Sentence evaluation apparatus and sentence evaluation method
View Patent ↗A sentence evaluation apparatus evaluates a sentence which is input. The sentence evaluation apparatus includes an acquisition device and a processor. The acquisition device acquires information indicating a first input sentence and information indicating a second input sentence. The processor executes information processing on the information acquired by the acquisition device, using an algorithm based on machine learning. The processor includes a first encoder that recognizes the first input sentence and a second encoder that recognizes the second input sentence, in the algorithm based on the machine learning. The processor generates evaluation information indicating evaluation on the first input sentence with reference to the second input sentence, based on a result of recognition by the first encoder on the first input sentence and a result of recognition by the second encoder on the second input sentence.
1. A sentence evaluation apparatus for evaluating a sentence which is input, the sentence evaluation apparatus comprising:
an acquisition device that acquires information indicating a first input sentence and information indicating a second input sentence; and
a processor that executes information processing on the information acquired by the acquisition device, using an algorithm based on machine learning,
wherein
the processor includes a first encoder that recognizes the first input sentence and a second encoder that recognizes the second input sentence, in the algorithm based on the machine learning, and
the processor generates evaluation information indicating evaluation on the first input sentence with reference to the second input sentence, based on a result of recognition by the first encoder on the first input sentence and a result of recognition by the second encoder on the second input sentence.
2. The sentence evaluation apparatus according to claim 1 , wherein the first and second encoders perform machine learning based on different learning parameters.
3. The sentence evaluation apparatus according to claim 1 , wherein
the first encoder
performs recognition, based on a predetermined data structure including a plurality of elements, on each element of the data structure corresponding to the first input sentence, and
refers to the result of the recognition by the second encoder on the second input sentence, to determine an importance of each element in the first input sentence with reference to the second input sentence.
4. The sentence evaluation apparatus according to claim 3 , wherein the processor generates the evaluation information indicating the evaluation so as to reflect the determined importance.
5. The sentence evaluation apparatus according to claim 3 , wherein the predetermined data structure includes at least one of a tree structure including a parent node and child nodes, and a data structure including a plurality of tree structures with child nodes overlapping each other.
6. The sentence evaluation apparatus according to claim 1 , wherein
the second encoder
performs recognition, based on a predetermined data structure including a plurality of elements, on each element of the data structure in the second input sentence, and
refers to the result of the recognition by the first encoder on the first input sentence to determine an importance of each element in the second input sentence with reference to the first input sentence.
7. The sentence evaluation apparatus according to claim 1 , wherein the processor further includes a fully connected layer that executes calculation processing for integrating the result of the recognition by the first encoder on the first input sentence and the result of the recognition by the second encoder on the second input sentence, in the algorithm based on the machine learning.
8. The sentence evaluation apparatus according to claim 7 , wherein the fully connected layer executes the calculation processing based on a logistic function.
9. The sentence evaluation apparatus according to claim 1 , wherein
the first input sentence is a translated sentence as a result of machine translation by a translation machine, and
the second input sentence is any one of an original sentence and a reference sentence, the original sentence being a target of the machine translation by the translation machine and the reference sentence being an exemplary sentence to which the original sentence is correctly translated.
10. The sentence evaluation apparatus according to claim 1 , wherein the evaluation information includes information indicating at least one of similarity between the first and the second input sentences, consistency of the first input sentence with reference to the second input sentence, classification of the first input sentence based on a plurality of grades, and a predetermined part in the first input sentence.
11. A sentence evaluation method of evaluating a sentence which is input to a sentence evaluation apparatus, the sentence evaluation method comprising:
acquiring information indicating a first input sentence;
acquiring information indicating a second input sentence;
recognizing the first input sentence by a first encoder based on machine learning;
recognizing the second input sentence by a second encoder that is different from the first encoder; and
generating evaluation information indicating evaluation on the first input sentence with reference to the second input sentence, based on a recognition result obtained by the first encoder for the first input sentence and a recognition result obtained by the second encoder for the second input sentence.