JP6709963B2 - Translation device and translation method - Google Patents

Description

The present disclosure relates to a translation device and a translation method for machine-translating input first language data into second language data.

Generally, machine translation includes free text translation and fixed text translation. In the translation of a fixed phrase (hereinafter, referred to as a fixed phrase translation), a fixed phrase created in advance and a corresponding translated sentence are stored, and the translated sentence is output according to the fixed phrase. Therefore, in the standard sentence translation, although the pattern of the input original sentence is limited, it is easy to obtain a highly accurate translation as compared with the translation of the free sentence. Therefore, the standard sentence translation is particularly effective when the utterance content is standardized to some extent, such as an announcement in an aircraft, and translation accuracy is required.

For example, in a certain fixed phrase translation, a Japanese fixed phrase and the English fixed phrase of the translation result are created in advance, a Japanese fixed phrase similar to the input sentence is acquired, and an English sentence corresponding to the fixed phrase is extracted. It is output (for example, refer to Patent Document 1).

Japanese Patent No. 4393494

In the conventional fixed phrase translation, it is difficult to match the input sentence with the fixed phrase. Therefore, the translation accuracy is not sufficient.

The present disclosure provides a translation device that can improve the accuracy of translation in fixed-form sentence translation.

A translation device according to one aspect of the present disclosure includes an input unit and a control unit. The input unit acquires the first text data in the first language. The control unit generates second text data in a second language which is a parallel translation of the first text data. The control unit further generates the first replacement data by replacing the first term in the predetermined type included in the first text data with a parameter, and generates the second replacement data in the second language corresponding to the first replacement data. The second text data is generated by acquiring and replacing the parameter included in the second replacement data with the second term of the second language which is a parallel translation of the first term.

The translation device according to the present disclosure is effective for improving translation accuracy.

FIG. 1 is a configuration diagram of the entire system according to the first embodiment. FIG. 2 is a configuration diagram of the translation apparatus according to the first embodiment. FIG. 3 is a flowchart showing the operation of the translation device. FIG. 4 is a flowchart showing a translation process by the translation device. FIG. 5A is a diagram showing an example of an utterance sentence in the first language. FIG. 5B is a diagram showing a result of recognition of the utterance sentence of FIG. 5A by the translation device. FIG. 5C is a diagram showing a result of parameter substitution performed on the sentence of FIG. 5B. FIG. 5D is a diagram showing a result of parameter weighting performed on the sentence of FIG. 5C. FIG. 5E is a diagram showing a result of similarity calculation performed on the sentence of FIG. 5D. FIG. 5F is a diagram showing parallel translation of the sentence of FIG. 5E in the second language. FIG. 5G is a diagram showing the result of replacing the parameters in the sentence of FIG. 5F. FIG. 6 is a diagram showing the contents of the parameter information. FIG. 7A is a diagram for explaining the configuration of parameters. FIG. 7B is a diagram for explaining the configuration of parameters. FIG. 8 is a diagram showing the contents of the example sentence information. FIG. 9A is a diagram illustrating an example of a script function. FIG. 9B is a diagram showing an example of a script function. FIG. 10A is a diagram showing an example of an utterance sentence in the first language. FIG. 10B is a diagram showing a result in which the utterance sentence of FIG. 10A is recognized by the translation device. FIG. 10C is a diagram showing a result of similarity calculation performed on the sentence of FIG. 10B. FIG. 10D is a diagram showing parallel translation of the sentence of FIG. 10C in the second language. FIG. 11 is a diagram for explaining scene information and scene determination. FIG. 12 is a diagram showing original data for generating example sentence information. FIG. 13 is a configuration diagram of a translation device according to another embodiment. FIG. 14A is a diagram showing an example of an example sentence. FIG. 14B is a diagram showing an example of a regular expression. FIG. 14C is a diagram showing an example of data having parameters. FIG. 15A is a diagram showing an example of data having parameters. FIG. 15B is a diagram illustrating an example of the example sentence.

Hereinafter, embodiments will be described in detail with reference to the drawings as appropriate. However, more detailed description than necessary may be omitted. For example, detailed description of well-known matters or duplicate description of substantially the same configuration may be omitted. This is to prevent the following description from being unnecessarily redundant and to facilitate understanding by those skilled in the art.

It should be noted that the inventors have provided the accompanying drawings and the following description in order for those skilled in the art to fully understand the present disclosure, and are not intended to limit the subject matter described in the claims by these. Absent.

(Embodiment 1)
[1-1. Constitution]
FIG. 1 shows the overall configuration of a system 1 including a translation device 110 (see FIG. 2) according to an embodiment of the present disclosure. In the present embodiment, the system 1 will be described as an example of what is set in the aircraft. The system 1 includes a server device 10 (translation device 110) and a plurality of display devices 20 connectable to the server device 10 wirelessly or by wire. The display device 20 is a terminal computer (information terminal) provided in each seat in the aircraft.

[1-1-1. Configuration of server device]
The server device 10 includes a CPU (Central Processing Unit) 11 (an example of a control unit and an input unit), a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, an input I/F (interface) 14, and an output I/F. The computer device includes an F (interface) 15, a communication I/F (interface) 16, and a storage device 17.

The CPU 11 is a processor or circuit that executes processing according to a predetermined program. The ROM 12 stores a control program or the like that describes the processing procedure of the CPU 11. The RAM 13 temporarily stores a control program and data as a work memory. The input I/F 14 is a connection unit connected to the microphone 30, and receives the A/D converted audio data from the microphone 30. The output I/F 15 is a connection unit connected to the speaker 40, and transmits the D/A converted sound to the speaker 40. The communication I/F 16 is a communication circuit for communicating with the display device 20 wirelessly or by wire.

The storage device 17 is a storage device such as a magnetic storage device such as an HDD (hard disk drive) or a semiconductor memory, and stores various programs such as applications and OSs, and various data. The storage device 17 includes a database DB (see FIG. 2) that holds various data used by the CPU 11. The storage device 17 may be connected to the server device 10 as a storage device separate from the server device 10.

The microphone 30 is connected to the server device 10, receives voice, and outputs voice data. The microphone 30 is used by a flight attendant, and the sound input to the microphone is output from the speaker 40.

[1-1-2. Configuration of translation device]
FIG. 2 shows the configuration of the translation device 110 according to this embodiment. The translation device 110 is a device that translates Japanese (an example of a first language) data input by voice into text data of an English (an example of a second language).

The translation device 110 includes a voice acquisition unit 111, a voice recognition unit 112, a text generation unit 113, a translation processing unit 114, and a translation result output unit 115. Each unit of the translation device 110 is realized by the CPU 11 referring to the data in the storage device 17 and executing a predetermined program.

The voice acquisition unit 111 acquires voice data from the microphone 30. The voice recognition unit 112 recognizes the acquired voice data based on a predetermined model stored in advance. The text generator 113 converts the recognized voice data into text data. The text data includes character information or a character string that can be converted into voice or can be displayed on the screen.

The translation processing unit 114 performs translation processing described below on the generated text data.

The translation result output unit 115 outputs the result of the translation processing by the translation processing unit 114. The output translation result is transmitted to each display device 20 via the communication I/F 16 (connection unit). Here, the English translation result is transmitted to the display device 20 set to English.

The storage device 17 holds various data stored as a database DB.

[1-1-3. Database contents]
In the database DB, parameter information (see FIG. 6), scene information (see FIG. 11), example sentence information (see FIG. 8), script function information (see FIG. 9A and FIG. 9B), etc. are stored in association with each other. ..

<Parameter information>
As shown in FIG. 6, the parameter information is information including a parameter type described later, a regular expression of a term, a parallel translation of the term in English, and an importance of the term. The parameter information is used when the text data to be translated includes a term in a predetermined type. The predetermined type is a category of important information such as an airport name, an aircraft flight number, and time. Each given type includes many variations of terms.

The degree of importance is, as will be described later, a degree of weight set in advance according to the type of parameter and the degree of importance of the term.

As shown in FIG. 7A, each parameter includes a parameter type (for example, “airports”), weighting designation information (for example, “!”) in which weighting set according to the importance of a term is reflected, It includes parameter position information (eg, “$1”) indicating the position of the parameter in the sentence, and “{%” and “%}” indicating the start and end of the parameter.

The parameter position information is information indicating the order of terms in a sentence when the recognized text data includes a plurality of terms of the same type. For example, in the example shown in FIG. 7B, since there are three parameters indicating the airport name, {%airports!$1%}, {%airports!$2%}, and {%airports!$3%} are included in the text data. .. Similarly, since there are two parameters representing time, {%hours!$1%} and {%hours!$2%} are included in the text data.

The parallel translation of the parameter information includes a calling function (script function) as shown in FIG. As shown in FIGS. 9A and 9B, the script function is a term in a specific type, and a term whose content pattern (number of digits of number, conversion of temperature unit, etc.) is determined to some extent is converted into an appropriate expression. It is a function for converting. By performing conversion processing using these functions, the quality and accuracy of translation are improved.

For example, when a flight number is uttered as described later, the number of digits of the flight number recognized by voice recognition may be different from the number of digits of the actual flight number. The number of digits in the flight number is predetermined (4 digits in the example of FIG. 9A). Therefore, by using the script function “FlightNo” generated in advance as shown in FIG. 9A, it is possible to shape the flight number with an appropriate number of digits.

Further, as shown in FIG. 9B, a script function for converting a unit different for each culture area for passengers may be used. For example, as shown in FIG. 9B, by using a script function “Celsius2Fahrenheit” generated in advance, the temperature expressed in Celsius (° C.) can be expressed in Fahrenheit (° F) in the translation result data.

<Scene information>
As shown in FIG. 11, the scene information is information indicating the time and the environment in which the aircraft is placed, and includes information indicating the current scene. The search range of the example sentence information described later is determined according to the current scene.

<Example sentence information>
As shown in FIG. 8, the example sentence information (an example of a reference text data group) includes a Japanese example sentence set and a parallel translation corresponding to the same example sentence set. Here, the Japanese example sentence set is associated with the scene information. Each example sentence set is generated for each scene and is composed of a combination of a plurality of example sentences (example sentence subset).

[1-1-4. Display device configuration]
The display device 20 is a computer device installed in each seat of the aircraft and provided with a display for displaying information. As shown in FIG. 1, the display device 20 includes a CPU 21, a ROM 22, a RAM 23, a communication I/F (interface) 26, a memory 27, an input device 28, and a display 29.

The CPU 21 is a processor or circuit that processes and outputs instructions and data. The ROM 22 stores a control program or the like that describes the processing procedure of the CPU 21. The RAM 23 temporarily stores a control program and data as a work memory. The communication I/F 26 is, for example, a communication circuit capable of wired or wireless communication with the server device 10 and other devices in the aircraft. The memory 27 is a storage circuit such as a semiconductor memory and stores data such as images.

The input device 28 receives an input operation by a user of a seat input via a touch panel, a keyboard, or a pointing device (mouse or the like). The display 29 includes a liquid crystal panel, an organic EL panel, or a panel similar thereto, and includes a screen for displaying a processed image. The display 29 displays the text data of the translation result generated by the translation device 110.

[1-2. motion]
Hereinafter, the operation of the translation device 110 according to the present disclosure will be described.

[1-2-1. Overall operation of the translation device]
FIG. 3 shows an overall control flow by the translation device 110. Hereinafter, a process of translating the content of the announcement in Japanese spoken by the crew into English in the aircraft will be described.

S100: In the translation device 110, the input I/F 14 and the voice acquisition unit 111 acquire voice data from the microphone 30. Here, it is assumed that the utterance sentence D100 is input from the microphone 30, as shown in FIG. 5A, for example.

S200: The voice recognition unit 112 performs recognition processing on the acquired voice data. Here, it is assumed that the voice data corresponding to the text data D200 shown in FIG. 5B has been voice-recognized. Here, the flight number is changed to "9000900909" instead of the original "9999" because the number of each digit is recognized as a single number because the user speaks slowly.

S300: The voice-recognized data is generated by the text generation unit 113 as text data D200 (an example of the first text data) as shown in FIG. 5B.

S400: The converted text data D200 is translated by the translation processing unit 114 as described later.

S500: The translation result data subjected to the translation processing is output by the translation result output unit 115.

S600: The translation result data is transmitted to the display device 20 of each seat, and the translation result data D500 (an example of the second text data) as shown in FIG. 5G is displayed on each display 29.

[1-2-2. Translation process]
FIG. 4 shows a flow of translation processing by the translation processing unit 114 (see FIG. 2) in step S400 of FIG.

S401: The text data D200 generated by the text generation unit 113 in step S300 is acquired.

S402: It is determined whether the acquired text data D200 includes a term. In the present example, in the case of the text data D200 shown in FIG. 5B, with reference to the parameter information shown in FIG. 6, the term T1 is determined to be the term in the type “airports”, and the term T2 is determined to be the term in the type “flight_num”. It

S403: When it is determined that the text data D200 recognized in step S402 includes a term, the relevant term is replaced with a parameter. In this example, as shown in FIG. 5C, since the terms T1 and T2 are determined to be terms, the data D403 is acquired.

The term recognized before the replacement is associated with the parameter and stored separately in a memory or the like. Here, the term T1 is associated with “{%airports!$1%}” and the term T2 is associated with “{%flight_num!$1%}”.

S404: The parameters of the acquired data D403 are weighted according to the parameter weighting specification information (importance in the parameter information).

Specifically, the importance of the term T1 is acquired with reference to the parameter information shown in FIG. Here, the importance of the term T1 is “5”. Therefore, "!" of the weighting designation information in the parameter is replaced with a character string of numerical length corresponding to this importance (an example of the importance information). The longer the character string, the larger the ratio of the parameter to the entire size of the sentence. Therefore, the accuracy in the similarity determination can be improved.

As a result of the above weighting, the length of the character string of the parameter corresponding to the term T1 is set according to the importance “5”. In the data D404 of FIG. 5D (an example of the first replacement data), the length of the character string is represented by the number of underbars. Specifically, when the importance is 5, the number of underbars is set to 5 which is the same as the importance. Since the importance of “flight_num” is 0, “!” which is the weighting designation information is removed.

S405: Determine the current scene based on the scene information. As described above, the scene information is information indicating the environment including the time and the state of the aircraft. The current scene is switched under a predetermined condition or timing. The details of the scene information and its determination processing will be described later. Here, it is assumed that the current scene is “boarding” (scene 1) (see FIG. 11).

The scene information is stored in association with the example sentence searched for as described above.

S406: An example sentence set corresponding to the current scene is selected for the weighted data D404 including parameters. Here, the case where the example sentence set corresponding to the current scene (scene 1) is the search target is taken as an example. Specifically, the example sentence set (subset 1, subset 2, subset 3...) Corresponding to the scene number “1” in the example sentence information shown in FIG.

Even when it is determined in step S402 that the parameter is not included, the current scene is determined (step S405), and the example sentence set corresponding to the determined scene is selected. The processing by the translation processing unit 114 for data that does not include parameters will be described later.

In addition, a plurality of example sentence sets may be search targets for one scene.

S407: The degree of similarity (distance) between each subset of the selected example sentence set and the data D404 is calculated.

Here, for example, the minimum number of steps (also referred to as an edit distance) required to equalize two character strings to be compared is obtained, and the similarity between the character strings having the minimum number of times is defined as the highest similarity. judge. This procedure includes replacement, insertion, deletion, etc. of each character. For example, a known method for determining the degree of similarity, such as the Levenshtein distance or the Jaro-Winckler distance, can be applied.

The importance is set according to the term, but may be set according to the parameter position information. In FIG. 5E, the parameter weighting designation information “!” in the example sentence is replaced with an underbar according to the importance of the term.

Further, as shown in FIG. 5E, also in the parameter in the retrieved example sentence D408, underbar replacement and deletion of the designation information "!" are performed according to the degree of importance as in the case of the data D404.

S408: As a result of the similarity calculation in step S407, the example sentence subset that is equal to or higher than a predetermined threshold value (score 70) and has the highest similarity is determined. Here, the example sentence D408 shown in FIG. 5E is selected as the sentence having the highest similarity from the selected example sentence subset 1, subset 2, subset 3...

S409: The parallel translation of the example sentence D408 determined in step S408 is acquired. Here, the parallel translation D409 (an example of the second replacement data) illustrated in FIG. 5F is acquired.

S410: It is determined whether the parallel translation D409 includes a parameter. If the parameter is included, the process proceeds to step S411. If the parameter is not included, the process proceeds to step S500 of FIG. 3 and the parallel translation is output as the translation result.

S411: The acquired parameter of the parallel translation D409 is replaced with the parallel translation of the term that is the term before the replacement and is stored. Here, the parallel translation is determined by referring to the parameter information. For example, it is stored in step 403 that the term before replacement corresponding to {%airports!$1%} is the term T1. Then, referring to the parameter information in FIG. 6, it is understood that the parallel translation corresponding to the term T1 is the term t1 “Kansai International Airport”. The parallel translation of "{%flight_num!$1%}", which is a parameter indicating the flight number, is acquired by using a script function that is a calling function, as shown in the parameter information of FIG. Specifically, by using the script function “FlightNo” shown in FIG. 9A, the digit number of the term T2 which is the recognized data is shaped and converted into “9999”.

Further, as shown in FIG. 7B, when the parallel translation D409 includes a plurality of parameters, an English term may be selected according to the positional parameter information and each parameter may be replaced with the term.

As a result, the translation result data D500 is acquired. Then, in step S500 of FIG. 3, the translation result output unit 115 outputs the translation result data D500.

[1-2-3. Translation process when parameters are not included]
Processing by the translation processing unit 114 for data that does not include parameters (processing from “No” in step S402 to steps S405 to S410 in FIG. 4) will be described using another example.

Here, as shown in FIG. 10A, description will be given using an example in which the utterance sentence D100a is input from the microphone 30 as the announcement information.

In step S100 of FIG. 3, the voice acquisition unit 111 acquires the utterance sentence D100a from the microphone 30 as voice data. Then, in step S200 of FIG. 3, the speech recognition unit 112 recognizes the utterance sentence D100a. Then, in step S300 of FIG. 3, the voice-recognized data is converted by the text generation unit 113 into text data D200a as shown in FIG. 10B. Then, the process proceeds to the translation process of step S400.

Similar to the previous example, the translation processing unit 114 acquires the text data D200a (step S401) and determines whether the text data D200a includes a term (step S402). In this case, since the term is not included, the process proceeds to step S405.

Then, similarly to the previous example, the current scene is determined (step S405), and the example sentence set corresponding to the current scene is selected (step S406). Then, as shown in FIG. 10C, the similarity between the recognized text data D200a and each subset included in the selected example sentence set is calculated (step S407). The calculation of the degree of similarity is the same as in the previous example.

As a result of the similarity calculation in step S407, the example sentence subset that is equal to or higher than a predetermined threshold and has the highest similarity is selected. Here, the subset D408a is selected as the subset having the highest similarity (step S408). Then, the parallel translation of the subset D408a selected in step S408 is acquired (step S409). Here, the parallel translation D500a shown in FIG. 10D is acquired. It is determined that the parallel translation does not include parameters (step S410), and the process proceeds to step S500 in FIG. 3 to output the parallel translation as a translation result.

[1-2-4. Scene information and judgment processing]
Hereinafter, scene information and its determination processing will be described with reference to FIG. 11.

<Significance of scene information>
The example sentence information described above is generated by combining sentences of existing aircraft manuals, as will be described later. However, since the number of such example sentence combinations (example sentence sets) is huge, only example sentence sets corresponding to each flight scene are created. When actually executing the similarity calculation (search), a possible example sentence set corresponding to the “current flight scene” at the time of the announcement is set as the search target. This can reduce the number of example sentence sets that are the target of the utterance sentence.

As shown in FIG. 11, the scene information includes, for example, (1) boarding, (2) (before takeoff) runway movement, (3) takeoff, (4) stable flight, (5) landing, ( It is set according to the timeline of actual passenger aircraft operation, such as 6) (after landing) moving the runway and (7) getting off the aircraft. The scene information also includes (8) in-flight service information, in-flight sales information, and the like, although the announcement timing has been determined to some extent, but the order is not predetermined. The scene information further includes unpredictable announcement information, such as (9) emergency contact when a large sway of an airplane occurs.

<Selecting the example sentence set to be searched according to the scene determination>
The selection of the example sentence set as the search target according to the determination of the scene is performed as follows. That is, (a) an example sentence set that may be used is added as a search target, (b) an example sentence set that is not likely to be used is excluded as a search target, or (c) ) Add or remove the example sentence set as a search target under the condition of environment information including the state of the aircraft.

<Example of scene judgment and example sentence subset switching for search>
As described above, the range of the example sentence set that is the search target is extracted according to the determined current scene. In this case, the scene is determined and the example sentence subset to be searched is switched according to the following rules.

(I) Rule 1
For a scene that changes in time series, the translation processing unit 114 determines the scene according to the following rules. For example, as shown in FIG. 11, the scene changes such as during boarding→running on the runway→taking off→stable flight→landing→running on the runway→getting off.

At the start of the work of the crew member, the example sentence set that is the search target is the example sentence set corresponding to the current scene “in flight”. Therefore, the first search is performed for each subset of this example sentence set. For example, when "boarding" is scene 1, as shown in FIG. 11, a set of example sentences corresponding to scene 1 is to be searched. When the search is successful and the corresponding example subset can be acquired from the example sentence set of "boarding", the "current scene" can be determined as "boarding". Therefore, it is stored that the "current scene" is "boarding". At the same time, (a) “running on the runway” that may be spoken next is included in the search target. If it is determined that the "current scene" has changed from "boarding" to "running on the runway" while repeating the utterance and the example sentence search, (b) the example sentence set corresponding to "boarding" is the passenger plane. There is no possibility of being uttered after that due to the flight. Therefore, the example sentence set corresponding to “during boarding” is excluded from the search target.

In this way, an example sentence set with a possibility of the next scene of the "current scene" judged from the rule indicating the scene transition and the utterance content is added, or an example sentence set with no possibility is excluded. By doing, the search target is limited. As a result, it is possible to search the example sentence set more accurately and at high speed.

In addition, when there are a plurality of example sentence sets that may be uttered in one scene, those example sentence sets are also set as search targets. For example, as shown in FIG. 11, when it is determined that the current scene is “in stable flight”, the subsequent example sentence sets “in-flight service guide” and “in-flight sales guide” may be uttered, respectively. There is. Therefore, both are included in the search target.

The conditions or timings for switching the scenes (that is, the conditions or timings for switching the example sentence set to be searched) may be conditions other than the above scene determination, such as when the operating state of the aircraft or the state of the cabin changes. .. For example, information on the condition such as whether the vehicle is gliding can be acquired from the aircraft. In addition, information on the change in state can be obtained by using altitude, tilt/speed of the body, heading of the nose, lighting of a belt wearing sign, and the like. For example, as shown in FIG. 11, the scene information has a correspondence relationship between each scene and the in-machine state. Therefore, the scene may be determined using information such as speed, altitude, presence/absence of door opening/closing, shaking, and GPS. For example, if information indicating that a large vibration has occurred in the aircraft, regardless of the current scene, switch to scene (9) and automatically search for the “emergency contact” example sentence set. You can

Further, the current scene may be switched by a direct input to the server device 10 by a crew member.

(Ii) Rule 2
Scenes that occur simultaneously with other scenes are determined as follows. For example, the example sentence set corresponding to the in-flight service guide for scene (8) is set to be searched only during stable flight. Further, the example sentence set corresponding to the emergency notification of the scene (9) is set so as to be always searched regardless of the scene switching.

As described above, the range in which the example sentence information is referred to may be changed according to the scene.

[1-2-5. Generate original data of example sentence set]
The example sentence set is generated from original data generated from a manual or the like of announcement information of an airline company. Specifically, sentences in the same scene are connected to generate a combined sentence, and an example sentence subset forming an example sentence set (see FIG. 8) is generated. The example sentence subset is generated with the maximum number of combinations as the upper limit. For example, the maximum number of combinations is four.

FIG. 12 shows an example of original data for generating an example sentence subset. FIG. 12 shows example sentences i to v used in the scene 1 (during boarding). The “order” in each example sentence indicates the order in which the example sentences are combined. Based on the example sentences i to v, a plurality of example sentence subsets (for example, see FIG. 8) corresponding to all combinations such as i, i+ii, ii+iii, etc. are generated, and as example sentence sets corresponding to scenes. Remembered.

The example sentence subsets are generated according to the order (ascending order) (combinations such as the order from 4 to 3 are not generated). For example, the combination of the example sentence i+the example sentence ii is generated as an appropriate example sentence subset because the example sentence i has the order “1” and the example sentence ii has the order “2”. However, the combination of the example sentence v+the example sentence iii is not proper because the example sentence v has the order “4” and the example sentence iii has the order “3”, and thus the combination is not generated.

Further, since the example sentence iv and the example sentence v have the same order of “4”, whichever comes first. Furthermore, an example sentence set including the same order as the example sentence iv+the example sentence v may be generated. Alternatively, conversely, when the same order is assigned, an exclusion flag or the like may be set to allow only the generation of the subset including any of the example sentences.

[1-3. Effect, etc.]
In the translation device 110 according to the above embodiment, the translation processing unit 114 replaces a term in a predetermined type included in the text data of Japanese (an example of the first language) with a parameter. Then, the translation processing unit 114 obtains a parallel translation of English (an example of the second language) of the Japanese text data including the same parameter, and converts the parameter included in the parallel translation into a parallel translation of the term, thereby performing translation. Generate result data.

In particular, in the fixed sentence translation, the translation processing unit 114 replaces frequently changing terms (in the example of the present disclosure, airport name, flight number, time, etc.) with parameters, and then selects bilingual translation. Then, the translation processing unit 114 finally replaces the parameter part with a term. As a result, the accuracy of the similarity determination with respect to the fixed phrase can be increased, and the accuracy of translation can be improved.

Further, the parameters are weighted according to the type of the parameter and its importance (for example, the length of the character string is adjusted). As a result, the influence of the term particularly important in the similarity determination can be increased, so that the translation accuracy can be improved.

Further, the search range of the example sentence can be narrowed according to the determined scene. Therefore, the similarity determination can be efficiently performed, and the translation process can be speeded up.

(Other embodiments)
As described above, the first embodiment has been described as an example of the technique disclosed in the present application. However, the technique of the present disclosure is not limited to this, and is also applicable to the embodiment in which changes, replacements, additions, omissions, etc. are appropriately made. Further, it is also possible to combine the constituent elements and functions described in the first embodiment to form a new embodiment.

Therefore, other embodiments will be exemplified below.

[1]
In the above embodiment, the translation device 110 transmits the translation result data in English as the second language to each display device 20, but the translation device 110 may deliver the translation result data in the language assigned to each display device 20. May be.

FIG. 13 is a functional block diagram of the translation device 2110 according to this example. The translation device 2110 is different from the translation device 110 in that the recognized text data in the first language is translated into a plurality of second languages, and the translation result data in the plurality of second languages is transmitted to the corresponding display devices 20. . The translation device 2110 is different from the translation device 110 in that it holds a plurality of parallel translation information of the second language in the example sentence set of the database DB and the parameter information. The translation result output unit 2115 determines which of the second languages the translation result data is to be transmitted to each display device 20, by referring to the language setting information 217a stored in advance in the storage device 17. Then, the translation result output unit 2115 transmits the translation result data based on the determination result. The language setting information 217a is set by the user of each display device 20 through the operation of the input device 28 (see FIG. 1). The language setting information 217a is transmitted to the translation device 2110 and stored in the storage device 17.

The language setting information 217a may be transmitted and stored in advance in the translation device 2110 according to the seat reservation information, nationality, etc. of the user.

Other configurations and functions are similar to those of the first embodiment.

[2]
The translation result data transmitted to each display device 20 may be data in a plurality of languages. For example, when the user's desired language is unknown, a plurality of translation result data may be displayed according to the language of the aircraft registration, the language of the departure country, the language of the arrival country, and the like.

[3]
In the above embodiment, the translation result data is displayed on the display 29 of the display device 20, but the present invention is not limited to this. The translation result data may be converted into voice and output from an earphone (not shown) of the display device 20. Alternatively, the translation result data may be converted into voice in the server device 10 and broadcast via the speaker 40 in the aircraft.

The translation result data may be displayed on a large-screen display connected to the server device 10 so that a plurality of users can view it.

[4]
The translation device 110 is not limited to being used in an aircraft. The translation device 110 may be used in locations such as airport guidance, other transportation facilities, restaurants, hospitals, accommodation facilities, etc., where translation using standard phrases is possible.

The translation device 110 may be provided in a computer terminal. In this case, the translation processing may be performed on the data input by voice or text, and the translation result may be output via the speaker or the display. Here, when the data is input by voice, the voice recognition process is performed before the translation process.

[5]
In the above embodiment, the threshold value for determining the similarity is 70, but the threshold value is not limited to this. The threshold may be greater than or less than 70.

Further, the threshold may be set according to the scene to be judged. For example, the threshold value of an emergency scene may be set higher than the threshold values of other scenes to obtain more accurate translation result data.

[6]
In the above-described embodiment, the translation device 110 changes the search range of the example sentence information according to the determined scene, but in addition or instead of this, the search range of the example sentence information is changed according to the determined scene. The priority order may be changed. For example, it is assumed that the example sentence subset corresponding to “emergency contact” in (9) of FIG. 11 is a search target in any scene because of its importance. However, the example sentence subset corresponding to “in flight” in (1) has a lower occurrence probability than in “stable flight” in (4). From this, it is conceivable to lower the priority of the retrieval of the example sentence subset corresponding to (1) "in flight". In this case, if a score with a very high degree of similarity is obtained during the search, the search may be stopped and the translation process may be performed using the example sentence set. As described above, the order of referring to the example sentence information may be changed according to the scene.

[7]
The script function may be used depending on the type of the second language. For example, changing the unit of temperature shown in FIG. 9B does not need to be converted even if the language is different in a culture using the same unit. In this case, information including the correspondence between the type of the second language and the unit information is stored in the storage device 17 in advance. Then, in step S411 of FIG. 4, the translation processing unit 114 may determine the type of the second language, determine whether or not to use the script function, and then perform the replacement into the bilingual translation.

[8]
The execution order of the processes (the processes of FIGS. 3 and 4 etc.) in the above embodiment is not necessarily limited to the description of the above embodiment, and the execution order may be changed without departing from the gist of the invention. It is possible.

For example, in FIG. 4, the determination of the current scene (step S405) may be performed before the parameter determination (step S402) or may be performed simultaneously with the parameter determination.

[9]
Since the example sentence subset is obtained depending on the scene, the types of parameters included in the example can be enumerated by checking the example sentence. Therefore, it is possible to limit the types of parameters (airports, etc.) that are searched for using the regular expression for the voice recognition result (utterance) sentence. As a result, it is possible to reduce the problem that a part that should not be subjected to parameter replacement is also replaced. This will be described with reference to FIGS. 14A to 14C.

For example, let us consider a case where there is an example sentence D800 as an example sentence during (2) runway movement or (3) takeoff, and there is a regular expression D801 as a regular expression of a term relating to temperature in the parameter information. When the example sentence D800 is spoken, there is a possibility that the expression of the term T3 matches the temperature-related parameter and the example sentence D800 is replaced with the data D802.

However, actually, there is no example sentence including a term regarding temperature during (2) runway movement or (3) takeoff. Therefore, during (2) movement of the runway, or (3) during takeoff, the term relating to temperature is not detected during the term detection in step S402. As a result, more accurate translation can be realized.

[10]
The specific parameter information may be replaced with the information held by the aircraft in advance. This will be described with reference to FIGS. 15A and 15B.

For example, when the information that the system 1 is destined for Kansai Airport is acquired in advance, all the parameters regarding the destination in the example sentence are replaced with the term T1 in advance. In addition, the term T1 is excluded from detection when the term is detected in step S402. By performing this replacement, the text data after the parameter replacement is performed on the utterance sentence becomes the data D900. The example sentence having the highest similarity is the example sentence D901. As a result, when an airport name other than the destination such as “Haneda Airport” is spoken, the score of the similarity can be lowered. Therefore, the accuracy of translation is improved.

The system 1 holds information indicating which parameter in the example sentence is related to the destination and refers to it. For example, “{%airports_____$1%}” in the example sentence D408 (see FIG. 5E) is related to the destination, but “{%airports!$1%}” in FIG. 7B is not related to the destination. Is managed by the table. By referring to this table, "{%airports!$1%}" in the example sentence shown in FIG. 7B is not replaced with the term T1.

[11]
In the above embodiment, voice data is input, but the present invention is not limited to this. For example, text data may be acquired from an input device such as a touch panel or a keyboard via an input I/F, or text data may be acquired from an external device via a communication I/F. Thereby, the translation processing by the translation processing unit 114 may be executed.

[12]
The translation device 110 according to the present disclosure is not limited to being implemented by the above embodiment. A translation method executed by the translation device 110, a computer program executed by the translation device 110, and a computer-readable recording medium recording the program are included in the scope of the present disclosure. Here, examples of the computer-readable recording medium include a flexible disk, a hard disk, a CD-ROM, an MO, a DVD, a DVD-ROM, a DVD-RAM, a BD (Blu-ray (registered trademark) Disc), and a semiconductor memory. Can be mentioned.

The computer program is not limited to the one recorded in the recording medium, but may be transmitted via an electric communication line, a wireless or wired communication line, a network typified by the Internet, or the like.

[13]
In the above embodiment, it is not necessary to determine or replace the parameters. That is, the translation processing unit 114 performs the scene determination in step S405 after acquiring the text data in step S401 in FIG. 4, and selects the example sentence subset corresponding to the scene determined in step S406.

In this case, it is possible to narrow the search range of the example sentence information by selecting a subset of the example sentence according to the determined scene. Therefore, the translation process can be made efficient and the translation process can be speeded up.

Based on this, the invention according to another aspect can be expressed as follows.

"An input unit for acquiring input data in the first language,
A control unit that recognizes the input data, generates text data of the first language based on the recognized input data, and generates text data of a second language that is a parallel translation of the text data. Department is
When the similarity with the text data of the first language is equal to or more than a threshold value by referring to the reference text data group including the example sentence in the first language and the parallel translation of the example sentence in the second language. Acquiring example sentences, and generating text data in the second language based on parallel translation of the example sentences in the second language,
Judge scenes that change according to at least one of time and environment,
A translation device that changes at least one of a range and an order in which the reference text data group is referred to according to the determined scene. "

The present disclosure can be used as a translation device.

1: System 10: Server device 11: CPU
12: ROM
13: RAM
14: Input I/F
15: Output I/F
16: Communication I/F
17: storage device 20: display device 21: CPU
22: ROM
23: RAM
26: Communication I/F
27: Memory 28: Input device 29: Display 30: Microphone 40: Speaker 110: Translation device 111: Speech acquisition unit 112: Speech recognition unit 113: Text generation unit 114: Translation processing unit 115: Translation result output unit 2110: Translation device 2115: Translation result output unit

Claims (10)

An input unit for acquiring the first text data of the first language,
A control unit for generating second text data in a second language which is a parallel translation of the first text data,
The control unit is
Based on parameter information including a first term of the first language in a predetermined type, an importance of the first term, and a second term of the second language that is a parallel translation of the first term, the first term 1st substitution data is produced|generated by substituting the said 1st term contained in 1 text data by the parameter containing the importance information whose character string becomes long, so that the importance of the said 1st term is high ,
By selecting a reference text data group including an example sentence and a bilingual translation of the example sentence, the example sentence whose similarity to the first replacement data is a threshold value or more is selected,
Acquiring a parallel translation of the example sentence as the second replacement data of the second language corresponding to the first replacement data,
Based on said parameter information, by replacing the parameter included in the second replacement data in the second term, and generates the second text data,
Translation device. The parameter is a parameter position indicating a position of the first term different from a position of the other first term in the first text data when the first text data includes another first term of the predetermined type. Including information,
The control unit selects the second term that replaces the parameter in the second replacement data according to the parameter position information,
The translation device according to claim 1. A storage device for holding the parameter information and the reference text data group,
The translation device according to claim 1. The control unit determines a scene that changes according to at least one of time and surrounding conditions, and changes at least one of a range and an order in which the reference text data group is referred to according to the determined scene.
The translation device according to claim 1 . The control unit changes the threshold value according to the determined scene,
The translation device according to claim 4 . The second language includes a plurality of second languages,
The control unit generates the second text data for each of the plurality of second languages,
Translation apparatus according to any one of claims 1 to 5. Has a connection,
The control unit transmits the second text data to the information terminal set in the second language via the connection unit,
Translation device according to any of claims 1 6. The control unit is
Determining whether the first text data includes a first term in the predetermined type,
When it is determined that the first term is not included in the first text data, a parallel translation of the first text data is acquired by referring to the reference text data group ,
Translation apparatus according to any one of claims 1 to 7. The input unit acquires the first text data by recognizing the voice data in the first language,
Translation device according to any one of claims 1 to 8. Acquire the first text data of the first language,
The first term of the first language in a predetermined type, the importance of the first term, and the second term of the second language that is a parallel translation of the first term, based on the parameter information, 1st substitution data is produced|generated by substituting the said 1st term contained in 1 text data by the parameter containing the importance information whose character string becomes long, so that the importance of the said 1st term is high ,
By selecting a reference text data group including an example sentence and a bilingual translation of the example sentence, the example sentence whose similarity to the first replacement data is a threshold value or more is selected,
The parallel translation of the example sentence is acquired from the memory as the second replacement data of the second language corresponding to the first replacement data,
Based on said parameter information, by replacing the parameter included in the second replacement data in the second term, to produce a second text data of the second language,
How to translate.

Images