Translation text calculation device

Abstract

The present invention addresses the problem of computing a translated sentence in which a portion of an already-translated sentence has been incorporated. A translated sentence computation device 1 that uses a translation machine 2 configured from a recurrent neural network using an encoder-decoder model in which an encoder 20 receives an input of a sentence in a first language and a decoder 21 sequentially outputs word candidates of a sentence in a second language corresponding to the sentence in the first language, said translated sentence computation device being provided with: a first input unit 12 for inputting to the encoder 20 a subject sentence, which is a sentence in the first language; a second input unit 13 for sequentially inputting to the decoder 21 the words of a partial sentence, which is part of a created sentence obtained by translating the subject sentence into the second language; and a computation unit 14 that computes a translated sentence that is based on the word candidates output by the decoder 21 on the basis of the inputs by the first input unit 12 and the second input unit 13. The partial sentence may be a portion of the beginning of the created sentence, and the translated sentence may be a sentence continuing from the partial sentence.

Description

Technical Field

[0001] One aspect of the present disclosure relates to a translated text calculation device that calculates a translated text obtained by translating a text in a first language into a second language.

Background Art

[0002] Conventionally, Neural Machine Translation, which is machine translation using a neural network of an encoder-decoder model composed of an encoder and a decoder, is known. In the neural network of the encoder-decoder model, the encoder inputs a text in a first language (for example, Japanese), and the decoder outputs, as a translation result, a text in a second language (for example, English) corresponding to the text in the first language.

[0003] For example, Patent Document 1 below discloses an automatic translation device including an encoder and a decoder configured by a neural network.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The above automatic translation device outputs, as a translation result, a text in a second language "How do I get to Gangnam?" for a text in a first language "How do I get to Gangnam Station?" input by a user. However, for example, a usage method in which the user translates part of the text by himself / herself, such as "Would you please", and then outputs the subsequent translation to the above automatic translation device is not assumed.

[0006] Therefore, it is desired to calculate a translated text taking into account a part of the translated text.

Means for Solving the Problem

[0007] A translated text calculation device according to one aspect of the present disclosure is a translated text calculation device that uses a translator composed of a recurrent neural network of an encoder-decoder model in which an encoder inputs a text in a first language and a decoder sequentially outputs word candidates for a text in a second language corresponding to the text in the first language, and includes a first input unit that inputs a target text, which is a text in the first language, to the encoder, a second input unit that sequentially inputs words of a partial text, which is a part of a created text that is a text obtained by translating the target text into the second language, to the decoder, and a calculation unit that calculates a translated text, which is a text based on the word candidates output by the decoder based on the inputs by the first input unit and the second input unit.

[0008] In such an aspect, for example, a translated text in the second language that takes into account a partial text in the second language corresponding to the content of the target text in the first language is calculated. That is, it is possible to calculate a translated text that takes into account a part of the translated text.

Advantages of the Invention

[0009] According to one aspect of the present disclosure, it is possible to calculate a translated text that takes into account a part of the translated text.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Mode for Carrying Out the Invention

[0011] Hereinafter, embodiments in the present disclosure will be described in detail with reference to the drawings. In the description of the drawings, the same reference numerals are assigned to the same elements, and redundant descriptions are omitted. Also, the embodiments in the present disclosure in the following description are specific examples of the present invention, and are not limited to these embodiments unless otherwise specified to limit the present invention.

[0012] FIG. 1 is a diagram showing an example of the functional configuration of a translation text calculation system 3 including a translation text calculation device 1 according to the embodiment. As shown in FIG. 1, the translation text calculation system 3 includes a translation text calculation device 1 and a translator 2. The translation text calculation device 1 and the translator 2 are communicatively connected to each other by a network and can transmit and receive information to and from each other. The translation text calculation device 1 and the translator 2 are not necessarily independent configurations, and the translator 2 may be included inside the translation text calculation device 1. The translation text calculation device 1 and the translator 2 in the translation text calculation system 3 may have any system configuration as long as the translation text calculation device 1 uses the translator 2.

[0013] The translation text calculation device 1 is a computer device that calculates a translated text in a second language that takes into account a partial text, which is a part of a created text that is a translation of a target text in a first language, based on the target text in the first language and the partial text. The first language is, for example, Japanese, but can be any other language. The second language is a language different from the first language, for example, English, but can be any other language. The first language and the second language can be dialects of different regions (for example, standard Japanese and Kansai dialect in Japan). The language is not limited to natural languages and can also be artificial languages or formal languages (such as computer programming languages). The created text is intended to be created by a person such as a user of the translation text calculation device 1, but can also be a text created by someone other than a person. The translation text calculation device 1 uses a translator 2. Details of the translation text calculation device 1 will be described later.

[0014] The translator 2 is a computer device that translates a text in a first language into a text in a second language. The translator 2 may evaluate the translation quality of each word and the entire text of the translated text in the second language. As shown in FIG. 1, the translator 2 includes an encoder 20 and a decoder 21. The translator 2 is composed of a recurrent neural network (Recurrent Neural Network, RNN) of an encoder-decoder model (also known as an encoder-decoder translation model, Sequence to Sequence Model) in which the encoder 20 inputs a text in the first language and the decoder 21 sequentially outputs word candidates for the text in the second language corresponding to the text in the first language. The neural network is, for example, a recurrent neural network called LSTM (Long Short Term Memory). The translator 2 performs neural machine translation. The decoder 21 may sequentially output the likelihood for each word candidate together with the word candidates for the text in the second language.

[0015] The encoder 20 inputs a sentence in the first language and outputs a vector of the intermediate layer (hidden layer). More specifically, the encoder 20 divides a sentence in the first language into words by morphological analysis or the like, converts the word ID (Word ID) corresponding to each word into a word vector (vector of the input layer), and then sequentially inputs them (in order from the first word to the last word of the sentence), and sequentially outputs vectors of the intermediate layer based on the input content up to that point (performs neural network calculations). The encoder 20 indicates the end of the sentence with " <eos>When the "」" is input, the encoder 20 outputs (passes) a vector of the intermediate layer based on the input content up to that point to the decoder 21. Conceptually, it can be said that the encoder 20 semantically analyzes the sentence in the first language and extracts the semantic representation.

[0016] The decoder 21 inputs the vector of the intermediate layer output from the encoder 20, and sequentially calculates and outputs a vector of the output layer based on the vector of the intermediate layer, or based on the vector of the intermediate layer and the words in the second language input to the decoder 21. The vector of the output layer is information indicating, for example, a list of word candidates in the second language and the likelihood of the word candidates. An example of such a list is '(word candidate "It" and its likelihood "0.66", word candidate "Tomorrow" and its likelihood "0.33",...)' and so on.

[0017] The processing of the decoder 21 will be described more specifically. First, when the vector of the intermediate layer output from the encoder 20 is input, a vector of the output layer corresponding to the first word in the sentence in the second language to be finally output is output based on the input vector of the intermediate layer. Subsequently, the decoder 21 extracts the word with the highest likelihood from among the word candidates indicated by the vector of the output layer of the Nth word (N is an integer of 1 or more), inputs the extracted word to its own decoder (the decoder 21), and outputs a vector of the (N + 1)th output layer based on the input word and the vector of the intermediate layer used when outputting the vector of the Nth output layer. This process is repeated until the last word of the sentence in the second language. Conceptually, it can be said that the decoder 21 generates a sentence (in a second language different from the first language) from the semantic representation extracted by the encoder 20.

[0018] In this embodiment, as a specific example of the translation machine 2, it is assumed to be a creation text evaluation device disclosed in the following reference document, but it is not limited thereto. The translation machine 2 refers to and quotes the disclosure content of the following reference document. Reference document: International Publication No. 2019 / 225154 (International Application No.: PCT / JP2019 / 013792)

[0019] Next, with reference to FIG. 2, a usage example of the translated text calculation device 1 by the user will be described. FIG. 2 is a diagram showing a display example by the translated text calculation device 1. The display example shown in FIG. 2 shows a scene where the user translates a target text (problem text), which is a text in the first language (Japanese), "This is the shortest way to go to the post office.", displayed on the screen. Note that the initially assumed model answer is "This is the shortest way to go to the post office."

[0020] The usage procedure will be specifically described below. Except for the processing by the user, the main body of each process is the translation text calculation device 1. First, in the initial state, the target text A, an empty text box B, and a scoring button C are displayed on the screen. Next, the user translates the target text A (translates the target text A in the first language (Japanese) into the second language (English) using their own mind), and inputs the created text D, which is the translated text, into the text box B. As shown in Figure 2, the created text D created by the user is "This is the best road to go to the office." Next, using existing technologies such as the created text evaluation device of the above-mentioned reference document, the translation quality of the created text D is evaluated (scored). As a result of the evaluation, an evaluated text E reflecting the evaluation result and a first evaluation value F indicating the evaluation value (score) for the entire text of the evaluated text E are displayed on the screen. In the evaluated text E, words whose evaluation of each word does not meet a predetermined standard (the evaluation value is lower than the predetermined standard value) (words that should be corrected) are displayed in a different form from other words, such as in bold or with color, to indicate to the user that the predetermined standard is not met. In addition, as other display variations, the evaluation value is displayed near the word that does not meet the predetermined standard, the evaluation value of the word is displayed near each word of the evaluated text E, and it is displayed with a boldness or color degree (brightness, chroma, hue, etc.) corresponding to the evaluation value. In the display example shown in Figure 2, for the words "best", "road", and "office" in the evaluated text E, since the evaluation does not meet the predetermined standard, they are displayed in bold. Note that the evaluated text E may be displayed in a form that overwrites the created text D in the text box B. Also, in this embodiment, since the created text D and the evaluated text E are identical in content, the created text D and the evaluated text E may be regarded as the same. Also, the first evaluation value F may not be displayed.

[0021] Here, when the translation text calculation device 1 designates (points out) a word whose evaluation of the word among the words of the evaluation text E does not meet a predetermined standard (or one word included in the creation text D) as the target word, it calculates and displays a translation text that is the text obtained by correctly proofreading the text after the target word. Further, for the proofreading text (the result of overall proofreading of the text), which is the combined text of a partial text that is the text before the target word in the evaluation text E or the creation text D and the translation text, if the evaluation (reliability) of the translation quality as the entire text does not meet a predetermined standard, it repeatedly performs proofreading while going back to the word before the target word, and when the reliability meets a predetermined standard (when the reliability exceeds a predetermined reference value), it finally displays the proofreading text at that time as the translated text.

[0022] Returning to FIG. 2, when the user designates the word "office" in the evaluation text E or the created text D, the translated text G, which is "post office.", is calculated and displayed. That is, the translated text calculation device 1 makes a proposal that it is better to change "office" to "post office". Note that an evaluation value regarding the translated text G may be displayed together with the translated text G. Here, if only "office" is changed to "post office", the corrected text is "This is the best road to go to the post office.", and it is assumed that the evaluation (reliability) of the translation quality as a whole of the text does not yet meet the predetermined standard (there are still areas for improvement as a whole). In that case, the correction is attempted by going back to the previous word. In the display example shown in FIG. 2, it is assumed that when going back to the word "road" (the next word that does not meet the predetermined standard) and changing the part after "road" to "shortcut to the post office.", the reliability meets the predetermined standard. Therefore, "This is the best shortcut to the post office." is displayed as the corrected text H, which is the correction result for the whole text. At that time, a second evaluation value I indicating the evaluation value (score) for the whole text of the corrected text H is displayed. Note that the second evaluation value I may not be displayed. Although the corrected text H is different from the above exemplary answer "This is the shortest way to go to the post office.", it is possible to make a correction that takes advantage of the user's created text D (the user's answer) that uses "best" instead of "shortest".

[0023] The correspondence between the terms used in the embodiment and the specific examples in the display example shown in FIG. 2 is shown below. Target text: "This is the best shortcut to go to the post office." Created text: "This is the best road to go to the office." Target words: "office", "road", etc. A passage: "This is the best road to go to the" (when the target word is "office"), "This is the best" (when the target word is "road"), etc. Translated passage: "post office." (when the target word is "office"), "shortcut to the post office." (when the target word is "road"), etc. Revised passage: "This is the best shortcut to the post office."

[0024] Regarding the display example shown in Figure 2, different variations of the target passage, etc. will be briefly explained. Assume that the target passage A is "This novel is read by many young people." and the created passage D is "Many young people reads this book." Suppose that as a result of the evaluation, it is determined that the evaluations of the words "reads" and "book" do not meet the predetermined criteria. When the word "book" is specified as the target word, "novel." is displayed as the translated passage G. However, since "reads" is grammatically incorrect in the first place, the evaluation of the current revised passage does not meet the predetermined criteria. Therefore, an attempt is made to revise by going back to the previous word "reads", and when the translated passage "read this novel." is calculated, the evaluation of the revised passage H "Many young people read this novel." meets the predetermined criteria, so the revised passage H is displayed. Although the model answer starts with "This novel is", it is possible to display a passage that starts with "Many young people" created by the user and is correct (the evaluation meets the predetermined criteria).

[0025] As described above, when the user designates (for example, clicks on the screen) an arbitrary word (for example, a misspelled word), the translation text calculation device 1 composes the subsequent text as a proofreading function. Further, if the text before the designated word is incorrect, for example, it traces back to the previous incorrect word and performs proofreading. That is, the translation text calculation device 1 performs a translation that respects the user's answer as much as possible. In other words, the translation text calculation device 1 composes and displays the text after the word to be corrected using AI (Artificial Intelligence) while respecting the user's answer to the maximum extent (utilizing the user's answer as much as possible).

[0026] Subsequently, the details of the translation text calculation device 1 will be described.

[0027] As shown in FIG. 1, the translation text calculation device 1 includes a storage unit 10, an input / output unit 11 (reception unit, display unit), a first input unit 12 (first input unit), a second input unit 13 (second input unit), and a calculation unit 14.

[0028] Each functional block of the translation text calculation device 1 is assumed to function within the translation text calculation device 1, but it is not limited thereto. For example, a part of the functional block of the translation text calculation device 1 may be a computer device different from the translation text calculation device 1, and may function while appropriately transmitting and receiving information with the translation text calculation device 1 within a computer device network-connected to the translation text calculation device 1. Also, some functional blocks of the translation text calculation device 1 may be omitted, a plurality of functional blocks may be integrated into one functional block, or one functional block may be decomposed into a plurality of functional blocks.

[0029] Hereinafter, each function of the translation text calculation device 1 shown in FIG. 2 will be described.

[0030] The storage unit 10 stores any information used in calculations in the translation text calculation device 1 and the results of calculations in the translation text calculation device 1. The information stored by the storage unit 10 is appropriately referred to by each function of the translation text calculation device 1. The storage unit 10 may store the target text in advance.

[0031] The input / output unit 11 (display unit) outputs (displays) the target text. For example, when the input / output unit 11 acquires an instruction to display the target text on the screen from the user or the translation text calculation device 1, it acquires the target text stored by the storage unit 10 and displays the acquired target text on the screen of the translation text calculation device 1 which is the output device 1006 described later (hereinafter simply referred to as "screen"). The output of the target text by the input / output unit 11 corresponds to the display of the target text A in the display example of FIG. 2. Note that "output" in the present embodiment is not limited to display, and includes, for example, transmission to another computer via a network, output by voice, etc.

[0032] The input / output unit 11 (display unit) outputs (displays) the created text. For example, the input / output unit 11 displays on the screen the created text input by the user using an input device 1005 such as a keyboard or a microphone described later. The output of the created text by the input / output unit 11 corresponds to the display of the created text D or the evaluation text E in the display example of FIG. 2.

[0033] The input / output unit 11 (display unit) inputs (acquires) the created text. For example, when the input / output unit 11 acquires an instruction to input the created text from the user, it inputs the created text. The input / output unit 11 outputs the input created text to the second input unit 13 and the calculation unit 14. The input of the created text by the input / output unit 11 corresponds to the input of the created text D when the scoring button C is pressed in the display example of FIG. 2.

[0034] The input / output unit 11 (display unit) outputs (displays) the evaluation text. For example, the input / output unit 11 evaluates the translation quality of the created text input by the input / output unit 11 using the translation machine 2 and displays the evaluation text reflecting the evaluation result on the screen. When outputting the evaluation text, the input / output unit 11 may also output the value of the evaluation of the translation quality for the entire text of the evaluation text. The output of the evaluation text by the input / output unit 11 corresponds to the output of the evaluation text E and the first evaluation value F in the display example of FIG. 2.

[0035] The input / output unit 11 (reception unit) receives a designation of one word included in the generated sentence as the target word. The input / output unit 11 (reception unit) may receive a designation of a word designated by the user from among the generated sentences displayed by the input / output unit 11 (display unit) as the target word. The input / output unit 11 outputs the designated target word to the second input unit 13 and the calculation unit 14. The reception of the designation of the target word by the input / output unit 11 corresponds to the reception of the designation (such as a click) of the word "office" in the generated sentence D or the evaluation sentence E in the display example of FIG. 2.

[0036] The input / output unit 11 (display unit) outputs (displays) the translated sentence. More specifically, the input / output unit 11 (display unit) displays the translated sentence calculated by the calculation unit 14. For example, the input / output unit 11 displays the translated sentence calculated by the calculation unit 14 on the screen. When outputting the translated sentence, the input / output unit 11 may also output a value of the evaluation of the translation quality regarding the translated sentence. The output of the translated sentence by the input / output unit 11 corresponds to the output of the translated sentence G in the display example of FIG. 2.

[0037] The input / output unit 11 (display unit) outputs (displays) the proofreading sentence. For example, the input / output unit 11 displays on the screen a proofreading sentence that is a combined sentence of a partial sentence that is a sentence before the target word for which the designation has been received by the input / output unit 11 from among the generated sentences input by the input / output unit 11 and the translated sentence calculated by the calculation unit 14. When outputting the proofreading sentence, the input / output unit 11 may also output a value of the evaluation of the translation quality for the entire proofreading sentence. The output of the proofreading sentence by the input / output unit 11 corresponds to the output of the proofreading sentence H and the second evaluation value I in the display example of FIG. 2.

[0038] The first input unit 12 inputs the target sentence, which is a sentence in the first language, to the encoder 20 of the translation machine 2. More specifically, the first input unit 12 sequentially inputs the words constituting the target sentence to the encoder 20 in the order in which they appear in the sentence. A specific example of the process of the first input unit 12 will be described later.

[0039] The second input unit 13 sequentially inputs words of a partial sentence, which is a part of a created sentence (the created sentence input by the input / output unit 11) that is a translation of the target sentence into the second language, to the decoder 21. The created sentence may be a sentence created by the user translating the target sentence (output by the input / output unit 11) into the second language. The partial sentence may be a partial sentence from the beginning of the created sentence. The second input unit 13 sequentially inputs words of the partial sentence (in the order in which the words constituting the partial sentence appear in the sentence) to the decoder 21, and after the input is completed, may sequentially input word candidates sequentially output by the decoder 21 to the decoder 21. The second input unit 13 may sequentially input words of a partial sentence, which is a sentence before the target word (designated by the input / output unit 11) in the created sentence, to the decoder 21. A specific example of the processing of the second input unit 13 will be described later.

[0040] Note that the trigger for the input by the first input unit 12 and the second input unit 13 may be any timing, such as the timing when the input / output unit 11 receives the designation of the target word, the timing instructed by the user or the administrator of the translation sentence calculation device 1, or periodically (for example, once per minute).

[0041] The calculation unit 14 calculates a translated sentence, which is a sentence based on the word candidates output by the decoder 21 based on the inputs from the first input unit 12 and the second input unit 13. The translated sentence may be a sentence following a partial sentence. The calculation unit 14 repeats the following process until the sentence evaluation, which is an evaluation of the sentence based on the partial sentence and the calculated translated sentence, satisfies a predetermined criterion: setting a word before the target word in the generated sentence as a new target word and setting the sentence before the new target word in the generated sentence as a new partial sentence, and performing inputs by the first input unit 12 and the second input unit 13 to calculate a new translated sentence. The calculation unit 14 may set, as a new target word, a word before the target word in the generated sentence and whose evaluation in the generated sentence does not satisfy a predetermined criterion. The sentence evaluation may be based on the likelihood for the word candidate that is the same as the word in the partial sentence among the word candidates and likelihoods output by the decoder 21 based on the inputs from the first input unit 12 and the second input unit 13. The calculation unit 14 may output the calculated translated sentence to the input / output unit 11.

[0042] Subsequently, with reference to FIG. 3, a specific example of the processing of the first input unit 12, the second input unit 13, and the calculation unit 14 will be described. FIG. 3 is a diagram showing an example of the use of the translator 2 by the translated sentence calculation device 1. An upper part of FIG. 3 shows a schematic diagram of the encoder 20 and the decoder 21 of the translator 2. In the translator 2, an LSTM is used, and it is divided into an encoder 20 side and a decoder 21 side. The first input unit 12 divides the target sentence (input sentence) into words and sequentially inputs the word IDs corresponding to the respective words to the encoder 20. The encoder 20 converts the input word IDs into word vectors, performs calculations of a neural network, and " <eos>When "(end of article)" is input, the encoder 20 passes the intermediate layer vector to the decoder 21. The decoder 21 calculates the output layer from the passed intermediate layer vector and calculates the likelihood of the output word using the Softmax function.

[0043] The decoder 21 compares the likelihood of the output word from the output layer with the likelihood of the word in the partial sentence input by the second input unit 13 for the segmented words of the partial sentence, and calculates the confidence of the word in the partial sentence. In the usage example shown in FIG. 3, the decoder 21 should output "It is fine tomorrow.", and for the first word, "It" has the highest likelihood of "0.66". The partial sentence, which is the sentence entered by the user up to that point, is "Tomorrow will", and although it is desired to start with "Tomorrow", the likelihood of "Tomorrow" as the first word is "0.33". Therefore, the confidence of the word "Tomorrow" is "0.5", which is obtained by dividing the likelihood of "Tomorrow" by the likelihood of "It".

[0044] Next, the decoder 21 calculates the word following "Tomorrow" and calculates the likelihood of the output word in the same way as the first word. As the output of the decoder 21, "is" has the highest likelihood of "0.5". The likelihood of "will" in the partial sentence of the user is "0.4". Therefore, the confidence of "will" is "0.8", which is obtained by dividing the likelihood of "will" by the likelihood of "is". Furthermore, since there is no input as the partial sentence created by the user from the word following "will", the second input unit 13 sets the word following "will" to the word with the highest likelihood among the output words of the decoder 21 and inputs it to the decoder 21 to complete the sentence (partial sentence). Since the word with the highest likelihood as the word following "will" is "be", the second input unit 13 adds "be" to the translated sentence. The confidence of "be" is for the word output by the decoder 21, so it becomes "1". Similarly, the word following "be" is "fine", the word following "fine" is " <eos>」, becoming " <eos>When "]" appears, end the text (translated text) there. The reliability of the entire text (proofread text) shall be the average value of the reliabilities of each word. The calculation unit 14 calculates a proofread text that is a translated text output by the decoder 21, or a combination of a partial text and a translated text output by the decoder 21. Note that the evaluation in this embodiment is not limited to being based on the likelihood output by the decoder 21 of the translation machine 2, and may be an evaluation of translation quality or the like by other existing technologies. For the process of repeatedly calculating a new translated text by the calculation unit 14 until the text evaluation satisfies a predetermined criterion, refer to the description regarding FIG. 2 above and the description regarding FIG. 4 below.

[0045] Subsequently, an example of the process (translated text calculation method) executed by the translated text calculation device 1 (or the translated text calculation system 3) will be described with reference to FIG. 4. FIG. 4 is a flowchart showing an example of the process executed by the translated text calculation device 1 (or the translated text calculation system 3).

[0046] First, the input / output unit 11 displays the target text on the screen (step S1). Next, the input / output unit 11 acquires the created text translated by the user himself / herself from the target screen displayed in S1 (step S2). Next, the input / output unit 11 evaluates the created text acquired in S2 using the translation machine 2 or the like, and displays the evaluation result on the screen (step S3). Next, the input / output unit 11 accepts the designation of the target word in the created text by the user based on the evaluation result displayed in S3 (step S4).

[0047] Next, the first input unit 12 inputs the target sentence to the encoder 20 of the translator 2, and the second input unit 13 inputs to the encoder 20 of the translator 2 a partial sentence based on the generated sentence obtained in S2 and the target word that received the designation in S4 (step S5). Next, the calculation unit 14 calculates a translated sentence based on the word candidates output by the decoder 21 of the translator 2 based on the input in S5 (step S6). Next, the calculation unit 14 evaluates a corrected sentence, which is a combination of the partial sentence in S5 and the translated sentence calculated in S6, based on the likelihood and the like output by the decoder 21 in S6 (step S7). Next, the calculation unit 14 determines whether the evaluation in S7 satisfies a predetermined criterion (step S8).

[0048] When it is determined in S8 that the predetermined criterion is not satisfied (S8: NO), the calculation unit 14 sets a word before the target word as a new target word, and sets, as a new partial sentence, the sentence before the new target word among the generated sentences obtained in S2 (step S9), and returns to the process of S5 (repeating S5 to S9 until S8 becomes YES). On the other hand, when it is determined in S8 that the predetermined criterion is satisfied (S8: YES), the input / output unit 11 displays the corrected sentence in S7 (of the final loop) on the screen (step S10). Note that S1 to S4 may be omitted. In that case, the target word (at the beginning of the loop starting from S5) may be any word in the generated sentence (the last word of the generated sentence, a word whose evaluation does not satisfy the predetermined criterion among the words of the generated sentence, the word closest to the end of the generated sentence among the words whose evaluation does not satisfy the predetermined criterion, etc.). Also, S7 to S9 may be omitted and S10 may be executed following S6. Also, when calculating the translated sentence in S6, the input / output unit 11 may display the translated sentence on the screen (as an intermediate result). Also, when evaluating in S7, the input / output unit 11 may display the result of the evaluation on the screen. Also, when setting a new target word in S9, the input / output unit 11 may display the new target word on the screen, or may display the new target word in the generated sentence already displayed in S3 or the like in a different form (for example, in bold). Also, in S10, instead of or together with the corrected sentence, the translated sentence finally calculated in S6 (during the loop) may be displayed.

[0049] Next, the operation and effect of the translation text calculation device 1 according to the embodiment will be described.

[0050] According to the translation text calculation device 1, an encoder 20 inputs a sentence in a first language, and a decoder 21 sequentially outputs word candidates for a sentence in a second language corresponding to the sentence in the first language. The translation text calculation device 1 uses a translator configured with a recurrent neural network of an encoder-decoder model. The translation text calculation device 1 includes a first input unit 12 that inputs a target sentence, which is a sentence in the first language, to the encoder 20, a second input unit 13 that sequentially inputs words of a partial sentence, which is a part of a created sentence that is a translation of the target sentence into the second language, to the decoder 21, and a calculation unit 14 that calculates a translation text, which is a sentence based on the word candidates output by the decoder 21 based on the inputs by the first input unit 12 and the second input unit 13. With this configuration, for example, a translation text in the second language that takes into account a partial sentence in the second language corresponding to the content of the target sentence in the first language is calculated. That is, a translation text that takes into account a part of the translated sentence can be calculated.

[0051] Also, according to the translation text calculation device 1, the partial sentence is a partial sentence from the beginning of the created sentence, and the translation text is a sentence following the partial sentence. With this configuration, for example, while adopting a partial sentence from the beginning created by the user, the sentence following the partial sentence can be automatically calculated.

[0052] Also, according to the translation text calculation device 1, the second input unit 13 sequentially inputs words of the partial sentence to the decoder 21, and after the input is completed, sequentially inputs the word candidates sequentially output by the decoder 21 to the decoder 21. With this configuration, the translation text following the partial sentence can be calculated more reliably.

[0053] Also, according to the translation text calculation device 1, the created sentence is a sentence created by the user by translating the target sentence into the second language. With this configuration, a translation text that takes into account a part of the sentence created by the user's translation can be automatically calculated.

[0054] Further, according to the translated text calculation device 1, it further includes an input / output unit 11 that accepts a specification of one word included in the created text as a target word. The second input unit 13 sequentially inputs the words of a partial text, which is the text before the target word in the created text, to the decoder 21. With this configuration, it is possible to easily generate a partial text by simply specifying one word and calculate the translated text.

[0055] Further, according to the translated text calculation device 1, the calculation unit 14 sets a word before the target word in the created text as a new target word and sets the text before the new target word in the created text as a new partial text, and repeats the process of inputting by the first input unit 12 and the second input unit 13 to calculate a new translated text until the text evaluation, which is the evaluation of the text based on the partial text and the calculated translated text, satisfies a predetermined criterion. With this configuration, it is possible to automatically calculate a translated text that satisfies a predetermined criterion.

[0056] Further, according to the translated text calculation device 1, the calculation unit 14 sets, as a new target word, a word that is before the target word in the created text and whose evaluation in the created text does not satisfy a predetermined criterion. With this configuration, since the calculation process of the translated text is performed only on the words whose evaluation does not satisfy a predetermined criterion, it is possible to perform a faster and more reliable calculation process of the translated text.

[0057] Further, according to the translated text calculation device 1, the decoder 21 sequentially outputs word candidates of the text in the second language and the likelihoods for the word candidates. The text evaluation is based on the likelihoods for the word candidates that are the same as the words of the partial text among the word candidates and likelihoods output by the decoder 21 based on the inputs by the first input unit 12 and the second input unit 13. With this configuration, since the text evaluation based on the likelihoods output by the decoder 21 of the translation machine 2 is performed, it is possible to perform a more accurate evaluation that does not require equipment other than the translation machine 2.

[0058] Further, according to the translated text calculation device 1, it further includes an input / output unit 11 for displaying the created text. The input / output unit 11 accepts a designation of a word specified by the user among the created text displayed by the input / output unit 11 as the target word. With this configuration, the user can specify the target word from the displayed created text, thus improving the usability.

[0059] Also, according to the translated text calculation device 1, the input / output unit 11 further displays the translated text calculated by the calculation unit 14. With this configuration, since the text after the specified target word is displayed as the translated text newly calculated by the translated text calculation device 1, the user can easily check the new translated text, improving the usability.

[0060] As described above, the translated text calculation device 1 relates to a composition proofreading system using a neural network. As background art, in English learning, the learning of "writing" and "speaking" has attracted attention. These learnings can be effectively learned if there is someone to proofread or give guidance, but it is difficult to learn by self-study. Also, it costs money to have someone proofread or give guidance. Therefore, a technology that can perform proofreading or guidance similar to that of humans using AI is desired. However, there is a problem that English compositions are diverse, and simply comparing with model answers cannot perform proofreading or guidance. Especially in English compositions for Japanese-to-English translation, there are various expression methods even for the same meaning. When it comes to proofreading or guidance based on the learner's expression method, proofreading or guidance that understands the meaning of the problem sentence is required.

[0061] As a system for grading Japanese-to-English translation compositions, for example, as described in the above reference, there is an English composition grading system that uses a neural network learned by neural machine translation and emphasizes the meaning of the composed text. It can also be said that the translated text calculation device 1 is an English composition proofreading system that uses a neural network learned by neural machine translation to solve the above problems and emphasizes the meaning of the text for proofreading.

[0062] According to the translated text calculation device 1, the user can have the machine proofread any word in the English text created by himself / herself, and furthermore, while respecting the user's composition to the greatest extent so that the reliability of the entire text is increased, it can proofread from the most appropriate words. Therefore, the user can learn correct English while trying various expressions.

[0063] Generally, the scoring criteria for English composition grading are diverse. Especially in grading that emphasizes meaning, there are various ways of expression, and there is a problem that it is difficult to compare with model answers. Depending on the problem text, there are many possible types of expression, and it is difficult to prepare model answers for each of them.

[0064] According to the translated text calculation device 1, using the neural network learned by neural machine translation, when the word sequence of the text written by the user is input to the decoder 21, it is calculated based on the likelihood of the output of the decoder 21 whether the next word is appropriate. Since the encoder 20 has grasped the meaning of the problem text, it is possible to score each word while emphasizing the meaning and allowing freedom of expression in English composition. As a result, it becomes possible to grade English composition, point out words that are not so good, and present the most appropriate words, etc., and the grading or feedback of English composition can be automated.

[0065] The processing of the translated text calculation device 1 can also be said to be an AI proofreading method for English composition. The translated text calculation device 1 is similar to the evaluation (grading) of the translation result, but uses the user's composition up to a certain point, and the translation machine 2 outputs the continuation to complete the composition. Up to that point, the output of the translation machine 2 is rewritten into the words answered by the user for processing. The translated text calculation device 1 respects the user's answer to the greatest extent (uses the user's answer as much as possible), and composes and displays the text after the word to be corrected using AI.

[0066] The translated text calculation device 1 may have the following configuration. It is equipped with a neural network learned for machine translation from the first language to the second language, Means for splitting the problem text in the first language into words and sequentially inputting the words into the encoder of the neural network, means for splitting a part from the beginning of a sentence written in a second language into words and sequentially inputting the words into a decoder of the neural network; a sentence correction means for complementing a continuation of an input sentence to the decoder with an output word of the decoder; means for comparing the likelihood of a word output from the decoder of the neural network with the likelihood of a word in a partially input composition sentence and calculating a score of the corrected sentence; A composition correction system comprising:

[0067] The translated sentence calculation device 1 may have the following configuration. means for splitting a composition sentence in a second language written for a problem sentence in a first language into words and displaying them; means for correcting using the above composition correction system from a word specified in the composition sentence onwards; means for displaying the corrected sentence when the score of the corrected sentence exceeds a predetermined threshold; means for correcting using the above composition correction system from one or more previous words when the score of the corrected sentence does not exceed a predetermined threshold A composition correction display screen comprising:

[0068] The correction result display algorithm of the translated sentence calculation device 1 may be the following procedure. (1) Split the user's English composition into words. (2) Set the target word as the word specified by the user. (3) Present a corrected sentence from the target word. (4) Determine whether the reliability of the corrected sentence from the target word is equal to or greater than a threshold. (5) If it is determined in (4) that the value is equal to or greater than the threshold, present the corrected sentence from the target word. (6) If it is determined in (4) that the value is less than the threshold, set the target word as the word immediately before the target word and return to (4).

[0069] The above proofreading result display algorithm will be described in more detail. First, for the problem of Japanese-to-English translation, the English text (created text) written by the user is split into words. The user can try proofreading after any word, and after the specified word (target word), the translation machine 2 composes a proofread text (translated text) and presents it. If the reliability of the proofread text (proofread text) composed by the translation machine 2 is lower than a predetermined value, the word immediately before the specified word (target word) is set as the target word, and a request is made to the translation machine 2 to proofread the text after the target word. If the reliability of this proofread text (proofread text) is higher than a predetermined value, the proofread text is presented as the proofread text for the entire text. If the reliability is not higher than the predetermined value, a request is similarly made to the translation machine 2 to proofread with the word immediately before as the target word. When tracing back to the first word, since all the words in the proofread text created by the translation machine 2 are words generated by the translation machine 2, the reliability will surely become "1" (highest), and the process ends there.

[0070] The algorithm for the processing of the translation text calculation device 1 may be the following procedure. (1) Obtain the scoring result of the English composition. (2) For each word with a low score (incorrect word), make a proofreading request for the text after that word. (3) Set the target word as the last incorrect word. (4) Determine whether the score of the proofread text from the target word is equal to or higher than the threshold. (5) If it is determined in (4) above that the score is equal to or higher than the threshold, present the proofread text from the target word. (6) If it is determined in (4) above that the score is less than the threshold, set the target word as the incorrect word immediately before the target word and return to (4) above.

[0071] According to the above algorithm, incorrect words are picked up from the back and proofreading is performed on the text starting from that word. If the score of the text after proofreading is not higher than the predetermined value, the previous incorrect word is picked up and the same proofreading and scoring check is performed. Check is traced back to the previous incorrect word until the score becomes higher than the predetermined value. When a text that meets the conditions is obtained, it is presented.

[0072] The algorithm of the translation text calculation device 1 may be the following procedure. (1) Receive the designation of an arbitrary word in the translation text, (2) Proofread the text after the designated word, which is the designated word in the translation text, and create a proofread text that is the result of the proofreading, (3) Evaluate the created proofread text, (4) If the evaluation in (3) above does not meet the predetermined criteria, return to (2) above with a word before the designated word in the translation text as the new designated word, (5) If the evaluation in (3) above meets the predetermined criteria, output the proofread text. According to the above configuration, it is possible to obtain a proofread text in which the sentences before the designated word in the translation text (for example, the English text written by the user in English composition) are respected and the evaluation meets the predetermined criteria.

[0073] Note that the block diagram used in the description of the above embodiment shows blocks of functional units. These functional blocks (components) are realized by any combination of at least one of hardware and software. Also, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using one physically or logically combined device, or two or more physically or logically separated devices may be directly or indirectly (for example, using wired, wireless, etc.) connected and realized using these multiple devices. The functional block may be realized by combining software with the above one device or the above multiple devices.

[0074] The functions include, but are not limited to, judgment, decision-making, determination, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, solution, selection, selection determination, establishment, comparison, assumption, expectation, presumption, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating (mapping), assigning, etc. For example, a functional block (component) that enables transmission is called a transmitting unit or a transmitter. As described above, the implementation method is not particularly limited.

[0075] For example, a translated text calculation device 1 in an embodiment of the present disclosure may function as a computer that performs the processing of the translated text calculation method of the present disclosure. FIG. 5 is a diagram showing an example of the hardware configuration of the translated text calculation device 1 according to an embodiment of the present disclosure. Physically, the above-described translated text calculation device 1 may be configured as a computer device including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like.

[0076] In the following description, the term "device" can be read as a circuit, a device, a unit, etc. The hardware configuration of the translated text calculation device 1 may be configured to include one or more of each device shown in the figure, or may be configured without including some devices.

[0077] Each function in the translated text calculation device 1 is realized by causing the processor 1001 to perform operations by loading a predetermined software (program) onto hardware such as the processor 1001 and the memory 1002, and controlling the communication by the communication device 1004, or controlling at least one of the reading and writing of data in the memory 1002 and the storage 1003.

[0078] Processor 1001 controls the entire computer by operating an operating system, for example. Processor 1001 may be constituted by a central processing unit (CPU: Central Processing Unit) including an interface with peripheral devices, a control device, an arithmetic device, registers, and the like. For example, the above-described input / output unit 11, first input unit 12, second input unit 13, calculation unit 14, and the like may be realized by processor 1001.

[0079] Further, processor 1001 reads a program (program code), software module, data, etc. from at least one of storage 1003 and communication device 1004 into memory 1002, and executes various processes according to these. As the program, a program for causing a computer to execute at least a part of the operations described in the above-described embodiments is used. For example, storage unit 10 may be stored in memory 1002 and realized by a control program operating in processor 1001, and other functional blocks may be realized in the same manner. Although it has been described that the above-described various processes are executed by one processor 1001, they may be executed simultaneously or sequentially by two or more processors 1001. Processor 1001 may be mounted by one or more chips. Note that the program may be transmitted from a network via a telecommunication line.

[0080] Memory 1002 is a computer-readable recording medium, and may be constituted by at least one of, for example, ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), RAM (Random Access Memory), and the like. Memory 1002 may be referred to as a register, a cache, a main memory (main storage device), and the like. Memory 1002 can store a program (program code), software module, etc. executable for implementing the wireless communication method according to an embodiment of the present disclosure.

[0081] Storage 1003 is a computer-readable recording medium and may be composed of at least one of, for example, an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disc, a digital versatile disc, a Blu-ray (registered trademark) disc), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy (registered trademark) disk, a magnetic strip, etc. Storage 1003 may be referred to as an auxiliary storage device. The above-described storage medium may be, for example, a database, a server, or other appropriate medium including at least one of the memory 1002 and the storage 1003.

[0082] The communication device 1004 is hardware (a transmission / reception device) for performing communication between computers via at least one of a wired network and a wireless network, and is also referred to as, for example, a network device, a network controller, a network card, a communication module, etc. The communication device 1004 may be configured to include, for example, a high-frequency switch, a duplexer, a filter, a frequency synthesizer, etc. in order to implement at least one of frequency division duplex (FDD: Frequency Division Duplex) and time division duplex (TDD: Time Division Duplex). For example, the above-described input / output unit 11, the first input unit 12, the second input unit 13, the calculation unit 14, etc. may be implemented by the communication device 1004.

[0083] The input device 1005 is an input device (e.g., a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that receives an external input. The output device 1006 is an output device (e.g., a display, a speaker, an LED lamp, etc.) that performs an output to the outside. Note that the input device 1005 and the output device 1006 may have an integrated configuration (e.g., a touch panel).

[0084] Also, each device such as the processor 1001 and the memory 1002 is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus or may be configured using different buses for each device.

[0085] Further, the translation text calculation device 1 may be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), and a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, the processor 1001 may be implemented using at least one of these hardware components.

[0086] The notification of information is not limited to the modes / embodiments described in the present disclosure, and other methods may be used.

[0087] The processing procedures, sequences, flowcharts, etc. of each mode / embodiment described in the present disclosure may be rearranged as long as there is no contradiction. For example, for the methods described in the present disclosure, the elements of various steps are presented using an exemplary order and are not limited to the specific order presented.

[0088] The input / output information, etc. may be stored in a specific location (e.g., memory) or may be managed using a management table. The input / output information, etc. may be overwritten, updated, or appended. The output information, etc. may be deleted. The input information, etc. may be transmitted to other devices.

[0089] The determination may be made based on a value represented by 1 bit (0 or 1), a boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0090] Each aspect / embodiment described in the present disclosure may be used alone, in combination, or switched and used during execution. Also, the notification of predetermined information (for example, the notification of "being X") is not limited to being explicitly performed, and may be performed implicitly (for example, by not performing the notification of the predetermined information).

[0091] As described above in detail, for those skilled in the art, it is obvious that the present disclosure is not limited to the embodiments described in the present disclosure. The present disclosure can be implemented as modified and changed aspects without departing from the spirit and scope of the present disclosure defined by the description of the claims. Therefore, the description of the present disclosure is for the purpose of illustration and has no restrictive meaning for the present disclosure.

[0092] Software should be broadly interpreted to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, functions, etc., whether called by the name of software, firmware, middleware, microcode, hardware description language, or other names.

[0093] Also, software, instructions, information, etc. may be transmitted and received via a transmission medium. For example, when software is transmitted from a website, server, or other remote source using at least one of wired technologies (such as coaxial cables, fiber optic cables, twisted pairs, digital subscriber lines (DSL)) and wireless technologies (such as infrared rays, microwaves), at least one of these wired technologies and wireless technologies is included within the definition of the transmission medium.

[0094] The information, signals, etc. described in this disclosure may be represented using any of a variety of different technologies. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0095] In addition, the terms described in this disclosure and the terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings.

[0096] The terms "system" and "network" used in this disclosure are used interchangeably.

[0097] Also, the information, parameters, etc. described in this disclosure may be represented using absolute values, relative values from a predetermined value, or corresponding other information.

[0098] The names used for the above-described parameters are not limiting in any way. Furthermore, the mathematical formulas, etc. using these parameters may be different from those explicitly disclosed in this disclosure.

[0099] The terms "determining" and "deciding" as used in this disclosure may encompass a wide variety of actions. "Determining" and "deciding" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up (e.g., searching in a table, database, or other data structure), ascertaining, and considering something as having been "determined" or "decided". Also, "determining" and "deciding" may include considering something as having been "determined" or "decided" after receiving (e.g., receiving information), transmitting (e.g., transmitting information), inputting, outputting, accessing (e.g., accessing data in a memory), etc. Further, "determining" and "deciding" may include considering something as having been "determined" or "decided" after resolving, selecting, choosing, establishing, comparing, etc. That is, "determining" and "deciding" may include considering that some action has been "determined" or "decided". Also, "determining (deciding)" may be read as "assuming", "expecting", "considering", etc.

[0100] The terms "connected" and "coupled," or any variations thereof, mean any direct or indirect connection or coupling between two or more elements, and can include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements can be physical, logical, or a combination thereof. For example, "connected" may be read as "accessed." As used in this disclosure, two elements can be considered to be "connected" or "coupled" to each other using at least one of one or more wires, cables, and printed electrical connections, as well as, by way of some non-limiting and non-exhaustive examples, electromagnetic energy having wavelengths in the radio frequency region, microwave region, and optical (both visible and invisible) region.

[0101] As used in this disclosure, the recitation "based on" does not mean "based only on" unless otherwise specified. In other words, the recitation "based on" means both "based only on" and "based at least in part on."

[0102] Any reference to an element using designations such as "first," "second," etc. used in this disclosure does not generally limit the quantity or order of those elements. These designations can be used in this disclosure as a convenient way to distinguish between two or more elements. Thus, a reference to a first and a second element does not mean that only two elements can be employed, or that the first element must precede the second element in any way.

[0103] In the configuration of each of the above devices, "means" may be replaced with "section," "circuit," "device," etc.

[0104] In the present disclosure, when terms such as "include", "including" and their variants are used, these terms are intended to be inclusive, similar to the term "comprising". Further, the term "or" used in the present disclosure is not intended to be an exclusive disjunction.

[0105] In the present disclosure, for example, when articles are added by translation, such as a, an and the in English, the present disclosure may include that the nouns following these articles are in the plural form.

[0106] In the present disclosure, the term "A is different from B" may mean that "A and B are different from each other". Note that the term may also mean that "A and B are each different from C". Terms such as "separate", "coupled" and the like may also be interpreted in the same way as "different".

Explanation of Reference Numerals

[0107] 1... Translation text calculation device, 2... Translator, 3... Translation text calculation system, 10... Storage unit, 11... Input / output unit, 12... First input unit, 13... Second input unit, 14... Calculation unit, 20... Encoder, 21... Decoder.< / eos> < / eos> < / eos> < / eos>

Claims

1. A translation text calculation device that uses a recurrent neural network of an encoder-decoder model in which an encoder inputs a sentence in a first language and a decoder sequentially outputs word candidates of a sentence in a second language corresponding to the sentence in the first language, a first input unit that inputs a target sentence, which is a sentence in the first language, to the encoder; a second input unit that sequentially inputs words of a partial sentence, which is a part of a created sentence that is a translation of the target sentence into the second language, to the decoder; a calculation unit that calculates a translated sentence, which is a sentence based on the word candidates output by the decoder based on the inputs by the first input unit and the second input unit; a reception unit that receives a designation of one word included in the created sentence as a target word; comprising: the second input unit sequentially inputs words of the partial sentence, which is a sentence before the target word in the created sentence, to the decoder as the partial sentence; the calculation unit, until a sentence evaluation, which is an evaluation of a sentence based on the partial sentence and the calculated translated sentence, satisfies a predetermined criterion, sets a word before the target word in the created sentence as the new target word and sets a sentence before the new target word in the created sentence as the new partial sentence, and performs the inputs by the first input unit and the second input unit to calculate a new translated sentence; repeats the above; a translation text calculation device.

2. the partial sentence is a partial sentence from the beginning of the created sentence, the translated sentence is a sentence following the partial sentence, the translation text calculation device according to Claim 1.

3. the second input unit sequentially inputs words of the partial sentence to the decoder, and after the input is completed, sequentially inputs word candidates sequentially output by the decoder to the decoder, the translation text calculation device according to Claim 2.

4. the created sentence is a sentence created by a user translating the target sentence into the second language, the translation text calculation device according to any one of Claims 1 to 3.

5. the calculation unit sets, as the new target word, a word before the target word in the created sentence, where the evaluation of the word in the created sentence does not satisfy a predetermined criterion, the translation text calculation device according to any one of Claims 1 to 4.

6. the decoder sequentially outputs likelihoods for the word candidates together with the word candidates of the sentence in the second language, The article evaluation is based on the likelihood for a word candidate that is the same as a word in the partial article among the word candidates and likelihoods output by the decoder based on the inputs from the first input unit and the second input unit. The translated article calculation device according to any one of claims 1 to 5. **Claim 7** The apparatus further includes a display unit that displays the created article. The reception unit receives, as the target word, a word specified by a user from among the created articles displayed by the display unit. The translated article calculation device according to any one of claims 1 to 6. **Claim 8** The display unit further displays the translated article calculated by the calculation unit. The translated article calculation device according to claim 7.

Images