JP2016218995A - Machine translation method, machine translation system and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a machine translation system capable of outputting a translation text desirable for a user with few correction on an input text or re-inputting thereof.SOLUTION: The machine translation system generates plural different forward translation texts in which translation object text of a first language received is translated into a second language, and generates plural reverse translation texts in which each of the plural different forward translation texts are reverse translated into the first language. When receiving an operation to select a reverse translation text from the plural reverse translation texts while an information output unit outputs plural reverse translation texts, the forward translation text corresponding to the reverse translation text is output.SELECTED DRAWING: Figure 3

Description

  The present invention relates to an apparatus for performing translation between a plurality of languages, a method for performing translation between a plurality of languages, and a system for performing translation between a plurality of languages.

  With globalization in recent years, machine translation devices and machine translation systems that enable communication between users whose native languages are different languages have been developed. In addition, the operation of a service that provides a machine translation function has been started, and it is actually being used in scenes such as travel conversations.

Japanese Patent No. 5653392 JP 2005-78318 A Japanese Patent No. 5097340 International Publication No. 2013/014877

Papineni, K .; Roukos, S .; , Ward, T .; , And Zhu, W .; J. et al. “BLEU: a method for automatic evaluation of machine translation”, Proc. of the Annual Meeting of the Association of Computational Linguistics (ACL), pp. 311-318, 2002 Natsume Takachi, Hideki Amagasaki, “Verification of automatic translation method considering scrambling with NTCIR-9 data”, 219th Natural Language Processing Research Presentation, Okayama Prefectural University, 2014 Ilya Suskever Papineni et al., “Sequence to Sequence Learning with Neural Networks”, Advances in Neural Information Processing Systems 27, p. 3104-3112, 2014 Dzmitry Bahdanau et al., “Neural Machine Translation by Joining Learning to Alignment and Translation”, arXiv: 1409.04733, ICLR 2015

  However, the machine translation apparatus and machine translation system described above require further improvement.

  One aspect of the present invention for solving the above problem is a machine translation method in a machine translation system that is connected to an information output device that outputs language information and performs a translation process between a first language and a second language. And receiving the translation target sentence in the first language, generating a plurality of different forward translation sentences obtained by translating the received translation target sentence into the second language, and each of the plurality of different forward translation sentences. Generating a plurality of back-translated sentences back-translated into the first language, and selecting one back-translated sentence from the plurality of back-translated sentences when the information output device outputs the plurality of back-translated sentences When the operation to perform is accepted, the forward translation sentence corresponding to the one reverse translation sentence is output.

  According to the above aspect, further improvement can be realized.

It is a figure which shows an example of the whole structure of the machine translation system in this Embodiment. It is a block diagram which shows the information display terminal structure in this Embodiment. It is a block diagram which shows the structure of the translation server in this Embodiment. It is a figure which shows the hardware constitutions of the machine translation system in this Embodiment. It is a flowchart which shows operation | movement of the machine translation system in this Embodiment. It is a flowchart which shows the specific operation | movement of the translation selection process in this Embodiment. It is a flowchart which shows the specific operation | movement of the back translation sentence selection process in this Embodiment. It is a flowchart which shows the specific operation | movement of a phrase evaluation process in this Embodiment. It is a flowchart which shows the specific operation | movement of a learning process in this Embodiment. It is a flowchart which shows the specific operation | movement of a learning process in this Embodiment. It is an example of the general phrase table in this Embodiment. It is explanatory drawing showing the outline | summary of the phrase division | segmentation in this Embodiment. (A), (B), (C) is a figure which shows an example of the display screen in this Embodiment, respectively. It is a figure which shows an example of the display screen in this Embodiment.

  Each of the embodiments described below shows a specific example of the present invention. Numerical values, shapes, components, steps, order of steps and the like shown in the following embodiments are merely examples, and are not intended to limit the present invention. In addition, among the constituent elements in the following embodiments, constituent elements that are not described in the independent claims indicating the highest concept are described as optional constituent elements. In all the embodiments, the contents can be combined.

(Knowledge that led to the present invention)
In the situation where translation equipment is actively developed, it is desired that the accuracy of machine translation is perfect in order to realize smooth communication in different languages. However, in the current machine translation, it is extremely difficult to translate an arbitrary sentence without error, and the translation accuracy is improved by limiting the fields (domains) that can be translated in the form of travel conversations. Is far away.

  Here, the machine translation technology is roughly classified into three types. In each of the main sections, 1) Rule-based machine translation (RBMT), 2) Statistical machine translation (SMT), 3) Model acquisition type machine translation (DNNNMT: Deep) It will be referred to as “Neural Network Machine Translation”.

1) Rule-based machine translation (RBMT)
Rule-based machine translation (RBMT) is a system that performs translation based on manual conversion rules (stored translation pairs as a database), and the database of the original text and the translated text is expressed as a translation memory. There is also.

  As an advantage, “Rules (patterns to be translated) can be defined precisely, so if the original text (part) exists in the rules, the translation accuracy of the corresponding parts will be high, and the translation expression in the rules. , It is easy to maintain consistency in the translation expression of the translation output.

  Disadvantages include: “If there are no rules, translation accuracy will be very low, or translation will not be possible at all, and adaptability to other industries and fields will be extremely low”. In addition, since the rules are basically constructed and expanded manually, the development costs are also high. Furthermore, it is necessary to add rules in order to improve translation performance. However, in order to customize the translation system by building rules on the user side, considerable knowledge of the rule design rules is required. It is not easy for general users to use. For this reason, the main development of rule-based translation (RBMT) is business translation (patent translation, etc.), and commercially available business translation software uses this rule-based translation or translation memory. Many.

2) Statistical machine translation (SMT)
Statistical machine translation (SMT) does not create rules (rules) like RBMT, but prepares a large number of “translation pairs in both source and target languages” called bilingual corpus, and statistically translates from this bilingual corpus This is a method of generating a translation by calculating the translation probability. It is widely used for translation engines provided on the web by Google (registered trademark), Microsoft (registered trademark), and the like.

  In this method, in calculating the translation probability, a language model (generally expressing the appearance probability between words in a single language by an n-gram model) and a translation model (word or phrase between bilinguals). It is statistically probable from the bilingual corpus for each of the two models that express the correspondence in units as an alignment model and the translational relationship between words at the vocabulary level as a vocabulary model. This is a technique that realizes statistical translation by obtaining the parameters (probability values) of each model. It is generally called an IBM model. (If Japanese is included, a model called “reorder model” may be added by parsing etc.).

  The advantage of this SMT is that it does not use rules like RBMT, so it can avoid the problem of “translation accuracy is very low when no rules exist” and has a generalization ability. Can be built ".

  On the other hand, there is a disadvantage that “because it is based on a probabilistic expression, it is necessary to prepare a large number of parallel corpora as a basis for probability calculation”. In addition, while the generalization performance is improved, the translation result is only an output with a “high probability score”, and therefore, there is no guarantee that the same translation accuracy can be obtained even in SMT for the case where the translation was successful in RBMT. Although there is a translation that is "relatively good as a whole / average", there are cases where individual translation cases are less reliable.

  Furthermore, when performing the probability calculation, it is necessary to multiply the probability values output by the internal models (language model, translation model, etc.). At this time, in order to tune the entire SMT translation engine, the machine learning is performed by weighting the probability value from each model and using the weight as a parameter. The machine evaluation value used here is literally “mechanical evaluation value”. (For example, an evaluation value called BLEU) ", it has been reported that it does not necessarily coincide with the subjective evaluation of the user (for example, see Non-Patent Document 2). That is, it also has a structural defect that “the machine evaluation value is high but is not linked to the user's actual evaluation”.

  In addition, even if you add several sentences of the bilingual corpus to improve performance on the user side or system side, the translation output depends on the stochastic behavior after statistical processing of the entire bilingual corpus, Another structural drawback is that it does not directly improve performance (it can also degrade performance).

3) Model acquisition type machine translation (DNNMT) by deep neural network
Model acquisition type machine translation (DNNNMT) using a deep neural network is a relatively new translation technology in which deep learning (DL: Deep Learning) technology is applied to machine translation (for example, see Non-Patent Document 3 and Non-Patent Document 4). ).

  Unlike a translation technique that applies / statistically processes an input sentence or a bilingual corpus to “rules and models designed by a person” such as RBMT and SMT, the system is characterized by learning an appropriate model itself. At present, this technology is in its early days and has not been put into practical use as machine translation. However, as a similar technology, it has already been put into practical use in speech recognition such as Siri (registered trademark) of Apple Inc.

  As an advantage, if an appropriate model is successfully learned, an improvement in translation performance is expected. In particular, a generalization performance that exceeds SMT, and a relatively long sentence (eg, 40 or more) that RBMT and SMT are not good at. It is expected that the translation result will be output without degrading the translation performance so much.

  Disadvantages include the difficulty of external control of how DNNNMT itself learns the model. There is no explicit “parameter” corresponding to so-called “tuning using parameters”, and although there are a large number of internal variables, it is unclear which variables are directly linked to translation performance. In other words, it is difficult to control how the system learns and what kind of performance comes out after deciding the neural network configuration and input data / teacher data, and it is unclear how the performance can be improved in principle. . It is difficult to apply a general tuning method, and it is difficult to cope with an attempt to improve performance on the user side or system side beyond SMT. Even if several sentences are added to the bilingual corpus, the translation output depends on the learning of the neural network, and there is a structural defect that what will come out cannot be known unless the internal variables of the neural network are determined.

  The above are the features, advantages, and disadvantages of machine translation technology when classified into three main types.

  On the other hand, when looking at machine translation technology from the user's perspective, if the user has insight into the target language, which is the language he wants to translate, even if the accuracy of machine translation is not very high, the conversation will proceed based on the translation output result Could be possible. In other words, it can be expected that the user can determine whether the output of machine translation is “correct / incorrect” and use a useful translation part.

  However, in practice, even if the user refers to the output result of the machine translation, it is easily imagined that the contents indicated by the translation result cannot be understood. For example, assuming that the user is Japanese, there are cases where you want to translate Japanese into a major language (English, etc.), while you want to translate into a relatively minor language (eg Malay, Vietnamese, etc.) There is also. For example, when the target language is English, a user who has knowledge about English may be able to use it for his own English conversation with reference to the result of machine translation. On the other hand, if the target language is a minor language, the user often has no knowledge of the target language, and the user can obtain any information as to whether the content of the machine translation result is correct. Absent. In other words, even if the target language translation is presented by machine translation, the content cannot be understood at all. Thousands of languages are said to exist, and for most users it can be said that the majority of languages are unknown. As described above, if the meaning of the translation result of the target language cannot be understood, the translation may be presented without confirming the intention transmitted to the other party, and communication may be broken.

  In order to improve the current machine translation accuracy as much as possible, it is desired that the machine translation system itself automatically learns and improves its performance. When a machine translation system automatically learns, it is expected that it will not put a burden on the user (source language speaker), and then the translated sentence while confirming the intention transmitted to the other party by a simple method. It is necessary to build a machine translation system so that can be generated and used. The machine translation system is required to automatically learn using the use result of the user. Naturally, it is also required to reduce the computer resources and development costs required for the system itself as much as possible.

  That is, what is required of a machine translation system is “(Requirement 1) that a user can generate and use a translated sentence while confirming the intention transmitted to the other party in a simple manner without burdening the user”, “(Requirement 2) Can be automatically learned by using the user's usage results (without newly burdening the user) "," (Requirement 3) Simultaneously realizing requirement 1 and requirement 2, simultaneously reducing computer resources and development costs 3 requirements. The realization of a machine translation system that satisfies these three requirements simultaneously has been an issue.

  As a means for solving this problem, for example, as in RBMT, an example that associates an example of the source language with an example of the target language is provided as a database or a rule / rule, and an example similar to the spoken input sentence. Obtain and present in both source and target languages (for example, Patent Document 1), calculate the distance between the input sentence and the result of reverse translation (translation from the target language to the source language) as translation reliability, and if necessary BLE (BiLingual, which is a method for evaluating the performance of a statistical machine translation (SMT) system, which presents the reliability and reverse translation results together, and allows the user to rephrase and paraphrase (for example, Patent Document 4). E (under Non-Patent Literature 1) using N (Evaluation Under study), N reverse translation sentences are acquired for the input sentence, and the comparative evaluation of the input sentence and the reverse translation sentence What performs valency (for example, Patent Document 2), and performs machine learning on the result of determining whether or not it is effective for the target translation device when registering the bilingual phrase in the bilingual dictionary (for example, Patent Document 3) and so on.

  Patent Document 1 discloses contents stored in an example storage unit in advance in association with a source language example and a target language example that is a translation of the source language example.

  Patent Document 1 discloses that an input source language character string and a target language character string corresponding to the source language character string are displayed on the display unit based on the information stored in the example storage unit. The contents displayed in different display areas are disclosed.

  Specifically, for the input source language character string, a plurality of similar example character strings similar to the source language character string are displayed in the same display area. A character string in the target language corresponding to each of the character string in the source language and a plurality of similar example character strings is displayed in different display areas.

  When the source language speaker and the target language speaker want to confirm the meaning of the input sentence, they can confirm the example. In addition, the similar sentence example selected by the target language speaker or the source language speaker is also displayed in highlight or the like on the partner speaker.

  In Patent Document 1, a source language character string as an input sentence is acquired, a similar example in a parallel translation example in the database is searched, and if the example has a similarity equal to or greater than a threshold, it is determined that the example is a similar example. Is output as a similar example for the input sentence. At this time, when a user's selection operation is performed on the output similar example, the content of highlighting the selected example is also disclosed.

  In Patent Document 2, the forward translation sentence output from the forward machine translation unit 11 is received with respect to the original sentence, and the backward translations A, B, and C are evaluated by the backward translation units 12a, 12b, and 12c. The content to be evaluated at 13 is disclosed.

  As an evaluation method at this time, the contents disclosed in Non-Patent Document 1 are generally known. In Non-Patent Document 1, after calculating the number of matching N-grams between the reference translation (correct translation created manually) and the translation output by machine translation, the influence of the length of the reference translation is calculated. The BLEU value is acquired by performing correction with consideration. BLEU is often used as an evaluation method for translation accuracy, but it is known that human translation and correlation are low particularly in translation between languages such as Japanese-English and English-Japanese that have very different word orders (for example, Non-Patent Document 2).

  Patent Document 2 further discloses the contents of outputting the reverse translation sentence having the maximum score and the original sentence after DP matching between the reverse translation sentences A, B, and C and the original sentence. The evaluator compares these to enable subjective evaluation of the forward translation.

  As described above, when translation is performed under a situation where machine translation accuracy is not perfect, the user (source language speaker) can easily confirm the intention transmitted to the other party without burdening the user repeatedly. It was necessary for the system itself to generate translation utterances and evaluate / learn the user's selection as an evaluation.

  Reprinting what is required of a machine translation system, “(Requirement 1) A user can generate and use a translated sentence while checking the intention transmitted to the other party in a simple manner without burdening the user”, “(Requirement 2) Can be automatically learned by using the user's usage results (without newly burdening the user) "," (Requirement 3) Simultaneously realizing requirement 1 and requirement 2, simultaneously reducing computer resources and development costs It is to satisfy the three requirements at the same time.

  As means for solving this problem, the following approaches are taken in the three types of technologies mentioned above.

1) Rule-based machine translation (RBMT)
As described above, in Patent Document 1, a similar sentence in a similar source language can be output from the input source language character string, and a corresponding translation sentence can be selected for the output similar sentence in the source language. Output to. As a result, for example, when a character string in the source language is input by voice, the influence of an input error due to a voice recognition error at the time of speech input is reduced, and the user can easily select an intention to be transmitted.

  In paragraph [0012] of Patent Document 1, “the example storage unit 105 associates a source language example (hereinafter also referred to as a source language example) with a target language example (hereinafter also referred to as a target language example). The example storage unit 105 stores a pair of the source language and the target language as a database for performing an example search. This part corresponds to a rule-based translation database.

  The paragraph number [0011] in Patent Document 1 states that “the machine translation unit 103 receives a source language character string from the speech recognition unit 102 and converts the source language character string into a character string of a target language (also referred to as a second language). Machine translation is performed to obtain a target language character string that is a character string of the translation result.Since the concrete process of the machine translation may be a general process, the description is omitted here. " Yes, the advantages and problems of using the SMT and DNNNMT for the machine translation unit 103 are not disclosed in Patent Document 1, and the entire speech translation system in Patent Document 1 is an RBMT by the example storage unit 105. The work is only disclosed characteristically.

  As mentioned earlier, RBMT has a significant decrease in translation accuracy or cannot be translated at all if there is an input sentence not described in the rules (rules) or an input from an industry or field that is not supported. Has the problem. Also in Patent Document 1, when a similar example is not described in the example storage unit 105 with respect to an utterance input, or when the field is different, there is a possibility that a translated sentence with extremely low translation accuracy may be presented, and there is knowledge about the target language. It may lead to a user who does not have the intention to communicate with the other party presenting the translation (or presenting the intention transmitted to the other party misunderstanding), but these issues are not disclosed and resolved There are no measures.

  Further, when there is no similar example in the example storage unit 105, the similar example itself is not presented, and the user has no choice but to use an input sentence (a lot of errors) of the speech recognition result. The translation result is only the translation result of the machine translation unit 103 for the input sentence based on the speech recognition result. In this case, the user cannot “simplely select” “sentence in accordance with the intention to be conveyed appropriately”, and a solution to this problem is not disclosed in Patent Document 1.

  That is, for (Requirement 1), the means for “simply selecting (sentence) by the user” is partially disclosed by providing means for presenting and selecting a plurality of similar sentence search results. When there is no similar example in the example storage unit 105 corresponding to RBMT, the option itself is lost, and a means for “a user can easily select (sentence)” is not provided completely.

  Furthermore, “confirmation of intention transmitted to the other party” is only limitedly solved when the search hits the content held in the example storage unit 105, and the translation result output by the machine translation unit 103 is output. Is not touched. That is, no mention is made as to whether or not the translation result of the machine translation unit 103 is in accordance with the contents of the original source language character string (in the paragraph number [0011] of Patent Document 1, “the machine translation unit 103 A source language character string is received from the speech recognition unit 102, the source language character string is machine-translated into a target language (also referred to as second language) character string, and a target language character string that is a translation result character string is obtained. The specific processing in FIG. 4 is not described here because general processing may be performed, and the correctness of the output of the machine translation unit 103 is not guaranteed), the source language Even if content that is far from the content of the character string appears as the output of the machine translation unit 103, the user has no reason to know it. That is, the above-mentioned problem that “a user without knowledge of the target language presents a translation without confirming the intention transmitted to the other party” has not been disclosed or solved.

  Further, (Requirement 2) and (Requirement 3) are not disclosed or suggested. Especially for requirement 3, in addition to reducing the weight of the entire system, in addition to the machine translation unit 103, an example storage unit 105 corresponding to RBMT is newly required, which increases both in terms of computer resources and development man-hours. doing.

  Regarding the correspondence to “(Requirement 2) that a user can automatically learn without newly burdening the user with the use result of the user” in RBMT, an input sentence and a translation sentence as in Patent Document 3 are used. An approach is disclosed in which machine learning is performed on the result of determining whether or not each translation device is effective when performing morphological analysis and registering the bilingual phrase in the bilingual dictionary.

  In paragraph [0020] of Patent Document 3, “5 is a machine translation engine, for example, in rule-driven machine translation”, RBMT is exemplified as a translation system under the name of rule-driven machine translation. Is disclosed as a method for determining the effectiveness of registering a word in a bilingual dictionary. Here, when registering a new word / phrase in the rule-based bilingual dictionary, whether or not the word / phrase registration is effective in the system is analyzed, and after analyzing the bilingual pair registered so far, a support vector machine Judgment is made by checking against the judgment space identified using. In other words, “registration of a new phrase to the rule base” is performed regardless of the user's use of translation, and no problem is disclosed or solved for automatic learning using the user's use result.

  In general, RBMT has the disadvantage that “if there is no rule, translation accuracy will be very low, or translation will not be possible at all, and adaptability to other industries and fields will be extremely low”. Is not essentially solved (for example, Patent Document 1).

  In order to improve translation performance, it is necessary to add rules, but this requires development costs. Furthermore, in order to customize the translation system by constructing rules on the user side, considerable knowledge about the rule design rules is required, so that it is not easy for general users to use. That is, as the rules of RBMT increase, the possibility of interference and side effects between rules increases dramatically, and adding a valid rule in one sentence example causes inconvenience in another sentence example. The phenomenon will occur frequently ", but in order to avoid this, it is necessary to add new rules after understanding all the rules of the system. In addition to “automatic learning using”, it is difficult to add simple rules. For example, in the above-mentioned Patent Document 3, only the word translation pairs of the translation rules possessed by the RBMT are automatically determined to be effective, and interference with syntactic translation rules is avoided. Not reached.

  Since RBMT has such a fundamental disadvantage, the efficiency and simplification of “word / example registration in the RBMT database by the user” is disclosed as an issue of RBMT. These require users to input learning data and make decisions, and do not make such requests to the users. “Automatic learning” has not been disclosed or solved.

2) Statistical machine translation (SMT)
In SMT, the distance of the result of the input sentence and reverse translation (the input sentence is once translated into the target language and then re-translated from the target language to the source language) is calculated as the translation reliability. Presenting the reverse translation result together and letting the user rephrase or paraphrase (see, for example, Patent Document 4. However, Patent Document 4 also mentions a case where SMT is not premised, and RBMT is also a machine translation engine) As for RBMT, in Patent Document 4, paragraph number [0009] is referred to as “grammatical rule type translation”.) As a technique for evaluating the translation accuracy of SMT, similarity in units of word n-grams is assumed. Automatic learning using degree (BLEU value) (see Non-Patent Document 1), N back-translated sentences for the translated sentence obtained for the input sentence are generated, and the input sentence and the reverse translation are generated Those comparative evaluation the quality of the translation from the similarity (e.g., Patent Document 2), and the like.

  Here, with regard to “(Requirement 1) that the translation can be generated and used while confirming the intention transmitted to the other party in a simple manner without imposing a burden on the user”, in these examples, for example, the user can By presenting the translation result (Patent Document 4), a method for confirming the “intention transmitted to the other party” to the user is disclosed. However, regarding “being able to generate and use a translated sentence in a simple form”, the user is requested to input again or paraphrase according to the result of reverse translation, and the problem has not been solved.

  Specifically, in paragraph [0013] of Patent Document 4, “the reliability of the translation result can be appropriately obtained, and if the reliability is low, the input side user is appropriately prompted to input again. Providing a reliability calculation device, a translation reliability calculation utilization method, and a translation engine program that can perform the above-mentioned ". In other words, when the system determines that the reliability of the translation result is low, the user is not only forced to re-input the original text, but also changes the expression of the original text until a translated text with sufficiently high reliability is output. It is necessary to continue to check the input to the translation system and the translation output while performing trial and error. In addition, it is important to note that the translation system “generates translated texts with what internal operation”, “calculates reliability based on what criteria”, and “how to trust” It means that they have no knowledge about the point of “whether a highly translated sentence can be obtained”. The user has no guidelines and is forced to input text in various ways in order to obtain highly reliable translation results. This is a system that is extremely difficult to use from a practical point of view. . Patent Document 4 does not present a problem or a solution for these problems, and it can be said that it is insufficient in that “translated sentences can be generated and used in a simple form”.

  In addition, regarding “(Requirement 2) to be able to automatically learn using the user's usage result (without newly burdening the user)”, “(Requirement 1) As described in regard to the ability to generate and use translated sentences while confirming the intention transmitted to the other party in a simple form, both the RBMT and the SMT are able to generate a simple form of the intention transmitted to the other party without burdening the user. "A translation sentence can be generated and used while confirming in" is not realized, and therefore a method for obtaining a "user usage result" that satisfies (Requirement 1) is not disclosed (Requirement 1) Regarding automatic learning using the “user usage result” that satisfies the above, no problem is disclosed or a solution is not presented.

  On the premise that there is no method for obtaining user usage results that satisfy (Requirement 1), the viewpoint of an automatic evaluation / learning method for a translation system when some other learning data is given As described above, “Those that perform automatic learning using BLEU values, which is a method for evaluating the performance of a statistical machine translation system (see Non-Patent Document 1)”, “N back-translated sentences for an input sentence Acquired and compared and evaluated from the similarity between the input sentence and the reverse translation sentence (for example, Patent Document 2), “When registering a bilingual phrase in the bilingual dictionary, whether or not it is effective for the target translation device There are those that machine-learn the determined results (see, for example, Patent Document 3).

  Non-Patent Document 1 prepares a reference translation (correct data) for an input sentence in advance, and compares the translation output result of the translation engine with the correct data based on n-gram in a single language. Is mechanically calculated as a value (BLEU score), and the system is tuned to increase the value. It is necessary to prepare an input sentence and a reference translation (translated sentence) as correct data in advance. Essentially, the internal model of the translation engine does not change at all by this tuning, and the weight itself changes. This method cannot be applied when learning is desired, or when correct data is not given or correct data is not uniquely determined.

  Even if the “use result of the user satisfying the requirement 1” is obtained, the result is “the user has generated and used the translated sentence while confirming the intention transmitted to the other party in a simple manner”. Of these “results”, disclosure of problems and solutions for “whether it is correct answer data”, “how to make the correct answer data uniquely”, and “how to learn the model afterwards” are also shown. Not.

  In Patent Document 2, the translation result is returned to the original language by a plurality of reverse-direction translators, and the similarity of the sentence between the input sentence and the plurality of reverse-translated sentences is calculated mechanically. The contents for evaluating the quality of the results are disclosed. In Patent Document 2, it is disclosed that the original translation result is evaluated by generating a plurality of back-translated sentences. However, the problem solved in Patent Document 2 is that some automatic processing of the translation result is performed. In this respect, there is essentially no difference from the previous Non-Patent Document 1. That is, Patent Document 2 includes some correct sentence (reference translation in Non-Patent Document 1, equivalent to input sentence in Patent Document 2) and translation result sentence (translation sentence in Non-Patent Document 1, translation result reverse in Patent Document 2). Only a score (score represented by the BLEU value of Non-Patent Document 1) representing the degree of coincidence or similarity with the translated back-translated sentence is calculated, and the quality of the translation result is evaluated with that value. Is disclosed.

  Although Patent Document 2 does not mention learning of the translation system itself, in Paragraph No. [0048] of Patent Document 2, “... Performs DP matching between three backward translated sentences and the original sentence, Since the maximum score can be the result of automatic evaluation of forward translations, it is possible to evaluate not only translations of example sentences but also all translations, reducing the labor of evaluation and increasing the reliability of the evaluation. The purpose of this is described in paragraph No. [0009], and it is possible to evaluate not only translated sentences of example sentences but also all translated sentences. This is a construction of an evaluation method for machine translation sentences and an evaluation apparatus for machine translation sentences ”. Even when there is no correct sentence (reference translation) required in advance in Non-Patent Document 1, Patent Document 2 discloses that evaluation can be performed based on a mechanical matching score between a back-translated sentence and an input original sentence. However, although Patent Document 2 does not mention learning, use of this score for learning is suggested in combination with Non-Patent Document 1.

  However, even when Patent Literature 2 and Non-Patent Literature 1 are combined, tuning is still impossible when it is desired to learn the model itself, when the matching score is incorrect, or when the matching score is not uniquely determined.

  In addition, as in Non-Patent Document 1, even if a “use result of a user satisfying Requirement 1” is obtained, the result is “translated text while confirming the user's intention transmitted to the other party in a simple manner. This is the result of generating and using As for how to judge such a result as an evaluation value and how to learn a model with the evaluation value, there is no disclosure or solution of any problem.

  In other words, regarding “generation and use of translated sentences while confirming the intention to be communicated to the other party in a simple manner”, neither assignments nor solution methods have been disclosed, and learning using them However, as with the previous Non-Patent Document 1, neither disclosure nor solution of the problem is shown.

  In addition, regarding “(Requirement 3) Realization of Requirement 1 and Requirement 2 and simultaneously reducing computer resources and development costs”, the calculation of BLEU values in Non-Patent Document 1 and the calculation of matching scores in Patent Document 2 are newly performed. It becomes necessary, and computer resources and development costs are increasing.

3) Model acquisition type machine translation (DNNMT) by deep neural network
Non-Patent Document 3 and Non-Patent Document 4 are LSTM (Long Short Term Memory), which is a kind of RNN (Recurrent Natural Network) and RNN among deep neural networks (DNN) as machine translation by neural networks. This is an example of DNNNMT using Both methods use the bilingual corpus as correct data (positive and negative examples) for the input and output layers of the neural network, and directly learn the intermediate layer of the neural network, so that the translation model is directly inside the neural network. To build. How a DNN has a translation model inside depends on the type and manner of learning data, the number of times of learning, the network configuration of the DNN itself, and the like. In either case, it is not disclosed how the translation performance changes when the internal state is changed (not clarified academically). Originally, the characteristic of neural networks is that non-linear output can be learned. However, the non-linearity has increased dramatically by becoming DNN, and a linear causal relationship between some internal parameters and output performance can be found. The current situation is not.

  In other words, DNNNMT is the same as the previous RBMT and SMT in the sense of “returning some translation result for the input sentence”. However, in terms of “why the translation result was obtained”, RBMT can understand the reason why the translation result was obtained by referring to the database describing the rule (the original rule). While DNNNMT shows that the translation result is the one with the highest probability selected from the model (probability of occurrence of each word / phrase, alignment probability, etc.) and the language model (n-gram probability), DNNNMT uses rules and models. Since the neural network constructs the equivalent of the above, the result of the output layer of the neural network was the sentence of the translation result, so we cannot obtain any knowledge about the internal model / operation.

  For this reason, DNNNMT is mainly in academic research and has not yet reached the practical phase. “(Requirement 1) Translated sentences while checking the user's intention in a simple form without placing a burden on the user. It has not yet reached a problem in terms of actual use, such as “to be able to generate and use“.

  In addition, “(Requirement 2) User learning result can be automatically learned using the user's usage result (without newly burdening the user)” is also obtained as “User usage result satisfying (Requirement 1)”. Even if the result is used, it is necessary to clarify the internal operation of “how to learn the model after that” using the result, and of course, no disclosure or solution of the problem is shown.

  In summary, the prior art has the following problems.

  -If similar examples and translations are not described in the bilingual database, and if the fields are different, there is a possibility that a translation with extremely low translation accuracy may be presented, or that translation may not be possible at all.

  -Since there is no way for the user to easily check and select the content of the translated text, the quality and reliability of the presented content is not good with the method of presenting the reverse translation text or the translation quality (reliability) for the input text. If it is low, the user is forced to re-input, but there is no guarantee that the quality will be improved by re-input, and the user will have to try and input the input.

  -Although the automatic tuning of a translation system based on a score mechanically calculated by some method such as a BLUE value is disclosed in the previous example, "translation while confirming the intention transmitted to the other party in a simple manner by the user" In the case of “generation / use of sentence”, no problem is disclosed about how to perform evaluation and learning based on the result, and no solution is disclosed.

  -Computer resources that generate similar examples and tuning data (e.g., evaluation score calculation) are required. In addition, development and human costs for creating similar examples are also required.

  Therefore, the following improvement measures were examined to improve the functions of the machine translation system.

  One aspect of the machine translation method of the machine translation system is a machine translation method in a machine translation system that is connected to an information output device that outputs language information and performs a translation process between a first language and a second language, The translation target sentence in the first language is received, a plurality of different forward translation sentences are generated by translating the received translation target sentence into the second language, and each of the different forward translation sentences includes the first translation sentence. An operation of generating a plurality of back-translated sentences back-translated into a language and selecting one back-translated sentence from the plurality of back-translated sentences when the information output device outputs the plurality of back-translated sentences. If accepted, the forward translation sentence corresponding to the one reverse translation sentence is output.

  According to the above aspect, a plurality of different forward-translated sentences obtained by translating the sentence to be translated in the first language into the second language are generated, and the plurality of reverse-translated sentences reversely translated into the first language for each of the plurality of forward-translated sentences. When an operation for selecting one reverse translation sentence from a plurality of reverse translation sentences is received when generating and outputting a plurality of reverse translation sentences in the information output device, the order corresponding to the one reverse translation sentence Output translated text.

  For example, a plurality of reverse translation sentences corresponding to the received translation target sentence are presented to the user, and a forward translation sentence corresponding to the reverse translation sentence selected by the user is presented. Accordingly, the user selects the reverse translation sentence closest to the intention of the translation target sentence input by the user from the plurality of reverse translation sentences. For example, the translation target sentence input in the first language is selected. Compared with a system in which only one translated sentence translated into the second language and one reverse translated sentence corresponding to the translated sentence are presented, the reverse translated sentence is different from the intended content of the translation target sentence. The number of scenes where correction of the translation target sentence or re-input is required is reduced.

  In addition, for example, since one reverse translation sentence is selected from a plurality of reverse translation sentences by the user, the machine translation system uses the plurality of presented reverse translation sentences as the intended contents of the input translation target sentence. Among them, it is possible to obtain feedback such as which back-translated sentence is most appropriate or whether it matches the user's favorite expression. Therefore, for example, when machine learning is applied to the machine translation system in the above aspect, in addition to the evaluation of whether or not the reverse translation sentence presented for the input translation target sentence is valid, It is possible to obtain an evaluation as to which of the reverse translated sentences is valid. At this time, feedback for a plurality of reverse translation sentences can be obtained by a single translation operation in the machine translation system, so that high learning efficiency of the machine translation system can be realized.

  Furthermore, according to the above aspect, for machine learning in the machine translation system, evaluation of translation accuracy by calculating a BLEU value, or evaluation of translation accuracy by a mechanical matching score between an input translation target sentence and a reverse translation sentence, etc. There is no need to generate learning data by selecting a reverse translation sentence by the user. Therefore, new computer resources are not required for generating learning data, and development costs can be reduced.

  In the above aspect, for example, the machine translation system is further connected to a voice input device that accepts a voice input by a user and a text input device that accepts a text input by a user, and the translation target sentence is a voice representing the translation sentence Information or text information, and depending on whether the speech information or text information is received as the translation target sentence, the forward translation sentence corresponding to the one reverse translation sentence The output form may be changed.

  According to the above aspect, the output form of the forward translation sentence is changed according to whether the sentence to be translated is received in the form of audio information or text information. As a result, for example, the output modal is determined in accordance with the input modal. Therefore, the user can freely determine the output form by changing the input form.

  In the above aspect, for example, the information output device includes an audio output device and a display, and when the translation target sentence is received in the form of audio information, the forward translation sentence corresponding to the one reverse translation sentence is displayed. If the sentence to be translated is received in the form of text information, the forward translation sentence corresponding to the one reverse translation sentence may be outputted via the display. Good.

  According to this, since the input format and the output format correspond to each other in the same modal, the user only has to input the sentence to be translated in his / her desired output format, and any input format is desired. There is no confusion as to whether the translated sentence is output in the output form.

  In the above aspect, for example, the translation target sentence is received as text information indicating the translation target sentence, and a plurality of different forward-translated sentences obtained by translating the translation target sentence into the second language based on the text information. It may be generated.

  In the above aspect, for example, the machine translation system is further connected to a text input device that accepts text input by a user, and the translation target sentence is received from the text input device as text information indicating the translation target sentence. It is good.

  In the above aspect, for example, the translation target sentence is received as voice information representing the translation target sentence, and the received voice information is subjected to voice recognition processing to generate text information indicating the translation target sentence. A plurality of different forward-translated sentences obtained by translating the sentence to be translated into the second language may be generated based on the text information.

  As a result, it becomes possible to input a translation target sentence using speech. For example, it is not necessary to input the translation target sentence using a keyboard, a touch display, or the like, so that the user can easily input the translation target sentence.

  In the above aspect, for example, the machine translation system is further connected to a voice input device that receives an input of a user's voice, and the translation target sentence is received as voice information representing the translation target sentence from the voice input device. You may do that.

  In the above aspect, for example, the information output device includes a display, and the plurality of reverse translation sentences are displayed in a first area of the display, and the translation is performed in a second area different from the first area of the display. The target sentence may be displayed.

  According to the said aspect, the area | region displayed by a reverse translation sentence and a translation object sentence is divided. Since the reverse translation sentence and the translation target sentence are sentences in the same language, the user can easily distinguish which is the reverse translation sentence and which is the translation target sentence, and there is no confusion.

  In the above aspect, for example, the forward translation sentence corresponding to the one reverse translation sentence may be displayed in the third area of the display.

  As a result, the reverse translation sentence, the translation target sentence, and the forward translation sentence are displayed in different areas, so that it is easy for the user to determine which sentence each is.

  In the above aspect, for example, the display direction of the forward translation sentence corresponding to the one reverse translation sentence may be changed according to an operation on the information output device.

  Thus, for example, when the user A and the user B speak in different languages, the forward translation is output in the language spoken by the user B corresponding to the translation target sentence input in the language spoken by the user A. When the sentence is displayed on the information output device, the direction of the forward-translated sentence can be changed, for example, in the reverse direction, the user A reads out the forward-translated sentence, or changes the direction of the information output apparatus itself, There is no need to convey the contents of the forward translation sentence to the user B, and communication between different languages can be achieved by two users facing each other looking into the information output device from above.

  In the above aspect, for example, the display direction of the forward translation sentence may be changed to a direction different from the display direction of the plurality of reverse translation sentences displayed in the first area.

  In the above aspect, for example, the display direction of the forward translation sentence may be changed to the same direction as the display direction of the plurality of reverse translation sentences displayed in the first area.

  In the above aspect, for example, the forward translation sentence corresponding to the one reverse translation sentence may be displayed in a direction different from the plurality of reverse translation sentences displayed in the first area.

  In the above aspect, for example, the machine translation system is a set of the forward translation sentences obtained by translating the received sentence to be translated into the second language, and includes a forward translation sentence group including the plurality of different forward translation sentences. Generating and determining whether each of the forward translation sentences included in the forward translation sentence group is classified into a question sentence, a positive sentence, a negative sentence, or a command sentence, and the plurality of different orders A translation sentence may be selected from the group of forward translation sentences based on the classified form.

  According to the above aspect, since a plurality of different forward-translated sentences are selected from the forward-translated sentence group based on the form of the sentence, for example, from the forward-translated sentence group mechanically generated based on the translation target sentence Only the forward translation sentence having the same form as the translation target sentence can be selected, and the final translation accuracy can be improved. Further, for example, a plurality of forward-translated sentences including a forward-translated sentence that is different from the form of the sentence to be translated may be selected, thereby being generated based on the forward-translated sentence and presented to the user. You can increase the variation of multiple back-translated sentences. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the plurality of different forward-translated sentences may include at least two or more forward-translated sentences classified into the different forms.

  Thereby, the variation of the several reverse translation sentence produced | generated based on a forward translation sentence and shown to a user can be increased. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the machine translation system generates a forward translation sentence group that is a set of the forward translation sentences obtained by translating the received translation target sentence into the second language, and the forward translation sentence group includes the forward translation sentence group, A plurality of different forward-translated sentences, a subject or a predicate of each of the forward-translated sentences included in the forward-translated sentence group is determined, and the plurality of different forward-translated sentences are included in the determined subject or previous description word Based on the forward translation sentence group, the forward translation sentence group may be selected.

  According to the above aspect, since a plurality of different forward translation sentences are selected from the forward translation sentence group based on the subject or predicate, for example, from the forward translation sentence group mechanically generated based on the translation target sentence Only the forward translation sentence having the same subject or predicate as the translation object sentence can be selected, and the final translation accuracy can be improved. In addition, for example, a plurality of forward translation sentences including a forward translation sentence having a subject or predicate different from the translation target sentence may be selected, thereby being generated based on the forward translation sentence and presented to the user. You can increase the variation of multiple back-translated sentences. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the plurality of different forward-translated sentences may include at least two or more forward-translated sentences that are determined to include different subjects or predicates.

  Thereby, the variation of the several reverse translation sentence produced | generated based on a forward translation sentence and shown to a user can be increased. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the plurality of different forward-translated sentences may be forward-translated sentences that are determined to include the same subject or predicate.

  As a result, for example, only forward translation sentences having the same subject or predicate as the translation target sentence can be selected from a group of forward translation sentences mechanically generated based on the translation target sentence. Can be improved.

  In the above aspect, for example, the machine translation system is a set of at least one reverse translation sentence generated for each of the plurality of different forward translation sentences, and includes the plurality of reverse translation sentences. A sentence group is generated, and for each of the reverse translation sentences included in the reverse translation group, an evaluation value that evaluates the similarity with the translation target sentence is calculated, and the plurality of reverse translation sentences are calculated based on the evaluation value. Based on the reverse translation sentence group, it may be selected.

  According to the above aspect, since a plurality of different back-translated sentences are selected from the back-translated sentence group based on the similarity to the translation target sentence, for example, only the reverse translation sentence having a high similarity to the translation target sentence Can be selected, and the final translation accuracy can be improved. In addition, for example, a plurality of back-translated sentences may be selected including a back-translated sentence having a low degree of similarity to the translation target sentence, thereby increasing variations of the plurality of back-translated sentences presented to the user. Can do. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the machine translation system is a set of at least one reverse translation sentence generated for each of the plurality of different forward translation sentences, and includes the plurality of reverse translation sentences. Generating a sentence group, determining whether each of the reverse-translated sentences included in the reverse-translated sentence group is classified into a question sentence, an affirmative sentence, a negative sentence, or an imperative sentence; The reverse translation sentence may be selected from the reverse translation sentence group based on the classified form.

  According to the above aspect, since a plurality of different back-translated sentences are selected from the back-translated sentence group based on the form of the sentence, for example, from the back-translated sentence group, the reverse is the same form as the form of the translation target sentence. Only the translation can be selected, and the final translation accuracy can be improved. In addition, for example, a plurality of back-translated sentences may be selected including a back-translated sentence that is different from the form of the sentence to be translated, thereby increasing variations of the plurality of back-translated sentences presented to the user. Can do. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the plurality of reverse translation sentences may include at least two or more reverse translation sentences classified into different forms.

  Thereby, the variation of the some reverse translation sentence shown to a user can be increased. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the machine translation system is a set of at least one reverse translation sentence generated for each of the plurality of different forward translation sentences, and includes the plurality of reverse translation sentences. Generating a sentence group, determining a subject or predicate of each of the reverse-translated sentences included in the reverse-translated sentence group, and the plurality of reverse-translated sentences are based on the determined subject or predescription word, The reverse translation sentence group may be selected.

  According to the above aspect, since a plurality of different back-translated sentences are selected based on the subject or predicate from the back-translated sentence group, for example, a reverse having the same subject or predicate as the translation target sentence from the back-translated sentence group. Only the translation can be selected, and the final translation accuracy can be improved. In addition, for example, a plurality of back-translated sentences including a back-translated sentence having a subject or predicate different from the translation target sentence may be selected, thereby increasing variations of the plurality of back-translated sentences presented to the user. Can do. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the plurality of back-translated sentences may include at least two or more back-translated sentences determined to include different subjects or predicates.

  Thereby, the variation of the some reverse translation sentence shown to a user can be increased. Therefore, for example, when applying machine learning to a machine translation system, if a user selects one reverse translation sentence from a plurality of reverse translation sentences having similar contents, the user inputs even a reverse translation sentence that has not been selected. When the intention indicated by the translated sentence is expressed, it is possible to prevent the machine translation system from learning that the translated sentence is wrong because it is not selected.

  In the above aspect, for example, the plurality of back-translated sentences may be back-translated sentences that are determined to include the same subject or predicate.

  Thereby, for example, only a reverse translation sentence having the same subject or predicate as the translation target sentence can be selected from the reverse translation sentence group, and the final translation accuracy can be improved.

  In the above aspect, for example, the machine translation system manages a probability model to be referred to in the translation process, applies machine learning in the translation process, and selects any one of the plurality of reverse translation sentences. The machine learning may be performed based on information indicating whether is selected as the one reverse translation sentence, and the parameters of the probability model may be updated.

  According to the above aspect, the machine learning is performed based on information indicating which one of the plurality of back-translated sentences is selected as the one back-translated sentence. Update. As a result, information indicating which reverse-translated sentence is selected from the multiple reverse-translated sentences presented for the translation target sentence is reflected in the system. The accuracy can be improved.

  In the above aspect, for example, the probability model includes a weight value assigned to each word or phrase used in the translation process, and the machine translation system is a forward translation sentence corresponding to the one reverse translation sentence. A word or phrase included in the selected order translation sentence is compared with a word or phrase included in a non-selected translation sentence that is a forward translation sentence corresponding to a reverse translation sentence other than the one reverse translation sentence. For a word or phrase included only in a translated sentence, a word or phrase included only in the non-selected ordered translated sentence, and a word or phrase included in both the selected ordered translated sentence and the non-selected ordered translated sentence The weight value is updated by applying different weight value update methods, and the updated weight value and the word or the phrase corresponding to the updated weight value are updated as a teacher data. May perform the machine learning is used as data.

  Thus, for example, machine learning can be performed with a difference in score between a word or phrase included in the selected order translation sentence and a word or phrase not included in the selection order translation sentence. Even a word or phrase included in a sentence may be positively evaluated in updating the weight value. Therefore, in a non-selection order translation sentence, when a partly correct translation is performed, the part can be correctly evaluated, and a user's evaluation result can be reflected.

  Furthermore, it is possible to perform learning while reflecting the user's selection result in units of words or phrases sequentially with respect to the probability model by machine learning, thereby improving translation accuracy.

  In the above aspect, for example, the probability model includes a weight value assigned to each word or phrase used in the translation process, and the machine translation system includes a word or phrase included in the one reverse translation sentence, A word or phrase included in a non-selected back-translated sentence that is a back-translated sentence other than the one back-translated sentence, and a word or phrase included only in the one back-translated sentence and the non-select back-translated sentence Applying different weight value update methods to the words or phrases included only in the sentence and the words or phrases included in both the one back-translated sentence and the non-selected back-translated sentence, the weights are applied. A value may be updated, and the machine learning may be performed using the updated weight value and the word or the phrase corresponding to the updated weight value as teacher data.

  Thus, for example, machine learning is performed with a difference in score between a word or phrase included in one selected reverse translation sentence and a word or phrase not included in one selected reverse translation sentence. Therefore, there is a case where a positive evaluation is performed in updating the weight value even for a word or a phrase included in the non-selected back-translated sentence. Therefore, in a non-selection reverse translation sentence, when a partly correct translation is performed, the part can be evaluated correctly and the evaluation result of the user can be reflected.

  Furthermore, it is possible to perform learning while reflecting the user's selection result in units of words or phrases sequentially with respect to the probability model by machine learning, thereby improving translation accuracy.

  In the above aspect, for example, the machine translation system includes a word or phrase included in the one reverse translation sentence and a word or phrase included in a non-selected reverse translation sentence that is a reverse translation sentence other than the one reverse translation sentence. And the word or phrase included only in the one reverse-translated sentence, the word or phrase included only in the non-selected reverse-translated sentence, the one reverse-translated sentence and the non-selected reverse-translated sentence The weight value is updated by applying different weight value update methods to words or phrases included in both, and the updated weight value and the word corresponding to the updated weight value Alternatively, the machine learning may be performed using the phrase as teacher data.

  In the above aspect, for example, the weight value is a value as a positive example for the word corresponding to only the one reverse translation sentence, and the word corresponding to only a sentence other than the one reverse translation sentence, It may be a negative example value.

  Thereby, both positive evaluation and negative evaluation can be reflected on the weight value.

  In the above aspect, for example, the machine learning may be learning using at least one of reinforcement learning, identification learning, and neural network learning.

  In another aspect, the machine translation apparatus performs a translation process between a first language and a second language, and includes an input unit that receives an input of a translation target sentence in the first language, and the translation target sentence. A translation unit that generates a forward translation sentence translated into the second language, a reverse translation sentence obtained by reverse-translating the forward translation sentence into the first language, and the reverse translation sentence and the one reverse translation sentence An output unit that outputs a forward-translated sentence; and a user input unit that accepts a user input, wherein the translation unit generates a plurality of different forward-translated sentences for the translation target sentence, and A plurality of back-translated sentences corresponding to each of the translated sentences are generated, and the output unit outputs one of the plurality of back-translated sentences from the plurality of back-translated sentences when outputting the plurality of back-translated sentences. If an input to select a reverse translation is received, To output the forward translation sentence corresponding to the reverse translation of.

  According to a second other aspect, there is provided a program for controlling an operation of a machine translation device connected to an information output device and performing a translation process between a first language and a second language. The computer receives the translation target sentence in the first language, generates a plurality of different forward translation sentences obtained by translating the received translation target sentence into the second language, and outputs the plurality of different forward translation sentences. When a plurality of reverse translation sentences reversely translated into the first language are generated for each, and the plurality of reverse translation sentences are displayed in the information output device, one reverse translation from the plurality of reverse translation sentences When an operation for selecting a sentence is accepted, a forward-translated sentence obtained by translating the one reverse-translated sentence into the second language is output.

(Embodiment)
Embodiments of the present invention will be described below with reference to the drawings.

  In the following embodiment, there are places where the source language, which is the language before translation, is described as Japanese, and the target language, which is the language after translation, is described as English. The word pair may be any combination of language pairs.

  A translation sentence obtained by translation from the source language to the target language is denoted as a forward translation sentence, and a translation sentence obtained by translation from the target language to the source language is denoted as a reverse translation sentence.

  Further, the reverse translation sentence presented to the user is referred to as a user presentation sentence, the sentence selected by the user is referred to as a (user) selected sentence, and the reverse translation sentence that is not selected is referred to as a (user) non-selected sentence.

  FIG. 1 is a diagram showing an example of the overall configuration of the system in the present embodiment. An information display terminal 100, a network 200, a translation server 300, a microphone 400, and a speaker 500 are provided. Examples of the information display terminal 100 include a smartphone, a tablet terminal, a dedicated display device terminal, and a personal computer (PC). Other than those listed here, any terminal that can exchange information with the user may be used.

  In addition, the user's input operation on the information display terminal 100 may be input by text, input by voice, or the like. In text input, for example, input using a touch panel or input using a keyboard can be considered. In the case of input by voice, for example, input by a microphone can be considered. In addition to this, for example, an input by a gesture may be used.

  When outputting a machine translation result or the like in the information display terminal 100, the result may be output via a display, or the result may be output using voice.

  The network 200 is connected to the information display terminal 100, the translation server 300, the microphone 400, and the speaker 500. An example of the connection method is wired or wireless LAN connection, but is not limited to this as long as each component is connected to be communicable.

  The translation server 300 performs machine translation processing on the translation target sentence received from the information display terminal 100. For example, a character string in the source language input from the information display terminal 100 is received and machine translation processing is performed. It also has a function of performing machine learning in response to feedback from the user about the machine translation result. The detailed configuration of the translation server 300 will be described later.

  For example, the information display terminal 100 and the translation server 300 may be realized integrally.

  The microphone 400 performs voice input to the machine translation system. The microphone 400 may be attached to the information display terminal 100 or may have a function of connecting to the network 200 alone. In addition, the configuration of the microphone 400 is not essential when no voice input is performed for the machine translation system.

  The speaker 500 performs audio output in the machine translation system. The speaker 500 may be attached to the information display terminal 100 or may have a function of connecting to the network 200 alone. In addition, in the machine translation system, the configuration of the speaker 500 is not essential when audio output is not performed.

  The input / output modality of the machine translation system may be provided with either voice input / output or text input / output, or both. When the user inputs voice to the machine translation system, the voice is output. Further, when input is made in a text format by the user, output is performed in text (screen display).

  FIG. 2 is a block diagram showing a configuration of information display terminal 100 in the present embodiment.

  The information display terminal 100 includes a communication unit 101, an input unit 102, an output unit 103, a control unit 104, a selected sentence detection unit 105, and a storage unit 106.

  The communication unit 101 communicates with the translation server 300 to transmit a translation target sentence input at the information display terminal 100, receive a translated sentence and a reverse translated sentence, which will be described later, and the like. Moreover, not only these information but various information is transmitted / received with the translation server 300. FIG.

  The input unit 102 receives input from the user. The input unit 102 receives input such as input of a translation target sentence and selection input of a reverse translation sentence described later. As an input form, speech input or text input is conceivable. When speech input is used, speech recognition processing is performed on a translation target sentence input by speech, and a character string output as a result of the speech recognition processing is input as an input sentence to the machine translation system. When input in text format is used, input of a character string by a keyboard, mouse, touch panel, etc. is accepted.

  The output unit 103 outputs a translation target sentence input through the input unit 102, a plurality of reverse translation sentences received via the communication unit 101, translation results, and the like. The output unit 103 may be realized as a display unit that performs screen display, such as a display. For example, a touch panel display or monitor used for a smartphone, a tablet terminal, or the like is assumed. Moreover, you may implement | achieve as an audio | voice output part which outputs audio | voices, such as a speaker. The control unit 104 controls operations of the communication unit 101, the input unit 102, the output unit 103, the control unit 104, the selected sentence detection unit 105, and the storage unit 106.

  The selected sentence detection unit 105 detects which reverse translation sentence the user has selected for the plurality of reverse translation sentences output by the output unit 103. For example, when the input unit 102 inputs that one reverse translation sentence is selected from a plurality of reverse translation sentences, user selection information indicating which reverse translation sentence is selected is selected by the selected sentence detection unit 105. Detected. The detected user selection information is transmitted to the translation server 300 via the communication unit 101. Further, the output content of the output unit 103 may be controlled based on the user selection information. For example, when the output unit 103 is realized by a display, the reverse translation sentence selected by the user may be highlighted, or the reverse translation sentence not selected by the user may be controlled to be deleted from the display screen.

  Here, the reverse translation sentence selected by the user and the forward translation sentence corresponding to the reverse translation sentence are set as the user selection sentence. Moreover, the reverse translation sentence which the user did not select and the forward translation sentence corresponding to the reverse translation sentence are made into a user non-selection sentence.

  The storage unit 106 temporarily stores information received from the translation server 300, stores various application programs executed in the information display terminal 100, and the like.

  FIG. 3 is a block diagram showing a configuration of translation server 300 in the present embodiment. The translation server 300 includes a communication unit 210, a control unit 220, a machine translation unit 230, and a storage unit 240. Further, the machine translation unit 230 includes a forward translation unit 231, a forward translation selection unit 232, a reverse translation unit 233, a reverse translation selection unit 234, a selection sentence determination unit 235, a phrase division unit 236, a selection result evaluation unit 237, learning Part 238.

  The communication unit 210 communicates with the information display terminal 100 to receive a translation target sentence input at the information display terminal 100, transmit a translated sentence and a reverse translated sentence, which will be described later, and the like. Moreover, not only these information but various information are transmitted / received with the information display terminal 100. FIG.

  The control unit 220 controls various operations of the communication unit 210, the machine translation unit 230, and the storage unit 240.

  The storage unit 240 stores a phrase table that the machine translation unit 230 refers to in various translation processes, phrase division processes, and the like. The phrase table will be described later.

  The machine translation unit 230 executes machine translation processing on the translation target sentence received via the communication unit. The machine translation unit 230 performs machine translation by rule-based machine translation (RBMT), statistical machine translation (SMT), model acquisition type machine translation (DNNMT) using a deep neural network, or the like. The machine translation unit 230 evaluates the translation result and obtains a score such as an automatic evaluation score (such as BLEU) and an internal score (such as manual evaluation).

  Moreover, in order to reflect the selection result by the user in machine learning, a phrase table as shown in FIG. 11 is prepared as necessary depending on the translation method.

  Since rule-based machine translation (RBMT) performs translation based on conversion rules (stored translation pairs as a database) constructed manually, there is a possibility that the phrase table as shown in FIG. 11 is not held. is there. However, when there is a parallel translation database in units of phrases or words, the learning result may be reflected in the parallel translation database, or a phrase table may be prepared separately.

  In statistical machine translation (SMT), since a phrase table as shown in FIG. 11 is held in advance, it may be used.

  In model acquisition type machine translation (DNNNMT) using a deep neural network, the model itself is automatically constructed, and therefore, the phrase table is often not held. Therefore, a separate phrase table may be prepared.

  Furthermore, the target for reflecting the user's selection in the learning result is not limited to this. For example, a database that represents a paraphrase pair of the source languages may be provided. Details of the machine translation process will be described later.

  The forward translation unit 231 executes machine translation processing from the language (source language) of the translation target sentence received via the communication unit 210 to the language (target language) output as a result of translating the translation target sentence. Here, a translation from the source language to the target language is a forward translation, and a translation sentence obtained by the forward translation is a forward translation sentence. At this time, a plurality of forward translation sentences are generated for the translation target sentence by the forward translation process. Further, the forward translation unit 231 performs machine translation processing with reference to the phrase table stored in the storage unit 240. A plurality of forward translation sentences generated in the forward translation unit 231 are defined as a forward translation sentence group. The forward translation unit 231 outputs the generated forward translation sentence group to the forward translation sentence selection unit 232.

  The forward translation sentence selection unit 232 performs a forward translation sentence selection process of selecting N forward translation sentences from the forward translation sentence group generated by the forward translation part 231. Details of the forward translation selection process will be described later. The forward translation sentence selection unit 232 outputs the selected N forward translation sentences to the reverse translation unit 233.

  The reverse translation unit 233 performs machine translation from the language of the forward translation sentence (target language) to the language of the translation target sentence (source language) for each of the N forward translation sentences selected by the forward translation sentence selection unit 232. Execute the process. Here, the translation from the target language to the source language is the reverse translation, and the translation sentence obtained by the reverse translation is the reverse translation sentence. At this time, one or more reverse translation sentences are generated for each forward translation sentence by the reverse translation process. Therefore, as a result, a plurality of reverse translation sentences are generated. Further, the reverse translation unit 233 performs machine translation processing with reference to the phrase table stored in the storage unit 240. A plurality of reverse translation sentences generated in the reverse translation unit 233 are set as a reverse translation sentence group. The reverse translation unit 233 outputs the generated reverse translation sentence group to the reverse translation sentence selection unit 234.

  The reverse translation sentence selection unit 234 performs reverse translation sentence selection processing for selecting M reverse translation sentences from the reverse translation sentence group generated by the reverse translation part 233. This reverse translation sentence selection process will be described later. The reverse translation sentence selection unit 234 transmits the selected M reverse translation sentences to the information display terminal 100 via the communication unit 210. In the output unit 103 of the information display terminal 100, M back-translated sentences are output so as to be selectable.

  Based on the user selection information received from the information display terminal 100 via the communication unit 210, the selected sentence determination unit 235 determines which reverse user the M of the reverse translation sentences selected by the reverse translation sentence selection unit 234 has. It is determined whether a translated sentence is selected, and the determined information is output to the phrase dividing unit 236.

  The phrase division unit 236 divides the reverse translation sentence into phrases or words for each of the plurality of reverse translation sentences based on the plurality of reverse translation sentences input from the reverse translation sentence selection unit 234. Also, the forward translation corresponding to the reverse translation sentence is also divided into phrases or words. At this time, information indicating which reverse translation sentence is selected from a plurality of reverse translation sentences input from the selected sentence determination unit 235 may also be used. Further, a phrase table stored in the storage unit 240 may be used. Information obtained by dividing the reverse translation sentence and the forward translation sentence into phrases or words and the user selection information are output to the selection result evaluation unit 237.

  Phrase division often uses a phrase table that shows the correspondence between phrases or words, as shown in Statistical Machine Translation (SMT), but it always uses a fixed phrase table. It is not necessary and may be similar. When a phrase table is used in machine translation, division may be performed using the phrase table. Alternatively, a separately prepared phrase table or the like may be used, or a bilingual dictionary or the like may be used.

  The selection result evaluation unit 237 evaluates the forward translated sentence and the reverse translated sentence based on the information input from the phrase dividing unit 236. At this time, different evaluations may be performed for the user selection sentence and the user non-selection sentence based on the user selection information. A detailed evaluation method will be described later. The selection result evaluation unit 237 outputs evaluation information obtained by evaluating the forward translation sentence and the reverse translation sentence to the learning unit 238.

  The learning unit 238 performs machine learning in the machine translation process by updating the phrase table stored in the storage unit 240 based on the evaluation information input from the selection result evaluation unit 237. That is, the evaluation information is reflected in the phrase table. The machine learning target may be a phrase table referred to by the forward translation unit 231 or a phrase table referred to by the reverse translation unit 233. Further, it is not always necessary to reflect the evaluation information in the phrase table. For example, the result may be reflected in a paraphrase dictionary or a word dictionary to perform machine learning in machine translation processing. A detailed reflection method of the evaluation information with respect to the phrase table will be described later.

  FIG. 4 is a diagram illustrating a hardware configuration of a computer that realizes the functions of the respective units of the information display terminal by a program. The computer 1000 includes an input device 1001 such as an input button and a touch pad, an output device 1002 such as a display and a speaker, a CPU (Central Processing Unit) 1003, a ROM (Read Only Memory) 1004, a RAM (Random Access Memory) 1005, and the like. Prepare. Further, the computer 1000 reads information from a recording medium such as a hard disk device, a storage device 1006 such as an SSD (Solid State Drive), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. 1007 may include a transmission / reception device 1008 that performs communication via a network. Each unit described above is connected by a bus 1009.

  Then, the reading device 1007 reads the program from a recording medium on which a program for realizing the functions of the above-described units is recorded, and stores the program in the storage device 1006. Alternatively, the transmission / reception device 1008 communicates with a server device connected to the network, and causes the storage device 1006 to store a program for realizing the function of each unit downloaded from the server device.

  Then, the CPU 1003 copies the program stored in the storage device 1006 to the RAM 1005, and sequentially reads and executes the instructions included in the program from the RAM 1005, thereby realizing the functions of the above-described units. Further, when executing the program, the RAM 1005 or the storage device 1006 stores information obtained by various processes described in each embodiment, and is used as appropriate.

  FIG. 5 is a flowchart showing the operation of the machine translation system in the present embodiment. For simplicity, in this flowchart, it is assumed that the input unit 102 and the output unit 103 of the information display terminal 100 are realized by a touch panel display. In the information display terminal 100, user input and output of translation results and the like are performed via a touch panel display. For example, a configuration in which the input unit 102 and the output unit 103 are independent, such as a keyboard and a display. It may be. Further, voice input / output may be performed. For explanation, there are places where the source language (native language) is Japanese and the target language is English. However, these are only examples, and the source language and the target language may be in any combination.

  First, in step S401, a translation target sentence input by the user is acquired. In step S402, machine translation processing is performed on the translation target sentence. The machine translation process here is a forward translation process for translating a source language translation target sentence into a target language character string (forward translation sentence). At this time, for example, statistical machine translation (SMT) scores the certainty as a translation from the translation model and the language model. The translation model is a statistical model that defines the likelihood of the translated word, and the language model is a statistical model that defines the likelihood of the arrangement of words in the output language. From these two models, the probability of translation is scored, and the translation results are output in the order of the scores, so that a plurality of forward translation sentences are generated. A translation group.

  Since the specific process of machine translation in the present embodiment is a general machine translation process, the description thereof is omitted here.

  In step S403, N forward-translated sentences are selected from the forward-translated sentence group based on a predetermined criterion. For example, an evaluation score may be assigned to each of a plurality of forward translation sentences included in the forward translation sentence group, and N may be selected from those having a high evaluation score. Further, N may be selected at random regardless of the evaluation score. Etc. In addition, in consideration of the contents indicated by a plurality of forward translation sentences included in the forward translation sentence group, the selected N forward translation sentences may be selected so as not to include the same meaning. In addition, when a plurality of forward translation sentences included in the forward translation sentence group have few different meanings, a process of adding forward translation sentences having different purposes may be performed as necessary. A detailed method for selecting a forward translation will be described later with reference to FIG.

  If it is determined that it is necessary to increase the number of forward translation sentences after the process of step S403 (YES in S404), the process returns to step S402, and the forward translation process is executed again. At this time, in order to obtain a forward translated sentence different from the already obtained forward translated sentence, a sentence having a lower score than before is selected. Moreover, you may use another translation parameter | index (for example, RIBES: rank-based intuitive biliary evaluation score). In addition, if a database of paraphrases in the source language is held, similar sentences can be created by applying them to the input sentence, and the forward translation process can be executed again. As a result, a sentence having the same meaning as that of the input sentence is inputted although it is different in the surface layer, so that a different forward translation sentence can be obtained.

  If the forward translation process is executed in step S403 after the forward translation process is performed again, the forward translation sentence may be selected based on a different standard from the previous one, or the forward translation sentence may be selected based on the same standard. You may choose.

  If it is determined that it is not necessary to increase the number of forward translation sentences after the process of step S403 is performed (No in S404), the flow proceeds to the reverse translation process of step S405. In step S405, reverse translation processing for performing reverse translation on the N forward-translated sentences obtained in step S403 is executed. If the translation from the source language to the target language is a forward translation, the reverse translation is a backward translation from the target language to the source language. In each of the N forward-translated sentences, a reverse translation process for generating an arbitrary reverse-translated sentence is performed. The reverse translation process for generating an arbitrary reverse translation sentence is a reverse translation process for generating a corresponding one-to-one reverse translation sentence for each of the N forward translation sentences, and a reverse translation process for the N forward translation sentences. This means reverse translation processing with a forward translation sentence that is not performed, reverse translation processing for generating a plurality of reverse translation sentences for one forward translation sentence, and the like. By this reverse translation process, a plurality of reverse translation sentences are generated. The plurality of reverse translation sentences are set as a reverse translation sentence group for convenience of explanation.

  In addition, the criteria for outputting the back-translated text may be determined based on some system criteria, or the user may determine them. The standard of the system here is to calculate the score of the forward translation sentence using evaluation such as BLEU or manual evaluation, for example, the reverse translation processing is not performed for the forward translation sentence having a low score (a certain forward translation sentence) The number of reverse translation sentences to be generated is 0), and an arbitrary number of reverse translation sentences having a high score are obtained (a plurality of reverse translation sentences are generated for a certain forward translation sentence). When the user determines the number of back-translated sentences, for example, setting how many back-translated sentences are generated for one forward-translated sentence can be considered, but this is not restrictive.

  Step S406 is a reverse translation sentence selection process for selecting M reverse translation sentences from the reverse translation sentence group obtained in step S405. In the reverse translation sentence selection process, a process substantially similar to the forward translation sentence selection process in step S403 is performed. A detailed selection method will be described later with reference to FIGS.

  If it is determined that it is necessary to increase the number of reverse translation sentences after the reverse translation sentence selection process in step S406 is executed (YES in S407), the process returns to step S402 and the forward translation process is performed again. If the forward translation process is executed in step S403 after the forward translation process is performed again, the forward translation sentence may be selected based on a different standard from the previous one, or the forward translation sentence may be selected based on the same standard. You may choose.

  If it is determined that it is not necessary to increase the number of reverse translation sentences after the reverse translation processing in step S406 (No in S407), the M reverse translation sentences selected from the reverse translation sentence group are the user-provided sentences. Is output at the information display terminal 100 in the next step.

  In step S408, M back-translated sentences are transmitted to the information display terminal 100 and displayed on the touch panel display.

  In step S409, the selected sentence detection unit 105 detects whether one reverse translation sentence is selected from the M reverse translation sentences displayed on the touch panel display of the information display terminal 100.

  If it is determined that there is no selection of the reverse translation sentence for a certain period, the machine translation system returns to the initial state and accepts the user's input sentence (No in S409). At this time, the display screen of the touch panel display is reset.

  Also, when the user performs any reset operation, the machine translation system similarly returns to the initial state and accepts the user's input.

  When the selected sentence detection unit 105 detects that any reverse translation sentence is selected (Yes in S409), user selection information indicating which reverse translation sentence is selected is transmitted to the translation server 300. The

  In step S411, the selected sentence determination unit 235 acquires M pieces of reverse translation sentences from the reverse translation sentence selection unit 234, and M pieces of information are displayed on the information display terminal 100 based on the user selection information received from the information display terminal 100. It is determined which of the reverse translation sentences has been selected. Subsequent steps will be described assuming that the selected back-translated sentence among the M back-translated sentences is a selected translated sentence, and the unselected back-translated sentence is a non-selected back-translated sentence. The selected sentence determination unit 235 outputs the selected back-translated sentence and the non-selected back-translated sentence to the phrase dividing unit 236.

  In step S <b> 412, phrase division processing is performed on the forward translated sentence corresponding to each of the selected reverse translated sentence and the non-selected reverse translated sentence, and the selected reverse translated sentence and the non-selected reverse translated sentence.

  Phrase division is to divide a target sentence into phrases or words that are shorter units. A specific example of the phrase dividing process will be described later.

  FIG. 11 is an example of a general phrase table in the present embodiment. The phrase table is a table that represents a correspondence relationship in units of phrases or words between the bilingual of the source language and the target language.

  In FIG. 11, when the source language is Japanese and the target language is English, from the left, the Japanese phrase, the English phrase, the English-Japanese translation probability of the phrase (the probability that the English phrase is translated into the Japanese phrase) ), English-Japanese word translation probabilities (the product of translation probabilities for each word in a phrase when English is translated into Japanese), phrase Japanese-English translation probabilities (Japanese phrases are in English (Probability of translation into a phrase) and product of translation probabilities of words in Japanese and English directions (product of translation probabilities for each word in the phrase when Japanese is translated into English). However, the phrase table does not necessarily include all these pieces of information, and is not limited to this notation method. Since this phrase table includes translation probabilities, it is also called a probability model in a broad sense.

  For example, in the phrase table shown in FIG. 11, the probability that the phrase PH1 is translated into the phrase PH1 is 0.38, the probability that the word SD3 is translated into the word SD1, and the probability that the word SD4 is translated into the word SD4. The product is 0.04, the probability that the phrase PH1 is translated into the phrase PH2 is 0.05, the product of the probability that the word SD1 is translated into the word SD3 and the probability that the word SD2 is translated into the word SD4 is 0.02. It represents that.

  Using such a phrase table, the phrase division processing is performed on the selected back-translated sentence, the non-selected back-translated sentence, and the forward-translated sentence corresponding to each of these back-translated sentences.

  FIG. 12 is an explanatory diagram showing an outline of phrase division.

  FIG. 12 shows reverse-translated sentences RS10, RS20, and RS30 expressed in the source language, and forward-translated sentences TS10, TS20, and TS30 of the target language corresponding to each of these reverse-translated sentences. When the phrase splitting process is performed on the reverse translation sentence in the source language, for example, when the reverse translation sentence RS10 is split into phrases, the phrase is divided into three phrases PH11, PH12, and PH12. Further, when the phrase dividing process is performed on the target language forward translation sentence, for example, when the forward translation sentence TS10 is divided into phrases, the phrase PH14, PH15, and PH16 are divided into three phrases (words).

  The phrase into which the target sentence to be subjected to the phrase division process is divided depends on the character strings of the source language and the target language described in the phrase table, and may not be uniquely determined.

  Step S413 is a phrase evaluation process for evaluating a score according to a predetermined standard for each phrase output by the phrase division process. Details of the phrase evaluation process will be described later.

  In step S414, the forward translation corresponding to the selected reverse translation sentence determined in S411 is transmitted to the information display terminal 100 and displayed on the touch panel display as the translation result. At this time, the selected reverse translation sentence displayed on the touch panel display may be highlighted. In addition, the display of the non-selected reverse translation sentence may be erased on the touch panel display, and if the displayed translation result is a forward translation sentence corresponding to the selected reverse translation sentence selected by the user, what will be done? May be displayed.

  In addition, since a series of processes of step S414 and steps S412 to S414 can be performed in parallel, step S414 can be any time before the phrase dividing process of step S412 and after the learning process of step S415. It may be executed at the timing.

  In step S415, machine learning such as reinforcement learning, identification learning, and neural network learning is performed based on the score for each phrase obtained in step S413. Details of this processing will be described with reference to FIGS.

  FIG. 6 is a flowchart showing a specific operation of the translated sentence selection process in the present embodiment.

  With reference to FIG. 6, the specific processes of the forward translation sentence selection process executed by the forward translation sentence selection unit 232 in step S403 and the reverse translation sentence selection process executed by the reverse translation sentence selection unit 234 in step S406 are described. explain. For convenience of explanation, the two processes of the forward translation sentence selection process and the reverse translation sentence selection process are collectively referred to as a translation sentence selection process.

  The forward translation sentence selection unit 232 executes a forward translation sentence selection process for selecting N forward translation sentences from the forward translation sentence group generated by the forward translation part 231, and the reverse translation sentence selection part 234 performs reverse translation. A reverse translation sentence selection process for selecting M reverse translation sentences from the reverse translation sentence group generated by the unit 233 is executed. Based on the evaluation score of each of the plurality of forward translation sentences included in the forward translation sentence group, N forward translation sentences are selected, and based on the evaluation score of each of the plurality of reverse translation sentences included in the reverse translation sentence group. Thus, M back-translated sentences are selected.

  The following explanation applies to both the forward translation sentence selection process and the reverse translation sentence selection process, so it is explained as the translation sentence selection process, and the forward translation sentence and the reverse translation sentence are collectively expressed as a translation sentence. To do. A forward translation sentence group and a reverse translation sentence group are referred to as a translation sentence group. Furthermore, in practice, N forward-translated sentences are selected and M reverse-translated sentences are selected, but in the following description, they are described as N without distinction.

  In step S501, Nk translation sentences having a high evaluation score (1 <= N, 0 <= k <= N) are selected from the translation sentence group. Here, as an example of the evaluation score, BLEU which is often used as a method for evaluating the translation accuracy can be cited. Other methods for evaluating translation accuracy include WER (Word Error Rate), METEOR (Metric for Evaluation of Translation Ordering), and RIBES (Rank-based Intrinsic Evaluation). Any of these may be used, and other methods may be used without being limited thereto.

  From the translation sentences, the remaining k sentences are selected from the translation sentences not selected in step S501 (S502).

  In order to select k translations, translations having an evaluation score within a predetermined threshold may be extracted, and k translations may be selected at random from the extracted translations. Alternatively, translation sentences having an evaluation score within a predetermined threshold may be extracted, and k pieces may be selected in order from the extracted translation sentences having the lowest score. When selecting Nk translations, a translation with a high evaluation score is selected. However, when selecting k translations, a translation with a high evaluation score is not necessarily selected. If only translated sentences having high evaluation scores that are mechanically given according to a specific evaluation criterion are selected, there is a high possibility that the selected translated sentences are all sentences having similar contents. In consideration of presenting a plurality of reverse translation sentences to the user on the information display terminal 100 and allowing the user to select one reverse translation sentence therefrom, a plurality of reverse translation sentences selected from a somewhat different viewpoint may be presented. preferable.

  If only similar reverse translation sentences are presented and the user selects them, there is a concern that a high learning effect cannot be obtained in a machine learning process for a machine translation system described later. As a correct example (correct answer) for a machine translation system when machine learning processing is executed by using selected back-translated sentences selected by the user and non-selected back-translated sentences not selected by the user as teacher data Since the selected back-translated sentence learned and the non-selected back-translated sentence learned as a negative example (incorrect answer) are similar, the machine translation system shows a significant difference between the positive and negative examples. I can not expect the effect of learning. For this reason, as described in the present embodiment, it is preferable to select a plurality of back-translated sentences from a somewhat different viewpoint.

  In addition, when only similar reverse translation sentences are presented, the user needs to select one reverse translation sentence in consideration of fine differences between the reverse translation sentences, so one reverse translation sentence is selected intuitively. There is also a concern that it may take time to select a reverse translation sentence. If a plurality of reverse translation sentences selected from a somewhat different viewpoint are presented as in the present embodiment, the user can intuitively select the translation content intended by the user.

  In addition, since the reverse translation sentence is generated from the forward translation sentence, it is considered preferable to select a plurality of forward translation sentences with a somewhat different idea in the same manner at the stage of selecting the forward translation sentence.

  Further, as a different method for selecting k translated sentences, k translated sentences may be selected based on past user selection information. For example, for each translated sentence, the number of times selected by the user in the past is stored (for the back-translated sentence, the number of directly selected is stored, and for the forward-translated sentence, the number of times the corresponding back-translated sentence is selected is stored. In other words, k translations may be selected in order from the highest stored number of translations in the translation group. Also, instead of selecting based on such direct number of times, based on the user's past use history of the machine translation system, the tendency of the translated sentence that the user can easily select for the sentence to be translated is analyzed. , K translations may be selected based on the analyzed tendency. Note that the selection method for selecting k translated sentences is not limited to this as long as a selection criterion different from the selection criterion for selecting N−k translated sentences is used.

  Furthermore, in the method for selecting these k translated sentences, a process of removing a translated sentence having the same meaning as the Nk translated sentences may be performed. Moreover, you may perform the process which excludes the translation sentence of the same meaning in k translation sentences. Alternatively, when k translations are not included in k translations or there are few translations having a different meaning from the Nk translations, a process of adding a translation having a different meaning is included. May be performed. Further, regarding the reverse translation sentence selection process, a process such as selecting a reverse translation sentence having the same purpose as compared with the translation target sentence may be performed.

  For example, it counts how many translated sentences (question sentences, positive sentences, negative sentences, and imperative sentences) are included in the translated sentence group (the number of different sentences). Perform forward or reverse translation.

  There are also techniques such as performing syntax analysis on translated sentences. Parse multiple translated sentences included in the translation sentence group, determine what each word indicates the subject, and the number of different subjects indicating the number of different words indicating the subject is less than or equal to the threshold If there is, the forward translation process or the reverse translation process is performed again. At this time, different numbers may be calculated for the predicates rather than the subject. Moreover, both may be sufficient.

  Here, the number of different sentences, the number of different subjects, the number of different verbs or objects, etc. are used as the evaluation score, and the forward translation sentence selection process and / or the reverse translation sentence selection so that the predetermined number of different numbers are included. Processing may be performed.

  When executing the forward-translated sentence selection process, for example, an evaluation criterion “make the number of sentence differences 2 or more” can be provided, and the forward-translated sentence can be selected using the number of sentence differences as an evaluation score. For each of a plurality of forward-translated sentences included in the forward-translated sentence group, analysis is performed sequentially using syntax analysis and / or semantic analysis, and any number of question sentences, affirmative sentences, negative sentences, and imperative sentences is selected. Whether a type appears may be scored as the number of different sentences, and l small sets including two or more types may be created.

  Further, for example, when the reverse translation sentence selection process is executed, an evaluation criterion “make the number of different subject numbers equal to or less than a predetermined number” can be provided, and the reverse translation sentence can be selected using the number of different subject words as an evaluation score.

  The forward translation sentence selection process or the reverse translation sentence selection process is not limited to these examples, and any forward translation sentence selection method or reverse translation sentence selection method may be used.

  Regarding the reverse translation sentence selection process, a comparison evaluation with the translation target sentence may be performed, and a value obtained by the comparison evaluation may be used as the evaluation score. For example, syntax analysis is performed on the translation target sentence and a plurality of reverse translation sentences in the reverse translation sentence group, and the similarity between the translation target sentence and each of the plurality of reverse translation sentences is determined, and the determined similarity degree Based on the above, k back-translated sentences may be selected.

  Moreover, you may use combining not only these examples but arbitrary evaluation scores. Furthermore, N may be set in advance as a predetermined number, and the score may be combined and used so that the different number included therein becomes a desired number.

  The above-described translation sentence selection method is an example, and is not limited thereto.

  When k = 0, all N sentences are selected in order from the translation sentence with the highest evaluation score. When k = N, all the N translated sentences are selected by a selection method other than selecting in order from the translated sentence having the highest evaluation score.

  The standard of the system here is to calculate the score of the forward translation sentence using evaluation such as BLEU or manual evaluation, for example, the reverse translation processing is not performed for the forward translation sentence having a low score (a certain forward translation sentence) The number of reverse translation sentences to be generated is 0), and an arbitrary number of reverse translation sentences having a high score are obtained (a plurality of reverse translation sentences are generated for a certain forward translation sentence). When the user determines the number of back-translated sentences, for example, setting how many back-translated sentences are generated for one forward-translated sentence can be considered, but this is not restrictive.

  FIG. 7 is a flowchart showing a specific operation of the reverse translation sentence selection process in the present embodiment.

  In the description of the reverse translation processing (step S405) using FIG. 5, it is assumed that there may be a forward translation sentence in which no reverse translation sentence is generated, and a plurality of reverse translation sentences are generated from one forward translation sentence. It was also good. Here, a process of selecting one reverse translation sentence from a plurality of reverse translation sentences obtained from one forward translation sentence will be described.

  The following processing is performed on all N forward-translated sentences selected by the forward-translated sentence selecting process.

  In step S601, the reverse translation sentence generated for the forward translation sentence A among the N forward translation sentences is extracted.

  In step S602, the number of extracted reverse translation sentences is determined. When no reverse translation sentence has been generated for forward translation sentence A, that is, when zero reverse translation sentences have been generated for forward translation sentence A (0 in step S602), Since the translated sentence A is not presented to the user, it is deleted (step S603).

  Next, when there is one reverse translation sentence for the forward translation sentence A (one in step S602), the reverse translation sentence is determined as a reverse translation sentence corresponding to the forward translation sentence A (step S604).

  Finally, when two or more reverse-translated sentences are generated for the forward-translated sentence A (two or more in step S602), one of the optimum reverse-translated sentences corresponding to the forward-translated sentence A is selected. Are determined (step S605). The determination method is selected, for example, by using a method of referring to an automatic evaluation score or a score by manual evaluation. These are repeated for all N forward-translated sentences.

  Finally, from the back-translated sentences obtained by these processes, remove the back-translated sentence with the same meaning, or add a translated sentence with a different purpose when there are few back-translated sentences with a different purpose. May be performed as necessary (step S606). The processing here is the same as the processing described in the description of FIG.

  In the above description, one reverse translation sentence is selected for one forward translation sentence in step S605 of FIG. However, a plurality of back-translated sentences may be selected for one forward-translated sentence. In this case, among the plurality of reverse translation sentences presented to the user, there exists a reverse translation sentence having the same corresponding forward translation sentence.

  FIG. 8 is a flowchart showing a specific operation of the phrase evaluation process in the present embodiment.

  In the initial state of the flow, when the selected translated sentence, the non-selected reverse translated sentence, and the forward translated sentence corresponding to each of the selected reverse translated sentence and the non-selected reverse translated sentence that have been divided into phrases in advance are acquired. To do. In addition, it is assumed that a phrase table corresponding to a phrase obtained by dividing the reverse translation sentence and the forward translation sentence into phrases is acquired in the same manner. This phrase evaluation process is performed for each of the reverse translation sentence and the forward translation sentence. For convenience of explanation, a phrase division process for the reverse translation sentence will be described as an example.

  In step S701, with respect to the selected back-translated sentence and the non-selected back-translated sentence, the phrases included in these back-translated sentences are compared for each back-translated sentence, and the presence or absence of a phrase that exists only in the selected back-translated sentence is confirmed To do.

  When there is a phrase that exists only in the selected reverse translation sentence (Yes in step S701), a user selection score is added to the phrase that exists only in the selected reverse translation sentence (step S702). This user-selected score is a score that favors phrases that appear only in the selected sentence. Finally, in the process of FIG. 9, in the case of a forward translation sentence (if the phrase evaluation process target is a forward translation sentence, the translation target sentence) ) Are reflected in the phrase table Japanese-English translation probability or English-Japanese translation probability. Any method may be used as the score addition method at this time. For example, a method of adding points uniformly to the corresponding phrase or adding points depending on the length of the corresponding phrase is assumed. When the score addition is completed, the flow proceeds to step S703. If there is no phrase that exists only in the selected reverse translation sentence (No in step S701), the flow proceeds directly to step S703 without performing any special processing.

  Similarly, in step S703, the phrase included in these reverse translation sentences is compared for each reverse translation sentence with respect to the selected reverse translation sentence and the non-selection reverse translation sentence, and the phrase exists only in the non-selection reverse translation sentence. Check if there is any.

  If there is a phrase that exists only in the non-selected reverse translation sentence (Yes in step S703), the user selection score is deducted for the phrase that exists only in the non-selected reverse translation sentence (step S704). Any method may be used as a score deduction method at this time. For example, a method of deducting points uniformly for the corresponding phrase or deducting points depending on the length of the corresponding phrase is assumed.

  In steps S702 and S704, addition or deduction of the user selection score is not always essential. That is, in the flowchart of FIG. 8, it is assumed that score addition or deduction is not performed for a phrase that exists only in a selected reverse translation sentence, and a user selection score is deducted for a phrase that exists only in a non-selected reverse translation sentence. Also good. Alternatively, the user selection score may be added to a phrase that exists only in the selected reverse translation sentence, and the user selection score may not be added or reduced to a phrase that exists only in the non-selected reverse translation sentence.

  In addition, scores may be added for phrases existing in the selected back-translated sentence and some non-selected back-translated sentences. In this case, for example, a value considering the user selection score of a phrase included only in the selected reverse translation sentence and a user selection score of a phrase included only in the non-selected reverse translation sentence (for example, a phrase included only in the selected reverse translation sentence) It is conceivable to add a user selection score of (2) and an average value of user selection scores of phrases included only in the non-selected reverse translation sentence. These score addition methods are merely examples, and are not limited thereto.

  Hereinafter, a specific example will be described with reference to FIG. For example, a case will be described in which three reverse translation sentences are presented on the information display terminal 100 and the user selects one of the reverse translation sentences. At this time, the three reverse translation sentences presented on the information display terminal 100 are the reverse translation sentences RS10, RS20, and RS30, and the reverse translation sentence selected by the user is the reverse translation sentence RS10. For convenience of explanation, a reverse translation sentence selected by the user is defined as a selected reverse translation sentence, and a reverse translation sentence not selected by the user is defined as a non-selected reverse translation sentence. In the initial state of the flow, the reverse translation sentence RS10 that is the selected reverse translation sentence is divided into phrases PH11, PH12, and PH13. Similarly, for the non-selected reverse translation sentence, the reverse translation sentence RS20 is divided into phrases PH21, PH22, and PH23, and the reverse translation sentence RS30 is divided into phrases PH31, PH32, and PH33.

  In S701, the presence / absence of a phrase that exists only in the reverse translation sentence RS10 that is the selected reverse translation sentence among the reverse translation sentences RS10, RS20, and RS30 is confirmed. Then, since the phrase PH12 is included only in the reverse translation sentence RS10 that is the selected reverse translation sentence, the user selection score of the phrase PH12 is added to “+1”.

  Similarly, in S702, the presence / absence of a phrase that exists only in the reverse translation sentences RS20 and RS30, which are non-selected reverse translation sentences, is confirmed. Then, since the phrase PH22 (PH32) and the phrase PH31 are included only in the non-selected reverse translation sentence, the user selection scores of the phrase PH22 (PH32) and the phrase PH31 are decremented to “−1”.

  Here, for the phrases PH11 (PH21) and PH13 (PH23, PH33) included in both the selected back-translated sentence and the non-selected back-translated sentence, the user-selected score is not added or reduced.

  With the above-described processing, the amount of points added or deducted for the final user-selected score for each phrase is as follows. “± 0” for phrase PH11 (PH21), “−1” for phrase PH31, “−1” for phrase PH22 (PH32), “+1” for phrase PH12, phrase For PH13 (PH23, PH33), the added amount or deducted amount is “± 0”. Since the score addition and deduction points here are only examples, points may be added or deducted in an order larger than this, or points may be added or deducted in an order smaller than this.

  Moreover, the phrase evaluation process with respect to the forward-translated sentence which is a target language is demonstrated with a specific example below using FIG. The above-described reverse translation sentences RS10, RS20, and RS30 correspond to the forward translation sentences TS10, TS20, and TS30, respectively. In the initial state of the flow, the forward translation TS10 is divided into phrases PH14, PH15, and PH16. Further, the forward translation TS20 is divided into phrases PH24, PH25, and PH26, and TS30 is divided into phrases PH34, PH35, and PH36.

  In S701, the presence or absence of a phrase that exists only in the forward translation sentence 1 corresponding to the selected reverse translation sentence 1 is confirmed. Then, since the phrase PH16 is included only in the forward translation sentence TS10, the user selection score of the phrase PH16 is added to “+1”.

  Similarly, in S702, the presence / absence of a phrase that exists only in one of the forward-translated sentences TS20 and TS30 corresponding to the reverse-translated sentences RS20 and RS30 that are non-selected reverse-translated sentences is confirmed. Then, since the phrase PH26 (PH36) and the phrase PH34 are included only in the forward translation TS20 or TS30, the user-selected scores of the phrase PH26 (PH36) and the phrase PH34 are decremented to “−1”.

  Phrase PH24 and phrase PH15 (PH25, PH35) included in the forward translation sentence TS10 corresponding to the selected reverse translation sentence and included in either the forward translation sentence TS20 or the forward translation sentence TS30 corresponding to the non-selected reverse translation sentence With respect to, the user-selected score is not added or reduced.

  With the above-described processing, the final score addition or deduction amount for each phrase is “± 0” for the phrase PH24 (PH34), “−1” for the phrase PH34, and the phrase PH26 (PH36). Is "-1" for the phrase PH16, "+1" for the phrase PH16, and "± 0" for the phrase PH15 (PH25, PH35). Since the points added and deducted for the user-selected score here are examples, points may be added or deducted in an order larger than this, or points added or deducted in an order smaller than this.

  FIG. 9 is a flowchart showing a specific operation of the learning process in the present embodiment.

  In step S801, a phrase pair of a phrase included in the forward translation sentence corresponding to the selected reverse translation sentence and a phrase included in the selected reverse translation sentence, or a phrase included in the translation target sentence corresponding to the selected reverse translation sentence; A phrase pair with a phrase included in the forward translation corresponding to the selected reverse translation is acquired. A phrase pair is two phrases that have correspondence (having the same meaning) between the source language and the target language during machine translation. Furthermore, the user selection score obtained by the process of FIG.

  In this phrase pair, if the user selection score for the value of the phrase table that is referred to when translated from the source language to the target language is defined using the example shown in FIG. 12, for example, phrase PH31 → phrase PH34: −1 / Phrase PH22 (PH32) → Phrase PH26 (PH35):-1 / Phrase PH11 (PH21) → Phrase PH14 (PH24): 0 / Phrase PH13 (PH23, PH33) → Phrase PH15 (PH25, PH35): 0 / Phrase PH12 → Phrase PH16: +1.

  Further, when a user selection score is defined for a phrase table value that is referred to when a target language is translated into a source language in a phrase pair, for example, phrase PH34 → phrase PH31: −1 / phrase PH26 (PH36) → phrase PH22 (PH32): -1 / phrase PH14 (PH24)-> phrase PH11 (PH21): 0 / phrase PH15 (PH25, PH35)-> phrase PH13 (PH23, PH33): 0 / phrase PH16-> phrase PH12: +1 Become.

  In step S802, the above-mentioned user selection score is reflected on the English-Japanese translation probability or the Japanese-English translation probability of the phrase table stored in the storage unit 240. In addition, the user selection score may be given a slope or weight, for example, by multiplying the user selection score by a certain value and reflecting the result in the phrase table.

  Using these, the machine translation unit 230 and the reverse translation unit 233 perform machine learning such as reinforcement learning, identification learning, and neural network learning.

  In conventional machine translation, the probability value of the phrase table as shown in FIG. 11 is tuned based on the bilingual corpus (data that collects pairs of sentences translated between two different languages). However, there has never been a method for performing machine learning with a difference in score between what is included in a user-selected sentence and what is not included in a user-selected sentence. Evaluation results can be reflected.

  Furthermore, it is possible to train a translation model or a language model by mixing user selection results in phrase units sequentially with a translation model or a language model created from a pre-prepared bilingual corpus by machine learning. Therefore, the accuracy can be improved.

  Further, by performing machine learning, an optimum parameter is selected based on the data. As a result, since the selection result of the person (user) is reflected in the translation system, it is possible to construct a translation system that is easy for a person to use.

  In addition to performing machine learning like these, it is also possible to generate a new corpus from the obtained phrases and use it as a parallel corpus for translation engine learning.

  FIG. 10 is a flowchart showing specific processing of the learning unit 238 in the present embodiment. The contents of the flowchart shown in FIG. 10 will be described using the reverse translation sentence and the forward translation sentence shown in FIG.

  In step S901, a phrase pair of a user-selected sentence and its forward translation sentence is acquired.

  For example, when the reverse translation sentence RS10 is selected by the user in a state where the reverse translation sentences RS10, RS20, and RS30 are displayed in the user presentation sentence display area 1102 (when the reverse translation sentence RS10 is a user selection sentence). explain. Since the reverse translation RS10 is selected by the user, phrase pairs such as the phrase PH11 and the phrase PH14, the phrase PH12 and the phrase PH16, and the phrase PH13 and the phrase PH15 are acquired.

  In step S902, a phrase pair in the input sentence and the forward translation sentence used in the machine translation in the machine translation unit 230 is acquired.

  For example, when the content indicated by the reverse translation RS30 is an input sentence, phrase pairs such as a phrase PH31 and a phrase PH34, a phrase PH32 and a phrase PH36, and a phrase PH33 and a phrase PH35 are acquired.

  In step S903, a phrase having the same target language character string is acquired for the phrase acquired in the input sentence and the user-selected sentence. For example, it is assumed that a phrase pair in the user-selected sentence is as follows. The phrase pairs in the user-selected sentence are phrase PH11 and phrase PH14, phrase PH12 and phrase PH16, phrase PH13 and phrase PH15. On the other hand, it is assumed that the phrase pair of the input sentence is as follows. The phrase pairs of the input sentence are phrase PH31 and phrase PH34, phrases PH32 and PH36, and phrases PH41 and PH42.

  At this time, in the phrase of the user selected sentence and the phrase of the input sentence, the phrase PH33 and the phrase PH41 in the source language are phrases having the same meaning but different expressions. The phrase PH33 and the phrase PH41 are phrases corresponding to each other between the phrase of the user-selected sentence and the phrase of the input sentence.

  Finally, it is checked whether the target languages are the same and the source languages are different, and different ones are held as paraphrases (paraphrases) (Yes in S904, S905). That is, since the source language is different even if the target language is the same, these can be regarded as paraphrases in the source language.

  For example, the phrase PH33 and the phrase PH41 can be regarded as paraphrases, and these are held as paraphrases between the source languages.

  This paraphrase can be referred to when performing machine translation in the machine translation unit 230, or can be referred to as paraphrasing on the source language side before translation.

  FIG. 13 is a diagram illustrating an example of a display screen in the present embodiment.

  For example, as shown in FIG. 13A, when the input of the original text to be translated is received from the user, the input original text OS1 content is displayed in the input text display area 1101.

  Next, as shown in FIG. 13B, the reverse translation result for the translated sentence obtained by translating the original sentence OS1 is displayed in the user-presented sentence display area 1102.

  As an example, a mode in which three reverse translation sentences are output will be described here. As the reverse translation result, for example, the reverse translation sentence RS1, the reverse translation sentence RS2, and the reverse translation sentence RS3 are output and displayed in the user presentation sentence display area 1102. At this time, the reverse translation sentences RS1 to RS3 displayed in the user presented sentence display area 1102 in FIG. 13B are similar sentences having the same meanings in the source language. Although these are expected to be similar sentences having the same meaning due to the characteristics of the reverse translation processing, they may be implemented so as to output sentences having different meanings.

  Next, as shown in FIG. 13C, the user confirms the reverse translation result displayed in the user-presented sentence display area 1102, and selects the reverse translation sentence closest to the input content intended by the user. . Here, for example, a reverse translation sentence RS1 is selected for the original sentence OS1.

  When the user selects a reverse translation sentence, a translation sentence corresponding to the selected reverse translation sentence is displayed in the translation result display area 1103. Here, the translated sentence TS1, which is a translated sentence corresponding to the reverse translation sentence RS1, is displayed.

  The screen display is not limited to the layout shown in FIGS. 13A, 13B, and 13C. Various buttons may be arranged as necessary. For example, after an original text to be translated is input, a translation process may be executed when an operation on the button is performed. In addition, a reverse translation sentence may be displayed in the user-sentence sentence display area 1102 by performing an operation on the button. Further, the arrangement, displayed contents, and orientation of the input sentence display area 1101, the user presentation sentence display area 1102, and the translation result display area 1103 are not limited to the above-described contents.

  FIG. 14 is a diagram showing an example of a display screen in the present embodiment.

  A part of the display is different from FIG. Here, the direction of the text displayed in the translation result display area 1201 is different from the direction of the text displayed in the input text display area 1202 and the user presented text display area 1203. This assumes a scene in which two users (a source language speaker and a target language speaker) communicate with each other across an information display terminal. That is, the sentences displayed in the input sentence display area 1202 and the user-provided sentence display area 1203 are displayed in a direction suitable for the source language speaker, and the sentences displayed in the translation result display area 1201 are the target language spoken words. It is displayed in the direction that suits the person. As a result, the source language speaker reads the translated sentence output for the input sentence to the target language speaker, or changes the direction of the information display terminal so that the target language speaker can easily confirm the translated sentence. Therefore, smooth communication is possible between users who speak different languages via an information display terminal or the like. The orientation of the translation result display area 1201 can be changed by an arbitrary operation by the user. Further, as in FIG. 13, the arrangement of each area, the displayed contents, and the orientation are not limited to these.

  The translation method according to one aspect of the present invention has been described based on the embodiments. However, the present invention is not limited to these embodiments. Unless it deviates from the main point of this invention, the form which carried out the various deformation | transformation which those skilled in the art can think to this embodiment, or the structure constructed | assembled combining the component in a different embodiment is also contained in the scope of the present invention. .

  For example, in the above description, one reverse translation sentence is selected by the user from a plurality of reverse translation sentences presented on the information display terminal 100, but a plurality of reverse translation sentences may be selected. . For example, one is selected as a user's selection and a forward translation sentence is presented, but other user non-selected sentences may also be evaluated, and the result may be reflected in the system as a learning result. As an evaluation method here, for example, the user selects a non-selected sentence that is allowed in the same order as the user-selected sentence, in which the user ranks the user-non-selected sentence in the order of good and bad. Evaluation is performed by a method such as the user selecting an unselected sentence that is not acceptable. By doing these, it is possible to evaluate a sentence that was not selected, and reflecting these in the system leads to learning of the system.

  Further, in the above description, as shown in FIG. 13, it is assumed that the user-provided sentence and the resulting translated sentence are output as text with respect to the input sentence. It doesn't matter. In that case, the user may select the user presented sentence by a method of selecting one from the user presented sentences through the microphone.

  The machine translation method according to the present invention is useful in a machine translation system that is connected to an information output device that outputs language information and performs a translation process between a first language and a second language.

DESCRIPTION OF SYMBOLS 100 Information display terminal 101 Communication part 102 Input part 103 Output part 104 Control part 105 Selected sentence detection part 106 Storage part 200 Network 210 Communication part 220 Control part 230 Machine translation part 231 Forward translation part 232 Forward translation sentence selection part 233 Reverse translation part 234 Reverse translation selection unit 235 Selection sentence determination unit 236 Phrase division unit 237 Selection result evaluation unit 238 Learning unit 240 Storage unit 300 Translation server 400 Microphone 500 Speaker 1000 Computer 1001 Input device 1002 Output device 1003 CPU
1004 ROM
1005 RAM
1006 Storage device 1007 Reading device 1008 Transmission / reception device 1009 Bus

Claims (21)

A machine translation method in a machine translation system that performs translation between a first language and a second language by connecting to an information output device that outputs language information,
Receiving the sentence to be translated in the first language;
Generating a plurality of different forward-translated sentences obtained by translating the received sentence to be translated into the second language;
Generating a plurality of back-translated sentences back-translated into the first language for each of the plurality of different forward-translated sentences;
When an operation for selecting one reverse translation sentence from the plurality of reverse translation sentences is received when outputting the plurality of reverse translation sentences in the information output device, the one corresponding to the one reverse translation sentence Output forward translation,
Machine translation method. The machine translation system is further connected to a speech input device that accepts speech input by a user, and a text input device that accepts text input by a user,
The translation target sentence is received in the form of audio information representing the translation sentence, or text information,
Changing the output form of the forward-translated sentence corresponding to the one reverse-translated sentence, depending on whether the sentence to be translated is received in the form of the audio information or the text information;
The machine translation method according to claim 1. The information output device has an audio output device and a display,
When the translation target sentence is received in the form of voice information, the forward translation sentence corresponding to the one reverse translation sentence is output via the voice output device,
When the translation target sentence is received in the form of text information, the forward translation sentence corresponding to the one reverse translation sentence is output via the display.
The machine translation method according to claim 2. The machine translation system is further connected to a text input device that accepts text input by a user,
The translation target sentence is received as text information indicating the translation target sentence from the text input device,
Based on the text information, generate a plurality of different forward-translated sentences obtained by translating the sentence to be translated into the second language.
The machine translation method according to claim 1. The machine translation system is further connected to a voice input device that accepts user voice input,
The translation target sentence is received as voice information representing the translation target sentence from the voice input device;
Generate text information indicating the translation target sentence by performing a speech recognition process on the received speech information,
Based on the text information, generate a plurality of different forward-translated sentences obtained by translating the sentence to be translated into the second language.
The machine translation method according to claim 1. The information output device has a display;
The plurality of reverse translation sentences are displayed in a first area of the display,
The translation target sentence is displayed in a second area different from the first area of the display;
The machine translation method according to claim 1. In the third area of the display, the forward translation sentence corresponding to the one reverse translation sentence is displayed.
The machine translation method according to claim 6. The forward translation sentence corresponding to the one reverse translation sentence is displayed in a different direction from the plurality of reverse translation sentences displayed in the first region.
The machine translation method according to claim 7. The machine translation system includes:
A set of forward translation sentences obtained by translating the received translation target sentence into the second language, and generating a forward translation sentence group including the plurality of different forward translation sentences;
For each of the forward-translated sentences included in the forward-translated sentence group, determine whether it is classified into a question sentence, an affirmative sentence, a negative sentence, or a command sentence,
The plurality of different forward-translated sentences are determined from the forward-translated sentence group based on the classified form.
The machine translation method according to claim 1. The plurality of different forward-translated sentences include at least two or more forward-translated sentences each classified into the different forms.
The machine translation method according to claim 9. The machine translation system includes:
Generating a forward translation sentence group that is a set of the forward translation sentences obtained by translating the received translation target sentence into the second language;
The forward translation sentence group includes the plurality of different forward translation sentences,
Determining the subject or predicate of each of the forward-translated sentences included in the forward-translated sentence group;
The plurality of different forward-translated sentences are determined from the forward-translated sentence group based on the determined subject or previous description word.
The machine translation method according to claim 1. The machine translation system includes:
A set of at least one reverse translation sentence generated for each of the plurality of different forward translation sentences, and generating a reverse translation sentence group including the plurality of reverse translation sentences;
For each of the reverse translation sentences included in the reverse translation group, calculate an evaluation value that evaluates the similarity with the translation target sentence,
The plurality of reverse translation sentences are selected from the group of reverse translation sentences based on the evaluation value.
The machine translation method according to claim 1. The machine translation system includes:
A set of at least one reverse translation sentence generated for each of the plurality of different forward translation sentences, and generating a reverse translation sentence group including the plurality of reverse translation sentences;
For each of the back-translated sentences included in the back-translated sentence group, determine whether it is classified into a question sentence, a positive sentence, a negative sentence, or a command sentence,
The plurality of reverse translation sentences are selected from the group of reverse translation sentences based on the classified form.
The machine translation method according to claim 1. The plurality of reverse translation sentences include at least two or more reverse translation sentences classified into the different forms.
The machine translation method according to claim 13. The machine translation system includes:
A set of at least one reverse translation sentence generated for each of the plurality of different forward translation sentences, and generating a reverse translation sentence group including the plurality of reverse translation sentences;
Determining the subject or predicate of each of the reverse translation sentences included in the reverse translation sentence group;
The plurality of back-translated sentences are selected from the back-translated sentence group based on the determined subject or predescription word.
The machine translation method according to claim 1. The plurality of back-translated sentences include at least two or more back-translated sentences that are determined to include different subjects or predicates.
The machine translation method according to claim 15. The plurality of back-translated sentences are back-translated sentences determined to include the same subject or predicate,
The machine translation method according to claim 15. The probability model includes a weight value given to each word or phrase used in the translation process,
The machine translation system includes:
A word or phrase included in a selected translated sentence that is a forward translated sentence corresponding to the one reverse translated sentence, and an unselected translated sentence that is a forward translated sentence corresponding to a reverse translated sentence other than the one reverse translated sentence Compare with the word or phrase it contains
A word or phrase included only in the selected translated sentence;
A word or phrase included only in the non-selected order translation, and
The weight value is updated by applying a different updating method of the weight value to words or phrases included in both the selected order translated sentence and the non-selected order translated sentence, and the updated weight value And performing the machine learning using the word or the phrase corresponding to the updated weight value as teacher data,
The machine translation method according to claim 15. The probability model includes a weight value given to each word or phrase used in the translation process,
The machine translation system includes:
Comparing a word or phrase included in the one reverse translation sentence with a word or phrase included in a non-selected reverse translation sentence that is a reverse translation sentence other than the one reverse translation sentence,
A word or phrase included only in the one reverse translation sentence;
A word or phrase included only in the unselected back-translated sentence;
For each word or phrase included in both the one back-translated sentence and the non-selected back-translated sentence, the weight value is updated by applying different weight value updating methods,
Performing the machine learning using the updated weight value and the word or the phrase corresponding to the updated weight value as teacher data;
The machine translation method according to claim 15. A machine translation device that performs a translation process between a first language and a second language,
An input unit for receiving input of the translation target sentence in the first language;
A translation unit that generates a forward translation sentence obtained by translating the sentence to be translated into the second language, and a reverse translation sentence obtained by reverse translating the forward translation sentence into the first language;
An output unit that outputs the forward-translated sentence corresponding to the reverse-translated sentence and the one reverse-translated sentence;
A user input unit that accepts a user input, wherein the translation unit generates a plurality of different forward-translated sentences for the translation target sentence, and a plurality of reverse translations corresponding to each of the plurality of different forward-translated sentences Generate statements,
When the output unit receives the input for selecting one reverse translation sentence from the plurality of reverse translation sentences in the user input unit while outputting the plurality of reverse translation sentences, Outputting the forward translation corresponding to the translation,
Machine translation device. A program for connecting to an information output device and controlling the operation of a machine translation device that performs translation processing between a first language and a second language,
For the computer of the machine translation device,
Receiving a sentence to be translated in the first language;
Generating a plurality of different forward-translated sentences obtained by translating the received sentence to be translated into the second language;
Generating a plurality of back-translated sentences back-translated into the first language for each of the plurality of different forward-translated sentences;
In the information output device, when an operation for selecting one reverse translation sentence from the plurality of reverse translation sentences is received while the plurality of reverse translation sentences are displayed, the one reverse translation sentence is changed to the first reverse translation sentence. To output forward-translated sentences translated into two languages,
program.

Images