Generation apparatus, generation method and program

1. A generating device, comprising a processor configured to execute operations comprising:

receiving first data, wherein the first data comprises a plurality of letters included in a text string as a generation target; and

generating, based at least on operations performed by a recurrent neural network, second data,

wherein the second data comprises the plurality of letters of the text string,

the second data satisfies a plurality of predetermined constraints,

the plurality of predetermined constraints comprises:

a first predetermined constraint indicating plausibility of the sequence of words as a phrase in the second data based on the plurality of letters of the text string of the first data according to a prediction value,

the recurrent neural network predicts the prediction value, the prediction value represents plausibility of the sequence of words, the plausibility of the sequence of words indicates semantic naturalness of the sequence of words,

the prediction value indicates a conditional probability of the last word in the sequence of words of the phrase in the second data in the sequence of words from a forefront word of the second data to an immediately-prior-to the-last word in the sequence of words of the phrase in the second data,

a second predetermined constraint indicates a number of occurrences of respective letters of the last word of the sequence of words being no greater than a number of occurrences of the respective words of the last word in the text string of the first data,

a third predetermined constraint indicates the last word being distinct from other words in the sequence of words, and

the generating further comprises terminating the depth-first search of the sequence of words based on the plausibility of the sequence of words; and

transmitting at least a part of the second data to an application, wherein the application sequentially outputs the second data as one or more anagrams of the first data, wherein the at least a part of the second data excludes a set of letters of the plurality of letters as an anagram of the first data based on the prediction value.

2. The generating device according to claim 1 , wherein the generating second data further comprises searching for a text string that satisfies the constraint by a depth-first search, and wherein, in the depth-first search, in a case where a letter or a word that does not satisfy the constraints is searched as a letter or word composing the text string, subsequent searching for a letter or a word is not performed after the letter or the word.

3. The generating device according to claim 2 , wherein, on the basis of a number of occurrences of each letter included in a plurality of letters represented by the first data, data relating to the number of occurrences of each letter included in the plurality of letters is taken as first number-of-occurrences data, the generating device further comprising:

referring a vocabulary store in which a plurality of pieces of word data is stored,

acquiring word data of a word regarding which the number of occurrences of each letter included in the word is not more than the number of occurrences of each of the letters represented by the first number-of-occurrences data from the vocabulary store, and

generating data, wherein the data is based on the number of occurrences of each letter included in a word represented by each of word data that is acquired, and the generating the second number-of-occurrences data further comprises searching for a text string satisfying the constraints by the depth-first search, using the first number-of-occurrences data and the second number-of-occurrences data.

4. The generating device according to claim 1 , wherein the constraints indicating plausibility represent constraints based on a value representing plausibility of the sequence of letters, calculated by a predetermined language model or a rule-based technique.

5. The generating device according to claim 1 , wherein the constraints include:

a first constraint for converting a text string represented by the second data into a text string in which a plurality of letters represented by the first data are rearranged, and

a second constraint is based on a number of occurrences of a letter in the first letter.

6. The generating device according to claim 1 , wherein the plurality of letters represented by the first data are at least one of a sentence, word, phrase, clause, set of words, and set of letters and the number of the letters.

7. A computer implemented method for generating text data, comprising:

receiving first data, wherein the first data comprises a plurality of letters included in a text string as a generation target; and

generating, based at least on operations performed by a recurrent neural network second data,

wherein the second data comprises the plurality of letters of the text string,

the second data satisfies constraints comprises:

a first constraint indicating plausibility of the sequence of words as a phrase in the second data based on the plurality of letters of the text string of the first data according to a prediction value,

the recurrent neural network predicts the prediction value, the prediction value represents plausibility of the sequence of words, and the plausibility of the sequence of words indicates the plausibility of the sequence of words indicates semantic naturalness of the sequence of letters,

the prediction value indicates a conditional probability of the last word in the sequence of words of the phrase in the second data in the sequence of words from a forefront word of the second data to an immediately-prior-to the-last word in the sequence of words of the phrase in the second data,

a second constraint indicates a number of occurrences of respective letters of the last word of the sequence of words being no greater than a number of occurrences of the respective words of the last word in the text string of the first data,

a third constraint indicates the last word being distinct from other words in the sequence of words, and

the generating further comprises terminating the depth-first search of the sequence of words based on the plausibility of the sequence of words; and

transmitting at least a part of the second data to an application, wherein the application sequentially outputs the second data as one or more anagrams of the first data, wherein the at least a part of the second data excludes a set of letters of the plurality of letters as an anagram of the first data based on the prediction value.

8. A computer-readable non-transitory recording medium storing computer-executable program instructions that when executed by a processor cause a computer system to:

receiving first data, wherein the first data comprises a plurality of letters included in a text string as a generation target; and

generating, based at least on operations performed by a recurrent neural network, second data,

wherein the second data comprises the plurality of letters of the text string,

the second data satisfies a plurality of predetermined constraints,

the plurality of predetermined constraints comprises:

a first predetermined constraint indicating plausibility of a sequence of words as a phrase in the second data based on the plurality of letters of the text string of the first data according to a prediction value,

the recurrent neural network predicts the prediction value, the prediction value represents plausibility of the sequence of words, and the plausibility of the sequence of words indicates semantic naturalness of the sequence of words, and

the prediction value indicates a conditional probability of the last word in the sequence of words of the phrase in the second data in the sequence of words from a forefront word of the second data to an immediately-prior-to the-last word in the sequence of words of the phrase in the second data,

a second predetermined constraint indicates a number of occurrences of respective letters of the last word of the sequence of words being no greater than a number of occurrences of the respective words of the last word in the text string of the first data,

a third predetermined constraint indicates the last word being distinct from other words in the sequence of words, and

the generating further comprises terminating the depth-first search of the sequence of words, based on the plausibility of the sequence of words; and

transmitting at least a part of the second data to an application, wherein the application sequentially outputs the second data as one or more anagrams of the first data, wherein the at least a part of the second data excludes a set of letters of the plurality of letters as an anagram of the first data based on the prediction value.

9. The generating device according to claim 2 , wherein the constraints indicating plausibility are constraints relating to a value representing plausibility of the sequence of letters, calculated by a predetermined language model or a rule-based technique.

10. The generating device according to claim 2 , wherein the constraints include:

a first constraint for converting a text string represented by the second data into a text string in which a plurality of letters represented by the first data are rearranged, and

a second constraint is based on a number of occurrences of a letter in the first letter.

11. The generating device according to claim 2 , wherein the plurality of letters represented by the first data are at least one of a sentence, word, phrase, clause, set of words, and set of letters and the number of the letters.

12. The generating device according to claim 3 , wherein the constraints indicating are constraints relating to a value representing plausibility of the sequence of letters, calculated by a predetermined language model or a rule-based technique.

13. The generating device according to claim 3 , wherein the constraints include:

a first constraint for converting a text string represented by the second data into a text string in which a plurality of letters represented by the first data are rearranged, and

a second constraint is based on a number of occurrences of a letter in the first letter.

14. The generating device according to claim 3 , wherein the plurality of letters represented by the first data are at least one of a sentence, word, phrase, clause, set of words, and set of letters and the number of the letters.

15. The generating method according to claim 7 , wherein the generating further comprises searching for a text string that satisfies constraints by the depth-first search, and wherein, in the depth-first search, in a case where a letter or a word that does not satisfy the constraints is searched as a letter or word composing the text string, subsequent searching for a letter or a word is not performed after the letter or the word.

16. The generating method according to claim 7 , wherein the constraints indicating plausibility are constraints relating to a value representing plausibility of the sequence of letters, calculated by a predetermined language model or a rule-based technique.

17. The generating method according to claim 7 , wherein the constraints include a constraint for converting a text string represented by the second data into a text string in which a plurality of letters represented by the first data are rearranged.

18. The generating method according to claim 7 , wherein the plurality of letters represented by the first data are at least one of a sentence, word, phrase, clause, set of words, and set of letters and the number of the letters.

19. The generating method according to claim 15 , wherein, on the basis of a number of occurrences of each letter included in a plurality of letters represented by the first data, data relating to the number of occurrences of each letter included in the plurality of letters is taken as first number-of-occurrences data, the processor further configured to execute operations comprising:

referring a vocabulary store in which a plurality of pieces of word data is stored,

acquiring word data of a word regarding which the number of occurrences of each letter included in the word is not more than the number of occurrences of each of the letters represented by the first number-of-occurrences data from the vocabulary store, and

generating second number-of-occurrences data, wherein the second number-of-occurrences data indicates the number of occurrences of each letter included in a word represented by each of word data that is acquired, and the generating the second number-of-occurrences data further comprises searching for a text string satisfying the constraints by the depth-first search, using the first number-of-occurrences data and the second number-of-occurrences data.

20. The computer-readable non-transitory recording medium of claim 8 , wherein the generating further comprises searching for a text string that satisfies constraints by a depth-first search, and

wherein, in the depth-first search, in a case where a letter or a word that does not satisfy the constraints is searched as a letter or word composing the text string,

subsequent searching for a letter or a word is not performed after the letter or the word.