JP4588657B2 - Translation device - Google Patents

Description

  The present invention relates to a translation apparatus, a translation program, and a translation method for searching and translating an example sentence in the source language and a translation sentence in the translation language of the example sentence as a parallel translation example sentence based on an input sentence in the source language.

  In the field of industrial translation (also called “practical translation”, “business translation”, etc.), it is necessary to translate a large amount of documents efficiently, at high speed and uniformly. Homogeneous translation is a translation in which terms and expressions are unified, and it is necessary to guarantee consistency with terms and expressions in the target field as well as in the work target document.

  In such industrial translation where unity is important, there is a system that accumulates bilingual translation examples created in the past in the same field and supports the translation work while referring to this. It has been used. Such a system is often referred to as a “translation memory” system, but is hereinafter referred to as a “translation example sentence search” in this specification.

The bilingual example sentence search system has a function to accumulate bilingual sentences created from documents of the same content translated in the past as a database, and the translator translates while searching for example sentences similar to the input sentence to be worked on from this database. Proceed with work. Furthermore, the translator can register the translation translated by himself / herself in the database as a parallel translation along with the original sentence, so that the terms and expressions can be unified within the scope of his / her work.

  The conventional bilingual example sentence search system is constructed on the assumption that sentences similar to each other at the sentence level are included in the bilingual example sentence database accumulated in the past with respect to the input of the translation target sentence. For example, when an input sentence and a search target sentence are compared, only one noun is different, and the others are exactly the same. Based on this premise, the ranking of the search results has also been adopted as an effective ranking for the parallel translation example sentences at the sentence level. In addition, the system user uses the system only in an ideal scene where the similarity in the whole sentence can be expected.

  On the other hand, there are overwhelmingly more cases in which actual bilingual example sentences contain sentences that do not match the input sentence at the sentence level but include reusable phrases (compound words, phrases, etc.). Even when diverting the conventional sentence example search system assuming similarities at the sentence level, it is possible as a framework for the user to find a reusable phrase by trying to input a substring of the input sentence by trial and error. It is. However, such a usage method has a large burden in terms of man-hours, and is rarely used in an actual scene where work efficiency is regarded as important.

In addition, the following patent document is known as a reference technique of a prior art (for example, refer patent document 1).
JP-A-10-116286

  As described above, the conventional bilingual example sentence search system is premised on matching at the sentence level, so it takes a lot of trial and error to manually search for sentences containing reusable phrases by diverting them. Therefore, there was a problem that it could not be used in practical situations.

  For example, in the conventional system, when translating the input sentence "This is the pen which I love very much." If there was an ideal bilingual sentence pair like "I once loved a pencil," it was mainly supposed to be searched and reused.

However, in reality, it is rare that similar sentences exist in the database as a whole . For example, “ This is the pen . ”, “They saw the pen which I love. saw the pen that I love. "," she loves it very much. = she is such loves it very much. ", multiple statements such as those containing phrases that make up the said input print statements to the input sentence There are more cases where there is.

  Theoretically, these phrases can be combined to complete the translation, but it takes a lot of work to search for such multiple parallel phrases, and in practical situations such reuse is possible. No method has been made.

  For example, in order to investigate what is in the database as a phrase constituting the input sentence “This is the pen which I love very much.” This is because it is tried and input whether or not each partial sequence such as “This is”, “This is the”...

  This operation is not only troublesome due to repetitive work, but the ranking display of the conventional system is unsuitable for displaying the result of such partial matching, thus reducing the work efficiency.

  Also, even if the ranking method is improved for phrases, trial and error using the conventional system can calculate the appropriate combination of phrases that make up the input sentence, even though it can find individual phrase candidates. I couldn't. In this regard, a simple combination of conventional systems is not sufficient, and a framework for finding a combination of phrase candidates is required.

  Due to such technical problems, the scope of the system has been greatly narrowed. The conventional system for retrieving similar sentences at the sentence level is only for the translation field where there is data suitable for this, but this is only a part of the entire translation market. From the viewpoint of the market as a whole, there are overwhelmingly more example sentences that can be reused at the phrase level than example sentences that can be reused at the sentence level, and there is a problem that the market targeted by the system is greatly limited.

  An object of the present invention is to provide a translation apparatus, a translation program, and a translation method that can efficiently perform translation work regardless of the translation market, and can perform quick and reliable translation. .

  In order to solve the above-described problem, the present invention provides a translation apparatus that searches and translates an example sentence in the first language and a translation sentence of the example sentence in the second language as a parallel translation example sentence based on an input sentence in the first language. An input sentence receiving unit that receives the input sentence, a partial sequence creating unit that creates a partial sequence of example sentences received by the input sentence receiving unit, and a partial sequence of example sentences created by the partial sequence creating unit A bilingual example sentence search unit that uses bilingual example sentences to search for bilingual example sentences in the first language and the second language, a bilingual example sentence searched by the bilingual example sentence search unit, and a partial sequence of example sentences related to the bilingual example sentence An evaluation value providing unit that evaluates the bilingual example sentence and assigns an evaluation value; and a predetermined bilingual example sentence from the bilingual example sentence searched by the bilingual example sentence searching unit based on the evaluation value provided by the evaluation value providing unit Is extracted as a phrase candidate Obtained by a candidate candidate extracting unit, a phrase candidate organizing unit that selects a predetermined phrase from the phrase candidates extracted by the phrase candidate extracting unit, a syntax analyzing unit that analyzes the syntax of an input sentence, and the syntax analyzing unit A semantic analysis unit that analyzes the meaning of the syntax and generates a conceptual structure including a corresponding translation; a corresponding translation search unit that searches for the corresponding translation with a word in the first language in the phrase candidate obtained by the phrase candidate organization unit; When the corresponding translated word is hit by the corresponding translated word search unit, the second language word in the corresponding translated word of the conceptual structure generated by the semantic analyzing unit corresponds to the first language word in the phrase candidate. A translation replacement unit that replaces words in two languages, and a translation generation unit that generates a translation using the translations replaced by the translation replacement unit.

  Here, the generated translated sentence may include a display unit that displays the translated sentence side phrase candidate and the corresponding original sentence side phrase in association with each other.

  The present invention is also a translation program for causing a computer to execute a process of translating an input sentence in a first language into a sentence in a second language, wherein the input sentence accepting step accepts the input sentence and the input sentence accepting step. A partial sequence creating step for creating a partial sequence of accepted example sentences; a parallel example sentence retrieving step for retrieving bilingual example sentences in the first language and the second language using the partial sequence of example sentences created in the partial sequence creating step; An evaluation value providing step of evaluating the parallel translation example sentence searched for based on the parallel translation example sentence searched by the parallel translation example sentence searching step and a partial sequence of the example sentence related to the parallel translation example sentence and assigning an evaluation value; and Based on the evaluation value given in the value assigning step, a predetermined translated example sentence is extracted as a phrase candidate from the parallel example sentence searched in the parallel example sentence searching step. Obtained in the phrase candidate extracting step, the phrase candidate organizing step for selecting a predetermined phrase from the phrase candidates extracted in the phrase candidate extracting step, the syntax analyzing step for analyzing the syntax of the input sentence, and the syntax analyzing step A semantic analysis step that analyzes the meaning of the syntax and generates a conceptual structure including the corresponding translation word, and searches for the first language word in the corresponding translation word from the first language word in the phrase candidate obtained in the phrase candidate organization step. When the first language word in the corresponding target word is hit by the corresponding target word searching step and the corresponding target word searching step, the second language word in the corresponding target word of the conceptual structure generated by the semantic analysis unit is hit. The translated word replacement step for replacing with the second language word corresponding to the first language word. When, it is executed and target generation step of generating a translation using a translation that was replaced by the translation replacing step, to the computer.

  The present invention also provides a translation method for searching and translating an example sentence in the first language and a translation sentence in the second language of the example sentence as a parallel example sentence based on the input sentence in the first language, , A step of creating a partial sequence of example sentences received by the input sentence receiving unit, and a bilingual example sentence of the first language and the second language using the partial sequence of example sentences created by the partial sequence creating unit A step of evaluating the bilingual example sentence retrieved by the bilingual example sentence and a partial sequence of the example sentence related to the bilingual example sentence, and assigning an evaluation value; A step of extracting a predetermined bilingual example sentence as a phrase candidate from the bilingual example sentence searched by the bilingual example sentence based on the evaluation value given by the evaluation value giving unit; and the phrase candidate extracting unit A step of selecting a predetermined phrase from the issued phrase candidates, a step of analyzing a syntax of the input sentence, a step of analyzing a meaning of the syntax obtained by the syntax analysis unit and generating a conceptual structure including a corresponding translation word; And the step of searching for the corresponding translation word with a word in the first language in the phrase candidate obtained by the phrase candidate organizing unit, and when the corresponding translation word hits by the corresponding translation word search unit, the semantic analysis unit generates Replacing a second language word in the corresponding translated word of the conceptual structure with a second language word corresponding to the first language word in the phrase candidate, and using the translated word replaced by the translated word replacing unit, Generating.

  As described above, according to the present invention, a candidate for a translated phrase having a high reusability is automatically calculated in advance for an input sentence, and its existence is presented to the user. A corresponding bilingual example sentence is displayed for the bilingual phrase. Thus, according to the present invention, for the translator who is trying to translate the input sentence in the form of the parallel translation example sentence, the translator can work quickly with high visibility and can efficiently perform the translation work.

  In the present invention, by automatically collecting operations corresponding to trial and error by manual work as mentioned in the problems of the prior art, and configured to present the obtained candidates to the user as an appropriate combination, The time for the user to actually operate can be greatly reduced. In order to realize this, the search system performs a search based on a combination of all subsequences of the translated sentence in advance, and presents the user with a high reuse value as a parallel translation phrase. In addition, an actual search result can be quickly referred to from the presented parallel phrase candidates. Furthermore, when an appropriate combination of phrase candidates for the input sentence is found, a configuration is made such that a translation that uses these phrase candidates as components is generated, thereby further improving the work efficiency. This function can be realized by using conventional machine translation sentence analysis and sentence generation techniques.

Embodiments of the present invention will be described below with reference to the drawings.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing the configuration of the translation support apparatus according to Embodiment 1 of the present invention. The translation support apparatus according to Embodiment 1 includes an input sentence receiving unit 1 that inputs a sentence to be translated as an input sentence (source language: first language), and all possible parts as a partial sequence of input sentences from the input sentence. An input sentence subsequence creating unit 2 that creates a combination of columns, a bilingual example sentence searching unit 3 that searches the databases 10 and 11 for bilingual example sentences from the input subsequence, and a partial candidate searched by the bilingual example sentence searching unit 3 An evaluation value providing unit 4 that performs evaluation on the image and assigns an evaluation value that is an index of appropriateness as a candidate is provided.

  Further, the translation support apparatus includes a phrase candidate extraction unit 5 that extracts a suitable phrase to be presented as a parallel translation phrase candidate from the search results to which the evaluation value is assigned, and a plurality of partial strings in an inclusive relationship. If there are candidates, the phrase candidate organizing unit 6 for organizing the phrase candidates by deleting the duplication while leaving the most suitable candidate among them, and the phrase candidates remaining after organizing are candidates for translation Input sentence creation unit 7 with a phrase candidate for creating an input sentence with a phrase candidate added to the input sentence, a display unit 8 for displaying the input sentence with a phrase candidate on the screen, and a parallel translation based on the phrase candidate for the input sentence A bilingual example sentence display data creation unit 9 that extracts and creates example sentence display data from the databases 10 and 11 is provided.

  The parallel translated example sentence databases 10 and 11 include a source language side database 10 that stores a source language in a translated example sentence and a target language side database 11 that stores a target language in a translated example sentence. The source language example sentence and the target language example sentence which are stored in the databases 10 and 11 and have a parallel translation relationship with each other are translated in a sentence unit.

  The operation of the present invention will be described below with reference to FIGS. FIG. 2 is a flowchart showing the operation of the first embodiment, and FIGS. 3 to 12 are explanatory diagrams showing states in each operation. This operation will be described using English-Japanese translation as an example. That is, here, the input sentence and the source language side example sentence are English, and the target language side example sentence is Japanese. However, this language pair is merely an example, and it goes without saying that the present invention can be applied to any language pair in principle.

  First, the input sentence receiving unit 1 receives one input sentence that is a translation target sentence (step S1). Here, “This is the pen which I love very much.” Is input as an input sentence.

  Next, a partial word string having all lengths starting with each component of the input sentence is created by the input sentence partial string creating unit 2 (step S2). In a language in which the input sentence is divided in units of words such as English, a combination of word strings is created. In FIG. 3, “partial word string” is shown using English word breaks. Create a partial word string of all lengths starting from the first word of the input sentence, then create a partial string of all lengths starting from the second word from the beginning, and so on. All combinations are created up to the substring starting from the last word of.

  When the input sentence is Japanese, a partial sequence may be created in units of characters or a morpheme sequence.

  Next, the bilingual example sentence search unit 3 extracts one partial word string as shown in FIG. 4 from the list of partial strings created in step S2 (step S3), and inputs the extracted partial word string as an input. In the databases 10 and 11, a similar sentence search is performed for the source language side sentence (step S4). As a search result, a candidate list of pairs of source language sentences similar to the input partial word string and target language sentences associated with the retrieved source language sentences is extracted. These candidates are hereinafter referred to as “partial candidates”. When the partial candidates are extracted, the evaluation value assigning unit 4 then assigns a “partial evaluation value” to each partial candidate for each partial sequence, and the “maximum part” is the maximum value of the partial evaluation values in each partial sequence. An “evaluation value” is obtained (step S5).

  In FIG. 4, one partial word string (“This is the pen which”) is extracted from the partial word string list created in FIG. 3, and this is used as a search key to search the parallel example sentence database.

  The lower half of FIG. 4 represents the source language side sentence candidates of the search results and places where evaluation values are assigned to them. The “partial candidate” is a list of source language side sentences of search results for the input partial word string. In the partial candidates, words that match the partial word string are underlined.

For example, in the source language candidate “This is the pen”, since all four words match the partial word string, the “match number” is four. “Candidate side” indicates that 4 words (all words) of 4 words in the source language side candidate sentence are matched. “Input side” indicates that 4 words out of 5 words in the input subsequence match.

  The “partial evaluation value” is an evaluation value for each partial candidate, and here is the product of the “number of matches”, the “candidate side” ratio, and the “input side” ratio. This evaluation formula is an index indicating whether it is appropriate as a phrase presented by a candidate, and the above formula of the present embodiment is an example for realizing it.

  In this way, partial evaluation values are calculated for all partial candidates for the partial word string, and the maximum value among the partial evaluation values becomes the “maximum partial evaluation value” for the partial word string.

  Next, in the above processing, it is determined whether or not the extracted partial sequence is the last entry in the partial sequence list (step S6). If it is the last entry, the process proceeds to the next step S7. If not, the processes in steps S3 and S4 are repeated until the last entry is reached.

  When the loop processing from step S3 to step S6 is completed for all entries, the maximum partial evaluation value is added to the search results of all partial strings. Therefore, as a next process, the phrase candidate extraction unit 5 takes out the partial sequence search results given the maximum partial evaluation value equal to or greater than the threshold value as a phrase candidate (step S7), and the phrase candidate. When there is an overlap among the extracted phrase candidates equal to or higher than the evaluation value, the organizer 6 deletes the overlapped phrase and arranges it (step S8).

  FIG. 5 shows each partial word string and its maximum partial evaluation value. Here, if the threshold value is 3, in the partial word string starting with “This ...”, two of “This is the” and “This is the pen” are equal to or more than the threshold value. None of the partial word strings starting with "is ..." exceed the threshold. Since “This is the” and “This is the pen” are subsequences of the latter, they are arranged, and the latter, which is the longest, remains.

  FIG. 6 shows a list of results obtained by extracting and organizing partial word string candidates having a maximum partial evaluation value equal to or greater than a certain threshold (here, 3). Since each candidate has a corresponding target language side sentence in the parallel translation database, these target language side sentences are also taken out together. These candidates are called “translation phrase candidates”.

  When the translated phrase candidate is acquired, the input sentence creation unit 7 with phrase candidate creates an input sentence with the phrase candidate added to the input sentence to be translated (step S9), and the result is displayed. (Step S10).

  FIG. 7 shows a state where the information of the parallel phrase candidate created in FIG. 6 is added to the input sentence, and shows the display result on the display unit 8. The part of the input sentence corresponding to “This is the pen”, which is the match part in the first candidate in FIG. 6, is displayed with an underline. Similarly, “which I love”, which is the match part of the second candidate, and “very much”, which is the match part of the third candidate, are indicated by underlines.

  FIG. 8 shows a state where the user has further operated the display screen for the phrase-added input sentence shown in FIG. The user recognizes that this subsequence may be a phrase candidate because “This is the pen” in the input sentence is underlined. This partial sequence is configured as a hyperlink, and when the user traces the hyperlink, a bilingual example sentence including the partial sequence is displayed. This bilingual example sentence is created and displayed by the bilingual example sentence display data creation unit 9.

FIG. 9 shows a state in which candidate hyperlinks are traced as in FIG.

  As a display method in the present embodiment, an input sentence can be displayed with phrase candidates as shown in FIG. In FIG. 10, a list of phrase candidates is displayed at the bottom of the input sentence. Each phrase candidate is a hyperlink.

  FIG. 11 shows a display when the user further performs an operation on the display of FIG. In the figure, the user designates “this is the pen” among the phrase candidates. Then, the hyperlink of the phrase candidate is traced, and the corresponding bilingual example sentence is popped up and displayed.

  As another display method in the present embodiment, it is possible to display as shown in FIG. The source language side of the bilingual example sentence is "this is the pen", but as a result of searching the bilingual dictionary, it turns out that the words in these source language side sentences correspond to "this is a pen" on the target language side ing. Therefore, in order to represent this, in the display method shown in FIG. 12, the underlined display is performed on the corresponding portion of the target language side sentence.

Embodiment 2. FIG.
The second embodiment will be described below. In the second embodiment, a translation device that generates a natural sentence translation using the phrase candidate obtained in the first embodiment as a constituent element will be described.

  FIG. 13 is a block diagram showing the second embodiment. The translation device of the second embodiment analyzes the input sentence receiving unit 1 to the phrase candidate organizing unit 6 and the display unit 8, the example sentence databases 10 and 11, and the syntax of the input sentence in the translation support device described in the first embodiment. The syntactic analysis unit 20, the semantic analysis unit 21 that analyzes the meaning of the syntax, the corresponding translation search unit 24 that searches for the corresponding translation from the phrase candidates obtained by the phrase candidate organization unit 6, and the analysis result of the syntax semantics are searched. A translation change unit (translation replacement unit) 22 that changes a translation based on the corresponding translation, a translation generation unit 25 that generates a translation using the changed translation, and information analyzed by the semantic analysis unit 21 is stored. And an analysis result storage database 23.

  The translation device according to the second embodiment outputs a sentence in the form of a sentence using a machine translation sentence generation technique for a target language side phrase group corresponding to a phrase candidate constituting an input sentence.

  The syntax analysis unit 20 is a syntax analysis device conventionally used in the field of natural language processing. The syntax analysis unit 20 analyzes the syntax structure after decomposing the input character string from the input sentence reception unit 1 into morpheme strings, and constructs the sentence syntax. Outputs a syntax tree representing information.

  The semantic analysis unit 21 is a semantic analysis unit as used in conventional machine translation, and outputs a conceptual structure by performing semantic analysis on the syntax tree obtained from the syntax analysis unit 20. The output conceptual structure is accumulated in the analysis result storage database 23 and is simultaneously transferred to the translated word changing unit 22.

  The corresponding translated word search unit 24 extracts only the first candidate from the target language candidate group corresponding to each source language phrase candidate output from the phrase candidate organizing unit 6. Here, matching is performed between the word in the phrase candidate on the source language side and the word in the first candidate on the target language side to find the corresponding translated word. Further, the source language word in each concept node in the concept structure of the analysis result storage database 23 is searched using the source language side word of the corresponding translation word obtained in this way.

  When the search for the concept node is hit in the corresponding translated word search unit 24, the translated word changing unit 22 replaces the corresponding target language side word string with the target language side word obtained from the phrase candidate.

  The translation generation unit 25 outputs the translation using the existing sentence generation device from the conceptual structure in which the translation (target language side word) is replaced by the translation change unit 22. As the sentence generation device, one conventionally used for machine translation or the like is used.

  In this way, in the present embodiment, the syntax analysis unit 20 extracts the syntax structure from the source language sentence using the syntax analysis, converts the obtained syntax structure into the conceptual structure by the semantic analysis unit 21, and translates the sentence structure. The generation unit 25 generates a target language translation from the converted conceptual structure. At the syntax structure stage, the target language side word of the phrase candidate is generated by the translation change unit 22 and the corresponding translation search unit 24. The translation is generated using the phrase. Needless to say, it may be similarly applied using machine translation by other methods.

The operation of the second embodiment will be described below using the flowchart shown in FIG.
First, an input sentence is accepted (step S21). Here, the process for the input sentence branches to steps S22 and S27. Step S22, which is one of the branches, is a process corresponding to steps S1 to S8 shown in FIG. 2, and a phrase candidate is created from the input sentence.

  In step S23, one source language side phrase is extracted from the created phrase candidates, and in step S24, each word is extracted one by one from the extracted phrase candidates, and the target language side phrase group is set using each word as a key. Are searched for. By step S25, the process is repeated for all words for each phrase on the source language side. In step S26, the phrase pairs in which the words are hit and the word pairs themselves are counted in the repetition process up to step S25.

  In step S27 which is the other of the branches, syntax analysis is performed on the input sentence in step S21, and a syntax tree is output as a result.

  FIG. 15 is an explanatory diagram illustrating a state in which the syntax analysis unit 20 performs syntax analysis on the input sentence. In terms of internal processing, the syntax structure of the input sentence is stored in a tree structure. The parsing method and the result storage method are generally used in natural language processing such as machine translation, and a detailed description of the processing is omitted.

  In FIG. 15, S represents a sentence (Sentence), NP represents a noun phrase, VP represents a verb phrase, “this”, “is”, etc. represent end nodes of the tree structure, Overall, it represents the grammatical structure of the source language.

  In step S28, semantic analysis is performed on the syntax tree, and a conceptual structure is generated as a result.

  FIG. 16 is an explanatory diagram illustrating a state in which the conceptual structure is generated by the semantic analysis unit 21 from the tree structure illustrated in FIG. 15. Also in the conversion from the syntax structure to the conceptual structure, a well-known method generally used in machine translation or the like is used.

  In the figure, concept nodes are represented by circles, relational arcs are represented by arrows, and concept names are represented by square brackets. The concept in the concept node has both a source language side word and a target language side word group. For example, the concept “this” is associated with “this” which is a source language word in the original sentence. Note that these concepts also have target language side words such as “m” and “here” (not shown).

  These target language side word groups are initially stored in the parallel translation dictionary for machine translation, and information is transmitted to each concept node when the concept structure is generated. The concept structure represents a connection between concepts, and a translation can be generated by applying a target language generation rule thereto.

  In the case of normal machine translation, as described above, words of a target language corresponding to each concept are stored, and generation is performed using these words. For example, target language words such as “I” and “I” are stored for the concept “I”.

  In this embodiment, instead of using the target language word string prepared by machine translation, the target language side word string of the translated phrase is assigned to the concept nodes using the phrase candidates arranged by the phrase candidate organizing unit 6, Translation is generated using the target language side word.

  FIG. 17 is an explanatory diagram showing a state in which only the first candidate phrase is extracted, and is in a state in which word correspondence is attached by dictionary lookup of the bilingual dictionary. For example, in FIG. 8, there are three target language sentences, Japanese sentences, as a parallel translation of “This is the pen”, and “this is the pen” at the top of the sentences is adopted.

  As ranking for multiple candidates, the similarity between the source language side character string and the target language side character string can be used, and the similarity degree itself can be calculated by various known methods such as "edit distance". Can do. Similarly, the target language character string of the first candidate is adopted for “which I love” and “very much”.

  In this way, after assigning a target language character string to each phrase one by one, correspondence between words is taken on the source language side and the target language side. This uses a bilingual dictionary such as a machine translation bilingual dictionary. For example, if “this” is included as a translation of “this” in the bilingual dictionary, these are associated. Similarly, association processing is performed for other phrase pairs.

  In this example, “I” appearing in the parallel translation phrase is assigned to the concept “I” instead of “Watakushi” and “I” prepared by the machine translation dictionary. This allocation process will be described below.

  In step S29, the phrase pair associated with the word from step S26 and the input sentence in the form of the conceptual structure from step S28 are input. Here, one phrase is taken out. In step S30, the source language word of each concept node of the concept structure is searched for in each word on the source language side of the phrase.

  In step S31, if there is a hit in the search in step S30, the target language side word of the concept node is replaced with the corresponding target language side word in the phrase.

  In step S32, the process is repeated for each phrase. In step S23, a translated sentence generation process is performed on the conceptual structure in which the target language side word has been replaced, and an output sentence is generated.

  FIG. 18 is an explanatory diagram showing a state in which the translated word side phrase is associated with each concept of the conceptual structure shown in FIG. 16 from the word correspondence information of the phrase candidate.

  As described above, each concept node stores a source language word and a target language side word group. Here, the target language side word group stored originally is not used. When the source language word of each concept node is searched with the source language side word shown in FIG. 17 and hit, the word of the associated target language side word phrase is assigned to the concept node. Actually, since the concept node is centered on a noun concept or a verb concept, an article such as “the” and a relative noun such as “which” do not hit.

  In this way, as shown in FIG. 18, the words in the parallel translation phrase are assigned to each concept node. For nodes that do not hit a word in the phrase, a word can be selected from the target language side word group originally prepared by machine translation.

  A translation generated by using this translated side phrase as a designated translated word is shown in the lower part of FIG. For the translation generation from the conceptual structure, a conventional translation generation technique is used.

  FIG. 19 shows a display state in which the translation relationship of phrase candidates is displayed for the translation generated with the translated word side phrase candidate as the designated translated word.

  As described above, the embodiment of the present invention has been described by taking an example when English is translated into Japanese. However, the present invention is all capable of searching a parallel example sentence in each section by dividing an input sentence into substrings. Needless to say, all existing methods can be applied to the above-described syntax analysis method and semantic analysis method.

  The translation program of the present invention can be recorded on a computer-readable recording medium. Here, the computer-readable recording medium includes a CD-ROM, a flexible disk, a DVD disk, and a magneto-optical disk. It also includes a portable storage medium such as an IC card, a database holding a computer program, another computer and its database, and a transmission medium on a line.

It is a block diagram which shows the structure of the translation assistance apparatus which concerns on Embodiment 1 of this invention. 3 is a flowchart showing the operation of the first embodiment. It is the state figure which represented the input sentence by the partial word string. It is explanatory drawing for demonstrating the acquisition method of the largest partial evaluation value. It is explanatory drawing which shows the largest partial evaluation value for every partial word string. It is a state diagram which shows the list of the result of having taken out and rearranging the partial word sequence candidate with the maximum partial evaluation value of 3 or more. It is a figure which shows the state which provided the information of the bilingual phrase candidate produced in FIG. 6 to the input sentence as a display result. It is a figure which shows the state which the user performed further operation with respect to the display screen of the input sentence with a phrase candidate shown in FIG. FIG. 9 is a diagram illustrating a state in which candidate hyperlinks are traced, as in FIG. 8. It is a figure which shows the display state which displayed the input sentence with the phrase candidate. It is a figure which shows the display state when a user further operates with respect to the display of FIG. It is a figure which shows the display state which performed the display by the underline to the part corresponding to a target language side sentence. It is a block diagram which shows the structure of the translation apparatus which concerns on Embodiment 2 of this invention. 10 is a flowchart showing the operation of the second embodiment. It is explanatory drawing which shows the state which is performing the syntax analysis with respect to an input sentence by a syntax analysis part. It is explanatory drawing which shows the state which produced | generated the conceptual structure by the semantic analysis part from the tree structure shown in FIG. It is explanatory drawing which shows the state which extracted only the 1st candidate of a phrase candidate. It is explanatory drawing which shows the state which matched the translation side phrase from the word corresponding information of a phrase candidate with respect to each concept of the conceptual structure shown in FIG. It is a figure which shows the state which displayed the translation relationship of the phrase candidate about the translated sentence produced | generated as the translation target side phrase candidate.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Input sentence reception part, 2 Input sentence subsequence preparation part, 3 Bilingual example sentence search part, 4 Evaluation value provision part, 5 Phrase candidate extraction part, 6 Phrase candidate arrangement part, 7 Phrase candidate input sentence preparation part, 8 Display part , 9 Bilingual example sentence display data creation unit, 10, 11 Bilingual example sentence database, 20 syntax analysis unit, 21 semantic analysis unit, 22 translation change (replacement) unit, 23 analysis result storage database, 24 corresponding translation search unit, 25 translation generation unit .

Claims (2)

A translation device that searches and displays an example sentence of a first language and a translated sentence of the example sentence in a second language as a bilingual example sentence based on an input sentence in a first language composed of a plurality of words inputted as electronic data. ,
An input sentence receiving unit for receiving the input sentence consisting of a plurality of words as electronic data;
A partial sequence creating unit that creates a partial sequence of the input sentence accepted as electronic data by the input sentence accepting unit, using electronic data;
A bilingual example sentence search unit for retrieving from the database a bilingual example sentence in a first language similar to the partial string and a second language corresponding to the first language, using the partial sequence of the input sentence created by the partial sequence creating unit; ,
The partial candidate of the first language with the number of matches, which is the number of words constituting the partial candidate of the first language similar to the partial sequence retrieved from the database by the bilingual example sentence search unit, which matches the word of the partial sequence An evaluation value that calculates a ratio between the number of words that constitutes and a ratio between the number of words in the partial sequence and the number of matches, and calculates and assigns evaluation values of the partial candidates based on the calculated values of these ratios A granting unit;
A phrase candidate extraction unit that extracts, as a phrase candidate, a bilingual example sentence in which the evaluation value given by the evaluation value giving unit is a predetermined value or more;
A phrase candidate organizing unit that deletes duplicated corresponding partial sequences from the phrase candidates extracted by the phrase candidate extracting unit;
A syntax analysis unit that parses the input sentence by machine translation and generates a syntax tree;
A semantic analysis unit that analyzes the meaning of the syntax tree obtained by the syntax analysis unit by machine translation and generates a conceptual structure including a corresponding translation word;
A corresponding word search unit for searching for a first language in the concept obtained by the semantic analysis unit with a word in the first language in the phrase candidate obtained by the phrase candidate organizing unit;
When the corresponding translated word is hit by the corresponding translated word search unit, the second language word in the corresponding translated word of the conceptual structure generated by the semantic analyzing unit corresponds to the first language word in the phrase candidate. A translation replacement unit that replaces words in two languages;
A translation apparatus comprising: a translation generation unit that generates a translation as display data using the translation replaced by the translation replacement unit. The translation device according to claim 1,
A translation apparatus comprising a display unit that displays a translation-side phrase candidate and a corresponding original-sentence phrase in association with each other in the generated translation.

Images