JP2022032319A - Created sentence evaluation device - Google Patents

Abstract

To provide a created sentence evaluation device capable of evaluating a created sentence created in a predetermined language.SOLUTION: There is provided a created sentence evaluation device that uses a neural network part of an encoder decoder model where an encoder part inputs a sentence of a first language, and a decoder part outputs word candidates of a sentence of a second language corresponding to the sentence of the first language and a likelihood of the word candidates one after another in a predetermined word order different from a word order from a starting word to an ending word of the sentence. The decoder part comprises: a decoder input part which inputs words of the created sentence created in the second language one after another in the predetermined word order; an evaluation part which evaluates the words of the created sentence on the basis of the word candidates of the second language and the likelihood of the word candidates that the decoder part outputs on the basis of input by the decoder input part; and an output part which performs output based on the evaluation by the evaluation part.SELECTED DRAWING: Figure 1

Description

One aspect of the present disclosure relates to a created sentence evaluation device that evaluates a sentence created by a user translating a sentence in a first language into a second language.

Conventionally, neural machine translation (Neural Machine Translation), which is a 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 sentences in the first language (for example, Japanese), and the decoder outputs the sentences in the second language (for example, English) corresponding to the sentences in the first language as translation results. ..

For example, Patent Document 1 below discloses an automatic interpreter including an encoder and a decoder configured by a neural network.

Japanese Unexamined Patent Publication No. 2018-5218

The above automatic interpreter will input the second language sentence "How do I get to Gangnam?", Which is the translation result, in response to the first language sentence "How do you get to Gangnam Station?" Entered by the user. Output. However, with the automatic interpreter, for example, it is not possible to evaluate the appropriateness of translation of a second language sentence created by the user by translating the first language sentence by himself / herself.

Therefore, it is desired to evaluate the created sentences created in a predetermined language.

In the created sentence evaluation device according to one aspect of the present disclosure, the encoder inputs a sentence in the first language, and the decoder determines the word candidate of the sentence in the second language corresponding to the sentence in the first language and the likelihood of the word candidate. Is a created sentence evaluation device using a recurrent neural network of an encoder decoder model that sequentially outputs in a predetermined word order different from the word order from the first word to the last word of the sentence. A decoder input unit that sequentially inputs the words of the created sentence created in two languages in a predetermined word order, a second language word candidate that the decoder outputs the words of the created sentence based on the input by the decoder input unit, and the relevant word. It includes an evaluation unit that evaluates based on the likelihood of word candidates, and an output unit that outputs based on the evaluation by the evaluation unit.

In this aspect, words in written sentences created in a second language can be evaluated. That is, it is possible to evaluate a created sentence created in a predetermined language.

According to one aspect of the present disclosure, it is possible to evaluate a written sentence written in a predetermined language.

It is a figure which shows an example of the functional structure of the created sentence evaluation apparatus which concerns on 1st Embodiment. It is a figure which shows the Example of translation and evaluation (the 1). It is a figure which shows the example (No. 2) of translation and evaluation. It is a flowchart which shows an example of a part of the processing executed by the created sentence evaluation apparatus which concerns on 1st Embodiment. It is a figure which shows an example of an interface. It is a figure which shows an example of the functional structure of the created sentence evaluation apparatus which concerns on 2nd Embodiment. It is a figure which shows an example of the hardware composition of the computer used in the created sentence evaluation apparatus which concerns on 1st Embodiment and 2nd Embodiment.

Hereinafter, embodiments in the present disclosure will be described in detail with reference to the drawings. In the description of the drawings, the same elements are designated by the same reference numerals, and duplicate description will be omitted. Further, the embodiments (including the first embodiment and the second embodiment) in the present disclosure in the following description are specific examples of the present invention, and unless otherwise stated to limit the present invention, these embodiments are described. It is not limited to the embodiment.

[First Embodiment]
FIG. 1 is a diagram showing an example of a functional configuration (functional block) of the created text evaluation device 1 according to the first embodiment. As shown in FIG. 1, the created sentence evaluation device 1 includes a neural network unit 10, a bilingual corpus storage unit 11, a problem sentence preparation unit 12, an encoder input unit 13, an output unit 14 (output unit), a created sentence input unit 15, and a decoder. It includes an input unit 16 (decoder input unit) and an evaluation unit 17 (evaluation unit). Further, the neural network unit 10 includes an encoder unit 100 and a decoder unit 101.

The created sentence evaluation device 1 is a computer device that evaluates a created sentence, which is a sentence created by translating a sentence in a first language into a second language. The first language is, for example, Japanese, but may be any other language. The second language is a language different from the first language, for example English, but may be any other language. The first language and the second language may be different local dialects (for example, standard language and Kansai dialect in Japan). The language is not limited to a natural language, but may be an artificial language, a formal language (computer programming language), or the like. The created text is intended to be created by a person such as a user of the created text evaluation device 1, but may be a text created by someone other than the person. The evaluation indicates that the appropriateness (good or bad, goodness of fit) as a translation of the created sentence is determined.

Each functional block of the created sentence evaluation device 1 is assumed to function in the created sentence evaluation device 1, but is not limited to this. For example, a part of the functional block of the created sentence evaluation device 1 is a computer device different from the created sentence evaluation device 1, and in the computer device connected to the created sentence evaluation device 1 via the network, the created sentence evaluation device 1 and the created sentence evaluation device 1 It may function while transmitting and receiving information as appropriate. Further, some functional blocks of the created sentence evaluation 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. May be good.

Hereinafter, each functional block of the created sentence evaluation device 1 shown in FIG. 1 will be described.

The neural network unit 10 is a recurrent neural network (RNN) of an encoder decoder model (also known as an encoder-decoder translation model, Sequence to Sequence Model) composed of an encoder unit 100 and a decoder unit 101. The neural network unit 10 is, for example, a recurrent neural network called LSTM (Long Short Term Memory). The neural network unit 10 performs neural machine translation. The neural network unit 10 may not be in the created sentence evaluation device 1 but in another device that can access each other via the created sentence evaluation device 1 and the network.

The encoder unit 100 inputs a sentence in the first language and outputs a vector of an intermediate layer (hidden layer). More specifically, the encoder unit 100 divides a sentence in the first language into words by morphological analysis or the like, converts a word ID (Word ID) corresponding to each word into a word vector (vector of an input layer), and then converts it into words. Then, input sequentially from the first word to the last word of the sentence, and sequentially output the vector of the intermediate layer based on the input contents so far (calculate the neural network). When the encoder unit 100 inputs "<EOS>" indicating the end of the sentence, the encoder unit 100 outputs (passes) the vector of the intermediate layer based on the input contents up to that point to the decoder unit 101. It can be said that the encoder unit 100 conceptually analyzes the meaning of a sentence in the first language and extracts the semantic expression.

The decoder unit 101 inputs the vector of the intermediate layer output from the encoder unit 100 and is based on the vector of the intermediate layer, or is based on the vector of the intermediate layer and the word of the second language input to the decoder unit 101. , The vector of the output layer is calculated sequentially and output. The vector of the output layer is, for example, information indicating a list of a set of a word candidate of a second language and the likelihood of the word candidate. As an example of the list, "(word candidate" I "and its likelihood" 0.7 ", word candidate" We "and its likelihood" 0.2 ", word candidate" You "and its likelihood" 0.1 " ")" And the like.

Here, the processing of the decoder (hereinafter referred to as “general decoder”) when performing neural machine translation in the recurrent neural network of the general encoder decoder model will be described. In a general decoder, first, when the vector of the intermediate layer output from the encoder is input, the first word (first word) of the second language sentence finally output based on the input vector of the intermediate layer. ) Corresponds to the output layer vector. After that, the general decoder extracts the word with the highest likelihood from the word candidates indicated by the vector of the output layer of the Nth word (N is an integer of 1 or more), and uses the extracted word as its own decoder (the general decoder). ), And the vector of the output layer of the (N + 1) word is output based on the input word and the vector of the intermediate layer used when outputting the vector of the output layer of the Nth word. Is repeated until the last word of the sentence in the second language. It can be said that a general decoder conceptually generates a sentence (in a second language different from the first language) from a semantic expression extracted by the encoder.

On the other hand, unlike a general decoder, the decoder unit 101 sets the word candidates of the second language sentence corresponding to the sentence of the first language and the likelihood of the word candidates in order from the first word to the last word of the sentence. It outputs sequentially in a predetermined word order different from the word order (the order adopted by a general decoder). The predetermined word order may be a word order (reverse order) from the last word of the sentence to the first word. Specifically, when the sentence is composed of M words (M is an integer of 1 or more), the Mth word, the (M-1)th word, the (M-2)th word, ..., the third word, The order is the second word and the first word. In addition, the predetermined word order may be a random order, the first word, the M word, the second word, the (M-1) word, the third word, and the (M-2) word. The order may be gradually toward the middle word of the sentence, such as the fourth word, ..., Or any other order. In the embodiment of the present disclosure, the predetermined word order is the reverse order described above, but the order is not limited to this, and may be appropriately read in any order.

As described above, in the created sentence evaluation device 1, the encoder unit 100 inputs the sentence in the first language, and the decoder unit 101 is the word candidate of the sentence in the second language corresponding to the sentence in the first language and the word candidate of the word candidate. The neural network unit 10 of the encoder / decoder model that sequentially outputs the likelihood in the reverse order (a predetermined word order different from the word order from the first word to the last word in the sentence) is used. Other functions of the decoder unit 101 will be described later.

The bilingual corpus storage unit 11 stores the bilingual corpus between the first language and the second language. A bilingual corpus is a corpus (sentence database) in which sentences in different languages are summarized in the form of bilingual translations, which is constructed for use as learning data for machine translation. The bilingual corpus storage unit 11 has a sentence in the first language (arbitrary) and a sentence in the second language corresponding to the sentence in the first language (translated from the sentence in the first language) in reverse order. Stores a reverse translation corpus, which is a translation corpus composed of pairs. For example, as an example of the group, the sentence "I like AI" in the first language (Japanese) and the sentence ".AI like I" in the reverse order of the second language (English) (normal sentence that is not in reverse order). Can be paired with "I like AI."). The neural network unit 10 may be learned based on a reverse-order bilingual corpus (a corpus of a sentence in a first language and a sentence in a second language corresponding to the sentence in the first language in reverse order). .. The learning may be performed based on the instruction of the administrator of the created sentence evaluation device 1, or may be performed regularly (for example, once a week).

The question sentence preparation unit 12 prepares a question sentence, which is a sentence in the first language to be presented to the user as a question sentence. More specifically, the question sentence preparation unit 12 extracts sentences in the first language included in the reverse-order bilingual corpus stored by the bilingual corpus storage unit 11. The sentences in the first language to be extracted may be random or may follow a predetermined procedure. The question sentence preparation unit 12 may acquire the question sentence from a different means, a different device, or the like, not from the reverse translation corpus stored by the bilingual corpus storage unit 11. The question sentence preparation unit 12 outputs the prepared question sentence to the encoder input unit 13 and the output unit 14.

The encoder input unit 13 inputs the problem text output from the problem text preparation unit 12 to the encoder unit 100.

The output unit 14 outputs the question sentence output from the question sentence preparation unit 12. More specifically, the output unit 14 displays the problem text to the user via the output device 1006 (for example, a screen) described later. Other functions of the output unit 14 will be described later.

The created text input unit 15 inputs a created text created in a second language. More specifically, the created sentence input unit 15 describes the created sentence created by the user translating the displayed problem sentence into a second language after the problem sentence is displayed to the user by the output unit 14. Input is performed via the input device 1005 (for example, a keyboard). That is, the created sentence may be a sentence created by the user translating the sentence in the first language input to the encoder unit 100 into the second language. In the embodiment of the present disclosure, it is assumed that the created text is created by the user, but it may be created by a person or device other than the user. The created text input unit 15 outputs the created text to the decoder input unit 16 and the evaluation unit 17.

The decoder input unit 16 sequentially inputs the words of the created sentence into the decoder unit 101 in the reverse order. More specifically, the decoder input unit 16 divides the created sentence output from the created sentence input unit 15 into words by morphological analysis or the like, and the word ID corresponding to each word is a word vector (vector of the input layer). After converting to, it is input to the decoder unit 101 in reverse order (from the last word to the first word in the created sentence).

Here, the decoder input unit 16 sequentially inputs the words of the created sentence into the decoder unit 101 in the reverse order as described above, instead of the word candidates that the decoder unit 101 sequentially outputs (in the reverse order). That is, the general decoder uses the word output from the general decoder (word candidate with the maximum likelihood) as the next input of the general decoder, but the decoder unit 101 of the creation sentence evaluation device 1 creates. The word of the sentence is used as the input of the decoder unit 101.

The processing of the decoder unit 101 based on the input of the decoder input unit 16 will be specifically described. First, when the vector of the intermediate layer output from the encoder unit 100 is input, the decoder unit 101 corresponds to the last word in the second language sentence finally output based on the input vector of the intermediate layer. Output the vector of the output layer. After that, the decoder unit 101 inputs the word of the Pth word (P is an integer of 2 or more) of the created sentence input by the decoder input unit 16 into the decoder unit 101, and the input word and the vector of the previous output layer. Based on the vector of the intermediate layer used when outputting, the process of outputting the vector of the output layer corresponding to the (P-1) word of the created sentence is performed by the first word (1) of the created sentence. Repeat until the word).

When the encoder input unit 13 inputs the problem text to the encoder unit 100, the encoder unit 100 performs a calculation, outputs the vector of the intermediate layer to the decoder unit 101, and the decoder unit 101 is based on the input vector of the intermediate layer. Then, the vector of the output layer corresponding to the last word in the sentence of the second language to be finally output is output, but the subsequent processing is the sequential (reverse order) of the words of the created sentence from the decoder input unit 16. Hold until there is input.

The evaluation unit 17 evaluates the words in the created sentence based on the word candidates of the second language output by the decoder unit 101 based on the input by the decoder input unit 16 and the likelihood of the word candidates (scoring, scoring). Then, the evaluation result is output to the output unit 14. The evaluation unit 17 in the created sentence is based on the word candidate of the second language output by the decoder unit 101 and the likelihood of the word candidate for the input word which is the word of the created sentence input by the decoder input unit 16. You may evaluate the next word in the reverse order of the input word.

More specifically, first, the evaluation unit 17 outputs a second language word candidate and the word candidate of the second language output by the decoder unit 101 with respect to the P-th word (input word) of the created sentence input by the decoder input unit 16. Get the likelihood of a word candidate. Next, the evaluation unit 17 acquires the word (P-1) of the next word in the reverse order of the word of the P word in the created sentence output from the created sentence input unit 15. Next, the evaluation unit 17 determines the likelihood of associating the acquired word candidate and the likelihood of the word candidate with the same word candidate as the word of the (P-1) th word in the (P-1) th word. Get as the likelihood of the word. Next, the evaluation unit 17 acquires the maximum likelihood (the likelihood of the most appropriate word candidate for translation) among the acquired word candidates and the likelihood of the word candidate. Next, the evaluation unit 17 divides the likelihood of the acquired (P-1) word by the acquired maximum likelihood (likelihood ratio) into the (P-1) word of the created sentence. Calculated as the evaluation value of. Next, the evaluation unit 17 outputs the calculated evaluation value to the output unit 14. The calculation of the evaluation value is not limited to the above-mentioned division, and for example, the likelihood of the word (P-1) and the maximum likelihood (or the obtained word candidate and the likelihood of the word candidate). Of these, any calculation may be made based on the comparison with the likelihood obtained according to a predetermined standard, or any other calculation based on the likelihood may be used.

The evaluation unit 17 may calculate the evaluation value of the created sentence (whole) based on the evaluation value of each word of the created sentence and output it to the output unit 14. For example, the evaluation unit 17 calculates an average evaluation value obtained by averaging the evaluation values of each word of the created sentence as the evaluation value of the created sentence.

The evaluation unit 17 may output the acquired word candidate and a part or all of the likelihood of the word candidate to the output unit 14. For example, the evaluation unit 17 outputs, among the acquired word candidates and the likelihoods of the word candidates, a word candidate having a likelihood higher than a predetermined reference or a word candidate having the maximum likelihood to the output unit 14.

The output unit 14 outputs based on the evaluation by the evaluation unit 17. More specifically, the output unit 14 outputs based on the evaluation result output from the evaluation unit 17. The output unit 14 may display the evaluation value (score) of the word of the created sentence output from the evaluation unit 17 (for each word). The output unit 14 may display the evaluation value (total score) of the created sentence (overall) output from the evaluation unit 17. The output unit 14 may display the word based on the evaluation value of the word in the created sentence output from the evaluation unit 17. For example, the output unit 14 displays a word whose evaluation value is lower than a predetermined reference by changing the color, size, or the like of the word. The output unit 14 is a word candidate whose likelihood is larger than a predetermined standard or a word candidate whose likelihood is the maximum, which is output from the evaluation unit 17 as another candidate (optimal word for each word) for the word in the created sentence. May be displayed. The output unit 14 may display a combination of one or more of the above display methods. For example, the output unit 14 may also display a word candidate having the maximum likelihood as another candidate for a word whose evaluation value is lower than a predetermined criterion. The output unit 14 may acquire and display a model answer to the question sentence by referring to the reverse translation corpus stored by the bilingual corpus storage unit 11. In addition, the output unit 14 shall appropriately acquire various information necessary for output from other functional blocks. Further, a part of the functions of the output unit 14 may be incorporated in the evaluation unit 17. That is, the evaluation unit 17 may perform evaluation and directly output the evaluation result.

2 and 3 are diagrams showing examples of translation and evaluation in the created sentence evaluation device 1. The example of the neural network unit 10 shown in FIGS. 2 and 3 should be shown in one figure, but due to the size, about the left half of the example of the neural network unit 10 is shown in FIG. , About the right half of an example of the neural network unit 10 is shown in FIG.

In the examples shown in FIGS. 2 and 3, an example of scoring (evaluation) when the question sentence "I like AI" is presented to the user and the user creates the created sentence "My favorite is AI." show. The model answer to the question sentence "I like AI." Is "I like AI.". As shown in Fig. 2, first, if the question sentence "I like AI" is divided into words, "I", "ha", "AI", "ga", "like", "is", "is", " ."become. These words are input to the encoder unit 100 in order (from the first word to the last word). After the word ".", Enter the "<EOS>" symbol indicating the end of the sentence. When the "<EOS>" symbol is input, the neural network unit 10 starts to output the word candidate and the likelihood of the word candidate from the decoder unit 101 (the output layer is calculated with the Softmax function or the like).

Subsequently, as shown in FIG. 3, the word candidate output first is ".", Which has the highest likelihood. Since the last word of the created sentence created by the user is also ".", The score is "1.00" because the likelihoods are the same (by division between the same likelihoods). Next, the (last) word "." Created by the user is input as an input of the decoder unit 101 (the created sentence created by the user is sequentially input to the decoder unit 101 for each word in reverse order). Then, the word candidate output by the decoder unit 101 becomes "AI" having the highest likelihood. Since the penultimate word of the created sentence created by the user is also "AI", the score is "1.00" because the likelihood is the same. Next, the user-created (second to last) word "AI" is input as input to the decoder unit 101. Then, as for the output of the decoder unit 101, "like" has the highest likelihood, and the likelihood is "0.75". However, the user-created word (in reverse order, following "AI") was "is", which had a likelihood of "0.15". Therefore, the likelihood of "is" is divided by the likelihood of "like", and "0.20" is the score (evaluation value) of "is". Next, "is" is input instead of "like" as the input of the decoder unit 101. Then, the output of the decoder unit 101 has the highest likelihood of "favorite". Since the user-created (third from the last) word is also "favorite", the score is "1.0". In this way, the score is calculated for each word in reverse order.

FIG. 4 is a flowchart showing an example of a part of the process (created sentence evaluation method) executed by the created sentence evaluation device 1. First, the encoder input unit 13 inputs the problem sentence (word) to the input layer of the encoder unit 100 (step S1). Next, when the encoder unit 100 determines whether or not the next input word is "<EOS>" (step S2), and determines that the next input word is not "<EOS>" (S2 :). NO), the process returns to S1 for the next input word. On the other hand, when it is determined that the next input word is "<EOS>" (S2: YES), the decoder unit 101 subsequently outputs the answer sentence (finally output second language) from the output layer with Softmax. The likelihood of the word (candidate word) of the sentence) is calculated (step S3). Next, the evaluation unit 17 compares the words in the created sentence with the likelihood, and the likelihood ratio is calculated as a word score (step S4). Next, the decoder unit 101 determines whether or not there is the next word in the reverse order of the created sentence (step S5), and if it is determined that there is the next word (S5: NO), the decoder input unit 16 Then, the word of the created sentence is input as the next input of the decoder unit 101 (step S6), and the process returns to the process of S3. On the other hand, when it is determined that there is no next word (S5: YES), the evaluation unit 17 averages the word scores of each word and calculates the total score as a sentence (created sentence) (step S7). Next, the output unit 14 outputs the word score for each word and the total score of the created sentence (step S8).

FIG. 5 is a diagram showing an example (screen example) of the interface of the created sentence evaluation device 1. The interface shown in FIG. 5 presents a Japanese (first language) question sentence to the user, inputs the English (second language) composition sentence written by the user to the question sentence, and scores the created sentence. It is something to do. Reference numeral A indicates the question sentence "that is, it may improve the student's grades." Reference numeral B indicates a created sentence "Because, there is a possibility that the student's grade will improve." Entered by the user in the text box. The scoring is executed when the user presses the "scoring" button indicated by the reference code C. The information indicated by the reference numerals D, E, F and H described below is displayed after the scoring is executed.

Reference numeral D indicates "8.21" points out of "10.00" points, which is the total score of the prepared sentence as a scoring result. A character image corresponding to the total score may be displayed around the reference numeral D. The reference code E is a created sentence, and as a result of scoring, words whose evaluation value is lower than a predetermined standard (word “Because” in FIG. 5) are displayed in bold (or in different colors). Is shown. Reference numeral F indicates a compound word (in FIG. 5, the compound word “In short”) composed of the word candidate having the maximum likelihood as another candidate for the word whose evaluation value is lower than the predetermined criterion. When the user presses the "view model answer" button indicated by the reference code G, the model answer to the question sentence is displayed. Reference numeral H indicates the displayed model answer "In short, there is a possibility that the student's grade will improve."

Next, the operation and effect of the created sentence evaluation device 1 according to the first embodiment will be described.

According to the created sentence evaluation device 1, the decoder input unit 16 sequentially inputs the words of the created sentence created in the second language into the decoder unit 101 in the reverse order, and the evaluation unit 17 inputs the words of the created sentence into the decoder unit. Based on the input by the unit 16, evaluation is performed based on the word candidate of the second language output by the decoder unit 101 and the likelihood of the word candidate, and the output unit 14 outputs based on the evaluation by the evaluation unit 17. More specifically, the evaluation unit 17 determines the word candidate of the second language output by the decoder unit 101 and the likelihood of the word candidate with respect to the input word which is the word of the created sentence input by the decoder input unit 16. Based on, evaluate the next word in reverse order of the input word in the created sentence. This makes it possible to evaluate words in a written sentence created in a second language. That is, it is possible to evaluate a created sentence created in a predetermined language. In addition, the created sentence created by the user can be evaluated word by word.

Further, according to the created sentence evaluation device 1, the decoder input unit 16 sequentially inputs the words of the created sentence into the decoder unit 101 in reverse order instead of the word candidates sequentially output by the decoder unit 101. As a result, the encoder unit 100 can estimate the most suitable word that comes next in the reverse order with respect to the word string of the created sentence created by the user while grasping the meaning of the problem sentence. In addition, since the next word is estimated in the reverse order based on the word of the expression selected by the user while grasping the meaning of the problem sentence, it is possible to evaluate with a degree of freedom of expression.

Further, according to the created sentence evaluation device 1, the created sentence is a sentence created by the user translating the sentence of the first language input to the encoder unit 100 into the second language. Thereby, the user can evaluate the translation of the text of the second language, which is the translation of the text of the first language.

Further, according to the created sentence evaluation device 1, the neural network unit 10 is based on a corpus of a set of sentences in a first language and sentences in a second language corresponding to the sentences in the first language in reverse order. To be learned. As a result, when the sentence of the first language is input to the encoder unit 100, the decoder unit 101 determines the word candidate of the sentence of the second language corresponding to the sentence of the first language and the likelihood of the word candidate in the reverse order. It is possible to output sequentially (from the last word to the first word in the sentence of the second language).

Further, according to the created sentence evaluation device 1, the predetermined word order is a word order from the last word to the first word in the sentence. This makes it possible to evaluate (scoring, correction, etc.) not only from the beginning of the sentence but also from the back. If it becomes possible to evaluate from the back of a sentence, it will be possible to correct only the front of a created sentence that is wrong only in the front (evaluation value is lower than a predetermined standard).

Hereinafter, a specific description will be given. In general, there are many written sentences in which only the beginning of the sentence is wrong and the end is correct. When evaluating from the beginning of a sentence on a word-by-word basis, all the wrong words at the beginning are evaluated first, so the sentence created by the user (backward) is not utilized. For example, if the user-created text is "Because, there is a possibility that the student's grade will improve." For the question text "That is, it may improve the student's grade." , The score is "5.92" out of a maximum of "10.00", and the corrected text is "This means that it can increase their academic performance." In this way, the created sentences created by the user are generally correct except for the translation of "that is" among the problem sentences, but the score is low, and the corrected sentences are completely rewritten. In other words, you only have to fix "Because", but you fix everything.

On the other hand, when the above-mentioned created sentence is evaluated by the created sentence evaluation device 1, for example, the corrected sentence becomes "In short, there is a possibility that the student's grade will improve.". In other words, since only the "Because" part is modified, the user's answer is utilized.

Further, according to the created sentence evaluation device 1, the evaluation unit 17 and the output unit 14 display the evaluation value of the word in the created sentence, display the word based on the evaluation value of the word in the created sentence, and display the created sentence. As another candidate for the word, the word candidate whose likelihood is higher than the predetermined criterion or the word candidate whose likelihood is the maximum is displayed, and the evaluation value of the created sentence based on the evaluation value of each word of the created sentence. Is displayed. This enables not only evaluation but also feedback learning to the user, such as giving advice to the user for better translation.

In the created sentence evaluation device 1, the created sentence input unit 15 inputs each word of the created sentence in real time into the decoder input unit 16 and the evaluation unit 17 without waiting for the user to finish inputting all the words. It may be output to. Then, the output by the output unit 14 may be performed in real time each time the evaluation result is output from the evaluation unit 17. As a result, the evaluation and the optimum word at that time are displayed while the user is inputting the created sentence, so that the evaluation and the optimum word are displayed after a long time has passed after inputting all the created sentences. It can play a role as a learning tool with a high learning effect for the user.

[Second Embodiment]
FIG. 6 is a diagram showing an example of a functional configuration (functional block) of the created text evaluation device 1A according to the second embodiment. As shown in FIG. 6, the created sentence evaluation device 1A includes a neural network unit 10, a neural network unit 10A, a bilingual corpus storage unit 11A, a problem sentence preparation unit 12, an encoder input unit 13A, an output unit 14A (output unit), and a created sentence. It includes an input unit 15, a decoder input unit 16A (decoder input unit), an evaluation unit 17A (evaluation unit, additional evaluation unit), and a reception unit 18A (reception unit). Further, the neural network unit 10 includes an encoder unit 100 and a decoder unit 101. The neural network unit 10A includes an encoder unit 100A and a decoder unit 101A.

Hereinafter, the main differences between the created text evaluation device 1A and the created text evaluation device 1 according to the first embodiment will be described, and the description of the same contents will be omitted as appropriate.

The neural network unit 10A, the encoder unit 100A, and the decoder unit 101A are substantially the same as the neural network unit 10, the encoder unit 100, and the decoder unit 101, respectively. The main difference is that the decoder unit 101A, like a general decoder, sets the word candidates of the second language sentence corresponding to the sentence of the first language and the likelihood of the word candidates from the first word to the last word of the sentence. It is a point to output sequentially in the order of words toward the words (hereinafter referred to as "normal order").

The bilingual corpus storage unit 11A is substantially the same as the bilingual corpus storage unit 11. The main difference is that the bilingual corpus storage unit 11A corresponds to the (arbitrary) sentence of the first language and the sentence of the first language (translated the sentence of the first language) in addition to the reverse bilingual corpus. The point is to store a forward-translated corpus, which is a bilingual corpus composed of pairs with sentences in a second language (words are in normal order). The above-mentioned neural network unit 10A may be learned based on a forward-order bilingual corpus.

The encoder input unit 13A is substantially the same as the encoder input unit 13. The main difference is that the encoder input unit 13A further inputs the problem sentence output from the problem sentence preparation unit 12 to the encoder unit 100A.

The output unit 14A will be described later.

The decoder input unit 16A is substantially the same as the decoder input unit 16. The main difference is that the decoder input unit 16A sequentially inputs the words of the created sentence into the decoder unit 101A in the normal order, and the decoder unit 101A replaces the word candidates sequentially output by the decoder unit 101A. As mentioned above, the points of the created sentences are input in order.

The evaluation unit 17A is substantially the same as the evaluation unit 17. The main difference is that the evaluation unit 17A evaluates the word of the created sentence based on the word candidate of the second language output by the decoder unit 101A based on the input by the decoder input unit 16A and the likelihood of the word candidate (score). The point where the evaluation result is output to the output unit 14A, and the second language word output by the decoder unit 101A with respect to the input word which is the word of the created sentence input by the decoder input unit 16A. The point is to evaluate the next word in the correct order of the input word in the created sentence based on the candidate and the likelihood of the word candidate. That is, the evaluation unit 17A evaluates the words in the created sentence in the normal order (word order from the first word to the last word in the created sentence).

The reception unit 18A either performs an evaluation based on the evaluation by the evaluation unit 17A based on the output of the decoder unit 101 (hereinafter referred to as "reverse order evaluation") (hereinafter referred to as "reverse order output"), or the decoder unit 101A. It accepts the user's selection of whether to perform an output (hereinafter referred to as "normal order output") based on an evaluation by the evaluation unit 17 based on the output (hereinafter referred to as "normal order evaluation"). For example, the output unit 14A displays a radio button capable of selecting either reverse order output or forward order output via an output device 1006 (for example, a screen) described later, and when the user selects either one, the reception unit 18A Acquires selection information regarding the selection.

The output unit 14A is substantially the same as the output unit 14. The main difference is that the output unit 14A performs an output (normal order output) based on the evaluation (normal order evaluation) by the evaluation unit 17 based on the output of the decoder unit 101A, which is higher than the reverse order evaluation and the forward order evaluation. The point of outputting based on the evaluation, the point of outputting based on the forward evaluation when the reverse evaluation does not meet the predetermined criteria, or the point of outputting based on the reverse evaluation when the forward evaluation does not meet the predetermined criteria. And, the output (reverse order output or forward order output) according to the user's selection (selection information) received by the reception unit 18A is performed.

For example, the output unit 14A may output the sum or average of the evaluation value of the forward evaluation and the evaluation value of the reverse order evaluation. For example, the output unit 14A may output the evaluation value of the reverse order evaluation if the evaluation value of the reverse order evaluation is higher than the evaluation value of the forward order evaluation, or the evaluation value of the reverse order evaluation is higher than the evaluation value of the normal order evaluation. If it is low, the evaluation value of the forward evaluation may be output. For example, the output unit 14A may output based on the forward evaluation as long as words having an evaluation value lower than a predetermined threshold are included in a predetermined ratio or more behind the center of the created sentence in the reverse evaluation. For example, the output unit 14A may output based on the reverse order evaluation as long as a predetermined ratio or more of words whose evaluation value is lower than a predetermined threshold value is included in front of the center of the created sentence in the forward order evaluation. For example, the output unit 14A may output in reverse order when the reception unit 18A accepts the selection to output in reverse order, and may output in forward order when the reception unit 18A accepts the selection to output in forward order. good.

Next, the operation and effect of the created sentence evaluation device 1A according to the second embodiment will be described.

According to the created sentence evaluation device 1A, the evaluation unit 17A evaluates the words in the created sentence in the normal order. This makes it possible to evaluate from the front of the created text.

Further, according to the created sentence evaluation device 1A, the output unit 14A further outputs based on the normal order evaluation (in addition to the reverse order evaluation). As a result, the forward evaluation can be reflected in the output result of the created sentence evaluation device 1, so that more accurate output can be performed.

Further, according to the created sentence evaluation device 1A, the output unit 14A outputs based on the higher evaluation of the reverse order evaluation and the forward order evaluation. As a result, more accurate output can be performed, and at the same time, it is possible to output by making the best use of the created text created by the user.

Further, according to the created sentence evaluation device 1A, the output unit 14A outputs based on the forward evaluation when the reverse evaluation does not meet the predetermined criteria, or when the forward evaluation does not meet the predetermined criteria. Output based on reverse evaluation. As a result, more accurate output can be performed, and at the same time, it is possible to output by making the best use of the created text created by the user.

Further, according to the created sentence evaluation device 1A, the reception unit 18A accepts the user's choice of whether to output based on the reverse order evaluation or the output based on the forward order evaluation, and the output unit 14A receives the reception unit. Output is performed according to the user's selection accepted by 18A. This improves user convenience.

As described above, it can be said that the created sentence evaluation device 1 or the created sentence evaluation device 1A executes the correction method from the rear of the sentence in the English composition AI scoring. That is, it is possible to score and correct not only from the beginning of the sentence but also from the back. If scoring and correction can be done from the back of the text, it will be possible to correct only the front even if the answer is wrong only in the front. As a result, the user's answer can be utilized. In the created sentence evaluation device 1 or the created sentence evaluation device 1A, for example, at the time of learning, a model (neural network unit 10) is generated in which the words of the correct sentence (English sentence) are input in the reverse order and the English sentence is output in the reverse order. I will do it. If user answers are entered in the model in reverse order, scoring and correction can be performed from the rear.

In the created sentence evaluation device 1A, for example, scoring / correction can be performed by fusing scoring from the front and scoring from the back. The created sentence evaluation device 1A may score from both the front and the back, and the one with the higher score may be adopted. The created sentence evaluation device 1A may try both forward correction and backward correction, and may correct by the one with higher reliability. The created sentence evaluation device 1A scores from the front, and if there is an error only in the front, it may be graded / corrected from the back, or if it scores from the back and there is an error only in the back. , You may score and correct from the front. The created sentence evaluation device 1A may also allow the user to select forward correction or backward correction. For example, if the user's answer to the question sentence "That is, it may improve the student's grade" is "Because, there is a possibility that the student's grade will improve." When scoring from the front, it gets "5.92" out of a perfect score of "10.00", and when scoring from the back, it gets "7.00" out of a perfect score of "10.00". Has a higher evaluation value.

The following sentence scoring correction system is shown as a modification of the created sentence evaluation device 1 or the created sentence evaluation device 1A.

It is a neural network of an encoder / decoder model, and in a network in which sentences in the first language are input to the encoder and sentences in the second language to be translated are output from the decoder, sentences in which the order of sentences in the second language is reversed. It is characterized by training to output from the decoder, and using the trained model, input the sentence of the first language to the encoder and reverse the word order of the translation to the second language. A sentence scoring correction system characterized by scoring by comparing the output word of the decoder with the word of the translated sentence while inputting it to the decoder, and making corrections by replacing it with the output word of the decoder.

In the above sentence scoring correction system, scoring is performed by both the scoring system that scores and corrects sentences from the front and the scoring system that scores and corrects the above sentences from the back, and the scoring result and correction of the higher score A sentence scoring correction system characterized by displaying the results.

In the above sentence scoring correction system, if the sentence is scored from the front and there is an error only in the front of the sentence, the above sentence is scored / corrected again with the scoring system that scores / corrects from the back, and from the back. A text scoring correction system characterized by displaying scoring / correction results.

The block diagram used in the description of the above embodiment shows a block of functional units. These functional blocks (components) are realized by any combination of at least one of hardware and software. Further, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using one physically or logically coupled device, or two or more physically or logically separated devices can be directly or indirectly (eg, for example). , Wired, wireless, etc.) and may be realized using these plurality of devices. The functional block may be realized by combining the software with the one device or the plurality of devices.

Functions include judgment, decision, judgment, calculation, calculation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, solution, selection, selection, establishment, comparison, assumption, expectation, and assumption. There are broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating, mapping, assigning, etc., but only these. I can't. For example, a functional block (component) that makes transmission function is called a transmitting unit or a transmitter. In each case, as described above, the realization method is not particularly limited.

For example, the created text evaluation device 1 and the created text evaluation device 1A in one embodiment of the present disclosure may function as a computer that processes the text evaluation method of the present disclosure. FIG. 7 is a diagram showing an example of the hardware configuration of the created text evaluation device 1 and the created text evaluation device 1A according to the embodiment of the present disclosure. The above-mentioned created sentence evaluation device 1 and created sentence evaluation device 1A are physically 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. May be done.

In the following description, the word "device" can be read as a circuit, a device, a unit, or the like. The hardware configuration of the created text evaluation device 1 and the created text evaluation device 1A may be configured to include one or more of the devices shown in the figure, or may be configured not to include some of the devices. good.

Each function in the created text evaluation device 1 and the created text evaluation device 1A is performed by the processor 1001 by loading predetermined software (program) on the hardware such as the processor 1001 and the memory 1002, and is performed by the communication device 1004. It is realized by controlling communication and controlling at least one of reading and writing of data in the memory 1002 and the storage 1003.

Processor 1001 operates, for example, an operating system to control the entire computer. The processor 1001 may be configured by a central processing unit (CPU) including an interface with a peripheral device, a control device, an arithmetic unit, a register, and the like. For example, the above-mentioned problem sentence preparation unit 12, encoder input unit 13, encoder input unit 13A, output unit 14, output unit 14, created text input unit 15, decoder input unit 16, decoder input unit 16A, evaluation unit 17, evaluation unit. The 17A, the reception unit 18A, and the like may be realized by the processor 1001.

Further, the processor 1001 reads a program (program code), a software module, data, and the like from at least one of the storage 1003 and the communication device 1004 into the memory 1002, and executes various processes according to these. As the program, a program that causes a computer to execute at least a part of the operations described in the above-described embodiment is used. For example, the problem sentence preparation unit 12, the encoder input unit 13, the encoder input unit 13A, the output unit 14, the output unit 14, the created text input unit 15, the decoder input unit 16, the decoder input unit 16A, the evaluation unit 17, the evaluation unit 17A, and the like. The reception unit 18A may be realized by a control program stored in the memory 1002 and operating in the processor 1001, and may be realized in the same manner for other functional blocks. Although it has been described that the various processes described above 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. The program may be transmitted from the network via a telecommunication line.

The memory 1002 is a computer-readable recording medium, and is composed of at least one such as a ROM (Read Only Memory), an EPROM (Erasable Programmable ROM), an EEPROM (Electrically Erasable Programmable ROM), and a RAM (Random Access Memory). May be done. The memory 1002 may be referred to as a register, a cache, a main memory (main storage device), or the like. The memory 1002 can store a program (program code), a software module, or the like that can be executed to implement the wireless communication method according to the embodiment of the present disclosure.

The storage 1003 is a computer-readable recording medium, and is, for example, an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, an optical magnetic disk (for example, a compact disk, a digital versatile disk, or a Blu-ray). It may consist of at least one (registered trademark) disk), smart card, flash memory (eg, card, stick, key drive), floppy (registered trademark) disk, magnetic strip, and the like. The storage 1003 may be referred to as an auxiliary storage device. The storage medium described above may be, for example, a database, server or other suitable medium containing at least one of the memory 1002 and the storage 1003.

The communication device 1004 is hardware (transmission / reception device) for communicating 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, or the like. The communication device 1004 includes, for example, a high frequency switch, a duplexer, a filter, a frequency synthesizer, and the like in order to realize at least one of FDD (Frequency Division Duplex) and Time Division Duplex (TDD). It may be composed of. For example, the above-mentioned output unit 14, output unit 14A, created text input unit 15, reception unit 18A, and the like may be realized by the communication device 1004.

The input device 1005 is an input device (for example, a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that receives an input from the outside. The output device 1006 is an output device (for example, a display, a speaker, an LED lamp, etc.) that performs output to the outside. The input device 1005 and the output device 1006 may have an integrated configuration (for example, a touch panel).

Further, 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 by using a single bus, or may be configured by using a different bus for each device.

Further, the created sentence evaluation device 1 and the created sentence evaluation device 1A include a microprocessor, a digital signal processor (DSP: Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array). ) And the like, and a part or all of each functional block may be realized by the hardware. For example, processor 1001 may be implemented using at least one of these hardware.

The notification of information is not limited to the embodiments / embodiments described in the present disclosure, and may be performed by other methods.

Each aspect / embodiment described in the present disclosure includes LTE (Long Term Evolution), LTE-A (LTE-Advanced), SUPER 3G, IMT-Advanced, 4G (4th generation mobile communication system), and 5G (5th generation mobile communication). system), FRA (Future Radio Access), NR (new Radio), W-CDMA (registered trademark), GSM (registered trademark), CDMA2000, UMB (Ultra Mobile Broadband), IEEE 802.11 (Wi-Fi (registered trademark)) )), LTE 802.16 (WiMAX®), IEEE 802.20, UWB (Ultra-WideBand), Bluetooth®, and other systems that utilize and extend based on these. It may be applied to at least one of the next generation systems. Further, a plurality of systems may be applied in combination (for example, a combination of at least one of LTE and LTE-A and 5G).

The order of the processing procedures, sequences, flowcharts, etc. of each aspect / embodiment described in the present disclosure may be changed as long as there is no contradiction. For example, the methods described in the present disclosure present elements of various steps using exemplary order, and are not limited to the particular order presented.

The input / output information and the like may be stored in a specific place (for example, a memory), or may be managed using a management table. Information to be input / output may be overwritten, updated, or added. The output information and the like may be deleted. The input information or the like may be transmitted to another device.

The determination may be made by a value represented by 1 bit (0 or 1), by a boolean value (Boolean: true or false), or by comparing numerical values (for example, a predetermined value). It may be done by comparison with the value).

Each aspect / embodiment described in the present disclosure may be used alone, in combination, or may be switched and used according to the execution. Further, the notification of predetermined information (for example, the notification of "being X") is not limited to the explicit one, but is performed implicitly (for example, the notification of the predetermined information is not performed). May be good.

Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described in the present disclosure. The present disclosure may be implemented as amendments and modifications without departing from the spirit and scope of the present disclosure as determined by the description of the scope of claims. Therefore, the description of this disclosure is for purposes of illustration and does not have any limiting meaning to this disclosure.

Software, whether referred to as software, firmware, middleware, microcode, hardware description language, or other names, is an instruction, instruction set, code, code segment, program code, program, subprogram, software module. , Applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, features, etc. should be broadly interpreted.

Further, software, instructions, information and the like may be transmitted and received via a transmission medium. For example, a website that uses at least one of wired technology (coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL), etc.) and wireless technology (infrared, microwave, etc.). When transmitted from a server or other remote source, at least one of these wired and wireless technologies is included within the definition of transmission medium.

The information, signals, etc. described in the present disclosure may be represented using any of a variety of different techniques. For example, data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description are voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, light fields or photons, or any of these. It may be represented by a combination of.

The terms described in the present disclosure and the terms necessary for understanding the present disclosure may be replaced with terms having the same or similar meanings.

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

Further, the information, parameters, etc. described in the present disclosure may be expressed using absolute values, relative values from predetermined values, or using other corresponding information. It may be represented. For example, the radio resource may be one indicated by an index.

The names used for the parameters described above are not limited in any respect. Further, mathematical formulas and the like using these parameters may differ from those expressly disclosed in this disclosure.

The terms "determining" and "determining" used in this disclosure may include a wide variety of actions. "Judgment" and "decision" are, for example, judgment (judging), calculation (calculating), calculation (computing), processing (processing), derivation (deriving), investigation (investigating), search (looking up, search, inquiry). It may include (eg, searching in a table, database or another data structure), ascertaining as "judgment" or "decision". Also, "judgment" and "decision" are receiving (for example, receiving information), transmitting (for example, transmitting information), input (input), output (output), and access. (Accessing) (for example, accessing data in memory) may be regarded as "judgment" or "decision". In addition, "judgment" and "decision" are considered to be "judgment" and "decision" when the things such as solving, selecting, selecting, establishing, and comparing are regarded as "judgment" and "decision". Can include. That is, "judgment" and "decision" may include considering some action as "judgment" and "decision". Further, "judgment (decision)" may be read as "assuming", "expecting", "considering" and the like.

The terms "connected", "coupled", or any variation thereof, mean any direct or indirect connection or connection between two or more elements and each other. It can include the presence of one or more intermediate elements between two "connected" or "combined" elements. The connection or connection between the elements may be physical, logical, or a combination thereof. For example, "connection" may be read as "access". As used in the present disclosure, the two elements use at least one of one or more wires, cables and printed electrical connections, and as some non-limiting and non-comprehensive examples, the radio frequency domain. Can be considered to be "connected" or "coupled" to each other using electromagnetic energy having wavelengths in the microwave and light (both visible and invisible) regions.

The phrase "based on" as used in this disclosure does not mean "based on" unless otherwise stated. In other words, the statement "based on" means both "based only" and "at least based on".

Any reference to elements using designations such as "first" and "second" as used in this disclosure does not generally limit the quantity or order of those elements. These designations can be used in the present disclosure as a convenient way to distinguish between two or more elements. Therefore, references to the first and second elements do not mean that only two elements can be adopted, or that the first element must somehow precede the second element.

The "means" in the configuration of each of the above devices may be replaced with a "part", a "circuit", a "device" and the like.

When "include", "including" and variations thereof are used in the present disclosure, these terms are as inclusive as the term "comprising". Is intended. Moreover, the term "or" used in the present disclosure is intended not to be an exclusive OR.

In the present disclosure, if articles are added by translation, for example a, an and the in English, the disclosure may include the plural nouns following these articles.

In the present disclosure, the term "A and B are different" may mean "A and B are different from each other". The term may mean that "A and B are different from C". Terms such as "separate" and "combined" may be interpreted in the same way as "different".

1.1A ... Creation text evaluation device, 10.10A ... Neural network unit, 11.11A ... Bilingual corpus storage unit, 12 ... Problem text preparation unit, 13 / 13A ... Encoder input unit, 14.14A ... Output unit, 15 ... Created text input unit, 16.16A ... Decoder input unit, 17.17A ... Evaluation unit, 18A ... Reception unit, 100 / 100A ... Encoder unit, 101 / 101A ... Decoder unit.

Claims (10)

The encoder inputs a sentence in the first language, and the decoder sets the word candidates of the sentence in the second language corresponding to the sentence in the first language and the likelihood of the word candidates in order from the first word to the last word of the sentence. It is a created sentence evaluation device that uses the recurrent neural network of the encoder / decoder model to sequentially output in a predetermined word order that is different from the word order to which the word is directed.
A decoder input unit that sequentially inputs words of a created sentence created in a second language into the decoder in the predetermined word order, and a decoder input unit.
An evaluation unit that evaluates the words in the created sentence based on the word candidates of the second language output by the decoder based on the input by the decoder input unit and the likelihood of the word candidates.
An output unit that outputs based on the evaluation by the evaluation unit, and an output unit.
Created text evaluation device equipped with. The decoder input unit sequentially inputs the words of the created sentence into the decoder in the predetermined word order instead of the word candidates sequentially output by the decoder.
The created sentence evaluation device according to claim 1. The evaluation unit is based on the word candidate of the second language output by the decoder and the likelihood of the word candidate with respect to the input word which is the word of the created sentence input by the decoder input unit. Evaluates the next word in the predetermined word order of the input word in
The created sentence evaluation device according to claim 1 or 2. The created sentence is a sentence created by the user translating the sentence of the first language input to the encoder into the second language.
The created sentence evaluation device according to any one of claims 1 to 3. The recurrent neural network is learned based on a corpus of a set of sentences in a first language and sentences in a second language corresponding to the sentences in the first language in the predetermined word order.
The created sentence evaluation device according to any one of claims 1 to 4. The predetermined word order is a word order from the last word to the first word in a sentence.
The created sentence evaluation device according to any one of claims 1 to 5. Further equipped with an additional evaluation unit that evaluates the words of the created sentence in the order of words from the first word to the last word of the created sentence.
The output unit further outputs based on the evaluation by the additional evaluation unit.
The created sentence evaluation device according to any one of claims 1 to 6. The output unit outputs based on the higher evaluation of the evaluation by the evaluation unit and the evaluation by the additional evaluation unit.
The created sentence evaluation device according to claim 7. Further equipped with an additional evaluation unit that evaluates the words of the created sentence in the order of words from the first word to the last word of the created sentence.
The output unit outputs based on the evaluation by the additional evaluation unit when the evaluation by the evaluation unit does not meet the predetermined standard, or the evaluation when the evaluation by the additional evaluation unit does not meet the predetermined standard. Output based on evaluation by the department,
The created sentence evaluation device according to any one of claims 1 to 6. An additional evaluation unit that evaluates the words in the created sentence in the order of words from the first word to the last word in the created sentence.
A reception unit that accepts the user's choice of whether to output based on the evaluation by the evaluation unit or output based on the evaluation by the additional evaluation unit.
Further prepare
The output unit outputs according to the user's selection accepted by the reception unit.
The created sentence evaluation device according to any one of claims 1 to 6.

Images