JP4488010B2 - Machine translation device, translation method thereof, and storage medium recording the machine translation program - Google Patents

Description

  The present invention relates to machine translation that automatically translates sentences, and more particularly, to a machine translation apparatus that translates using an example, a translation method thereof, and a storage medium that records the machine translation program.

  In general, translation processing of natural language by machine translation equipment first analyzes the input source language sentences, then converts the analysis results to the target language, which is the translation destination language, and finally the target language. Translation is performed by performing a generation process that correctly generates language sentences.

  Also, translation processing by machine translation devices requires the use of various complex and large-scale language knowledge and translation knowledge such as word information, grammar rules, and language conversion rules. A huge cost is necessary.

  In particular, linguistic phenomena with strong idiomatic factors were difficult to generalize and accounted for more of the cost of building translation systems.

  An example of a language phenomenon with such idiomatic factors is the translation of noun phrases. This will be explained using English-Japanese translation as an example.

  The Japanese translation of the English noun sequence may be translated as a series of nouns, or it may be translated by inserting function words that connect the nouns.

  For example, a Japanese translation of the English sentence “a baseball game” is inappropriate as a “baseball game” in which nouns obtained by translating the words “baseball” and “game” into Japanese are arranged in succession similarly to the English sentence. Between the nouns “baseball” and “game”, it is required to translate the word “baseball game” by inserting the case particle “no” between the nouns as a connected function word.

  Conversely, the Japanese translation of the English sentence “a baseball player” is not necessarily “baseball player”, but it is natural to translate it as “baseball player”.

  Also, the Japanese translation of the English text “Tokyo Station” is inappropriate as “Tokyo Station” and is required to be translated as “Tokyo Station” without inserting the case particle “no” between nouns. .

  Which form should be translated depends largely on the idiomatic factors for each word combination, and it was very difficult to build translation knowledge that could be automatically translated and separated. .

  For the purpose of reducing the construction cost of such translation knowledge, a method of using a large amount of examples collected as translation knowledge has been proposed.

  For example, in the machine translation method disclosed in Patent Document 1 (Japanese Patent Laid-Open No. 03-276367), a method for converting the analysis result of the source language into the structure of the target language using a parallel translation example has been proposed (hereinafter referred to as “No. 1”). 1).

  In this method, as in normal translation processing, the input sentence is first analyzed to derive the dependency structure. Then, the obtained dependency structure is compared with the examples, and examples having similar source language structures are extracted. Further, the dependency structure of the input sentence is converted into the dependency structure of the target language in accordance with the parallel translation relationship of the example obtained here, and the translation process proceeds.

  By using an example that is easier to construct in this way instead of conversion knowledge, we aim to realize a high-quality translation apparatus at a low cost.

  In addition, the machine translation system disclosed in Patent Document 2 (Japanese Patent Laid-Open No. 11-259482) proposes a method of directly converting a source language morpheme string into a target language morpheme string using a bilingual example ( Hereinafter, it is referred to as a second conventional example).

  Here, the translation object is limited to the compound noun, and an example in which the correspondence relationship of the morpheme string is maintained is used. In the first conventional example, the dependency structure analysis of the input sentence is necessary to apply the example, whereas in this method, the example can be applied without performing the dependency structure analysis, and the translation process can be performed. It is said.

  As a result, the aim is to realize a high-quality translation device at a lower cost.

  However, in both the first conventional example and the second conventional example, an example used as knowledge needs to be given a bilingual relationship between elements as its internal structure.

  That is, in the example used in the first conventional example, it is necessary to provide a correspondence for each node with respect to the dependency relationship between two languages in a parallel translation relationship, and in the example used in the second conventional example, It is necessary to assign a correspondence relationship for each morpheme to the morpheme strings of the two languages in the bilingual relationship.

  As described above, since the creation of the example involves an associating operation between these elements, the creation of the example still requires a high cost.

For this reason, a very large cost is required to output a high-quality translation result, and in reality it was difficult to obtain a high-quality translation result.
Japanese Patent Laid-Open No. 03-276367 JP-A-11-259592

  In the first conventional example and the second conventional example, the example used for the translation process is an example for bilingual translation, and it is necessary to associate elements inside the translation process, so that a large amount of high-quality translation process can be performed. Collecting bilingual examples still requires a lot of cost, and it was difficult to collect practically sufficient examples.

  This problem also applies to translation processing that involves determining whether or not a specific connected function word should be inserted in a phrase such as a noun phrase, and accumulates example data necessary for appropriate determination. To do so, it required a huge amount of manual costs.

  For this reason, a translation device that performs appropriate translation by referring to example data only in the translation destination language that does not require parallel translation relationship setting processing, and automatically uses this example data based on sentence data in the translation destination language. There is a need for a translation device that can be edited and generated.

  A first object of the present invention is to provide a machine translation device that outputs a natural translation result by appropriately translating each phrase in a sentence with reference to example data of connected function words.

  A second object of the present invention is to provide a machine translation apparatus that refers to example data only in a target language, rather than using an example in parallel translation from a source language to a target language.

  A third object of the present invention is to provide a machine translation apparatus that automatically edits and generates such example data simply by inputting text data in a target language.

  The fourth object of the present invention is to detect information on both the examples including and not including the concatenation function word in the sentence phrase from the text data of the input target language, and appropriately determine whether or not the concatenation function word is used. It is an object of the present invention to provide a machine translation device that automatically edits and generates determined example data and executes translation based on the example data based on this appropriate determination.

  In order to achieve the above object, a machine translation device of the present invention is a machine translation device that translates a source language sentence that is a translation source language into a target language that is a translation destination language. An example database that records example data indicating word data of a sentence phrase and information on the suitability of a designated connected function word in each of the sentence phrases, and the suitability of the use of the connected function word in the target language sentence. Judging by referring to the example database, it is provided with a target language generating means for generating a sentence in which a connection function word is set at an appropriate position in the sentence and the use of the connection function word is appropriately corrected. .

  The machine translation apparatus of the present invention analyzes a source language sentence that is an input translation target sentence in a machine translation apparatus that translates a source language sentence that is a source language into a target language that is a destination language. Source language analysis means, and conversion data that is data obtained by converting each word and sentence structure of the source language sentence into a target language word and sentence structure as a translation destination language based on the analysis of the source language analysis means Example of recording and referring to example data indicating language conversion means for generating a word, word data of a sentence phrase of a specified type in a target language, and information on suitability of a designated connected function word in each sentence phrase Judgment whether or not the connection function word is used in the database and the conversion data is determined with reference to the example database, and the connection function word is set at an appropriate position in the conversion data. Characterized in that it comprises a target language generating means for generating a target language sentence that.

  In the machine translation apparatus according to the present invention, the target language generation means analyzes the structure of the sentence in the target language, and detects the phrase of the specified type of sentence phrase in the sentence as a phrase to be processed. Means, a connected function word inserting means for referring to the example data recorded in the example database, and inserting an appropriate connected function word at an appropriate place in an appropriate case for the phrase to be processed, and the connected According to the present invention, there is provided a target language sentence generation means for generating a sentence in the target language according to the insertion processing of the function word insertion means and the analysis result of the target phrase identification means.

  The machine translation apparatus of the present invention is characterized by comprising example database creation means for automatically editing and registering example data recorded in the example database based on text data in the input target language.

  In the machine translation apparatus of the present invention, the example database creation unit analyzes the input sentence data in the target language, and outputs the analysis data as a morpheme string, and the specified morpheme string A target phrase component extraction means for extracting data of types of text phrases as a component string; a correct example extraction means for detecting a correct example that is an example including the specified connected function word from the component string; It is characterized by comprising example registration means for generating the example data related to the legitimate example detected by the legitimate example extracting means and registering it in the example database.

  In the machine translation apparatus of the present invention, the example database creation unit includes a negative example extraction unit that detects a negative example that is an example that does not include the designated connected function word from the component element sequence, and the example registration unit Generates the example data related to the negative example detected by the negative example extraction unit and registers it in the example database, and the example database creation unit includes the positive example and the negative example in the registered example data. On the basis of the above information, there is provided an example adjusting means for determining whether or not the specified connected function word is used in the sentence phrase and registering the determination result of the suitability in the example database.

  In the machine translation device of the present invention, each of the example data in the example database includes the designation of data in front of the phrase, data in the phrase behind, the phrase between the phrase in front and the phrase behind. It is characterized by comprising data indicating the suitability of inserting the connected function word.

  In the machine translation device of the present invention, each of the example data in the example database includes the designation of data in front of the phrase, data in the phrase behind, the phrase between the phrase in front and the phrase behind. Data indicating the number of times that the positive example extraction means detects the correct example including the connected function word, and data indicating the number of times the negative example extraction means detects the negative example that does not include the designated connection function word; , Comprising data indicating the suitability of inserting the designated connected function word between the preceding word and the following word.

  In the machine translation device of the present invention, the example adjustment unit is configured such that when the number of times the positive example extraction unit detects the right example is larger than the number of times the negative example is detected by the negative example extraction unit, the example data It is determined that the designated connected function word is inserted between the preceding word and the following word, and if not, between the preceding word and the following word in the example data. It is determined that the designated connected function word is not inserted.

  In the machine translation apparatus according to the present invention, the morpheme analyzing means identifies words constituting the sentence data from the inputted sentence data of the target language, determines a part of speech of each word, and the word and its part of speech. Means for outputting the above information as a morpheme string.

  The machine translation device of the present invention uses one of the specified types of sentence phrases as a noun phrase, and the target phrase component extraction means uses a noun phrase as a continuous part of words that are nouns in the morpheme string. It has a means for recognizing and extracting it as a component string.

  The machine translation apparatus of the present invention is characterized in that the target language is Japanese, and one of the designated connected function words is the case particle "no".

  The machine translation method of the present invention is a machine translation method for translating a source language sentence, which is a source language, into a target language, which is a destination language. , An example reference step for recording and referring to an example data indicating information regarding the suitability of a designated connected function word in each sentence phrase, and the suitability of the use of a connected function word in a sentence in a target language. A target language generating step for determining based on the example data referred to in the example reference step, setting a connected function word at an appropriate position in the sentence, and generating a sentence in which use of the connected function word is appropriately corrected; It is characterized by that.

  The machine translation method of the present invention analyzes a source language sentence that is an input translation target sentence in a machine translation method that translates a source language sentence that is a source language into a target language that is a destination language. Source language analysis step and conversion data which is data obtained by converting each word and sentence structure of the source language sentence into a target language word and sentence structure as a translation destination language based on the analysis by the source language analysis step Record in the database and refer to the example data indicating the language conversion step for generating the word data of the specified phrase in the target language, and information on the suitability of the specified connected function word in each sentence phrase Determining whether or not to use the connected function word in the conversion data based on the example data referred to by the example reference step; Set the connecting function words in position, characterized in that it comprises a target language generation step of generating a target language sentence is a sentence translation result.

  In the machine translation method of the present invention, the target language generation step analyzes the structure of the sentence in the target language, and detects the phrase of the specified type of sentence phrase in the sentence as a phrase to be processed. A link function word insertion step that inserts an appropriate link function word at an appropriate location, if appropriate, for the phrase to be processed based on the example data referred to by the example reference step, and the link A target language sentence generation step of generating a sentence in the target language according to the insertion processing by the function word insertion step and the analysis result by the target phrase identification step is provided.

  The machine translation method according to the present invention includes an example database creation step for automatically editing, registering and updating new example data with respect to the database referred to by the example reference step, based on sentence data in an input target language. It is characterized by providing.

  In the machine translation method of the present invention, the example database creation step analyzes the input sentence data in the target language and outputs the analysis data as a morpheme string; A target phrase constituent element extracting step for extracting data of types of sentence phrases as a constituent element string; a correct example extracting step for detecting a correct example that is an example including the specified connected function word from the constituent element string; An example registration step of generating the example data relating to the example for use detected by the example for extracting the example for use and registering in the database referred to by the step for reference to the example is provided.

  In the machine translation method of the present invention, the example database creation step includes a negative example extraction step for detecting a negative example that is an example that does not include the specified connected function word from the component element sequence, and the example registration step Generates the example data related to the negative example detected by the negative example extraction step, registers the database in the database referred to by the example reference step, and the example database creation step Based on the information on the positive example and the negative example, the propriety of use of the designated connected function word in the sentence is determined, and the determination result of the propriety is registered in the database referred to by the example reference step. An example adjustment step is provided.

  A storage medium on which the machine translation program of the present invention is recorded is a storage medium on which a machine translation program for translating a sentence in a source language that is a translation source language into a target language that is a translation destination language. An example reference step for recording and referring to a database of example data indicating the word data of the type of sentence phrase and information on the suitability of the designated connected function word in each sentence phrase, and in the sentence of the target language Appropriateness of use of the connection function word was determined based on the example data referred to by the example reference step, the connection function word was set at an appropriate position in the sentence, and the use of the connection function word was appropriately corrected. A target language generation step for generating a sentence is provided.

  The storage medium recording the machine translation program of the present invention is a translation medium inputted with the machine translation program for translating a sentence in the source language as a source language into a target language as a destination language. A source language analysis step for analyzing a source language sentence that is a target sentence, and each word and sentence structure of the source language sentence based on an analysis by the source language analysis step, and a target language word that is a translation destination language A language conversion step for generating conversion data which is data converted into a sentence structure, word data of a specified type of sentence phrase in the target language, and information on the suitability of the designated connected function word in each of the sentence phrases; The example reference step refers to the example reference step for recording and referring to the example data indicating the above and the suitability of the use of the connected function word in the converted data. The example data is determined based on that, the set of coupled function words in the appropriate position in the conversion data, characterized in that it comprises a target language generation step of generating a target language sentence is a sentence translation result.

  In the storage medium storing the machine translation program of the present invention, the target language generation step analyzes the structure of the sentence in the target language, and uses the specified type of sentence phrase in the sentence as a phrase to be processed. Connected function word insertion that inserts an appropriate connected function word at an appropriate place when appropriate for the phrase to be processed based on the target phrase identifying step to be detected and the example data referred to by the example reference step And a target language sentence generation step of generating a sentence in a target language according to an insertion process by the step of inserting the connected function word and an analysis result by the target phrase identification step.

  The storage medium storing the machine translation program of the present invention automatically edits, registers, and updates new example data with respect to the database referred to by the example reference step, based on the text data of the input target language. An example database creation step is provided.

In the storage medium storing the machine translation program of the present invention, the example database creating step analyzes the input sentence data of the target language and outputs the analysis data as a morpheme string, and from the morpheme string A target phrase component extraction step for extracting data of the specified type of sentence / phrase as a component string, and a correct example for detecting a correct example that is an example including the specified connected function word from the component string An example extraction step, and an example registration step of generating the example data related to the actual example detected by the actual example extraction step and registering the example data in the database referred to by the example reference step are provided.
The storage medium storing the machine translation program of the present invention includes a negative example extracting step of detecting a negative example that is an example not including the designated connected function word from the component string in the example database creating step. The example registration step generates the example data related to the negative example detected by the negative example extraction step, registers in the database referred to by the example reference step, and the example database creation step is registered. Further, based on the information on the positive example and the negative example in the example data, it is determined whether or not the designated connected function word is used in the sentence phrase, and the determination result is referred to the example reference step. And an example adjustment step of registering in the database.

As described above, according to the present invention, the following effects can be obtained.
First, since the example data indicating the appropriateness of the use of the connected function word is used, it is possible to appropriately output the connected function word, appropriately translate each phrase in the sentence, and output a natural translation result.
Secondly, instead of using an example of translation from the source language to the target language, a process for correcting the sentence converted into the target language during the translation process to refer to the example data of only the target language. Can be executed.
Furthermore, for this reason, conventionally, it is possible to automatically edit and generate such example data only by inputting text data in the target language without requiring the processing for setting the parallel translation relationship of the example which has required enormous costs. Can be realized.
Fourth, it is possible to appropriately determine whether or not to use the concatenated function word by detecting information on both the example including and not including the concatenated function word in the sentence phrase from the sentence data of the input target language. In addition, it is possible to automatically edit and generate example data based on the appropriate determination and execute translation based on the appropriate determination.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  The machine translation apparatus of the present invention is characterized in that in natural language translation processing or the like, by referring to data of examples of connected function words, natural translation results are output in translation of various phrases in a sentence. .

  Hereinafter, the source language is referred to as a source language, and the destination language is referred to as a target language. For example, in English-Japanese translation, English is the source language and Japanese is the target language.

  The translation source sentence data (sentence or character string, word, etc.) to be translated is called a source language character string, and the sentence data generated as a translation result is called a target language sentence.

  FIG. 1 is a block diagram showing the configuration of the machine translation apparatus according to the first embodiment of this invention.

  Referring to FIG. 1, the machine translation apparatus of the present embodiment includes an input unit 10 that receives input of a source language character string that is sentence data of a translation source, and a source language character string that is input to the input unit 10. Language analysis unit 20, language conversion unit 30 that converts the analysis result output from source language analysis unit 20 to the structure of the target language, and target language that generates the target language sentence from the structure of the target language converted by language conversion unit 30 A generation unit 40 and an output unit 50 that outputs a target language sentence generated by the target language generation unit 40 are provided.

  In particular, the target language generating means 40 analyzes the structure of the target language converted by the language conversion means 30 and the connected function word example database 41 that stores phrase examples related to the predetermined connected function words in the target language, and sets them as processing targets. The target phrase identifying means 42 for identifying the partial structure corresponding to the power phrase, and the connected function word are inserted into appropriate phrases in the phrase identified by the target phrase identifying means 42 according to the examples stored in the connected function word example database 41. There is provided a connected function word inserting means 43 and a target language sentence generating means 44 for generating a target language sentence from the structure of the target language in accordance with the connecting function word inserting process of the connected function word inserting means 43 and the like.

  Next, the operation of the machine translation apparatus according to this embodiment will be described.

  FIG. 2 is a flowchart for explaining the translation processing of the machine translation apparatus of this embodiment.

  Here, the translation process using the example of the machine translation apparatus of the present embodiment will be described using a specific example in which a determination process is performed to determine whether or not a predetermined connected function word is inserted between two words.

  Referring to FIG. 2, the input unit 10 first receives an input of a source language character string to be translated (step 201).

  This input processing of sentence data to be translated can be realized by, for example, a keyboard, OCR, an input method using voice recognition, or by directly reading sentence data recorded as electronic data in a storage medium such as an external storage device.

  Next, the source language analyzing unit 20 analyzes the source language character string read from the input unit 10 and calculates the structure of the sentence (step 202).

  The language conversion means 30 refers to the translation knowledge data and converts the sentence structure to be translated output by the source language analysis means 20 into a sentence structure of the target language (step 203).

  Next, the target language generation unit 40 generates a target language sentence from the structure of the target language output by the language conversion unit 30.

  First, the target phrase identifying means 42 identifies the phrase part that is the target of the connected function word insertion processing from the data converted into the sentence structure of the target language by the language converting means 30 in step 203 (step 204). ).

  Specifically, each word constituting the phrase structure of the input sentence data is inspected, and when two words that are constituent elements of the phrase to be inspected have a direct dependency relationship, this part is designated as a connected function word. Extracted as a phrase to be inserted.

  The connected function word inserting means 43 refers to the examples stored in the connected function word example database 41, and inserts a predetermined case particle into the phrase identified as the processing target by the target phrase identifying means 42 when appropriate.

  The linked function word example database 41 stores phrase examples related to predetermined linked function words.

  Here, as a specific example of a method for storing examples in the connected function word example database 41, a set of two words constituting a phrase and a predetermined connected function word inserted between the two words (expressed) ) Can be configured using a table that records the correspondence with the connected function word expression flag indicating whether or not to be performed.

  Specifically, first, one phrase structure to be processed identified by the target phrase identifying means 42 is extracted (step 205). In this case, whether or not the insertion processing is to be performed is checked to see whether or not an unprocessed phrase remains (step 206), and if an unprocessed phrase remains, it is extracted. Then, two words constituting the extracted phrase to be processed are searched in the connected function word example database 41 (step 207).

  Next, the obtained search result is checked (step 208), and if an example is obtained as the search result, the connected function word expression flag of the example should indicate that the connected function word should be inserted. For example, a predetermined connected function word is inserted into the phrase to be processed (step 209).

  The process for determining whether or not to perform the above insertion processing is executed for the phrase structure to be processed identified by the target phrase identifying means 42, and the insertion processing is executed (step 206).

  After the processing for all the phrase structures to be processed is completed, the target language sentence generation unit 44 arranges the words included in the target language structure subjected to the connection function word processing by the connection function word insertion unit 43 and performs processing such as transformation To generate a target language sentence (step 210).

  The output means 50 outputs the target language sentence generated by the target language generation means 40 (step 211).

  This translation result output method can be realized by, for example, a display device such as a display, an output from a speaker by a speech synthesizer, or a direct output as electronic data to a storage medium such as an external storage device.

  Next, an example of a specific translation operation of the machine translation apparatus according to the present embodiment will be specifically described with reference to FIGS.

  In the following specific example, explanation will be given by an example of English-Japanese translation in which the source language that is the translation source language is English and the target language that is the translation destination language is Japanese, but the machine translation device of the present embodiment The translation processing is not limited to this, and can be applied to translation between arbitrary languages such as French-Japanese translation, German-Japanese translation, Russian-Japanese translation, and Japanese-English translation.

  In the following, a noun phrase is considered as a phrase to be processed, and a case particle “no” is considered as a predetermined connected function word, but the scope of the present invention is not limited to this, and adjective phrases, verb phrases, etc. It can be applied to any phrase structure and any connected function word such as “NA”, “TO”, “TO”, “NI”, “HA”, “HA” and “GA”.

  Here, consider a case where “Iwent to the baseball game.” Is input to the input unit 10 as an English sentence to be translated.

  The English text input here is analyzed into a syntax structure by the source language analysis means 20 and further converted into a syntax structure of the target language by the language conversion means 30.

  FIG. 3 is a diagram showing an example of the analysis of the syntax structure of an English sentence to be translated by the source language analysis means of the present embodiment. In FIG. 3, each word in the input sentence is assigned to each leaf node, and the dependency relationship between the words is shown in a tree structure.

  For example, referring to the structure for “baseball game” shown in FIG. 3, it can be seen that there is a direct dependency between the two words, and the two together form a noun phrase (NP).

  Further, in this figure, a head word, which is a representative word of the subordinate structure, is given to each node, and the relationship is indicated by a bold line.

  In other words, in the structure of the noun phrase “baseball game”, “game” is the headword, and the structure above it can be understood as a relationship between “the” and “game”.

  FIG. 4 is a diagram illustrating an example of a syntax structure of a Japanese sentence as a result of conversion by the language conversion unit according to the first embodiment of this invention. In FIG. 4, a Japanese translation for each word in the input English sentence is assigned to each leaf node in Japanese word order.

  Further, as in FIG. 3, the dependency relationship between the words is shown in a tree structure, but in FIG. 4, a concatenated word indicating the dependency relationship of each word is also given.

  For example, it is shown that there is a subject-predicate relationship connected by “ga” between “I” and “go”.

  FIG. 5 is a diagram illustrating an example of example data stored in the connected function word example database 41 according to the first embodiment of this invention.

  In FIG. 5, pairs of nouns that can appear in the case particle “no” are stored in the word 1 and word 2 columns.

  Further, each word set stores a connected function word expression flag indicating whether or not the case particle “no” should be expressed in between. Here, it is assumed that the connected function word expression flag should be expressed by “1” and should not be expressed by “0”.

  The syntax structure of the target language converted by the language conversion unit 30 is sent to the target phrase identification unit 42.

  The target phrase identifying means 42 examines each word in the received syntax structure to search for a noun or a similar word, and if the destination is a noun or similar word, it extracts it as a noun phrase to be processed. To do.

  Referring to FIG. 4, the word "I" is first examined here.

  “I” is a pronoun and can be recognized as a word similar to a noun. However, since it is a verb “go”, it is determined that it is not subject to processing.

  Next, the word “baseball” is examined, and since this is a noun and relates to the noun “game”, the syntactic structure for “baseball / game” is identified as a noun phrase to be processed.

  Similarly, all words are checked to identify all objects to be processed.

  The noun phrases to be processed identified here are sent one by one to the connected function word insertion means 43.

  The connected function word inserting unit 43 searches the connected function word example database 41 using two nouns included in the noun phrase to be processed identified by the target phrase identifying unit 42 or a similar word.

  Here, an example is searched in which the word 1 that is the source of the engagement is “baseball” and the word 2 that is the destination of the engagement is “game”.

  When the connected function word expression flag in the search result obtained here is checked, it is found that this is “1”, and that the case particle “no” should be inserted between them.

  The connected function word inserting means 43 inserts “no” as the relationship of the part where “baseball” is related to “game” according to the result obtained here.

  FIG. 6 is a diagram illustrating an example of a processing result of the connected function word insertion processing according to the present embodiment.

  The syntax structure obtained here is sent to the target language sentence generation means 44 and is generated as a Japanese sentence.

  Here, “G”, “No”, “Ni”, and “Ta” in the past, which represent the relationship between words, are expressed as particles, and are used and modified to say, “I went to a baseball game” Japanese sentences are generated.

  The Japanese sentence generated here is sent to the output means 50 and output.

  FIG. 7 is a block diagram showing a configuration of an example of the machine translation apparatus according to the first embodiment of this invention.

  Referring to FIG. 7, in this embodiment, in a computer processing apparatus 90 having an input / output function, an information processing function, and the like, a machine translation program having each function such as the target language generation means 40 of the present embodiment is loaded into a memory. Can be realized. The machine translation program is stored in a storage medium 100 that can be read by a magnetic disk, magnetic tape, optical disk, semiconductor memory, or other computer. Then, the above-described functions are realized by being loaded from the storage medium 100 to the computer processing device 90 and controlling the operation of the computer processing device 90.

  In other words, the machine translation program stored in the storage medium 100 is read from the storage medium 100 into the memory of the computer processing device 90, and the operation of the computer processing device 90 is controlled so that the connected function word is stored on the computer processing device 90. In addition to realizing a storage unit such as the example storage unit 941, the input unit 91, the source language analysis unit 92, the language conversion unit 93, the target phrase identification unit 942, the connected function word insertion unit 943, the target language sentence generation unit 944, and the output unit 95 Each function is realized.

  Here, in relation to the first embodiment of FIG. 1, the input unit 91 is the input unit 10, the source language analysis unit 92 is the source language analysis unit 20, the language conversion unit 93 is the language conversion unit 30, Each unit in the target language generation unit 94 corresponds to each unit in the target language generation unit 40, and the output unit 95 corresponds to the output unit 50.

  The present embodiment is characterized in that the computer processing apparatus 90 operates under the control of a machine translation program, and the basic functions are the same as those in FIGS.

  That is, the computer processing device 90 executes the following processing under the control of the machine translation program.

  By inputting the data to be translated into the input unit 91, the input unit 91 receives the input of the source language character string to be translated and activates the source language analyzing unit 92.

  The source language analysis unit 92 analyzes the source language character string input from the input unit 91 and activates the language conversion unit 93.

  The language conversion unit 93 activates the target phrase identification unit 942 after converting the structure of the analysis result for the translation target output by the language analysis unit 92 into the structure of the target language.

  The target phrase identifying unit 942 identifies the partial structure of the phrase that is the target of the connected function word insertion processing from the structure of the target language output by the language converting unit 93, and activates the connected function word inserting unit 943 for each. To do.

  The link function word insertion unit 943 inserts a predetermined case particle into the partial structure of the phrase identified by the target phrase identification unit 942 when appropriate with reference to the example stored in the link function word example storage unit 941. To do.

  Here, after performing the connected function word insertion processing on all the partial structures identified by the target phrase identifying unit 942, the target language sentence generating unit 944 is activated.

  The target language sentence generation unit 944 generates words in the target language by arranging words included in the structure of the target language that has been subjected to the connection function word processing by the connection function word insertion unit 943 and performing processing such as utilization modification.

  The output unit 95 outputs the target language sentence generated by the target language generation unit 94.

  Since the machine translation device of the present embodiment described above uses the example data indicating the suitability of the use of the connected function word, the concatenated function word is appropriately output, each phrase in the sentence is appropriately translated, The translation result can be output.

  Also, instead of using an example of parallel translation from the source language to the target language, refer to the example data of the target language only, so that the sentence converted to the target language during the translation process will be corrected to an appropriate sentence can do.

  FIG. 8 is a block diagram showing the structure of the machine translation apparatus according to the second embodiment of the present invention.

  The machine translation apparatus according to the present embodiment outputs a natural translation result at a low cost by using automatically collected examples of connected function words.

  The machine translation apparatus of the present embodiment is different from the machine translation apparatus of the first embodiment of FIG. 1 in that it includes an example database creation means 60 that automatically creates a linked function word example database 41. is there.

  An example database creation unit 60 in the present embodiment includes an example input unit 61 that receives an input of a target language character string, a morpheme analysis unit 62 that analyzes a character string input from the example input unit 61 and outputs a morpheme string, A target phrase component extraction unit 63 that extracts a component sequence of a target phrase from a morpheme sequence in the target language, and a component extracted by the target phrase component extraction unit 63 from the morpheme sequence analyzed by the morpheme analysis unit 62 Reference example extracting means 64 for extracting a connected function word correct example including a predetermined connected function word with reference to the column, and an example registration for registering the correct example extracted by the correct example extracting means 64 in the connected function word example database 41 And means 65.

  Here, the right-use example is an example that includes a connected function word between words in the example registered in the example database 41.

  Next, the operation of the machine translation apparatus according to this embodiment will be described.

  The operation of the machine translation apparatus according to the present embodiment is different from the operation of the machine translation apparatus according to the first embodiment in that a step of automatically creating the connection function word example database 41 by the example database creation means 60 is performed. The operations of the other input means 10, source language analysis means 20, language conversion means 30, target language generation means 40, and output means 50 are the same as those described in the operation of the embodiment of the first invention. It is. Hereinafter, the operation of the example database creating means 60, which is a feature of the present embodiment, will be described.

  FIG. 9 is a flowchart for explaining the operation of the example database creation means 60 of this embodiment. In the flowchart of FIG. 2, input / analysis of the source language character string of the source language for execution of translation is performed, but here the target language character string of the destination language for registration of example data is shown here. Perform input / analysis of.

  The example input means 61 receives an input of a target language character string that is a target of example extraction (step 901).

  This can be realized by, for example, an input method using a keyboard, OCR, voice recognition device, or the like, or a method of directly reading electronic data of a source language character string stored in an external storage device or the like.

  The morpheme analysis unit 62 performs morpheme analysis on the target language character string input from the example input unit 61 and decomposes it into a morpheme sequence (step 902). Here, the morpheme string is information indicating each word of the target language character string and the part of speech formed by each word as shown in the example of FIG.

  The correct example extracting unit 64 extracts an example to be registered in the connected function word example database 41 from the morpheme string output as an analysis result by the morpheme analyzing unit 62.

  Specifically, first, a predetermined connected function word (such as “no”) is searched from the morpheme string output as an analysis result by the morpheme analysis means 62 (step 903).

  The search result obtained here is checked (step 904), and if there are predetermined linked function words in the morpheme string, the following processing is executed for all of them.

  First, it is checked whether or not the morpheme a immediately before the connected function word to be processed is a constituent element of the phrase to be registered (step 905). This checks whether the part of speech of the morpheme a is a noun, an adjective or the like, and whether a phrase sandwiching the connected function words is formed.

  When the morpheme a is a constituent element of the target phrase, the morpheme string (or morpheme) after the connected function word is passed to the target phrase constituent element extraction unit 63, and the constituent element immediately after the connected function word A column is extracted (step 906).

  The target phrase component extraction means 63 checks whether or not the component string can be extracted (step 907), and if it can be extracted normally, for example, to record in accordance with the recording form shown in the example of FIG. A set of the morpheme a and the last morpheme b of its component string is extracted as an example and passed to the example registration unit 65 (step 908).

  The example registration unit 65 registers the set of the morpheme a and the morpheme b extracted by the regular example extraction unit 64 as an example in the connected function word example database 41 (step 909).

  Here, as an example of storing in the connected function word example database 41, whether or not a predetermined connected function word should be expressed between two words as shown in the example of FIG. 5 and the two words. And a connected function word expression flag representing.

  As a registration method to the connected function word example database 41 by the example registering means 65, the already registered example is used in the connected function word example database 41 in step 909 together with or instead of the connected function word expression flag. It is also possible to store information on the number of times registration is instructed as shown in FIG.

  Furthermore, for each set of words, instead of registering in the alternative form of using or not using the connected function word by the connected function word expression flag, a predetermined connected function word is used between the words. It is also possible to register with the information of the probability of occurrence of (and the information of the number of appearances and the number of non-appearances).

  In the translation processing of the present embodiment, the processing described in the operation of the first embodiment is executed using the connected function word example database 41 created by the example database creation means 60.

  Next, a specific operation of the machine translation apparatus according to the second embodiment will be described in detail with reference to the drawings.

  In the following description, the source language that is the source language is English and the target language that is the target language is Japanese, but the language that the machine translation device of the present embodiment translates is described here. The present invention is not limited, and can be similarly applied to translation between arbitrary languages.

  In the following description, the noun phrase is a processing target phrase and the case particle “no” is a predetermined connection function word. However, the processing target phrase and the connection processed by the machine translation device of the present embodiment will be described. The function word is not limited to this, and can be similarly applied to any phrase structure and connected function word.

  The machine translation apparatus according to the present embodiment is different from the first embodiment in the process of automatically creating the connected function word example database 41 by the example database creating means 60. The other input means 10, The operation of each part of the source language analysis unit 20, the language conversion unit 30, the target language generation unit 40, and the output unit 50 is the same as that described in the operation of the first embodiment. Therefore, the operation of the example database creation means 60 that is a feature of the present embodiment will be specifically described below.

  The Japanese sentence “I like to go to baseball games” will be explained as an example extraction target.

  First, the Japanese sentence to be processed is input to the example input means 61. Then, the morpheme analyzing means 62 analyzes the input Japanese sentence and converts it into a morpheme string.

  FIG. 10 is a diagram illustrating an example of a morpheme string obtained as a result of morpheme analysis by the morpheme analysis unit of the present embodiment.

  In the example of FIG. 10, one morpheme is shown in each line, and each of them shows a morpheme character string (word) and its part of speech.

  The morpheme analysis result obtained by the morpheme analysis unit 62 is sent to the regular example extraction unit 64, and the examples contained therein are extracted.

  First, the correct usage example extracting unit 64 searches for a predetermined connected function word from the obtained morpheme analysis result.

  Here, when the case particle “no” is searched, the fourth morpheme is the case particle “no”, and this position is obtained as a search result.

  Next, the correct example extracting means 64 examines the part of speech of “baseball” which is a morpheme (word) immediately before the obtained search result. When the part of speech is a noun or the like, it is determined that the example is to be extracted, and when the part of speech is a particle or the like, it is determined that the example is not a candidate to be extracted.

  Here, the part of speech of “baseball” is a noun and is a candidate to be extracted as an example. Therefore, the correct example extracting unit 64 sends the morpheme string starting from the next of the case particle “NO” to be processed, which is the fourth morpheme, to the target phrase component extracting unit 63. Then, the target phrase constituent element extracting unit 63 extracts a noun continuous portion from the morpheme string sent from the correct example extracting unit 64 as a constituent element string of the noun phrase to be processed.

  The length of the noun phrase extracted by the correct usage example extracting means 64 is a method of extracting a predetermined number of words or a noun phrase such as a noun by sequentially checking the part of speech of each word after the case particle “no”. A method of extracting a word in a range in which a part of speech continues, a method of extracting a word until a punctuation mark appears first after the case particle “no”, and the like are possible.

  Here, description will be made assuming that the morpheme string “game / to / go / no / ga / like / da”, for example, is sent to the target phrase constituent element extraction unit 63 from the correct example extraction unit 64.

  First, since the target phrase component extraction means 63 examines the part of speech of the first morpheme “game” after the case particle “no”, it is a noun, so this morpheme is part of a noun sequence (part of a noun phrase). It is judged that.

  Next, when the part of speech of the second morpheme “Ni” is examined, the part of speech is a case particle, so it is determined that this morpheme does not belong to the noun continuation, and the morpheme string up to the previous morpheme “match” is used as the noun continuation part. Extract.

  With the above processing, “game” is extracted as a noun continuous part and sent to the correct example extracting means 64.

  The correct usage example extracting unit 64 extracts the last morpheme of the noun continuous part extracted by the target phrase component extracting unit 63 as the morpheme of the connected function word currently being processed.

  Here, “game” is extracted, and as an example, a set of morphemes “baseball” and “game” is obtained, and this is sent to the example registration means 65.

  The example registering unit 65 registers the set of morphemes extracted as examples by the correct example extracting unit 64 in the connected function word database 41.

  Here, a set of morphemes of “baseball” and “game” is registered, but the one shown in FIG. FIG. 11 is a diagram illustrating an example of example data stored in the linked function word example database according to the second embodiment of this invention.

  Referring to FIG. 11, a pair of nouns that can appear in the case particle's “no” dependency source and the relationship destination is stored in the word 1 and word 2 columns.

  Further, each word set stores a connected function word expression flag indicating whether or not the case particle “no” should be expressed in between.

  Here, it is assumed that the connected function word expression flag indicates that it should be expressed by “1” and should not be expressed by “0”.

  Further, each set of words stores the number of examples registered in the connected function word example database 41 with respect to an example in which a predetermined connected function word appears between the words.

  For example, referring to a combination of “baseball” and “game”, it can be seen that an example that appears with a “baseball game” and a case particle “no” is registered 54 times.

  In the registration of a morpheme set of “baseball” and “game” in the connected function word database 41, first, the number of examples for which word 1 is “baseball” and word 2 is “game” is incremented by one to 55. .

  In addition, the linked function word expression flag is also updated as the number of regular examples increases. For example, the connected function word expression flag is set to 1 when the number of regular examples exceeds a predetermined threshold by the registration process of this example.

  The translation processing in the second embodiment executes the processing described in the operation of the embodiment of the first invention using the connected function word example database 41 created here.

  FIG. 12 is a block diagram illustrating a configuration of an example of the machine translation apparatus according to the second embodiment of this invention.

  Referring to FIG. 12, the present embodiment is a computer processing apparatus 90 having an input / output function, an information processing function, etc., and a machine translation having each function of the target language generation means 40, the example database creation means 60, etc. of the present embodiment. This can be realized by loading the program into memory. The machine translation program is stored in a storage medium 100 that can be read by a magnetic disk, magnetic tape, optical disk, semiconductor memory, or other computer. Then, the above-described functions are realized by being loaded from the storage medium 100 to the computer processing device 90 and controlling the operation of the computer processing device 90.

  In other words, the machine translation program stored in the storage medium 100 is read from the storage medium 100 into the memory of the computer processing device 90, and the operation of the computer processing device 90 is controlled so that the connected function word is stored on the computer processing device 90. In addition to realizing a storage unit such as the example storage unit 941, the input unit 91, the source language analysis unit 92, the language conversion unit 93, the target phrase identification unit 942, the connected function word insertion unit 943, the target language sentence generation unit 944, and the example input unit 961, morpheme analysis unit 962, target phrase component extraction unit 963, correct example extraction unit 964, example registration unit 965, and output unit 95 are realized.

  8, the input unit 91 is the input unit 10, the source language analysis unit 92 is the source language analysis unit 20, the language conversion unit 93 is the language conversion unit 30, and the target language generation unit. Each unit in 94 corresponds to each unit in the target language generation unit 40, the output unit 95 corresponds to the output unit 50, and each unit in the example generation unit 96 corresponds to various stages in the example database creation unit 60.

  The present embodiment is characterized in that the computer processing apparatus 90 operates under the control of a machine translation program, and the basic functions are the same as those in FIGS.

  That is, the computer processing device 90 executes the following processing under the control of the machine translation program.

  In the example registration process, first, according to an example input to the input unit 91, the example input unit 961 receives an input of a target language character string that is a target of example extraction, and activates the morpheme analysis unit 962.

  The morpheme analysis unit 962 performs morpheme analysis on the target language character string input from the example input unit 961, decomposes it into a morpheme sequence, and then activates the correct example extraction unit 964.

  The correct usage example extraction unit 964 extracts a usage example to be registered in the connected function word usage example storage unit 941 from the morpheme sequence output as an analysis result by the morpheme analysis unit 962, and activates the usage example registration unit 965.

  When extracting an example, a predetermined connected function word is searched from the morpheme string output as an analysis result by the morpheme analysis unit 962, and the target morpheme immediately after the obtained connected function word and the immediately following morpheme string The phrase component element sequence extracted by the phrase component extraction unit 963 is used.

  The example registration unit 965 registers the set of two morphemes extracted as examples by the regular example extraction unit 964 in the connected function word example storage unit 941.

  Further, in the translation process, the process described in the operation of the first embodiment is performed using the connected function word example storage unit 941 created by the example creation unit 96.

  With the machine translation apparatus of the present embodiment described above, it is possible to automatically edit and generate example data used for translation simply by inputting text data in a target language.

  FIG. 13 is a block diagram illustrating a configuration of the machine translation apparatus according to the third embodiment of this invention.

  The machine translation apparatus according to the present embodiment is characterized in that a natural translation result is output at a low cost by using automatically collected examples of connected function words.

  The machine translation apparatus according to the present embodiment is different from the machine translation apparatus according to the second embodiment in FIG. 8 in the internal configuration of the example database creation means 60a. The negative example extraction means 66 and the example adjustment means 67 is newly provided.

  Here, the negative example is an example in which the connected function word is not included between words in the example registered in the example database 41 contrary to the positive example.

  That is, the example database creation unit 60a of the present embodiment includes an example input unit 61 that receives input of a target language character string, and a morpheme analysis unit that analyzes a character string input from the example input unit 61 and outputs a morpheme string. 62, a target phrase component extraction unit 63 that extracts a component sequence of a target phrase from a morpheme sequence in the target language, and a target phrase component extraction unit 63 that is extracted from the morpheme sequence analyzed by the morpheme analysis unit 62 A reference example extracting means 64 for extracting a connected function word correct example including a predetermined connected function word with reference to the component string, and a correct example extracted by the correct example extracting means 64 are registered in the connected function word example database 41. And a predetermined connected function word by referring to the constituent element sequence extracted by the target phrase constituent element extracting means 63 from the morpheme sequence analyzed by the morpheme analyzing means 62 Comprises a negative example extraction means 66 for extracting Manai connecting function words negative examples, the example adjusting means 67 for adjusting the content of the coupling function words example database 41 by using the negative examples extracted with negative example extraction means 66.

  Next, the operation of the machine translation apparatus according to this embodiment will be described.

  FIG. 14 is a flowchart for explaining the translation processing of the machine translation apparatus according to the third embodiment of this invention.

  The difference of the operation of the machine translation apparatus of the present embodiment from the machine translation apparatus of the second embodiment is the process of automatically creating the linked function word example database 41 by the example database creating means 60a. The operations of the other input means 10, source language analysis means 20, language conversion means 30, target language generation means 40, and output means 50 are the same as those described in the second embodiment and in the first embodiment. Is the same. For this reason, the operation of the example database creating means 60, which is a feature of the present embodiment, will be described below.

  First, the example input means 61 receives an input of a target language character string that is a target of example extraction (step 1401).

  This input processing of sentence data to be translated can be realized by, for example, a keyboard, OCR, an input method using voice recognition, or by directly reading sentence data recorded as electronic data in a storage medium such as an external storage device.

  The morpheme analysis unit 62 performs morpheme analysis on the target language character string input from the example input unit 61 and decomposes it into a morpheme sequence (step 1402). Here, information on the part of speech of each word is also given to each morpheme of the analysis result.

  The correct example extracting unit 64 extracts a correct example to be registered in the connected function word example database 41 from the morpheme string output as an analysis result by the morpheme analyzing unit 62.

  Specifically, first, a predetermined connected function word is searched from the morpheme string output as an analysis result by the morpheme analysis means 62 (step 1403).

  Then, the obtained search result is checked (step 1404), and if there is a predetermined connected function word in the morpheme string, the following processing is executed for all of them.

  First, it is checked whether or not the morpheme a immediately before the connected function word to be processed is a constituent element of the target phrase (step 1405).

  When the morpheme a is a constituent element of the target phrase, the morpheme string after the connected function word is passed to the target phrase constituent element extracting means 63, and the constituent element string immediately after the connected function word is extracted (step) 1406).

  It is checked whether or not the component string can be extracted by the target phrase component extraction means 63 (step 1407). If extraction is possible, for example, a pair of morpheme a and the last morpheme b of the component string is extracted as an example. It passes to the registration means 65 (step 1408).

  The example registration unit 65 registers the set of the morpheme a and the morpheme b extracted by the correct example extraction unit 64 as the correct example in the connected function word example database 41 (step 1409).

  Here, as an example of storing in the connected function word example database 41, for example, a set of two words and a connected function word expression flag indicating whether or not a predetermined connected function word should be expressed between the two words, Can be configured.

  Also, together with or instead of the connected function word expression flag, a correct example in which a predetermined connected function word appears between those words, or these words appear continuously without going through the specified connected function word About the negative example, you may store the number of the examples registered into the connection function word example database 41. FIG.

  Furthermore, the probability that a predetermined connected function word appears between each pair of words may be stored.

  The negative example extracting unit 66 extracts a negative example to be registered in the connected function word example database 41 from the morpheme string output as an analysis result by the morpheme analyzing unit 62.

  Specifically, first, the constituent elements of the target phrase are extracted from the morpheme string output as the analysis result by the morpheme analyzing means 62 using the target phrase constituent element extracting means 63 (step 1410).

  Here, if two or more morpheme component strings are obtained, the following processing is executed for all of them.

  First, one morpheme combination in the component row is extracted (step 1411).

  When the combination of two morphemes can be extracted (step 1412), the morpheme c that appears in the front and the morpheme d that appears in the rear are extracted as negative examples (step 1413) and passed to the example adjustment unit 67.

  The example adjustment unit 67 registers the set of the morpheme c and the morpheme d extracted as the negative example by the negative example extraction unit 66 in the connected function word example database 41, and if already registered, the connected function word table of the example. The out flag and the like are adjusted (step 1414).

  Next, the operation of the present embodiment will be specifically described with reference to the drawings.

  In the following description, the source language that is the source language is English and the target language that is the target language is Japanese, but the language that the machine translation device of the present embodiment translates is described here. The present invention is not limited, and can be similarly applied to translation between arbitrary languages.

  In the following description, the noun phrase is a processing target phrase and the case particle “no” is a predetermined connection function word. However, the processing target phrase and the connection processed by the machine translation device of the present embodiment will be described. The function word is not limited to this, and can be similarly applied to any phrase structure and connected function word.

  The difference between the machine translation apparatus of the present embodiment and the machine translation apparatus of the second embodiment is the process of automatically creating the linked function word example database 41 by the example database creation means 60a. The operations of the input unit 10, the source language analysis unit 20, the language conversion unit 30, the target language generation unit 40, and the output unit 50 are the same as those described in the operations of the second embodiment and the first embodiment. . Therefore, the operation of the example database creation unit 60a will be specifically described below.

  The Japanese sentence “I saw a baseball player” will be explained as an example extraction target.

  First, the Japanese sentence to be processed is input to the example input means 61. Then, the morpheme analyzing means 62 analyzes the input Japanese sentence and converts it into a morpheme string.

  FIG. 15 is a diagram illustrating an example of a morpheme string obtained as a result of morpheme analysis by the morpheme analysis unit of the present embodiment.

  In the example of FIG. 15, one morpheme is shown in each line, and each of them shows a morpheme character string (word) and its part of speech.

  The morpheme analysis result obtained by the morpheme analysis unit 62 is sent to the correct example extraction unit 64, and the correct example contained therein is extracted.

  First, as in the second embodiment, a predetermined connected function word is searched from the obtained morpheme analysis result.

  However, in the Japanese sentence “I saw a baseball player” here, it cannot be obtained even by searching for the case particle “no”. In the same manner as described in the operation of the embodiment, the extraction and registration processing of the right-hand example is performed.

  Next, the result of morpheme analysis obtained by the morpheme analysis unit 62 is sent to the negative example extraction unit 66, and the negative example contained therein is extracted.

  First, the target phrase component extraction means 63 is activated to extract a component string of the target phrase. The target phrase component extraction unit 63 extracts a target phrase component string from the input morpheme string. Here, it is considered that a morpheme string “baseball / player / watch / ta” has been input.

  First, when the part of speech of the first morpheme “baseball” is examined, since it is a noun, it is determined that this morpheme is a constituent element of a noun phrase.

  Next, when the part of speech of the second morpheme “player” is examined, since that part of speech is also a noun, it is determined that this morpheme is also a constituent element of the noun phrase.

  Further, when the part of speech of the third morpheme “O” is examined, it is a case particle, so it is determined that this morpheme does not belong to a noun phrase component, and the morpheme string up to the previous morpheme is extracted as a noun phrase component string. To do.

  As described above, “baseball / player” is extracted as a component row and sent to the negative example extracting means 66.

  The negative example extracting unit 66 checks whether or not two or more morphemes are included in the constituent element sequence extracted by the target phrase constituent element extracting unit 63, and extracts a combination of the morphemes therein as a negative example.

  Here, a combination of two morphemes “baseball” and “player” can be extracted as negative examples, and these are sent to the example adjusting means 67.

  The example adjustment unit 67 adjusts the data in the connected function word database 41 using the morpheme set extracted by the negative example extraction unit 66 as a negative example.

  First, a set of morphemes of “baseball” and “player” is registered, but the one shown in the example of FIG.

  FIG. 16 is a diagram illustrating an example of example data stored in the connected function word example database according to the present embodiment.

  Referring to FIG. 16, a pair of nouns that can appear in the case particle “no” and the word noun are stored in the word 1 and word 2 columns.

  Further, each word group stores a connected function word expression flag indicating whether or not the case particle “no” should be expressed in between.

  The connected function word expression flag indicates that it should be expressed by “1” and should not be expressed by “0”.

  Each word set stores the number of appearances of the correct example registered in the connected function word example database 41 with respect to the number of appearances of the correct example including a predetermined connected function word between the words.

  For example, referring to a combination of “baseball” and “player”, it can be seen that an example that appears with a case particle “no” is registered as a baseball player “no” eight times as a proper example.

  Furthermore, each word set is also registered in the connected function word example database 41 in the same manner with respect to the number of negative examples that appear in the noun series without a predetermined connected function word between them. The number of occurrences of negative examples is stored.

  For example, referring to the combination of “baseball” and “player”, it can be seen that an example that appears in the noun sequence without the case particle “no” is registered 40 times as a negative example “baseball player”.

  In addition, in the registration process of the “baseball” and “player” morpheme pairs in the connected function word database 41 described here, first, the number of negative examples of the example in which the word 1 is “baseball” and the word 2 is “player”. Is incremented by 1 to 41.

  Further, the linked function word expression flag is updated as the number of negative examples increases. For example, when the value of “number of positive examples−number of negative examples” falls below a predetermined threshold value (for example, “0 times”), the update process of the connected function word expression flag is performed. Processing such as setting the output flag to 0 is performed.

  Further, the translation processing of the present embodiment is performed in the same manner as described in the operations of the second embodiment and the first embodiment using the connected function word example database 41 created here. .

  FIG. 17 is a block diagram illustrating a configuration of an example of the machine translation apparatus according to the third embodiment of this invention.

  Referring to FIG. 17, this embodiment is a computer translation apparatus 90 having input / output functions, information processing functions, and the like, and machine translation having each function such as target language generation means 40 and example database creation means 60a of the present embodiment. This can be realized by loading the program into the memory. The machine translation program is stored in a storage medium 100 that can be read by a magnetic disk, magnetic tape, optical disk, semiconductor memory, or other computer. Then, the above-described functions are realized by being loaded from the storage medium 100 to the computer processing device 90 and controlling the operation of the computer processing device 90.

  In other words, the machine translation program stored in the storage medium 100 is read from the storage medium 100 into the memory of the computer processing device 90, and the operation of the computer processing device 90 is controlled so that the connected function word is stored on the computer processing device 90. In addition to realizing a storage unit such as the example storage unit 941, the input unit 91, the source language analysis unit 92, the language conversion unit 93, the target phrase identification unit 942, the connected function word insertion unit 943, the target language sentence generation unit 944, and the example input unit 961, morphological analysis unit 962, target phrase component extraction unit 963, positive example extraction unit 964, example registration unit 965, negative example extraction unit 966, example adjustment unit 967, and output unit 95 are realized.

  13, the input unit 91 is the input unit 10, the source language analysis unit 92 is the source language analysis unit 20, the language conversion unit 93 is the language conversion unit 30, and the target language generation unit. Each unit in 94 corresponds to each unit in the target language generation unit 40, the output unit 95 corresponds to the output unit 50, and each unit in the example generation unit 96a corresponds to various stages in the example database creation unit 60a.

  This embodiment is characterized in that the computer processing device 90 operates under the control of a machine translation program, and the basic functions are the same as those in FIGS. 13 and 14.

  That is, the computer processing device 90 executes the following processing under the control of the machine translation program.

  In the example registration process, first, according to an example input to the input unit 91, the example input unit 961 receives an input of a target language character string that is a target of example extraction, and activates the morpheme analysis unit 962.

  The morpheme analysis unit 962 performs morpheme analysis on the target language character string input from the example input unit 961, decomposes it into a morpheme sequence, and then activates the correct example extraction unit 964.

  The correct example extracting unit 964 extracts a correct example to be registered in the connected function word example storage unit 941 from the morpheme sequence output as an analysis result by the morpheme analyzing unit 962, and starts the example registering unit 965.

  When extracting the correct usage example, the morpheme analysis unit 962 searches for a predetermined connected function word from the morpheme string output as the analysis result, and from the morpheme immediately before and the morpheme string immediately after the obtained connected function word. The target phrase component extraction unit 963 uses the target phrase component string extracted by the target phrase.

  The example registration unit 965 registers the set of two morphemes extracted as examples by the correct example extraction unit 964 in the connected function word example storage unit 941.

  The negative example extraction unit 966 extracts a negative example to be registered in the connected function word example storage unit 941 from the morpheme string output as the analysis result by the morpheme analysis unit 962, and starts the example adjustment unit 967.

  In the extraction of the negative example, first, the target phrase constituent element extraction unit 963 extracts a constituent element string from the morpheme strings output as the analysis result by the morpheme analyzing unit 962, and the combination of the morphemes in the obtained constituent element string is obtained. Do by taking.

  The example adjustment unit 967 registers the set of two morphemes extracted as an example by the negative example extraction unit 966 in the connection function word example storage unit 941 and adjusts the connection function word flag of the example.

  Further, in the translation process, the process described in the operation of the example of the first and second embodiments is executed using the connected function word example storage unit 941 created by the example creation unit 96.

  By using the machine translation device of the present embodiment described above, information such as the number of detections of both a positive example including a concatenated function word in a sentence phrase and a negative example not included is detected from the input sentence data in the target language. Thus, it is possible to appropriately determine whether or not to use the concatenated function word, to automatically edit and generate example data based on this appropriate determination, and to execute translation based on this appropriate determination.

  In each of the embodiments described above, the example in which the phrase to be processed is a noun phrase and the connected function word to be processed is the case particle “no” is described, but the present invention is not limited to this. The type of target phrase and the number of connected function words to be processed are not limited to one. When there are a plurality of types of phrases to be processed and the number of connected function words to be processed, a plurality of tables for recording example data are provided in the connected function word example database 41, each of which is based on various phrases and connected function words. As described in each of the above-described embodiments, a method for recording example data and a method for recording the phrase data and the type of the connected function word together with the example data in the connected function word example database 41 are used. It can be implemented in the same way as with one type.

  In each of the embodiments described above, as a recording method of the example data in the example database, a description has been given of a method in which a sentence phrase is recorded separately in the front part and the rear part from the position where the connected function word is inserted. Is not to be done. In addition, a method of recording the data based on the data of the entire text phrase and the data at the position where the connected function word is inserted (such as the number of characters from the beginning of the text phrase) can be implemented in the same manner as in each of the above embodiments. Moreover, although the example which makes the number of words of the front and back of a connection function word especially "1" is demonstrated, both the number of words of the front and back is not limited to "one". With the method described in each of the embodiments, it is possible to handle translation using example data of sentence phrases using arbitrary plural words and automatic registration of example data.

  Furthermore, as another embodiment of the present invention, the connected function words described above are words of a word indicating an auxiliary meaning or role that may not be directly described or described in the translation source sentence. Applicable in general, not limited to words that are inserted between the preceding and following words, but can be used for words used at the beginning of sentences and words used at the end, such as English articles. It is done. In other words, this example is also intended for example data when no word is recorded in either the sentence before or after the connected function word in the example data (the number of words is “0”).

  In the automatic registration processing of the example data of the second and third embodiments, the words before and after the connected function word (immediately before and after the connected function word) are checked. If the part of speech is a word used at the beginning of a sentence phrase such as an article, the process of checking the front word is omitted (steps 905 and 1405 are omitted), whereby the second and third The example data can be automatically registered in the same manner as described in the embodiment. In addition, the translation processing using the example data without the words preceding or following the connected function words can be translated by the method described in the first embodiment. As a result, processing such as use / nonuse of articles and selection of “the” and “a” in Japanese-English translation of noun phrases can be executed with reference to the example data.

  Although the present invention has been described with reference to the preferred embodiments and examples, the present invention is not necessarily limited to the above-described embodiments and examples, and various modifications can be made within the scope of the technical idea. Can be implemented.

It is a block diagram which shows the structure of the machine translation apparatus of the 1st Embodiment of this invention. It is a flowchart for demonstrating the translation process of the machine translation apparatus of the 1st Embodiment of this invention. It is a figure which shows an example of the analysis result of the syntax structure of the English sentence of the translation object by the source language analysis means of the 1st Embodiment of this invention. It is a figure which shows an example of the syntax structure of the Japanese sentence of the conversion result by the language conversion means of the 1st Embodiment of this invention. It is a figure which shows an example of the example data stored in the connection function word example database of the 1st Embodiment of this invention. It is a figure which shows an example of the process result of the connection function word insertion process of the 1st Embodiment of this invention. It is a block diagram which shows the structure of one Example of the machine translation apparatus of the 1st Embodiment of this invention. It is a block diagram which shows the structure of the machine translation apparatus of the 2nd Embodiment of this invention. It is a flowchart for demonstrating operation | movement of the example database preparation means of the 2nd Embodiment of this invention. It is a figure which shows an example of the morpheme row | line | column obtained as a result of the morpheme analysis by the morpheme analysis means of the 2nd Embodiment of this invention. It is a figure which shows an example of the example data stored in the connection function word example database of the 2nd Embodiment of this invention. It is a block diagram which shows the structure of one Example of the machine translation apparatus of the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the machine translation apparatus of the 3rd Embodiment of this invention. It is a flowchart for demonstrating the translation process of the machine translation apparatus of the 3rd Embodiment of this invention. It is a figure which shows an example of the morpheme row | line | column obtained as a result of the morpheme analysis by the morpheme analysis means of the 3rd Embodiment of this invention. It is a figure which shows an example of the example data stored in the connection function word example database of the 3rd Embodiment of this invention. It is a block diagram which shows the structure of one Example of the machine translation apparatus of the 3rd Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Input means 20 Source language analysis means 30 Language conversion means 40 Target language generation means 41 Concatenated function word example database 42 Target phrase identification means 43 Concatenated function word insertion means 44 Target language sentence generation means 50 Output means 60, 60a Example database creation means 61 Example input unit 62 Morphological analysis unit 63 Target phrase component extraction unit 64 Positive example extraction unit 65 Example registration unit 66 Negative example extraction unit 67 Example adjustment unit 90 Computer processor 91 Input unit 92 Source language analysis unit 93 Language conversion unit 94 Target language generation unit 941 Linked function word example storage unit 942 Target phrase identification unit 943 Linked function word insertion unit 944 Target language sentence generation 95 Output units 96, 96a Example creation unit 961 Example input unit 962 Morphological analysis unit 963 Target phrase component extraction 964 Positive example extraction unit 965 Example registration unit 966 Negative example extraction 967 Example adjustment unit 100 Storage medium

Claims (12)

In a machine translation device that translates a source language sentence that is a source language into a target language that is a destination language,
The two words of the phrase in which the two words of the target language are in a direct dependency relationship;
Information relating to the suitability of inserting the predetermined connected function word between the two words calculated in advance based on the appearance information of the right example that is an example including the predetermined connected function word between the two words; ,
An example database that holds example data indicating
The appropriateness of the use of the connection function word in the sentence of the target language is determined with reference to the example database, and when using the connection function word as a result, the correspondence in the sentence is referred to with reference to the applied example. And a target language generating means for generating a sentence in which a connected function word is set in a portion to be used and the use of the connected function word is corrected,
The target language generation means includes
Analyzing the structure of the sentence in the target language and detecting the phrase part in the sentence as a phrase to be processed;
Linked function word insertion means for referring to the example data recorded in the example database and inserting the connected function word at a position close to the word of the origin with respect to the phrase to be processed;
In accordance with the insertion processing of the connected function word insertion means and the analysis result of the target phrase identification means, target language sentence generation means for generating a sentence in the target language,
A machine translation device comprising: In a machine translation device that translates a source language sentence that is a source language into a target language that is a destination language,
A source language analysis means for analyzing a source language sentence that is a sentence to be translated;
Based on the analysis of the source language analysis means, a language conversion means for generating conversion data that is data obtained by converting each word and sentence structure of the source language sentence into a target language word and sentence structure as a translation destination language; ,
The two words of the phrase in which the two words of the target language are in a direct dependency relationship;
Information relating to the suitability of inserting the predetermined connected function word between the two words calculated in advance based on the appearance information of the right example that is an example including the predetermined connected function word between the two words; ,
An example database that holds example data indicating
Whether or not the connection function word is used in the converted data is determined by referring to the example database, and as a result, when using the connection function word, the correspondence in the conversion data is referred to by referring to the applied example. A target language generation means for setting a connection function word in a portion to be generated, correcting the use of the connection function word, and generating a target language sentence that is a translation result sentence;
The target language generation means includes
Analyzing the structure of the sentence in the target language and detecting the phrase part in the sentence as a phrase to be processed;
Linked function word insertion means for referring to the example data recorded in the example database and inserting the connected function word at a position close to the word of the origin with respect to the phrase to be processed;
In accordance with the insertion processing of the connected function word insertion means and the analysis result of the target phrase identification means, target language sentence generation means for generating a sentence in the target language,
A machine translation device comprising: Accepts sentence data in the target language as input, extracts positive examples and negative examples that appear in the document data, and includes them in the examples based on the appearance information in the document data of the extracted positive examples and negative examples It claims that between two words to create the example data by calculating information about the advisability of inserting a predetermined connecting function words, characterized in that it comprises the example database creation means for registering the example database Item 3. The machine translation device according to Item 1 or 2 . 4. The machine translation apparatus according to claim 1, wherein one of the sentence phrases is a noun phrase. The target language is Japanese,
5. The machine translation apparatus according to claim 1, wherein one of the predetermined connected function words is a case particle “no”. 6. In a machine translation device that translates a source language sentence that is a source language into a target language that is a destination language,
Two words of a phrase in which the two words of the target language have a direct dependency relationship;
Information regarding the suitability of inserting a predetermined connected function word between the two words;
An example database that holds example data indicating
A correct example extracting means for accepting sentence data of a target language as an input and extracting a correct example that is an example including a predetermined connected function word between the two words;
Based on the appearance information of the correct example extracted by the correct example extracting means, information on the suitability of inserting the predetermined connected function word between the two words is calculated and stored in the example database together with the correct example. An example registration means;
The appropriateness of the use of the connection function word in the sentence of the target language is determined with reference to the example database, and when using the connection function word as a result, the correspondence in the sentence is referred to with reference to the applied example. A target language generating means for setting a connected function word in a portion to be generated and generating a sentence corrected for the use of the connected function word;
A machine translation device comprising: In a machine translation method in a machine translation device that translates a source language sentence that is a source language into a target language that is a destination language,
The machine translation device, wherein the two words of the sentence phrase in which the two words of the target language are in a direct dependency relationship;
Information relating to the suitability of inserting the predetermined connected function word between the two words calculated in advance based on the appearance information of the right example that is an example including the predetermined connected function word between the two words; ,
An example database that holds example data indicating, and target language generation means,
The target language generation means determines whether or not the connection function word is used in the target language sentence by referring to the example database, and when using the connection function word as a result, refer to the applied example. And setting a connected function word in a corresponding part in the sentence, including a target language generating step for generating a sentence corrected for the use of the connected function word,
The target language generation means comprises a target phrase identification means, a connected function word insertion means, and a target language sentence generation means,
The target language generation step includes:
The target phrase identifying means analyzes the structure of a sentence in a target language and detects a part of the sentence phrase in the sentence as a phrase to be processed,
The connected function word insertion means refers to the example data recorded in the example database and inserts the connected function word at a position close to the original word for the phrase to be processed. Steps,
The target language sentence generation means generates a target language sentence according to the insertion processing of the connected function word insertion means and the analysis result of the target phrase identification means; and
A machine translation method comprising: In a machine translation method in a machine translation device that translates a source language sentence that is a source language into a target language that is a destination language,
The machine translation device, wherein the two words of the sentence phrase in which the two words of the target language are in a direct dependency relationship;
Information relating to the suitability of inserting the predetermined connected function word between the two words calculated in advance based on the appearance information of the right example that is an example including the predetermined connected function word between the two words; ,
An example database that holds example data indicating
Language analysis means, language conversion means, and target language generation means,
The source language analysis means for analyzing a source language sentence that is an input sentence to be translated;
Based on the analysis of the source language analysis unit, the language conversion unit converts conversion data that is data obtained by converting each word and sentence structure of the source language sentence into a word and sentence structure of a target language that is a translation destination language. A language conversion step to generate,
The target language generation means determines whether or not the connection function word is used in the conversion data by referring to the example database, and when using the connection function word as a result, refer to the applied example. A target language generation step of setting a connected function word in a corresponding part in the converted data and correcting the use of the connected function word to generate a target language sentence that is a sentence of a translation result,
The target language generation means comprises a target phrase identification means, a connected function word insertion means, and a target language sentence generation means,
The target language generation step includes:
The target phrase identifying means analyzes the structure of a sentence in a target language and detects a part of the sentence phrase in the sentence as a phrase to be processed,
The connected function word insertion means refers to the example data recorded in the example database and inserts the connected function word at a position close to the original word for the phrase to be processed. Steps,
The target language sentence generation means generates a target language sentence according to the insertion processing of the connected function word insertion means and the analysis result of the target phrase identification means; and
A machine translation method comprising: The machine translation device includes example database creation means,
The example database creation means accepts sentence data in a target language as input, extracts positive examples and negative examples that appear in the document data, and appearance information in the document data of the extracted positive examples and negative examples The example database creation step of creating the example data by calculating information on the suitability of inserting a predetermined connected function word between two words included in the example based on the example database and registering it in the example database The machine translation method according to claim 7, wherein the machine translation method is included. In a storage medium recording a machine translation program in a machine translation device that translates a source language sentence that is a source language into a target language that is a destination language,
The machine translation device, wherein the two words of the sentence phrase in which the two words of the target language are in a direct dependency relationship;
Information relating to the suitability of inserting the predetermined connected function word between the two words calculated in advance based on the appearance information of the right example that is an example including the predetermined connected function word between the two words; ,
An example database that holds example data indicating, and target language generation means,
The target language generation means determines whether or not the connection function word is used in the target language sentence by referring to the example database, and when using the connection function word as a result, refer to the applied example. And setting a connected function word in a corresponding part in the sentence, including a target language generating step for generating a sentence corrected for the use of the connected function word,
The target language generation means comprises a target phrase identification means, a connected function word insertion means, and a target language sentence generation means,
The target language generation step includes:
The target phrase identifying means analyzes the structure of a sentence in a target language and detects a part of the sentence phrase in the sentence as a phrase to be processed,
The connected function word insertion means refers to the example data recorded in the example database and inserts the connected function word at a position close to the original word for the phrase to be processed. Steps,
The target language sentence generation means generates a target language sentence according to the insertion processing of the connected function word insertion means and the analysis result of the target phrase identification means; and
A storage medium having a machine translation program recorded thereon. In a storage medium recording a machine translation program in a machine translation device that translates a source language sentence that is a source language into a target language that is a destination language,
The machine translation device, wherein the two words of the sentence phrase in which the two words of the target language are in a direct dependency relationship;
Information relating to the suitability of inserting the predetermined connected function word between the two words calculated in advance based on the appearance information of the right example that is an example including the predetermined connected function word between the two words; ,
An example database for holding example data indicating, a language analysis unit, a language conversion unit, and a target language generation unit,
A source language analysis step in which the language analysis means analyzes a source language sentence that is an input sentence to be translated;
Based on the analysis of the source language analysis unit, the language conversion unit converts conversion data that is data obtained by converting each word and sentence structure of the source language sentence into a word and sentence structure of a target language that is a translation destination language. A language conversion step to generate,
The target language generation means determines whether or not the connection function word is used in the conversion data by referring to the example database, and when using the connection function word as a result, refer to the applied example. A target language generation step of setting a connected function word in a corresponding part in the converted data and correcting the use of the connected function word to generate a target language sentence that is a sentence of a translation result,
The target language generation means comprises a target phrase identification means, a connected function word insertion means, and a target language sentence generation means,
The target language generation step includes:
The target phrase identifying means analyzes the structure of a sentence in a target language and detects a part of the sentence phrase in the sentence as a phrase to be processed,
The connected function word insertion means refers to the example data recorded in the example database and inserts the connected function word at a position close to the original word for the phrase to be processed. Steps,
The target language sentence generation means generates a target language sentence according to the insertion processing of the connected function word insertion means and the analysis result of the target phrase identification means; and
A storage medium having a machine translation program recorded thereon. The machine translation device includes example database creation means,
The example database creation means accepts text data in a target language as input, extracts positive examples and negative examples that appear in the document data, and appearance information in the document data of the extracted positive examples and negative examples The example database creation step of creating the example data by calculating information on the suitability of inserting a predetermined connected function word between two words included in the example based on the example database and registering it in the example database The storage medium according to claim 10 or 11 , wherein a machine translation program recorded therein is recorded.

Images