WO2000049523A1

WO2000049523A1 - Translating method and translating device

Info

Publication number: WO2000049523A1
Application number: PCT/JP2000/000907
Authority: WO
Inventors: Masato Shimakawa; Masanori Omote; Naoto Iwahashi
Original assignee: Sony Corporation
Priority date: 1999-02-18
Filing date: 2000-02-17
Publication date: 2000-08-24
Also published as: JP2000242642A

Abstract

A translating device (10) stores in a translation information database (18) in addition to an original language text and its object language text, a user confirmation original language text written in the original language and edited by making an addition or a change to or in the expression on the basis of the object language text, a situation describing text describing the situation where the expression of the user confirmation original language text is used, a situation describing image and a situation describing speech relating to the situation where the expression is used, and translation information including information concerning the original language text and the object language text, that is, the application limit applied to the candidates of the results of translation of the original language text. The original language text is translated into an object language text by using the translation information.

Description

TECHNICAL FIELD The present invention relates to a translation processing method and a translation processing device for translating an input source language sentence into a target language sentence. BACKGROUND ART In recent years, translation systems for translating an input arbitrary language sentence into a different language sentence have been put to practical use. Improving the translation accuracy and the quality of the translated sentence is currently regarded as important in this translation system, and various research and development are being carried out for that purpose.

A so-called syntactic transfer type translation system focuses on the syntactical correspondence between the original sentence and the translated sentence, and performs translation based on a large number of manually described rules.

However, in this syntactic transfer-type translation system, describing the syntactic partial correspondence in detail does not necessarily lead to an improvement in the translation accuracy of the entire original sentence. Also, in a syntactic transfer type translation system, the entire translation is generated using information selected based on partial syntactical relationships instead of syntactical relationships that take into account the entire original sentence. It is difficult to get a natural translation. In addition, the so-called knowledge-based translation system expresses the meaning of the original sentence explicitly when translating the original sentence, efficiently organizes the knowledge of a relatively limited range of target fields, and then converts this concept system. Perform the inference used and translate. This knowledge-based translation system realizes high-quality translation by performing inference using a concept system, thereby eliminating ambiguity associated with analysis and generation.

However, in a knowledge-based translation system, it is difficult to maintain the consistency of the conceptual system as its scale increases. Another problem with knowledge-based translation systems is that it takes an enormous amount of time and effort to explicitly describe the meaning of the original text.

Furthermore, the so-called example-based translation system enables the translation accuracy to be improved by collecting and using a large amount of corpus, and it is possible to generate a natural translation in comparison with other translation methods. It is said that there is.

However, in this example-based translation system, generally, in order to determine the similarity to the example, a distance is determined in a conceptual space using a thesaurus, or the syntax structure is analyzed after parsing. Judgment criteria are unique throughout the translation system, such as performing similarity judgment. Therefore, in the example-based translation system, if the number of corpora is not sufficient, the translation accuracy is significantly reduced. Therefore, the example-based translation system requires a huge amount of time and effort to maintain the corpus, except for specific uses such as compound word analysis, and is sufficient to construct a practical system. is not.

Furthermore, as described in, for example, Japanese Patent Application Laid-Open No. 5-29082 and Japanese Patent Application Laid-Open No. 110-14747, the translation example is abstracted. If the input sentence cannot be composed as a combination of these patterns, translation is performed using a set of less restrictive patterns, or the above-mentioned syntax transformer is used. There is a technique for performing the translation of the web-form. By doing so, this method enables natural translation of only specific expressions while maintaining the maintainability of the translation system as a whole.

However, in this method, if the original text cannot be composed as a combination of patterns, there is a problem that the translation accuracy is reduced. In addition, since the original text can be interpreted in multiple meanings depending on the context and the user's intention, this method does not always match the user's intended content even if it matches the pattern combination. Is not always available.

Such conventional translation systems are based on the premise that in any case, the input original text is completely automatically translated and output. For this reason, the above-mentioned problems associated with the conventional translation system stem from the fact that research and development are being conducted under these preconditions from the perspective of how to improve translation accuracy and translation quality. By the way, in conventional translation systems, although the translation accuracy is gradually improving due to the maintenance of language data, for example, users who do not have any knowledge of the target language, which is the language of the translated sentence, can use it with confidence. The translation quality is not

'Ivy. Disclosure of the invention The present invention has been made in view of the above circumstances, and has a problem in a conventional translation system that presumes that an input original text is completely automatically translated and output in any case. It is an object of the present invention to provide a translation processing method and a translation processing apparatus that can obtain a high-accuracy and high-quality translation result according to a user's intention that has been impossible in the past.

The translation processing method according to the present invention that achieves the above-described object includes a sentence or a part of a sentence in an input arbitrary source language and a sentence or a part of a sentence in an arbitrary target language different from the source language. Based on the bilingual information in which the sentence is matched, the matching or similarity between the sentence or part of the sentence in the source language and the sentence composed of the combination of the translated information is determined, and the sentence or the sentence in the source language is determined. A translation processing method in which a part is translated into a sentence or part of a sentence in the target language and output. The bilingual information includes a sentence or part of the sentence in the source language and a part of this sentence or part of the sentence. It is characterized by a sentence or part of a sentence in the target language that is a parallel translation, and a sentence or part of the sentence in the source language and related information related to the sentence or part of the sentence in the target language. I have.

Such a translation processing method according to the present invention includes a sentence or part of a sentence in the source language, a sentence or part of the sentence in the target language, and a sentence or part of the sentence in the source language and the purpose. Translates a sentence or part of a sentence into a target language or part of a sentence in a target language using bilingual information that describes relevant information related to a sentence or part of a sentence in a language ₌ A translation processing apparatus according to the present invention that achieves the above-described object includes: input means for inputting a sentence or a part of a sentence in an arbitrary source language; Sentence or part of a sentence, and a sentence or part of a sentence in an arbitrary target language different from the source language, which is a translation of this sentence or part of the sentence, are described in association with each other. A storage means for storing bilingual information in which a sentence or a part of a sentence and related information related to the sentence or a part of the sentence in the target language are written; a sentence or a part of the sentence in the source language based on the translated information; Analysis means for analyzing a sentence or a part of the sentence in the source language to generate a translation result candidate by judging the coincidence or similarity with the sentence composed of the bilingual information; Target language sentence generating means for generating a sentence or part of a sentence in the target language which is a translation of a sentence or a part of the sentence according to the target language, and a sentence or part of the sentence in the target language generated by the target language sentence generating means Output means for outputting It is characterized by a door.

Such a translation processing device according to the present invention can provide a sentence or a part of a sentence in the source language, a sentence or a part of the sentence in the target language, a sentence or a part of the sentence in the source language, and an object. The bilingual information in which relevant information related to a sentence or a part of the sentence in the language is described is stored in the storage unit, and the sentence or a part of the sentence in the source language is converted into the target language using the translated information. To a sentence or part of a sentence. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 is a block diagram that describes the configuration of the translation apparatus shown as an embodiment of the present invention _c

FIG. 2 is a diagram for explaining an example of the bilingual information held by the translation apparatus, and is a diagram for explaining the bilingual information in which link information is described. FIG. 3 is a diagram illustrating an example of the bilingual information held by the translation apparatus, and is a diagram illustrating the bilingual information in which field information and field limited information are described.

FIG. 4 is a block diagram illustrating a configuration of a bidirectional translation apparatus which is an application example configured by combining the translation apparatuses.

FIG. 5 is a diagram for explaining an example of the bilingual information held by the bidirectional translation device, and is a diagram for explaining the bilingual information in which response prediction information is described.

FIG. 6 is a plan view illustrating the appearance of a speech translator that more specifically configures the bidirectional translation device.

FIG. 7 is a block diagram illustrating the configuration of the speech translator.

FIG. 8 is a flowchart illustrating a series of processing steps in the speech translator.

FIG. 9 is a diagram illustrating a series of processing steps in the speech translator, and is a flowchart illustrating a processing step performed subsequent to the processing step illustrated in FIG.

FIG. 10 is a diagram for explaining an example of the bilingual information held by the speech translator.

Figure 11 is a diagram illustrating a specific example of bilingual information held by the speech translator:

FIG. 12 is a diagram illustrating an example of user confirmation information generated by the speech translator, and is a diagram illustrating a state when the user confirmation information is output from a liquid crystal unit.

FIG. 13 is a diagram for explaining another example of the user confirmation information generated by the speech translator. FIG. 6 is a diagram for explaining a state when output is performed by a speed and speed.

Fig. 14 is a diagram for explaining an example of the target language sentence and the response prediction information generated by the speech translator, when the target language sentence and the response prediction information are output by the liquid crystal unit and the speed. FIG.

FIG. 15 is a diagram for explaining another example of the target language sentence and the response prediction information generated by the speech translator.The target language sentence and the response prediction information are output from the liquid crystal unit and the speaker. It is a figure explaining the situation at the time. BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, specific embodiments to which the present invention is applied will be described in detail with reference to the drawings.

A translation device 10 shown in FIG. 1 as an embodiment to which the present invention is applied is intended to ensure the quality of a target language sentence that is a translation corresponding to a source language sentence. So-called example-based translation, pattern-based translation, As with the translation method called template-based translation, the combination of the source language sentence or a part of it and the corresponding target language sentence or part of it is part of the basic information (translation information) in the translation process. Furthermore, the translation processing is performed by describing information related to the source language sentence in the bilingual information, and performing a match determination or a similarity determination with a sentence formed by the combination of the bilingual information.

It should be noted that the translator 10 is capable of translating not only sentences in the source language, but also units constituting sentences such as arbitrary phrases and words. It is. Therefore, in the following description, these sentences, phrases, words, and the like to be translated are collectively used as sentences. In other words, the source language sentence indicates a sentence, phrase, word, or the like in the source language, and the target language sentence indicates a sentence, phrase, word, or the like in the target language.

As shown in FIG. 1, the translation device 10 includes, as shown in FIG. 1, a sentence input unit 11 which is an input means for inputting a source language sentence such as a voice or a character and converting the sentence into data that can be analyzed, and a bilingual information database described later. Sentence analysis unit 12 which is an analysis means for judging the similarity or similarity between the sentence composed of the bilingual information in 18 and the source language sentence, and the translation result for the source language sentence Constraint judging unit 13 which is an application constraint judging means for judging whether or not the candidate satisfies the application constraint given to the bilingual information, and user confirmation information generation for generating confirmation information to be presented to a user Unit 14, a user confirmation unit 15 for presenting the confirmation information generated by the user confirmation information generation unit 14 to the user, and a target language sentence which is a target language sentence generating means for generating a target language sentence. Generation unit 16 and generation of this target language sentence It has a translation result output unit 17 which is an output means for outputting the target language sentence generated by 16 as, for example, voice or characters, and a bilingual information database 18 which is a storage means for holding various types of bilingual information. .

Such a translation device 10 first performs, for example, digitization processing on the input source language sentence by the sentence input unit 11. The translation device 10 inputs the source language sentence processed to the sentence analysis unit 12, and the sentence analysis unit 12 causes the input source language sentence and the bilingual information database 18 Judgment of similarity or similarity with a sentence composed of a combination of bilingual information is performed. This allows the translation The device 10 generates, as a candidate, an expression similar to the syntax of the original language sentence, regardless of the context or the user's intention in the source language sentence.

Next, the translation device 10 determines, by the application constraint determination unit 13, whether or not the plurality of generated candidates satisfies all the application constraints individually assigned to each piece of translation information. Do for That is, the translation device 10 describes the application constraint for each piece of parallel translation information instead of using a unique criterion indicating the applicability of each piece of translation information: As a result of this determination processing, when it is recognized that there is an expression that does not satisfy all of the application constraints among a plurality of candidates, this expression is removed from the candidates. As a result, the translation device 10 can perform high-accuracy translation without having a huge number of corpora.

In the translator 10, this determination process may be performed by the sentence analyzer 12. Also, in the translation apparatus 10, as in the example-based translation system, it is possible to use a unique applicability determination criterion such as a distance in a conceptual space using a thesaurus. The accuracy of matching or similarity judgment between the source language sentence and the sentence composed of the bilingual information is slightly reduced.

Further, if the translation apparatus 10 recognizes that no candidate exists at this point, the translation apparatus 10 notifies the user that the source language sentence cannot be translated, for example, by voice or text, and expresses another expression. Ask the user to re-enter the source language sentence in. In this way, the translator 10 does not translate any input source language sentence, but handles it by performing applicability determination processing. You can reject source language sentences that cannot be done.

Next, the translation device 10 generates information to be presented to the user by the user confirmation information generation unit 14 based on the information given to the respective translation information constituting each selected candidate. The generated information is presented to the user by the confirmation unit 15. c In response to this, the user selects a candidate that best matches the context and intention. In other words, in the translator 10, if the source language sentence can be interpreted multiple times depending on the context and the user's intention, and the corresponding target language sentence can be generated multiple times, the final target language sentence can be obtained. Generates the information needed to confirm with the user to determine

Here, the “information required for confirmation to the user” is a source language sentence for user confirmation, which is a sentence in the source language in which the expression has been added or changed so that the meaning matches the target language sentence, or , A situation description sentence that explains the situation where the expression of the user confirmation source language sentence is used, a situation explanation image that is also an image related to the situation where the expression is used, and a situation where the expression is also used This refers to the situation explanation sound, which is the sound related to.

In other words, the source language sentence for user confirmation is a sentence in the source language in which the expression relating to the context or the user's intention is added or changed based on the target language sentence for the input source language sentence. This is a sentence in which the paraphrased sentence is expressed in the source language. A situation explanation is a sentence that explains a situation that can be considered from the context or the user's intention, in addition to the expression in the source language sentence. In addition, the situation explanation image and the situation explanation sound are not limited to the expression in the source language sentence, but the content of the target language sentence is appropriate in relation to the context or the user's intention. Image and sound. It should be noted that the situation description image is not limited to a still image or a moving image.

By using the source language sentence for user identification in place of the source language sentence at the time of user identification, the translation device 10 can generate a translation result that more accurately reflects the context or the intention of the user. Also, the translation device 10 can use the situation description, the situation description image or the situation description sound, and the source language sentence at the time of user confirmation, and can translate the context or the user's intention. Results can be generated. Note that if the input source language sentence and the input source language sentence for user confirmation to be described in the bilingual information corresponding to the source language sentence are the same, the translator 10 checks the user in the bilingual information. Description of source language sentence is omitted. As described above, when a plurality of target language sentences corresponding to the input source language sentence can be generated, the translation device 10 determines which of these candidates should be output as the final translation result. The user is asked for confirmation and is selected, and the meaning of the final target language sentence is determined accordingly. Therefore, the user does not recognize at the time of input that there are multiple interpretations in the source language sentence with a specific context or intention, but the translation result from the translation device 10 is not recognized. By presenting a plurality of candidates with the above information for confirmation, an appropriate one can be selected from among them. As a result, the translation device 10 can generate a correct translation result even when the context of the source language sentence or the user's intention cannot be determined by inference or the like, and can improve translation accuracy.

Then, after confirmation and selection by the user, the translation device 10 causes the target language sentence generation unit 16 to execute bilingual information related to the candidate selected by the user. Based on the report, a target language sentence is generated, and the generated target language sentence is output by the translation result output unit 17 using, for example, voice or characters, and a series of processing is completed.

By performing such a series of processes, the translation device 10 can independently process the translation process for each input source language sentence. Also, the translator 10 changes the contents of the bilingual information, so that each translation process is not independent, and the translation process for one source language sentence affects the translation process of the next input source language sentence. You can also translate. In other words, the source language sentence input after the source language sentence for a certain topic is likely to be about the same topic. In order to use this property, the translator 10 adds information for limiting the next source language sentence to the bilingual information, as shown in Figs. 2 and 3. I do.

Here, the “information for limiting the source language sentence to be input next” refers to link information to bilingual information, field information, field-limited information, and the like.

In other words, the link information shown in Fig. 2 is provided to limit the target bilingual information to a very small number when analyzing the sentence of the next input source language sentence. This is the information associated with the bilingual information that should be used in. In the translation device 10, a plurality of link information may be described in one piece of translation information. When using this link information in the translator 10, the bilingual information includes, in addition to the link information, the bilingual information that is always used even if the source language sentence is input after any source language sentence. The common information flag for indicating is described in _c. Specifically, as shown in Fig. 2, the link information in a certain bilingual information A a is linked to the bilingual information B, link information b is linked to the bilingual information C, and link information 1 is linked to the bilingual information N. When the bilingual information A is used, the translator 10 uses the link destinations from the link information a, b,..., N when analyzing the sentence of the next input source language sentence. It is only necessary to use the bilingual information B, C, ...

On the other hand, the field information and the field limitation information shown in FIG. 3 are information provided for limiting the target field when the sentence analysis of the next input source language sentence is performed. The field information describes information indicating the field in which the bilingual information is used. The field-limited information describes information indicating the field to which the bilingual information to be used after translation of the source language sentence using the bilingual information belongs. In the translation device 10, a plurality of these pieces of information may be described in one piece of bilingual information. The translator 10 uses the link information, the field information, and the field-specific information to translate relatively typical conversational content, such as travel conversations. Candidates generated during translation of source language sentence can be limited, and the burden of asking the user for selection can be reduced.

By combining the two translation devices 10 as described above, a translation device capable of bidirectional translation can also be configured-that is, as shown in FIG. 4, the translation device 10 has the same configuration as the translation device 10 described above. Yes, the translation device 10a that translates the input language A as the source language into the target language B, and also has the same configuration as the translation device 10; the source language and the object in the translation device 10a A translation device that replaces the language and translates the input language B as the source language into the target language A. Consider a bidirectional translator 20 that combines 10 b with 10 b.

— In such a bidirectional translator 20, there is a case where the other user inputs a response sentence in response to a result of translating and outputting an input from one user. Therefore, in the bidirectional translator 20, information for limiting the sentence to be input next by the other party in the bilingual information for one translation, that is, information for associating with the other bilingual information for translation is used. To achieve more accurate translation.

The bidirectional translator 20 adds the bilingual information in the bilingual information databases 18a and 18b to, for example, link information to bilingual information in bidirectional translation, and field information and field-limited information in bidirectional translation. Is described.

That is, the link information is provided, for example, to limit the translation information used when responding to the source language sentence input to one of the translators 10a from the other translator 10b to a very small number. Of course, the bidirectional translator 20 may describe a plurality of pieces of link information in one piece of bilingual information.

Here, for the sake of specific explanation, the link information a, b, and c in certain translation information in one translation device 10a are respectively different from each other in the other translation device 10b. The link information a, b, and c in certain bilingual information in the translator 10b are respectively linked to different bilingual information in the translator 10a. Shall be Then, at the time of the sentence analysis of the source language sentence in the language A input to one of the translators 10a. When bilingual information in the bilingual information database 18a is used, the bidirectional translator 20 Next, to the language B input to the translation device 10 b When the source language sentence is analyzed, the translation information in the translation information database 18 b, which is the link destination from each of the link information a, b, and c in the translation information database 18 a It is sufficient to use only In addition, as a response at this time, when the bilingual information in the bilingual information database 18 b is used during the sentence analysis of the source language sentence input to the translator 10 b, the bidirectional translator 20 Next, when analyzing the source language sentence input to the translator 10a, the link destinations from the link information a, b, and c in the bilingual information in the bilingual information database 18b It is sufficient to use only the bilingual information in the bilingual information database 18a.

On the other hand, field information and field-limited information are, for example, sentence analysis of a source language sentence input to the other translator 10b in response to a source language sentence input to one translator 10a. This information is sometimes provided to limit the target field. In the bidirectional translator 20, a plurality of such information may be described in one bilingual information in each translator 10 a, 10 b ₌

The bidirectional translator 20 uses these link information, field information, and field-limited information to translate relatively typical conversation contents such as travel conversations. For any of the source language sentences input to 10a and 10b, the translation information to be used can be limited and the candidates generated during translation can be limited, resulting in high accuracy and high quality Can be translated

The bidirectional translator 20 can also be used as a dialogue-assisted translator that translates with real time and supports dialogue. In this case, as shown in FIG. A prediction example regarding a response to the source language sentence input to the translation device 10a is described in the bilingual information as response prediction information such as a response sentence, a response image, and a response voice. Then, the bidirectional translator 20 outputs the translation result, and may respond to the user of the translator 10a by the dialogue partner of the translator 10b with what kind of response. Is illustrated. Note that the response image may be a still image or a moving image. _C By doing so, the bidirectional translator 20 can make the user's conversation smooth. Becomes possible.

The translator 10 to which the embodiment of the present invention is applied and the bidirectional translator 20 as an application of the translator have been described above. FIG. A dialogue-assisted speech translator 30 as shown in FIG.

As shown in FIG. 6, the speech translator 30 is used in a form in which a user and a conversation partner with this user face each other with the speech translator 30 interposed therebetween. The speech translator 30 is used as a shared part between the user and the conversation partner by providing information necessary for the conversation, such as translation results, by using characters, images, etc. in a direction suitable for the user and the conversation partner. It has a liquid crystal display 31 for display and a speaker 32 for outputting information necessary for dialogue by voice. Direction buttons 3 3 a, 3 3 b, selection buttons 3 4 a, 3 4 b for selecting OKZ CAN CEL, etc., and voice input buttons 3 5 a pressed by the user and the conversation partner when inputting voice, respectively. , 35b, and microphones 36a, 36b for the user and the conversation partner to input voice, respectively.

As shown in FIG. 7, the speech translator 30 The voice input unit 41a, which is an input unit for inputting and digitizing the source language sentence by voice from the step 36a, and the source language sentence digitized by the voice input unit 41a A speech recognition unit 42 a that is a speech recognition unit that performs continuous word recognition based on acoustic processing and a statistical language model, and a sentence analysis unit 43 that is an analysis unit that analyzes the output from the speech recognition unit 42 a. a, and an application constraint judging unit, which is an application constraint judging unit for judging whether or not the translation result candidate for the source language sentence satisfies the application constraint given to the bilingual information held in the bilingual information database 49 a described later. 44a, a user confirmation information generation unit 45a for generating confirmation information to be presented to the user, and the confirmation information generated by the user confirmation information generation unit 45a are transmitted to the liquid crystal unit 31 or the like. The user confirmation section 4 6 a that outputs to the speaker 3 2 Target language sentence generation unit 47 a, which is a target language sentence generation means for generating a target language sentence, and the target language sentence generated by the target language sentence generation unit 47 a to the liquid crystal unit 31 and speaker 32 The translation result output unit 48a, which is an output means for outputting the translation information, the translation information database 49a, which is a storage means for holding various translation information, and the output from the speech recognition unit 42a. Speech recognition target limiting unit 50 a, which is a speech recognition target limiting unit that holds the limiting conditions to be applied, and a bilingual information selection unit 5 1 that holds the translation information selection conditions used for the input source language sentence. a, a response prediction unit 52 a that is a response prediction unit that generates response prediction information, and a response prediction presentation unit that is a response prediction presentation unit that outputs the response prediction information generated by the response prediction unit 52 a 5 3 and a _a,:

Also, the speech translator 30 is a speech input unit 41b as a second input unit, a speech recognition unit 42b, and a second analysis unit on the conversation partner side, similarly to the user side. Sentence analysis unit 4 3 b and application constraint determination unit 4 4 b, a user confirmation information generation unit 45b, a user confirmation unit 46b, a target language sentence generation unit 47b which is a second target language sentence generation unit, and a second output unit A translation result output unit 48b, a bilingual information database 49b as a second storage means, a speech recognition target limiting unit 50b, a bilingual information selection unit 51b, and a response prediction unit 5 2b and a response prediction presentation section 53b.

Such a speech translator 30 performs a series of processes as shown in FIGS. 8 and 9, and performs bidirectional translation while supporting a dialog between a user and a conversation partner. Here, a case will be described in which the source language sentence in Japanese issued by the user is translated into the target language sentence in English which the conversation partner can understand.

First, as shown in FIG. 8, in step S1, the speech translator 30 is input by the user speaking into the microphone 36a while pressing the speech input button 35a. Digitizing the source language sentence in Japanese by the voice input unit 4 1a ₌

Next, the speech translator 30 uses the speech recognition unit 42 a for the source language sentence digitized by the speech input unit 41 a based on the acoustic processing and the statistical language model in step S2. Perform continuous word recognition: For this continuous word recognition process, various methods such as generally used HMM-based recognition technology can be applied. The speech translator 30 generates a wordgraph by this continuous word recognition process: Then, the speech translator 30 determines in step S3 that the limiting condition applied to the generated wordgraph is a speech recognition target. It is determined whether or not it is held in the limiting part 50a. This limiting condition is to limit the range that the speech recognition unit 42a should target in order to improve the accuracy of speech recognition. This is information that limits the source language sentence that the user will enter next, or information that limits the source language sentence that the other party will enter next. As the limiting condition, there can be various kinds of information such as, for example, field information of the source language sentence described in the bilingual information, but here, as will be described later, processing of the source language sentence previously input is performed. It is a set of related independent words generated from the bilingual information associated with the bilingual information used in (1).

Here, if the speech translator 30 determines that the limitation condition is not held in the speech recognition object limitation unit 50a, the speech translator 30 shifts the processing to step S5. When the input is made at the beginning, since the limiting condition is not held in the voice recognition target limiting unit 50a, the voice translator 30 shifts to the processing in step S5.

On the other hand, if the translator 30 determines that the restriction condition is held in the voice recognition target restriction unit 50a, the speech translator 30 determines in step S4 the restriction condition for the generated code graph. By applying this method, the size of the code graph is reduced to limit the recognition vocabulary: That is, the speech translator 30 generates, in the generated word graph, nodes that include independent words that are not included in the set of independent words described above. If it does, _c to reduce the size of the entire Wa one Dogurafu by removing that part

Further, in step S5, the speech translator 30 determines whether the size of the word graph exceeds a predetermined size.

Here, if the size of the word graph does not exceed the predetermined size, the speech translator 30 shifts to the processing in step S7, and on the other hand, after the processing in step S4, the speech translator 30. If the size is large, the speech translator 30 performs the marking once by the so-called N-Best search in step S6, and leaves nodes related to those having a high probability of appearing. Reduce the size of the word graph by removing nodes not included in the path so that the size of the word graph does not exceed a certain size.

Next, in steps S7 to S9, the speech translator 30 analyzes the word graph output from the speech recognition unit 42a by the sentence analysis unit 43a, and applies The constraint judging unit 44a performs the application constraint judging process. These processes will be described in detail with reference to the bilingual information shown in FIGS. 10 and 11.

The speech translator 30 holds bilingual information as shown in FIG. 10 in a bilingual information database 49a. Actually, among the items in the bilingual information described in the figure, except for the source language sentence and the target language sentence, it is information that is described as necessary, and is included in all bilingual information entries. On the other hand, it should be noted that not all items are described = Speech translator 30 is basically a device based on the example-based translation system here. Translation is performed by determining the similarity with a sentence composed of bilingual information.However, in order to perform accurate translation without a large number of examples, general similarity determination processing is performed. First, the speech translator 30 performs a similarity determination on the structure of a sentence as a first change, as described below. In other words, in the case of an ordinary example-based translation system, the ability to separately perform syntactic analysis and determine the similarity based on the results The speech translator 30 has a grammatical allowance for each example such as part of speech. By describing the category of This eliminates errors in similarity judgment. In addition, the speech translator 30 performs a similarity determination regarding the meaning of the word as a second change. In other words, in the case of a normal example-based translation system, the similarity is determined by defining the distance between words in the concept space using a thesaurus and confirming that the distance is equal to or less than a certain value. The speech translator 30 eliminates errors in similarity determination by describing a semantic category including the maximum number of words within an allowable range for each example.

By making such a change to the processing, the speech translator 30 can avoid a decrease in accuracy by describing further information in each bilingual information on a portion that may decrease the accuracy. In addition, by treating the description of the grammatical category as a non-terminal symbol, it is possible to handle each piece of bilingual information as a generally widely used context-free grammar rule.

A specific example of the translation information in the speech translator 30 is as shown in FIG. The meaning of each item used in the figure is as shown in Table 1.

SRC original eyes δ entrance

SRCCOND Restrictions (source language)

TGT target language sentence

TGTCOND application restriction (target language sentence)

UISRC user confirmation source language sentence

STATUSMSG Status description

STATUSVIDEO Status description image

STATUSAUDIO Status description audio

LI concealed XT Information related to next sentence

LINKRESP Information related to the next response statement

RESPSRC response sentence (source language)

ESPTGT response statement (target language)

ESPVIDEO response image

RESPAUDIO response voice

In speech translator ³ 0, as shown in FIG. 1 1, for example, ": S @ 1" and terms as used _"Train:: ΝΡ: Transportation @ 1" like in parallel translation information, the right side and the left side Both are <notation>: <grammar category>: <ieature>: <semantic category> @ <link number>, and unnecessary items and delimiters can be omitted. Here, the grammar category> corresponds to the first change described above, and the semantic category> corresponds to the second change.

The ring number> indicates the link information number to which the part described in each sentence is related.If the same number is assigned to each sentence, the link information is related to the same link information. It indicates that

The <feature> is the conjugation form of the verb and auxiliary verb, It is used for describing various other constraints.

<LINK NEXT> ^ The number described in <LINKRESP> etc. is the bilingual information number.

Points to what is described on the line starting with "#". That is, the translation information number is a file name that guarantees that the translation information is unique. This translation information number may make it difficult to maintain the translation information if it is written as it is, but the speech translator 30 uses a separate translation information authoring system. When writing these information, we do not use direct human resources to avoid the difficulty of maintaining bilingual information.

Such a speech translator 30 transmits the source language sentence of the bilingual information selected in step S7 in FIG. 8 based on the bilingual information selection condition described later held in the bilingual information selecting section 51a. Based on only the part, the sentence analyzing unit 43a determines which combination of the bilingual information the input source sentence can be composed. The speech translator 30 considers, for example, the source language sentence part of the bilingual information as a context-free grammar rule, and performs analysis using a so-called chart parser inputting a code graph: For example, as described in “Hozumi Tanaka,“ Basics of Natural Language Analysis ”, Sangyo Tosho, 1989”. This is a method generally used as a parsing method. Then, the explanation is omitted.

In step S8, the speech translator 30 generates a plurality of sentence candidates configured as a set of bilingual information from the word graph of the speech recognition result by performing the analysis using the charter.

Then, the speech translator 30 outputs the same bilingual translation among the plurality of generated candidates. If there is more than one combination consisting of information, in step S9, similar to the ordinary example-based translation system, based on the similarity calculation, other than the best candidates are deleted, and If the information describes the application constraint, the application constraint judging unit 44a performs the application constraint judgment, and deletes candidates that do not satisfy the conditions.

By performing such processing, the speech translator 30 generates an optimal candidate for the source language sentence.

Specifically, when the speech translator 30 holds the bilingual information shown in FIG. 11, for example, the user inputs, "How much does a train via Futakotamagawaen take?" If the speech recognition is performed correctly, the sentence analysis section 43a causes the sentence to ask for the time using a combination of the bilingual information 120 and the bilingual information 283 and the bilingual information 130 and the bilingual information. then _c generates a statement "asking cost in combination with the multi-address 2 8 3 as a candidate, the speech translation device 3 0, as shown in FIG. 9, whether Oite, candidate exists in step S 1 0 Is determined.

Here, if there is no catch, the speech translator 30 shifts the processing to step S 20, for example, displays characters and images on the liquid crystal unit 31, and outputs speech to the speaker 3. By outputting from step 2, the user is informed that no candidate exists, and the series of processing ends.

On the other hand, if there is a candidate, the speech translator 30 generates user confirmation information by the user confirmation information generation unit 45a based on the generated candidate in step S11. That is, the speech translator 30 generates a user confirmation source language sentence based on the user confirmation source language sentence described in each piece of bilingual information and the link information described therein. In the speech translator 30, this user confirmation In some cases, the source language sentence is described for each piece of bilingual information, and sometimes it is not written for each piece of bilingual information. For example, in the case of the speech translator 30, if the input source language sentence and the source language sentence for user confirmation to be described in the bilingual information corresponding to this source language sentence are the same, the The description of the source language sentence for user confirmation is omitted. If the translation information is not described for each piece of bilingual information, the speech translator 30 uses the source language sentence as it is instead.

Specifically, the speech translator 30 uses the combination of the bilingual information 1 20 and the bilingual information 2 8 3 shown in Fig. 11 to say, `` How long does it take for a train to travel through Futako Tamagawaen. Is generated, and the combination of the bilingual information 130 and the bilingual information 283 is used to determine how much money is required for a train traveling through Futako Tamagawaen. Generate an approved source language sentence.

In addition, the speech translator 30 can also generate a situation description-a situation description image and a situation description voice as user confirmation information. The status description, the status description image, and the status description sound explain the bilingual information independently of the source language text. Therefore, the speech translator 30 uses these situation explanations, situation explanation images when the situation explanations, situation explanation images, and situation explanation speech are described in the bilingual information related to the source language sentence. Instead of combining the situation explanation voices into one like the source language sentence for user confirmation, they are simply generated as a set of information.

In this way, the speech translation device 3 0, in response to the generated each candidate, the user confirmation information corresponding to the meaning possessed by the candidate to generate _c Subsequently, the speech translator 30 displays the generated user confirmation information on the liquid crystal unit 31 via the user confirmation unit 46a in step S12 in FIG. By outputting it, it is presented to the user.

In response to this, in step S13, the speech translator 30 allows the user to select, from the user confirmation information, the most appropriate candidate along the context of the source language sentence and the user's intention. Selected. Specifically, for example, as shown in FIG. 12, the speech translator 30 generates the “Futako Tamagawaen” generated from the combination of the bilingual information 120 and the bilingual information 283 shown in FIG. The time required for a train via Futako-Tamagawaen to be generated from a combination of a source language sentence for user confirmation that a train via Is displayed on the liquid crystal unit 31. In addition, as shown in FIG. 13, for example, the speech translator 30 uses a source language sentence for user confirmation of “Sorry” in the erroneous situation generated earlier from the bilingual information 457 shown in FIG. The source language sentence for user confirmation “Sorry,” in the thank-you situation generated from the information 458, is displayed on the LCD unit 31 and an image explaining the situation is displayed on the LCD unit 31. A sound explaining the situation is output from the speaker 32. In response to the output of the user confirmation information, the speech translator 30 allows the user to operate the direction button 33 a to select a candidate to be selected, and the selection button 34 a Press to confirm.

In this way, the speech translator 30 uses the source language sentence for user confirmation, the situation description, the situation description image, and the situation description voice to translate the speech. The meaning of the target language sentence can be correctly conveyed to the user from various angles, and when the user makes a selection based on these, the interpretation is ambiguous due to the intention of the contextual user. Language sentences can be translated correctly.

After the selection by the user, the speech translator 30 generates the target language sentence by the target language sentence generation unit 47a in step S14 in FIG. Further, when the response sentence, the response image, and the response speech are described in the bilingual information, the speech translator 30 uses the response prediction unit 52 , And the response voice is also generated. When the link information is described for the response sentence (source language and target language) in the bilingual information, the speech translator 30 is described in each bilingual information in the same manner as the user confirmation source language sentence. A response sentence is generated based on the response sentence and the link information described therein. When the link information is not described for the response sentence (source language and target language) in the bilingual information, the speech translator 30 combines the response sentence into one like the response sentence, just like the situation explanation. A response sentence is generated by assembling the sentence as it is without performing the combining process. Further, the speech translator 30 generates a response image and a response voice by assembling the response image and the response voice as they are, similarly to the situation description image and the situation description voice. Specifically, the speech translator 30 responds to the source language sentence “How long does a train from Futakotamagawa Sono take on the train?” “How long wi1 1 the train which via ru akoTakot amaga aen t ake? ”生成 Generated as a T target language sentence. a Fut ak otamagawaen cos t? ”as the target language sentence. Also, voice translation If the machine 30 translates “Sorry,” in a situation where thank you as a source language sentence, it refers to the bilingual information 458 shown in Fig. 11 earlier and responds with “You are wel come In addition to predicting and generating a response sentence "," it also predicts and generates a response sentence, "I am happy."

In the speech translator 30, since there is a possibility that a plurality of these response sentences, response images, response voices, and the like exist for each piece of translation information, when generating the response prediction information, Needless to say, it is necessary to handle all response sentences, response images, and response voices in information.

Next, the speech translator 30 sets bilingual information selection conditions in step S16 in FIG. That is, the speech translator 30 performs a process for limiting the bilingual information to be used for the next input source language sentence according to the current translation result.

Specifically, when the number is described in <LINKNEXT>, as in the translation information 120 shown in FIG. 11 earlier, the speech translator 30 is described in this way. Using only the bilingual information and general-purpose bilingual information used under any circumstances, the related bilingual information is translated so that the next input source language sentence can be processed. Record in the information selection section 51a. In this way, the speech translator 30 enters a state in which the translation information selection condition is set at the time of the translation processing of the next input source language sentence, and based on the translation information selection information, The processing is performed by limiting the target range at the time of selection. If the speech translator 30 cannot perform the process correctly using such a limiting condition, it removes the limiting condition and performs the process again. This By doing so, the speech translator 30 can flexibly cope with a situation change.

Similarly, when the number is described in <LINKRESP>, as in the bilingual information 120 shown in FIG. 11, the speech translator 30 executes step S 17 in FIG. As shown in the above, based on the information described in LINKRESP>, the bilingual information selection condition in the reverse translation for translating the source language sentence input from the conversation partner is set and recorded in the bilingual information selection section 51a. .

Further, in the following step S18, the speech translator 30 sets a limitation condition for speech recognition based on the bilingual information selection condition, and records it in the speech recognition target limitation section 50a. In the speech translator 30, since the speech recognition unit 42a cannot use the bilingual information as it is, a list of independent words included in the related bilingual information recorded in the speech recognition target limiting unit 50a. Is generated and stored in the voice recognition target limiting unit 50a, and this list of independent words is used as a voice recognition limiting condition.

Then, in step S19, the speech translator 30 outputs the generated target language sentence and the response prediction information via the translation result output unit 48a and the response prediction presentation unit 53a, respectively. A series of processes ends: Specifically, the speech translator 30 converts the generated target language sentence and the response prediction information into characters in a liquid crystal unit, for example, as shown in FIGS. 14 and 15. Displayed on 3 1 or output from voice 3 2 by voice: After such a series of processing, the voice translator 30 performs voice input button 35 b or voice input button depending on the conversation partner or the user. By pressing 3 a, the source language sentence is input, the translation process for this source language sentence is performed, and thereafter, these processes are repeated. In addition, Although omitted, the speech translator 30 naturally has the same processing for the source language sentence input from the conversation partner as the above-described series of processing.

As described above, the translation device 10, the bidirectional translation device 20, and the speech translator 30 shown as the embodiments of the present invention are in accordance with the intention of the user, which was previously impossible. High-accuracy and high-quality translation results can be obtained, and a high-accuracy translation system is realized. In particular, the speech translator 30 can support a smooth conversation and realize a highly accurate and easy-to-use speech translation system. It should be noted that the configurations shown as the above-described embodiments are shown as examples embodied to realize the present invention, and the present invention is not limited to these configurations. Absent. For example, it goes without saying that the present invention can be easily applied not only when the source language sentence is input by voice but also when it is input by character.

In the above-described embodiment, an example in which the configuration is based on an example-based translation system has been mainly described. However, the present invention may be applied to a translation method such as a pattern-based translation or a template-based translation. Of course.

Furthermore, it goes without saying that other modifications can be made as appropriate without departing from the spirit of the present invention. INDUSTRIAL APPLICABILITY As described in detail above, the translation processing method according to the present invention is different from the input sentence or any part of the sentence in any source language. Based on the bilingual information that correlates the sentence or part of the sentence in any target language, the consistency or similarity between the sentence or part of the sentence in the source language and the sentence composed of the combination of the bilingual information is determined. This is a translation processing method in which a sentence or part of the sentence in the source language is translated into a sentence or part of the sentence in the target language and output. Part, a sentence or part of a sentence in the target language which is a translation of this sentence or part of the sentence, and relevant information relating to the sentence or part of the sentence in the source language and the sentence or part of the sentence in the target language Is described.

Therefore, in addition to the sentence or part of the sentence in the source language and the sentence or part of the sentence in the target language, the translation processing method according to the present invention provides the sentence or part of the sentence or source sentence in these source languages. By using bilingual information in which relevant information related to a sentence or part of a sentence in a language is described, only the sentence or part of the sentence in the source language and the sentence or part of the sentence in the target language are described. It is more convenient than using bilingual information, and can provide high-accuracy and high-quality translation results in the source language or part of the context of the sentence or the user's intention, and facilitates dialogue. Can help.

Further, the translation processing device according to the present invention comprises: an input means for inputting a sentence or a part of a sentence in an arbitrary source language; a sentence or a part of the sentence in the source language; Is described in association with a sentence or part of a sentence in any target language that is different from the source language, and a sentence or part of a sentence in the source language and a sentence or part of a sentence in the target language. A storage means for storing bilingual information in which related information is described, and, based on the bilingual information, matching or similarity between a sentence or a part of the sentence in the source language and a sentence formed by combining the bilingual information Determining gender Means for analyzing a sentence or part of a sentence in the source language to generate translation result candidates, and generating a sentence or part of the sentence in the target language that is a translation of the sentence or part of the sentence in the source language A target language sentence generating means for outputting a sentence or a part of the sentence in the target language generated by the target language sentence generating means.

Therefore, in addition to the sentence or part of the sentence in the source language, the sentence or part of the sentence in the target language, the translation processing device according to the present invention, the sentence or part of the sentence or source sentence in these source languages, By using bilingual information in which relevant information related to a sentence or part of a sentence in a language is described, only the sentence or part of the sentence in the source language and the sentence or part of the sentence in the target language are described. It is more user-friendly than translating using parallel translation information, and it is possible to obtain high-accuracy and high-quality translation results according to the context of the sentence or part of the sentence in the source language and the user's intention. When using it for dialogue, it is possible to support the progress and realize a smooth dialogue.

Claims

The scope of the claims

1. Based on the translation information in which the input sentence or part of the sentence in any source language is correlated with the sentence or part of the sentence in any target language different from the above source language, Judge the consistency or similarity between the sentence or part of the sentence and the sentence composed of the above-mentioned translation information, and replace the sentence or part of the sentence in the source language with the sentence or sentence in the target language. A translation processing method for translating and outputting a part of

The bilingual information includes a sentence or part of the sentence in the source language, a sentence or part of the sentence in the target language which is a translation of the sentence or part of the sentence, and a sentence or sentence in the source language. And the relevant information related to the sentence or part of the sentence in the target language described above.

A translation processing method characterized by the following.

2. The relevant information must be information that assists in interpreting the meaning of the input sentence or part of the sentence in the target language, which is a translation of the sentence or part of the sentence in the source language.

The translation processing method according to claim 1, wherein:

3. The relevant information includes a confirmation source language sentence in the source language, in which the input sentence or a part of the sentence in the target language, which is a translation of the sentence or part of the sentence, is paraphrased. thing

3. The translation processing method according to claim 2, wherein:

4. If the input sentence or part of the sentence in the source language is identical to the source language for confirmation to be described in the bilingual information corresponding to the sentence or part of the sentence in this source language. Must omit the description of the original language sentence for confirmation in the bilingual information. 4. The translation processing method according to claim 3, wherein:

5. The relevant information may be a situation description in the source language that describes the situation in which the confirmation source language expression is used, a situation description image associated with the situation in which the confirmation source language expression is used, or At least one of the situational explanation sounds related to the situation where the above confirmation source language sentence expression is used

4. The translation processing method according to claim 3, wherein:

6. The relevant information is the link information to the bilingual information related to the sentence or part of the sentence in the source language, which is likely to be input next to the sentence or part of the sentence in the source language. There is

The translation processing method according to claim 1, wherein:

7. The related information is field information for limiting the field to which the sentence or part of the sentence in the source language that is likely to be input after the input sentence or part of the sentence in the source language belongs. Being

The translation processing method according to claim 1, wherein:

8. The relevant information is a response sentence that predicts a response to the input sentence or part of the sentence in the source language, a response image that predicts a response to the input sentence or part of the sentence in the source language, or At least one of the response voices that predicts the response to the input sentence or part of the sentence in the source language

The translation processing method according to claim 1, wherein:

9. The relevant information is bilingual information in which the input sentence or part of the sentence in the source language is associated with a sentence or part of the sentence in the target language that is a translation of this sentence or part of the sentence. Based on the sentence or part of the sentence in the source language described above, Applicable constraints used to determine lethality or similarity

The translation processing method according to claim 1, wherein:

10. Judge whether the input translation result candidate for the sentence or part of the sentence in the source language satisfies the application constraint,

If there is no candidate that satisfies the above application restrictions, notify that the sentence or part of the sentence in the above source language cannot be translated.

10. The translation processing method according to claim 9, wherein:

1 1. Inputting a sentence or part of a sentence in the above target language and translating it into a sentence or part of a sentence in the above source language

The translation processing method according to claim 1, wherein:

1 2. The relevant information is translated into the bilingual information related to the sentence or part of the sentence in the target language which is likely to be input next as a response to the input sentence or part of the sentence in the source language. Must be related information

The translation processing method according to claim 11, wherein:

13 3. The related information limits the field to which the sentence or part of the sentence in the target language which is likely to be input next as a response to the input sentence or part of the sentence in the source language belongs. Field information for

The translation processing method according to claim 11, wherein:

1 4. Sentence or part of sentence in the above source language is input and recognized by voice. Generated sentence or part of sentence in the above target language, which is a translation of the sentence or part of sentence in the above source language. Output by voice

The translation processing method according to claim 1, wherein:

15 5. The relevant information is a sentence or a sentence in the input source language. Part of the sentence or part of the sentence in the source language that is likely to be input next to the section,

Based on the association information, generate information that limits the sentence or part of the sentence in the source language to be subjected to speech recognition.

15. The translation processing method according to claim 14, wherein:

16. Predict the response to the input sentence or part of the sentence in the source language,

Presenting response prediction information generated by prediction

The translation processing method according to claim 1, wherein:

17. The response prediction information is a response sentence described as the related information in the bilingual information corresponding to the input sentence or a part of the sentence in the source language, and one of the input source sentence or sentence in the source language. The response image described as the related information in the bilingual information corresponding to the part, or the response voice described as the relevant information in the bilingual information corresponding to the input sentence or part of the sentence in the source language described above. At least one of

The translation processing method according to claim 16, characterized by:

18. Input means for inputting a sentence or part of a sentence in an arbitrary source language, and a sentence or part of a sentence in the above source language and the source language which is a translation of this sentence or part of a sentence are different A sentence or a part of a sentence in an arbitrary target language is described in association with the sentence or a part of the sentence in the source language and related information related to a sentence or a part of the sentence in the target language. Storage means for storing the described bilingual information;

Based on the bilingual information, a sentence or a sentence in the source language is determined by determining the consistency or similarity between the sentence or a part of the sentence in the source language and the sentence composed of the combination of the bilingual information. Part of Analysis means for analyzing and generating translation result candidates;

A target language sentence generating means for generating a sentence or part of a sentence in the target language which is a translation of a sentence or part of the sentence in the source language;

Output means for outputting a sentence or a part of the sentence in the target language generated by the target language sentence generating means.

A translation processing device characterized by the above-mentioned.

19. The relevant information is information that is input by the input means and that assists in interpreting the meaning of the sentence or part of the sentence in the target language, which is a translation of the sentence or part of the sentence in the source language. There is

19. The translation processing device according to claim 18, wherein:

20. The above-mentioned related information is for confirmation in the above-mentioned source language in which a sentence in the above-mentioned target language which is a bilingual sentence or a part of a sentence in the above-mentioned source language inputted by the above-mentioned input means is reworded. Include source language sentences

10. The translation processing device according to claim 19, wherein the translation processing device is characterized in that:

2 1. The sentence or part of the sentence in the source language input by the input means and the source language sentence for confirmation to be described in the bilingual information corresponding to the sentence or part of the sentence in the source language. If they are the same, omit the description of the confirmation source language sentence in the bilingual information.

20. The translation processing device according to claim 20, wherein:

2 2. The related information is a situation description sentence in the source language that explains the situation in which the expression of the confirmation source language sentence is used, and a situation explanation image related to the situation in which the expression of the confirmation source language sentence is used. Or, at least one of the situation explanation sounds related to the situation where the expression of the confirmation original language sentence is used 20. The translation processing device according to claim 20, wherein:

23. The relevant information is related to a sentence or part of a sentence in the source language which is likely to be input next to a sentence or part of the sentence in the source language input by the input means. The information must be related to the bilingual information

19. The translation processing device according to claim 18, wherein:

24. The related information indicates the field to which the sentence or part of the sentence in the source language that is likely to be input next to the sentence or part of the sentence in the source language input by the input means belongs. Field information to limit

19. The translation processing device according to claim 18, wherein:

25. The related information is a response sentence that predicts a response to the sentence or part of the sentence in the source language input by the input means, or one of the sentence or sentence in the source language input by the input means. At least one of the response image that predicted the response to the part or the response voice that predicted the response to the sentence or part of the sentence in the source language input by the input means

19. The translation processing device according to claim 18, wherein:

26. The related information corresponds to the sentence or part of the sentence in the source language input by the input means and the sentence or part of the sentence in the target language which is a translation of the sentence or part of the sentence. Based on the attached bilingual information, it must be an applicable restriction used to determine the consistency or similarity between a sentence or part of the sentence in the source language and a sentence composed of a combination of the above-mentioned translated information.

19. The translation processing device according to claim 18, wherein:

27. The apparatus further comprises an application constraint determination unit configured to determine whether a translation result candidate for the sentence or part of the sentence in the source language input by the input unit satisfies the application constraint,

If there is no candidate that satisfies the application constraint as a result of the determination by the application constraint determination means, information indicating that the sentence or part of the sentence in the source language cannot be translated is output.

27. The translation processing apparatus according to claim 26, wherein:

28. A second input means for inputting a sentence or a part of a sentence in the target language,

A sentence or part of a sentence in the target language described above is associated with a sentence or part of the sentence in the source language, which is a translation of the sentence or part of the sentence, and a sentence in the target language is written. Or a second storage means for storing bilingual information in which a part of a sentence and a sentence in the original language or related information related to the part of the sentence are described;

Determining, based on the bilingual information stored in the second storage means, a match or similarity between a sentence or a part of the sentence in the target language and a sentence composed of a combination of the translated information. A second analysis means for analyzing a sentence or a part of the sentence in the target language to generate a translation result candidate,

A second target language sentence generating means for generating a sentence or part of a sentence in the source language which is a translation of a sentence or a part of the sentence in the target language, and a second target language sentence generating means 19. The translation processing apparatus according to claim 18, further comprising: a second output unit that outputs a sentence or a part of the sentence in the source language.

2 9. The relevant information is the source language input by the input means. Then, the response to the sentence or part of the sentence

-It should be the information that is likely to be input by the target language and the bilingual information related to the sentence or part of the sentence.

29. The translation processing apparatus according to claim 28, wherein:

30. The relevant information is a response to the sentence or part of the sentence in the source language input by the input means, which is likely to be input next by the second input means in the target language. Field information to limit the field to which the sentence or part of the sentence belongs

The translation processing device according to claim 28, characterized by the following _:

31. Speech recognition means for recognizing a sentence or a part of the sentence in the source language input by the input means by voice,

The output means outputs a generated sentence or a part of the sentence in the target language, which is a bilingual sentence or a part of the sentence in the source language.

The translation processing device _{c according} to claim 18, characterized in that:

3 2. The relevant information is a bilingual translation related to a sentence or part of a sentence in the source language that is likely to be input next to a sentence or part of the sentence in the source language input by the input means. A voice recognition target limiting unit configured to store information for limiting a sentence or a part of a sentence in the source language to be generated by the voice recognition unit, the information being generated based on the association information. To prepare

The translation processing device _{c according} to claim 31, wherein

33. Response prediction means for predicting a response to a sentence or a part of a sentence in the source language input by the input means,

The response prediction information predicted and generated by the response prediction means is presented. Response prediction presenting means

19. The translation processing device according to claim 18, wherein:

34. The response prediction information is a response sentence described as the related information in the bilingual information corresponding to the sentence or part of the sentence in the source language input by the input means, and is input by the input means. A response image described as the related information in the bilingual information corresponding to a sentence or part of the sentence in the source language, or a bilingual sentence or part of the sentence in the source language input by the input means. At least one of the response voices described as related information in the information

The translation processing device according to claim 33, characterized by: