning dictionary,
A translation engine that translates a source language document into a target language document by using the morpheme frequency dictionary , a grammatical dictionary / grammar rule with a new grammar rule added from the document environment learning unit, and the translation pattern learning dictionary A machine translation apparatus comprising: a section. A bilingual document input unit for inputting a bilingual document consisting of a source language document and a target language document in natural language, a translation morpheme dictionary including a large number of morphemes constituting a sentence, and a storage unit storing grammar rules; A translation knowledge learning apparatus comprising a translation pattern dictionary including a translation pattern in which a source language and a target language phrase correspond to each other,
The input source language document and the target language document are analyzed to be classified into morphemes based on the grammatical rules, and each morpheme is associated with the number of appearances indicating the number of occurrences of the morpheme in the document. A source language morpheme frequency dictionary and a target language morpheme frequency dictionary are generated, and a translation pattern dictionary in which the number of appearances indicating the number of occurrences of the translation pattern is associated with the translation pattern in which the phrases of both documents are associated Vocabulary knowledge learning department to
Another storage unit for storing the source language morpheme frequency dictionary, the target language morpheme frequency dictionary, and the translation pattern dictionary for each bilingual document;
The morpheme frequency dictionary of another bilingual document in which at least one of the generated source language morpheme frequency dictionary and the generated target language morpheme frequency dictionary has a predetermined number or more of morphemes in common among the other storage units And adding a weight count value obtained by adding a weighting coefficient to the number of appearances to each morpheme in each of the generated morpheme frequency dictionaries and the number of appearances to each translation pattern in the generated translation pattern dictionary A weight count value obtained by adding a weighting coefficient to the generated morpheme frequency dictionary, and the morpheme weight count value in each of the generated morpheme frequency dictionaries is smaller than the number of appearances of the same morpheme in each morpheme frequency dictionary of the other bilingual document. The morpheme dictionary for translation is updated with each morpheme in each of the generated morpheme frequency dictionaries, the number of appearances, and the number of weights after changing the weight number value to the number of appearances. If the weight count value of the translation pattern in the generated translation pattern dictionary is smaller than the appearance count of the same translation pattern in the translation pattern dictionary of the other parallel translation document, the weight count value is changed to the appearance count and the generation is performed. A translation knowledge learning apparatus, comprising: a translation environment learning unit that updates each translation pattern in each translation pattern dictionary and the number of appearances and the number of weights and updates the pattern dictionary for translation. A bilingual document input unit for inputting a bilingual document consisting of a source language document and a target language document in natural language, a translation morpheme dictionary including a large number of morphemes constituting a sentence, and a storage unit storing grammar rules; A machine translation device comprising a translation pattern dictionary including a translation pattern in which a source language and a target language phrase correspond to each other,
The input source language document and the target language document are analyzed to be classified into morphemes based on the grammatical rules, and each morpheme is associated with the number of appearances indicating the number of occurrences of the morpheme in the document. A source language morpheme frequency dictionary and a target language morpheme frequency dictionary are generated, and a translation pattern dictionary in which the number of appearances indicating the number of occurrences of the translation pattern is associated with the translation pattern in which the phrases of both documents are associated Vocabulary knowledge learning department to
Another storage unit for storing the source language morpheme frequency dictionary, the target language morpheme frequency dictionary, and the translation pattern dictionary for each bilingual document;
The morpheme frequency dictionary of another bilingual document in which at least one of the generated source language morpheme frequency dictionary and the generated target language morpheme frequency dictionary has a predetermined number or more of morphemes in common among the other storage units And adding a weight count value obtained by adding a weighting coefficient to the number of appearances to each morpheme in each of the generated morpheme frequency dictionaries and the number of appearances to each translation pattern in the generated translation pattern dictionary A weight count value obtained by adding a weighting coefficient to the generated morpheme frequency dictionary, and the morpheme weight count value in each of the generated morpheme frequency dictionaries is smaller than the number of appearances of the same morpheme in each morpheme frequency dictionary of the other bilingual document. The morpheme dictionary for translation is updated with each morpheme in each of the generated morpheme frequency dictionaries, the number of appearances, and the number of weights after changing the weight number value to the number of appearances. If the weight count value of the translation pattern in the generated translation pattern dictionary is smaller than the appearance count of the same translation pattern in the translation pattern dictionary of the other parallel translation document, the weight count value is changed to the appearance count and the generation is performed. A document environment learning unit for updating each translation pattern in each translation pattern dictionary and the translation pattern dictionary with the appearance count and the weight count value;
When a source language document to be translated is input, the translation pattern dictionary, the translation morpheme dictionary, and a translation engine unit that translates the document to be translated into a target language document based on the grammar rules A machine translation device characterized by that. A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in natural language;
A morpheme dictionary / grammar rule part indicating morphological analysis information and grammar rules provided in advance for use in translation processing from the source language to the target language;
A bilingual document-specific morpheme frequency dictionary / translation pattern dictionary showing the number of appearances of morphemes by bilingual document and translation patterns;
Using the morpheme dictionary / grammar rules, the morpheme, the number of appearances thereof, the translation pattern thereof, and the number of appearances thereof are extracted from the bilingual document, and the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary is extracted from the extracted information as dictionary information. Vocabulary knowledge learning section to output to
From the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary, an optimal dictionary is selected based on the extracted information , the optimal dictionary and the extracted information are merged, and the resulting morpheme frequency dictionary A document environment learning unit that outputs a grammatical rule extracted from the bilingual document based on the grammatical rule in the morpheme dictionary / grammar rule as a bilingual document-specific document environment setting file;
Wherein based on the translated document by morpheme frequency dictionary translation pattern dictionary and bilingual document by document configuration file, environmental adaptation output as additional information into a form Taimoto dictionary and grammar rules by creating a morpheme frequency dictionary and grammar rules A translation knowledge learning apparatus comprising a dictionary construction unit. The translation knowledge learning apparatus according to claim 5 ,
A translation knowledge learning apparatus comprising a dictionary / environment setting file editing unit for editing a bilingual document-specific morpheme frequency dictionary / translation pattern dictionary and a bilingual document-specific document environment setting file. A bilingual document input unit for inputting a bilingual document composed of a source language and a target language in natural language;
A morpheme dictionary / grammar rule part indicating morphological analysis information and grammar rules provided in advance for use in translation processing from the source language to the target language;
A bilingual document-specific morpheme frequency dictionary / translation pattern dictionary showing the number of appearances of morphemes by bilingual document and translation patterns;
Using the morpheme dictionary / grammar rules, the morpheme, the number of appearances thereof, the translation pattern thereof, and the number of appearances thereof are extracted from the bilingual document, and the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary is extracted from the extracted information as dictionary information. Vocabulary knowledge learning section to output to
From the bilingual document-specific morpheme frequency dictionary / translation pattern dictionary, an optimal dictionary is selected based on the extracted information , the optimal dictionary and the extracted information are merged, and the resulting morpheme frequency dictionary A document environment learning unit that outputs a grammatical rule extracted from the bilingual document based on the grammatical rule in the morpheme dictionary / grammar rule as a bilingual document-specific document environment setting file;
Together with the basis on the bilingual document by morpheme frequency dictionary translation pattern dictionary and bilingual document by document configuration file, and outputs the additional information to create and form Taimoto dictionary and grammar rules and a morpheme frequency dictionary and grammar rules, create a translation pattern dictionary, and environmental adaptation dictionary building unit for outputting the translation pattern dictionary as a translation pattern learning dictionary,
The morpheme dictionary and grammar rules in which the morpheme frequency dictionary and grammar rules are added by the environment adaptive dictionary construction unit, and a translation engine unit that translates a source language document into a target language document using the translation pattern learning dictionary; A machine translation device comprising: The machine translation device according to claim 7 ,
A machine translation apparatus comprising a dictionary / environment setting file editing unit for editing a bilingual document-specific morpheme frequency dictionary / translation pattern dictionary, a bilingual document-specific document environment setting file, and a translation pattern learning dictionary.

Images