JP2012053858A - Machine translation device and machine translation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To create a translation in which the terms of a relevant technical field are standardized.SOLUTION: The morphemic analysis of a document in first language is performed by using translation dictionary information of a translation dictionary section and parallel translation information of a technical term dictionary section, attribute information of morphemes and translation word information are acquired as analytic information, and a translation is created based on the translation word information. In this case, when syntax analysis fails due to the use of the entry words of a technical term dictionary, the entry words are rejected and the translation is obtained. When the translation is obtained by rejecting the entry words of the technical term dictionary, it is determined whether or not any of the translation word candidates of the entry words in the first language divided into word units is matched with the translation words of the entry words divided into word units, and when there is any matched translation word candidate, the section which is equivalent to the translation words of the entry words in the translation obtained by document analyzing means is replaced with the translation word candidate.

Description

  Embodiments described herein relate generally to a natural language processing technique, and more particularly, to a machine translation apparatus and a machine translation program that process a natural language document and translate a first language into a second language.

  In recent years when machine translation has become popular, the level of quality expected of machine translation has also increased. As one of these efforts, there is a translation that suits each field. In machine translation processing, in order to obtain translation results using terms in the field of the document to be translated, in addition to the standard dictionary used regardless of the field, a technical term dictionary in the corresponding field or registered as required by the user Translation is performed using a user dictionary constructed in the same way. The importance of such terminology dictionaries is emphasized in various areas. This is because human translators do not have all the knowledge of translations related to vast terminology, and if human translators translate patents without knowing the terminology, they will mistranslate the terminology Because there are things.

  On the other hand, technical term dictionaries have abundant headings, but some character strings that can be interpreted multiple times become headwords (usually compound words). By registering, the interpretation of the morpheme is given priority, the syntax analysis may fail, and the analysis result may be deteriorated. Particularly in English, since there are many words having the same form and having a plurality of parts of speech, such a phenomenon is likely to occur when the first language is English. In such a case, conventionally, one of the following methods has been used.

  The first method is to identify a headword that causes a failure of analysis in a sentence that fails to be analyzed by visual check, and to reject the headword and translate it. This creates the possibility of successful analysis. Typically, a headword composed of a plurality of words is individually divided and reanalyzed.

  The second method is to specify multiple technical term dictionaries, and if the analysis fails with the highest priority dictionary, the entry word that failed to be analyzed is registered with a different part of speech in the lower priority dictionary. Search in order. If there is a headword that has failed in the analysis, the translation is performed using the part of speech, and if successful, a translated sentence is generated using a translation of a dictionary with a low priority. On the other hand, if such a candidate is not found, the headword part is treated as an unknown word, and the character string as it is in the first language is inserted into the translated sentence.

  In order to avoid an analysis failure due to a specific headword (morpheme), a plurality of dictionaries are switched in this way. Specifically, the translation dictionary used in the order of the conversion history of the currently selected technical term dictionary, the currently selected technical term dictionary, and the technical term dictionary with the next highest priority is automatically switched.

  However, in the first method, it is not easy to specify a headword that causes a failure in analysis in machine translation because at least knowledge of the first language is required. The tendency of failure in machine translation differs from the tendency of failure in human translation, and detection requires a certain skill. Also, assuming that such a headword can be identified, if the headword is registered as at least rejected in the same document, the sentence that appears later in the document is analyzed with the headword and part of speech. Will not be used even if it succeeds. On the other hand, if not registered, if the headword and the part of speech fail in a sentence that appears later in the document, it is necessary to issue a rejection command in the same manner, resulting in a waste of work.

  Moreover, even if the analysis is successful, if it is not given priority, it will not be a translation in the field of the selected specialized dictionary, and instead a translation from the standard dictionary will be adopted, resulting in insufficient term unification. In other words, whether or not technical terms are used depends on whether or not the analysis fails. In the worst case, the user remains in the first language, and suffers from understanding the sentence for the user who understands only the second language.

Japanese Patent No. 2807586 JP 2000-3364 A

  Can avoid the failure of analysis that occurs when the selected technical term dictionary is simply applied, and can create a translation that uses the technical term dictionary as much as possible to unify the terms in the relevant technical field. A machine translation apparatus and a machine translation program are provided.

  A machine translation device according to an embodiment includes a translation dictionary unit that stores parallel translation information of a first language and a second language in a word unit in a storage device, a headword of a technical term of a first language composed of a plurality of word strings, and A technical term dictionary unit that stores translated words as parallel translation information in a storage device. The technical term adjustment means performs a syntax analysis using the translation dictionary information of the translation dictionary unit and the parallel translation information of the technical term dictionary unit, and when the syntax analysis fails due to a headword in the technical term dictionary, The headword and its translation are dissected into words and a dictionary is searched to determine whether any of the candidate translations of the headword divided into words corresponds to the translation divided into words. If there is something that is translated, the portion corresponding to the translation of the headword in the obtained translation is replaced with the translation candidate.

The functional block block diagram of an example of the machine translation apparatus which concerns on embodiment. The block block diagram which shows the hardware constitutions of the machine translation apparatus which concerns on embodiment. Explanatory drawing of an example of the 1st language document used as translation object. The flowchart which shows an example of the processing content of the machine translation apparatus which concerns on embodiment of FIG. Explanatory drawing which shows an example of the morpheme analysis information after performing the process of step S1-step S8 of FIG. 4 with respect to the sentence of the sentence number 2 of FIG. Explanatory drawing which shows an example of the analysis result at the time of performing a syntax analysis with respect to the sentence of the sentence number 2 of FIG. Functional block configuration diagram of another example of the machine translation device according to the embodiment The flowchart which shows an example of the processing content of the list creation of the translation word candidate of the headword in the sub technical term creation means in embodiment. The flowchart which shows an example of the processing content of the machine translation apparatus which concerns on embodiment of FIG. The functional block block diagram of another example of the machine translation apparatus which concerns on embodiment. Explanatory drawing of the syntax analysis failure environment knowledge stored in the syntax analysis failure environment knowledge part in embodiment. The flowchart which shows an example of the processing content of the syntax analysis failure cause determination means in embodiment. Explanatory drawing which shows an example of the corpus search result in embodiment. The functional block block diagram of another example of the machine translation apparatus which concerns on embodiment. Explanatory drawing of an example of the screen which displayed the reference information by the reference information display means in embodiment.

  Hereinafter, embodiments will be described with reference to the drawings. FIG. 1 is a functional block configuration diagram of an example of a machine translation apparatus according to the embodiment, and FIG. 2 is a block configuration diagram illustrating a hardware configuration of the machine translation apparatus according to the embodiment.

  In FIG. 2, the machine translation device 11 is realized by installing a software program such as a machine translation program in a general computer and executing the software program in the processor 13 of the arithmetic control device 12.

  The arithmetic and control unit 12 performs various calculations related to machine translation. The arithmetic and control unit 12 includes a processor 13 and a memory 14. The memory 14 stores a machine translation / proofreading support program 15. The work area 16 is used when processing is executed. The calculation results of the calculation control device 12 are displayed / output on the display device 17 which is an output device. Although the display device 17 is shown as an output device, the output device is not limited to the display device 17, but is a printing device such as a printing machine, an output device to a computer-readable medium such as a magnetic tape, a magnetic disk, or an optical disk. It is also possible to employ a transmission device that transmits a document to other media.

  The input device 18 is used to input information to the arithmetic control device 12, and is composed of, for example, a mouse 19, a keyboard 20, and a disk drive 21, and also an OCR (optical character reader), magnetic tape, magnetic disk, It is also possible to employ a reading device from a computer-readable medium such as an optical disk.

  For example, the mouse 19 and the keyboard 20 as the input device 18 input various commands to the arithmetic control device 12 via the display device 17, and the keyboard 20 and the disk drive 21 input a document to be supported for machine translation / proofreading. That is, the disk drive 21 inputs / outputs a file of a machine translation / proofreading support target document to / from a storage medium. Further, a hard disk drive (HDD) 22 is provided for storing a calculation result of the calculation control device 12 and a translation dictionary storing knowledge and rules necessary for machine translation / proofreading support.

  In FIG. 1, each functional block in the arithmetic and control unit 12 shown in FIG. 1 corresponds to each program constituting the above-described machine translation program. That is, when the processor 13 executes each program constituting the machine translation program, the arithmetic control device 12 functions as each functional block. Each block of the storage device 25 corresponds to a storage area of the memory 14 and the hard disk drive 22 in the arithmetic control device 12.

  The input device 18 inputs electronic data of a first language document to be translated, and inputs electronic data of the first language document to be translated based on a user input operation. The first language document to be translated input by the input device 18 is input and captured by the input processing unit 23 of the arithmetic control device 12 and stored in the document storage unit 26 of the storage device 25 via the control unit 24. Is done. The input device 18 gives various commands to the control unit 24 via the input processing unit 23. The control unit 24 controls the input processing unit 23, the output processing unit 33, the document analysis unit 28, and the technical term adjustment unit 30, and also controls data exchange with the storage device 25.

  The storage device 25 stores a translation dictionary unit 27 and a technical term dictionary unit 34. The translation dictionary unit 27 stores various dictionary data used when the document analysis unit 28 analyzes the first language document to be translated, a dictionary for translating from the first language to the second language, And a dictionary for translating from the second language to the first language.

  That is, the dictionary for translating from the first language to the second language is a first language utilization change dictionary 27a for converting a word / idiom having a change in the ending to its original form, for analyzing the first language. The first language analysis grammar dictionary 27b storing the grammar of the first language, the first language word / idiom dictionary 27c in which the translation of the second language corresponding to the first language word / idiom together with the part of speech information is stored, The first language conversion grammar dictionary 27d storing the conversion information to the second language, the second language generation grammar dictionary 27e for determining the structure of the sentence in the second language, and further changing the word form such as the ending and completing the translated sentence The second language morpheme generation grammar dictionary 27f to be stored is stored.

  In addition to them, the second language utilization change dictionary 27h, the second language analysis grammar dictionary 27i in which the grammar for analyzing the second language is stored, and the translation of the first language corresponding to the second language word / idiom, The second language word / idiom dictionary 27j stored together with the part of speech information, the second language conversion grammar dictionary 27k stored with the conversion information from the second language to the first language, and the first language sentence structure are determined. A one-language generation grammar dictionary 27l and a first language morpheme generation grammar dictionary 27m for changing a word form such as a ending and completing a translated sentence are stored.

  The technical term dictionary unit 34 stores a technical term dictionary of a first language composed of a plurality of word strings and a technical term dictionary using the translated words as parallel translation information.

  The document analysis means 28 is input to the document storage unit 26 by the input device 18 using the translation dictionary information of the translation dictionary unit 27 of the storage device 25 and the parallel translation information of the technical term dictionary unit 34 in accordance with an instruction from the control unit 24. The morpheme analysis of the stored first language document to be translated is performed, and the attribute information and the translated word information of the morpheme are stored in the analysis information storage unit 29 of the storage device 25 as analysis information.

  At the same time, when the syntax analysis is performed and the syntax analysis is successful, a translation is generated based on the translation information of the analysis information. On the other hand, if syntax analysis fails without using the technical term dictionary entry word in the technical term dictionary unit 34, a translation is generated in an incomplete state, and the technical term dictionary entry word in the technical term dictionary unit 34 is used. If the parsing fails, the headline is rejected and a translation is obtained. Moreover, the analysis result and the translation of the document analysis means 28 are displayed / output on the display device 17 via the control unit 24 and the output processing unit 33 as necessary.

  The technical term adjustment means 30 detects a headword of the technical term dictionary that causes the failure when the parsing fails, decomposes the headword and its translation into a word unit, and between the first language and the second language To extract the word level correspondence. In other words, when the document analysis unit 28 obtains a translation by rejecting the headword in the technical term dictionary unit 34, the technical term adjustment unit 30 decomposes the headword and the translation into words and performs a dictionary lookup. It is determined whether any of the translation candidates of the first word headword divided into units matches the translation of the headword divided into word units. If there is a match, the part corresponding to the translation of the headword in the translation obtained by the document analysis means 30 is replaced with the translation candidate. The result is displayed / output on the display device 17 via the control unit 24 and the output processing unit 33 as necessary.

  The output processing unit 33 outputs the bilingual document of the first language document, the analysis result, the syntax analysis failure location, and the new translation result after the technical term adjustment to the display device 17 supplied via the control unit 24. Thus, the translation information screen is displayed on the display screen of the display device 17. The output processing unit 33 displays responses from the control unit 24 to various commands to the control unit 24.

  In the following description, the case where English is the first language and Japanese is the second language will be described as an example. FIG. 3 is an explanatory diagram of an example of a first language document to be translated used in the following description. Sentence numbers 1 to 3 are examples of sentences in the first language to be translated, sentence numbers. 4 is an example of a sentence in the first language used in later explanation. Since these sentences are examples for explanation, the sentences are not related to each other.

  In addition, it is assumed that “chemical dictionary” is selected as the technical term dictionary used in the following description. In this technical vocabulary dictionary, the sentence like sentence 1 in FIG. 3 is stored in mind, and the headword “mixed solution” is registered as “noun” as part of speech and “mixed solution” as translation. It shall be. Here, in order to simplify the explanation, it is assumed that there is one technical term dictionary other than the standard dictionary and nothing is registered in the user dictionary. In addition, in terms of using compound word registration content as much as possible, it can be applied not only to one technical term dictionary but also to a user dictionary. It may be implemented as including a dictionary.

  FIG. 4 is a flowchart showing an example of processing contents of the machine translation apparatus according to the embodiment of FIG. First, the control part 24 memorize | stores the 1st language document used as the translation object input via the input process part 23 from the input device 18 (S1). That is, the first language to be translated is stored in the document storage unit 26.

  Next, the control unit 24 activates the document analysis unit 28. When the document analysis unit 28 is activated, the document 24 is read from the document storage unit 26 and the technical term dictionary of the technical term dictionary unit 34 is read out. Each sentence of the language document is divided into morphemes, and attribute information such as part of speech is obtained (S2). This is to obtain syntactic features of each sentence of the first language document to be translated. In step S2, the dictionary for translation from the first language to the second language of the translation dictionary unit 27, specifically, the collation of the first language utilization change dictionary 27a and the first language analysis grammar dictionary 27b, Part of speech, original form, and attribute are assigned to each word, and a sentence structure (dependency relationship) indicating what relationship each morpheme has is obtained.

  Next, the document analysis means 28 uses the first language word / idiom dictionary 27c to obtain translation information defined in the translation dictionary unit 27 for each morpheme (S3). Further, syntax analysis is performed (S4), and it is determined whether the syntax analysis is successful (S5).

  If the syntax analysis is successful, the document analysis unit 28 converts the structure into the second language using the first language conversion grammar dictionary 27d, the second language generation grammar dictionary 27e, and the second language morpheme generation grammar dictionary 27f. The morpheme is generated to obtain a translation (S6).

  On the other hand, when the document analysis unit 28 determines that the result of the determination in step S5 is not successful (failure), it is possible whether the cause of the failure is a word registered in the technical term dictionary. In order to investigate the nature, it is checked whether or not a headword derived from the technical term dictionary is used in the morphological analysis result (S7). When a headword derived from the technical term dictionary is not used, a translation that can be generated in an incomplete state in which parsing has failed is obtained (S6). On the other hand, when a headword derived from the technical term dictionary is used, the headword is rejected in the morphological analysis (S8). Then, the process returns to step S4 and the syntax analysis is performed again.

  When there are a plurality of problematic morphemes, all of them are gradually deleted from the candidates instead of being deleted from the candidates all at once. In other words, morphemes that have no problem are used as much as possible. Now, if it is determined again as a failure in the determination in step S5, a translation that can be generated in an incomplete state in which parsing has failed is obtained (S6).

  Here, the specific example of the process of step S1-step S8 so far is demonstrated. The morpheme analysis information, the translation information, the dependency relationship, and the translation result obtained by the processes in steps S1 to S8 are stored in the analysis information storage unit 29 by the control unit 24. FIG. 5 shows an example of morpheme analysis information stored in the analysis information storage unit 29 in the case of sentence number 2 in FIG.

  FIG. 5 is an explanatory diagram showing an example of morpheme analysis information after the processing of step S1 to step S8 in FIG. 4 is performed on the sentence with sentence number 2 in FIG. As shown in FIG. 5, the sentence of sentence number 2 is divided into 10 morphemes and has items of English part of speech, original form, inflection form, Japanese translation, Japanese part of speech, and source dictionary name. When a plurality of Japanese translations are possible, they are stored in descending order of frequency. For example, it can be seen that two translations of “solution” and “solution” are stored in solution. For translations other than solution, only one translation is listed for simplicity of explanation.

  In the sentence of sentence number 2, the morpheme sequence that is first adopted for the syntax analysis includes the headword of the technical term dictionary (here, “mixed solution” of number 10). That is, 1-2-3-4-5-6-10. When this is expressed as an arrangement of parts of speech, it becomes “article (1) + noun (2) + verb (3) + preposition (4) + article (5) + adverb (6) + noun (10)”.

  When parsing, “article (1) + noun (2)” can be summarized as a noun phrase. This noun phrase can be connected if it is taken as the subject of the next verb (3). Also, verb (3) + preposition (4) can be connected. It is expected that a noun phrase follows the preposition (4). Since the part of speech at the beginning of a noun phrase can be an article, the following article (5) is not a problem. Although the remaining two morphemes are expected to form a noun phrase, adverbs cannot modify nouns. Therefore, “adverb (6) + noun (10)” is not recognized as a noun phrase. In this way, the role that these two morphemes play in the sentence is unknown in the analysis, and as a result, the parsing fails.

  If the technical term dictionary entry (“mixed solution”) is included, the syntax analysis fails, so the “mixed solution” of the noun (10) is excluded from the candidates (S7, S8), and the syntax analysis is performed again. (S4).

  The morpheme sequence used for this second parsing is 1-2-4-5-6-7-9 or 1-2-3-4-5-6-8-9 in number. . When expressed in parts of speech, “article (1) + noun (2) + verb (3) + preposition (4) + article (5) + adverb (6) + verb past tense (7) + noun (9)”, Or, “article (1) + noun (2) + verb (3) + preposition (4) + article (5) + adverb (6) + verb past participle form (8) + noun (9)”.

  Focusing on the subsequent part of article (5), which was impossible to analyze in the first candidate, the former “adverb + verb past tense + noun” cannot constitute a noun phrase by the English grammar, but the latter “ Adverb + verb past participle + noun can constitute a noun phrase, so "article (1) + noun (2) + verb (3) + preposition (4) + article (5) + adverb (6) + verb The past participle form (8) + noun (9) "can be adopted to succeed in the parsing.

  FIG. 6 is an explanatory diagram showing an example of the analysis result when the syntax analysis is performed on the sentence with sentence number 2 in FIG. As shown in FIG. 6, the parsing succeeds when the morpheme string is 1-2-3-4-5-6-8-9. As a result, “the reaction continues in a carefully mixed solution” is obtained as the translated sentence of sentence number 2 using the information shown in FIG.

  Next, it is determined whether the translation is obtained in step S5 due to rejection of the technical term (S9). If it is not due to rejection of technical terms, the process is terminated. If it is due to rejection of the technical terms, adjustment of the translated sentence “Reaction continues in a carefully mixed solution” obtained in step S6 is performed by the following processing.

  The control unit 24 activates the technical term adjustment means 30. When the technical term adjustment means 30 is activated, the headword and the translation are each divided into morphemes to obtain attribute information such as part of speech (S10).

  In the case of sentence No. 2, the headword in question is “mixed solution” (translation “mixed solution”, part of speech “noun”). Using the first language word / idiom dictionary 27c and the second language word / idiom dictionary 27j, dictionary lookup is performed for each morpheme of the headword and its translation word to obtain a translation (S11). That is, the headword and the translated word are decomposed into word units to perform dictionary lookup.

  The “mixed solution” in the headword part can be decomposed into mixed (<mix) and solution as shown in FIG. 5, and there are “mixed” and “solution, solution” as translations, respectively. On the other hand, when the translated word “mixed solution” is analyzed with a compound word consisting of “mixed” and “solution” and morphological analysis, it is determined that “mixed” is a sane noun and “solution” is a noun. The second language word / idiom dictionary 27j has "mix, blend, mingle" as a verb translation of "mix", mixing, mixture as a noun translation, and "solution" as a translation of "solution" And

  Here, in the above explanation, it was determined that “mixed” corresponds to “mixed” based on the fact that “mix” is in the English translation of “mixed”. You can take a look. That is, it can be determined that “mixed” corresponds to “mixed” even when the first translation of the standard dictionary mix is “mixed” and the second translated word or later is “mixed”. As a result, it can be seen that “mixed” corresponds to “mixed”, and the “mixed” and “mixed” can be derived from the nature of the weird variable noun.

  When the dictionary lookup in step S11 ends, it is determined whether each morpheme of the first language entry word matches one of the translation word candidates of the dictionary lookup result of the corresponding second language translation word in terms of the content word. (S12).

  As a result of this determination, the verb “mix” is included in the translation candidate for the first language of “mixed”, and the noun “solution” is included in the translation candidate for the first language for “solution”. , You can see that “solution” is in the second language translation of solution, so “mixed” and “mixed”, “solution” and “solution” correspond at the word level. Recognize.

  Therefore, the translation part of the problematic word string in the translation generated by the translation of the standard dictionary is replaced with these terms. That is, in the translated text obtained in step S6, the translated word is replaced with respect to the part that has been matched in step S12 (S13).

  This means that in “the reaction continues in a carefully mixed solution”, the “mixed solution” corresponding to the “mixed solution” is replaced with the “mixed solution”. As a final translation, the process ends with "The reaction continues in a carefully mixed solution." In this way, a translation using terms that match the selected chemical field is generated.

  In the above description, the case of the sentence with sentence number 2 in FIG. 3 was seen. Similarly, the sentence with sentence number 3 also causes a failure in parsing by adopting “mixed solution” as a noun. In the case of the sentence with sentence number 3, mixed functions as the main verb of the sentence.

  On the other hand, the sentence with sentence number 4 is a sentence from a document outside the chemical field. Therefore, the “chemical terminology dictionary” is not selected. In this sentence, “mixed solution” is a case where parsing succeeds even if it is treated as a noun, and solution should be interpreted as meaning “solution, solution”. Thus, sentence number 4 does not require adjustment of the translation in the chemical terminology dictionary, so the translation of the standard dictionary is used for solution and the solution is “solution”. 1 to 3 can be adjusted to “solution” based on the information in the “chemical terminology dictionary”, so that a translation corresponding to the field can be output.

  Next, another example of the machine translation apparatus according to the embodiment will be described. FIG. 7 is a functional block configuration diagram of another example of the machine translation device according to the embodiment. In another example, a sub technical term creating means 31 and a sub technical term dictionary unit 32 are added to the example shown in FIG. The same elements as those in FIG. 1 are denoted by the same reference numerals, and redundant description is omitted.

  The sub-technical term creating means 31 creates a list of candidate translations of the first language headwords that have been decomposed into units of words for all headwords in the technical term dictionary unit 34. Stores a list of candidate translations of headwords created by the sub-technical term creation means 31.

  FIG. 8 is a flowchart showing an example of processing contents for creating a list of candidate translations of headwords in the sub technical term creating means 31. As shown in FIG. 8, the sub-technical term creating means 31 performs the same processing (C1, C2,...) As shown in FIG. 8 for all headword / translation pairs of the selected technical term dictionary. C3) is performed.

  That is, the headword and the translation are each divided into morphemes to obtain attribute information such as part of speech (C1). Next, using the first language word / idiom dictionary 27c and the second language word / idiom dictionary 27j of the translation dictionary unit 27, dictionary lookup is performed for each morpheme of the headword and its translation word to obtain a translation (C2 ). Then, it is determined whether each morpheme of the first language entry word matches with any one of the translation word candidates of the dictionary lookup result of the corresponding morpheme of the second language translation word (C3).

  As a result of the determination, the matched pair is stored in the sub-terminology dictionary unit 32 together with the part of speech (C4). Then, it is determined whether or not there is a headword (C5). If there is a headword, the process returns to step C1 and the processes of steps C1 to C5 are repeated. As a result, a list of words in the first language, their parts of speech, and the corresponding translated words is created for the headword and stored in the sub-terminology dictionary unit 32. The list stored in the sub-terminology dictionary unit 32 can be regarded as a complementary dictionary that supplements the original technical term dictionary unit 34.

  FIG. 9 is a flowchart showing an example of processing contents of the machine translation apparatus according to the embodiment of FIG. Steps S14 to S17, which are new processes, are added to the processing contents shown in FIG. 4 after it is determined NO in step S12.

  That is, the technical term adjustment means 30 rejects, when the content word does not match any of the translation word candidates of the dictionary lookup result of the morpheme of the second language translation corresponding to each morpheme of the first language entry word. It is determined whether or not the words / parts of speech that have been successfully parsed in the headword equivalent portion are registered in the regular technical term dictionary (S14). In other words, if any of the translation candidates of the first language entry word divided into word units does not match the translation of the entry word divided into word units, the translation of the first language entry word decomposed into word units Is in the technical term dictionary unit 34.

  When it is registered in the technical term dictionary, the translated word is replaced for the part obtained in step S14 in the translated sentence obtained in step S6 (S15). That is, the portion corresponding to the word-by-word translation of the entry word in the translation obtained by the document analysis unit 28 is replaced with the translation of the technical term dictionary unit 34.

  On the other hand, if it is determined in step S14 that it is not registered in the technical term dictionary, it is determined whether a word / part of speech that has been successfully parsed in the rejected headword equivalent part is registered in the sub-terminology dictionary ( S16). That is, it is determined whether or not the translation of the first language headword decomposed into word units is in the sub-terminology dictionary unit 32.

  If it is registered in the sub-terminology dictionary, the translated word is replaced for the part obtained in step S16 in the translated sentence obtained in step S6 (S17). That is, the part corresponding to the word-by-word translation of the entry word in the translation obtained by the document analysis unit 28 is replaced with the translation of the sub-terminology dictionary unit 32.

  Now, the field of the technical term dictionary selected from the technical term dictionary unit 34 is “computer”, and “These formulas, including both physical and virtual channels, are also permitted” is used as an example sentence to be translated. In this computer terminology dictionary, the headword “virtual channel” is registered as a translated word “VC” and a part of speech “noun”.

  If the headline “virtual channel” is preferentially used for syntax analysis, the syntax analysis will fail. Focusing on the “physical and virtual channels” that are constituent elements, the plural “virtual channels” are treated as nouns, and they are called physical (adjective) + and (connective) + virtual channels (noun). This is because the part-of-speech string cannot be a constituent element in English, and therefore there is a contradiction.

  If the heading word “virtual channel” is not a noun and the word “virtual channel” is “adjective + noun” at the word level, then [[adjective + conjunction + adjective] + noun] is obtained and the parsing is successful. In order to adjust the translation of the technical term dictionary, it is assumed that the correspondence relationship between the “virtual channel” and the VC is taken in step S12 of FIG.

  This is because “virtual channel” consists of two words, but VC in Japanese can be regarded as consisting of V and C, but V and C itself have no meaning. On the other hand, other registered words of this computer dictionary include “virtual domain” (translation “virtual domain”, part of speech “noun”) and “virtual memory” (translation “virtual memory”, part of speech “noun”). Then, the headword virtual, the translation “virtual”, and the part of speech “adjective” are acquired in the process of step C1 to step C5 in FIG. 8 and stored in the sub-terminology dictionary of the computer.

  First, in step S14 of FIG. 9, it is determined whether or not a word / part of speech that has been successfully parsed in the rejected headword equivalent portion is registered in a regular technical term dictionary. In this case, if “channel” of “virtual channel” is registered in the regular technical term dictionary with the translated word “channel” and the part of speech “noun”, the translated word of the above-mentioned example channel is “channel” in step S15. "

  On the other hand, if virtual is not registered in the regular technical term dictionary, the process proceeds to step S16, and this time, the sub technical term dictionary is searched. Then, as described in the preconditions, it can be seen that virtual is registered in the sub-terminology dictionary, and the translation is “virtual” (S17). In this way, even if “virtual” and “virtual” are registered as translations of virtual (adjective) and “channel” is registered as a translation of channel (noun) in the standard dictionary, steps S14 to S14 are performed. Through the processing in step S17, a translation using computer technical terms can be obtained.

  Next, another example of the machine translation apparatus according to the embodiment will be described. FIG. 10 is a functional block configuration diagram of another example of the machine translation apparatus according to the embodiment. Another example of this is obtained by adding a syntax analysis failure cause determination means 35 and a syntax analysis failure environment knowledge storage unit 36 to the other example shown in FIG. The same elements as those in FIG.

  The syntax analysis failure cause determination means 35 searches the cause of the failure in the syntax analysis in step S5 of FIG. 4 by the document analysis means 28. The parsing failure environmental knowledge section 36 stores environmental knowledge in which parsing fails with a sequence of parts of speech of words in the first language. The syntax analysis failure cause determination means 35 is based on the registered part-of-speech of the headword that has failed in the syntax analysis and the constituent part-of-speech in which the parts of speech of the plurality of words constituting the headword are arranged. Determine the cause of parsing failure by referring to environmental knowledge.

  FIG. 11 is an explanatory diagram of syntax analysis failure environment knowledge stored in the syntax analysis failure environment knowledge unit 36. The syntax analysis failure environment knowledge is composed of the entry words “component part of speech”, “registered part of speech”, “environment in which syntax analysis fails”, and “part of speech string with registered part of speech”.

  For example, in Type 1, the headword is composed of two word strings, the part of speech of the two words is “adjective + noun”, and the registered part of speech is registered as “noun” as one headword. Show. In this case, when the headword is handled as one morpheme (one word), the part of speech is handled as a “noun”. Therefore, as shown in "Environment where parsing fails", the sequence of words such as "adjective and / or adjective noun" is originally a correct part-of-speech sequence, but "adjective noun" is a "noun" of one morpheme. As shown in “Part of speech string with registered part of speech”, it becomes “adjective and noun”, resulting in a part of speech string that fails parsing.

  The syntax analysis failure cause determination means 35 determines the cause of the failure in the syntax analysis based on such syntax analysis failure environment knowledge. For example, in the case of “adjective and noun” shown in “part of speech with registered part of speech”, if the parsing fails, the constituent part of speech of the headword “adjective + noun” will be used instead of “noun” which is the registered part of speech of the headword ”And parse again. Then, the "adjective and / or adjective nouns" shown in "Environment where parsing fails" will succeed, and the parsing will succeed. Thereby, it is determined that the cause of the syntax analysis failure is a headword in the technical term dictionary.

  An example in which syntax analysis fails for each type 1 to 6 of syntax analysis failure environment knowledge stored in the syntax analysis failure environment knowledge unit 36 shown in FIG.

(1) Type 1
(A) administrative personnel
Word strings that fail parsing: scientific, technical and administrative personnel
(B) Private interests
Word strings that fail parsing: sectarian or private interests
(2) Type 2
(A) punitive laws
Word strings that fail parsing: excessively punitive laws
(B) Toxic dose
Words that fail parsing: overtly toxic dose
(3) Type 3
(A) aggregate basis
Word strings that fail parsing: on a weighted, aggregate basis
(4) Type 4
(A) Earned capital
Word strings that fail parsing: They earned capital .
(5) Type 5
(A) shift operations
Word strings that fail parsing: They would shift operations to a country.
(6) Type 6
(A) anode effects
Word strings that fail parsing: The impact of the electrons with the anode effects generation of x-ray radiation.
FIG. 12 is a flowchart showing an example of processing contents of the syntax analysis failure cause determination means 35. When the syntax analysis fails in step S5 of FIG. 4 and the cause is investigated, the syntax analysis failure cause determination means 35 is activated. The syntax analysis failure cause determination means 35 first determines whether or not it corresponds to the “part of speech string by registered part of speech” of the syntax analysis failure environment knowledge (D1). When it corresponds to “part of speech string by registered part of speech”, it is determined whether or not the focused word (entry word of technical term dictionary) corresponds to “component part of speech” of syntax analysis failure environment knowledge (D2). When it falls under the “component part of speech” of the syntax analysis failure environmental knowledge, the syntax analysis is performed again using the “component part of speech” (D3). Then, it is determined whether the syntax analysis is successful (D4). If the syntax analysis is successful, the cause of the failure is determined (D5). That is, it is determined that the registered part-of-speech of the word of interest (keyword in the technical term dictionary) is the cause.

  If the result of determination in steps D1, D2, and D4 is NO, a normal method for investigating the cause of failure in parsing, such as checking for the presence or absence of multiple parts of speech or repartitioning of components, is used (D6). When another candidate is used, it is determined whether or not the syntax analysis is successful (D7). If the syntax analysis is successful, the cause of the failure is determined (D8). On the other hand, when the parsing fails, it is determined whether another candidate remains (D9). If another candidate remains, the process returns to step D7, and if no other candidate remains, The cause of failure is unknown (D10), and the process is terminated.

In this way, it is determined in Steps D1 and D2 whether or not there is a syntax analysis failure environment knowledge, and if the syntax analysis succeeds in Step D4, the syntax analysis is performed based on the difference between the syntax analysis failure time and the success time. The cause of this can be determined.

  In step D4, if the parsing failure environment knowledge does not lead to successful parsing, a normal parsing and elucidation method is used (step D6). Here, the loop processing is performed until the parsing of possible candidates succeeds or until all candidates are tried (steps D7 and D9). Steps D7 and D9 are portions that require more time as the number of candidates increases. Therefore, if it is possible to proceed to step D5 in the case of syntax analysis failure due to the introduction of syntax analysis failure environment knowledge, the processing of these steps D7 and D9 becomes unnecessary, leading to efficiency.

  This parsing failure environment knowledge can also be used to support dictionary construction. FIG. 13 is an explanatory diagram of an example of a corpus search result. For example, when the user tries to register the translation word “addictive dose” as a noun for “toxic dose”, toxic is an adjective and dose is a noun, which corresponds to types 1, 2, and 3 in FIG. Then, for example, if the corpus of the corresponding field is given to the system, “adjective and / or adjective noun” is an environment where type 1 morphological analysis fails, and “adverb adjective noun is an environment where type 2 morphological analysis fails. It is possible to examine whether or not “adjective, adjective noun”, which is an environment in which type 3 morphological analysis fails, appears in the corpus.

  If the location is presented to the user in the form of KWIC (keyword in context), the user can determine the necessity. In this process, for example, if you know that the word sequence corresponding to the environment where the type 2 morphological analysis of “potentially (adverb) toxic dose” fails appears in this corpus, “toxic dose” and two words In addition to registering compound words, it can be judged that it is necessary to register “potentially toxic dose”.

  In general, if such a long word string is registered, the possibility of failure in parsing becomes relatively low. If dictionary construction is supported in the above manner, terms suitable for a document to be translated are registered in the dictionary not only from the viewpoint of contents but also from the viewpoint of morphemes. In other words, this means that the registration of “adjective + noun” as “noun” has morphological analysis of types 1, 2, and 3 even if there is a risk of types 1, 2, and 3 in FIG. Means that if the environment fails, it doesn't happen to the document you are trying to translate in the first place. Here, the translation may be performed in a batch or may be performed interactively in a sentence.

  Next, still another example of the machine translation apparatus according to the embodiment will be described. FIG. 14 is a functional block configuration diagram of still another example of the machine translation apparatus according to the embodiment. Another example of this is provided by adding reference information display means 37 to another example shown in FIG. The same elements as those in FIG. 10 are denoted by the same reference numerals, and redundant description is omitted.

  The reference information display means 37 displays the translation of the headword in the technical term dictionary when the technical term adjustment means 30 adjusts the headword of the technical term in the translation obtained by the document analysis means 28. .

  As described above, the translation may be performed in a batch or interactively in a sentence, but in the case of batch translation, the syntax depends on the use of a technical term dictionary as a preparation for translation. It is also possible to detect in advance a part that may fail in the analysis and perform translation processing in two stages.

  That is, a case in which a part that may fail in the syntax analysis is detected in advance depending on whether or not the technical term dictionary is used, and the result in step S12 of FIG. 9 is NO, that is, a word in the selected technical term dictionary. Lists the items that could not be used for translation, inquires about what the user should translate with respect to them, and asks the obtained information for the first language word in the corresponding technical term dictionary in the translation dictionary section. It is reflected in the idiom dictionary. Thereby, in the actual translation process, translation can be performed smoothly.

  For example, in the terminology dictionary used, “private interests” is registered with the noun “private interest”, and when this is simply given priority, a sentence that fails parsing is encountered (eg “Education serves both public and private interests. ”), the system determines that private and interests are treated as two words, parsing is successful, and the translation of the terminology dictionary is used to make“ private interest ”and“ private interest ”smaller. If you try to break down into units and take correspondences between them, but you are unable to do so, “private interests” will use the standard dictionary translation.

  Thus, the translation of this part is "public" from public, "private" from private, "profit" from interests, for example, "public / private profit". At this time, the control unit 24 activates the reference information display means 37 and presents to the user as reference information that “private interests” and “private interest” are present in the selected technical term dictionary, and how to Inquire about what should be translated.

  FIG. 15 is an explanatory diagram of an example of the screen. This allows the user to manually modify “public and private interests” as “public and private interests” or register “pubic and private interests” as a noun “public and private interests” in the vocabulary dictionary in use. To help you.

  According to the embodiment, mainly four effects can be expected. First, it is possible to generate a translation in which the terms of the technical term dictionary are unified while avoiding failure in parsing. Secondly, the headword that caused the parsing failure in a certain part is not rejected throughout the document, but only the failed part can be rejected, and flexible translation can be performed. Therefore, it is possible to save labor for a person to determine whether or not to register such a headword in a dictionary storing headwords to be rejected. Thirdly, in pursuing the cause of failure in parsing, a significant part of what is caused by the use of a terminology dictionary is automatically detected, so the work of reviewing the translation of machine translation can be reduced. . Fourth, it is currently unregistered in the document in the field to be translated, but it contributes to the construction of a high-quality dictionary by prompting the user to register words that are desired to be registered in the dictionary for analysis.

  In the above description, the case of English-Japanese translation has been mainly explained, but it can be applied to other language pairs. Moreover, although some embodiment was described, these embodiment is shown as an example and is not intending limiting the range of invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 11 ... Machine translation apparatus, 12 ... Calculation control apparatus, 13 ... Processor, 14 ... Memory, 15 ... Program, 16 ... Work area, 17 ... Display device, 18 ... Input device, 19 ... Mouse, 20 ... Keyboard, 21 ... Disk Drive, 22... Hard disk drive (HDD), 2 3 .. Input processing unit, 2 4... Control unit, 25... Storage device, 26 ... Document storage unit, 27. , 30 ... Technical term adjustment means, 31 ... Sub technical term creation means, 32 ... Sub technical term dictionary part, 33 ... Output processing part, 34 ... Technical term dictionary part, 35 ... Syntax analysis failure cause determination means, 36 ... Syntax analysis Failure environment knowledge storage unit, 37 ... reference information display means

Claims (5)

Translation dictionary unit storing bilingual information of first language and second language in word unit in storage device, headword of technical term of first language consisting of plural word strings and its translation word as bilingual information in storage device When the syntactic analysis is performed using the stored technical term dictionary part, the translation dictionary information of the translation dictionary part and the parallel translation information of the technical term dictionary part, and the syntax analysis fails due to the entry word of the technical term dictionary Then, the headword and its translation are disassembled into words and a dictionary is searched to determine whether any of the candidate translations of the headword divided into words corresponds to the translation into words. A machine translation device comprising technical term adjustment means for replacing a part corresponding to the translation of the headword in the obtained translation with the translation candidate when there is a corresponding one. A sub-terminology dictionary unit that stores a list of candidate translations of first-word headwords that have been decomposed into word units in advance for all headwords in the technical term dictionary unit, and the terminology adjustment means includes word units If any of the translation candidates of the divided first language headword does not match the translated word headword, the translation of the first language headword divided into word units is the terminology dictionary. And if it is in the technical term dictionary part, the part corresponding to the word-by-word translation of the headword in the translation is replaced with the translation of the technical term dictionary part, and the technical term dictionary part If it is not, it is determined whether the translation of the headword in the first language decomposed into words is in the sub-terminology dictionary part, and if it is in the sub-terminology dictionary part, the headword in the translation The word Machine translation apparatus according to claim 1, wherein the replacing the portion corresponding to the position of translation in the translation of the sub terminology dictionary unit. A configuration in which environment analysis failure knowledge storage section storing environmental knowledge in which parsing fails due to an arrangement of parts of speech of words in the first language, and a registered part of speech of the headword and parts of speech of a plurality of words constituting the headword 3. The machine according to claim 1, further comprising: a syntax analysis failure cause determination means for determining a cause of the syntax analysis failure by referring to the syntax analysis failure environment knowledge of the syntax analysis failure environment knowledge section based on the part of speech. Translation device. The reference information display means for displaying the translation of the headword of the technical term dictionary when the technical term adjustment means adjusts the translation is provided. The machine translation device described in 1. Translation dictionary for machine translation program, knowledge information used for translation processing when translating first language document to be translated into second language, and bilingual information between first language and second language in word units A translation dictionary part stored as information and a technical term dictionary part containing a first language technical term consisting of a plurality of word strings and their translations as parallel translation information are stored, and the computer translates the translation dictionary part A function of performing morphological analysis of the first language document using dictionary information and bilingual information of the technical term dictionary unit, and storing attribute information and translated word information of the morpheme as analysis information in an analysis information storage unit of the storage device; A function that generates a translated sentence based on the translated word information when the parsing is successful after parsing a single language document, and is incomplete when the parsing fails because the entry word of the technical term dictionary is not used. A function for generating a translated sentence in an unsatisfactory state, a function for generating a translated sentence by rejecting the heading when the headword of the technical term dictionary is used and parsing fails, and rejecting the headword of the technical term dictionary When the translation is obtained by the above, the translation of the entry word that any of the translation candidates of the entry word of the first language divided into word units by decomposing the entry word and the translation word into word units and dividing it into word units A machine translation program for realizing a function for determining whether or not there is a match and a function for replacing a part corresponding to the translation of the headword in the translation with the translation candidate when there is a match .

Images