Natural language data generation using automated knowledge distillation techniques

1 . A computer-implemented method comprising:

retrieving, in response to an input query, a set of passages from at least one knowledge base by processing at least a portion of the input query using at least a first set of multiple artificial intelligence techniques, wherein retrieving the set of passages comprises;

identifying at least a given number of the passages, represented in vector form, having at least a designated amount of similarity to one or more parts of the input query represented in vector form by processing at least a portion of the input query using at least one dense passage retrieval model, at least one contextualized late interaction over bidirectional encoder representations from transformers (ColBERT) model, and at least one artificial intelligence-based keyword search model; and

combining, to create the set of passages, one or more portions of the at least a given number of passages identified using each of the at least one dense passage retrieval model, the at least one ColBERT model, and the at least one artificial intelligence-based keyword search model;

ranking at least a portion of the set of passages by (i) generating scores for the at least a portion of the set of passages using at least a second set of one or more artificial intelligence techniques, wherein the scores each comprise a probability of relevance value, trained with at least one cross-entropy loss, for a given passage from the at least a portion of the set of passages, and (ii) ranking the at least a portion of the set of passages in accordance with the scores;

generating at least one natural language answer, in response to the input query, by processing a subset of the set of passages in connection with one or more automated knowledge distillation techniques based at least in part on the ranking of the at least a portion of the set of passages; and

performing one or more automated actions based at least in part on one or more of the ranking of the at least a portion of the set of passages and the at least one generated natural language answer;

wherein the method is carried out by at least one computing device.

2 . The computer-implemented method of claim 1 , wherein processing a subset of the set of passages in connection with one or more automated knowledge distillation techniques based at least in part on the ranking of the at least a portion of the set of passages comprises providing at least a portion of the subset of the set of passages, using the one or more automated knowledge distillation techniques, from the at least a second set of one or more artificial intelligence techniques to the at least a first set of multiple artificial intelligence techniques, in conjunction with the at least a second set of one or more artificial intelligence techniques being trained.

3 . The computer-implemented method of claim 1 , wherein ranking at least a portion of the set of passages comprises processing the set of passages using at least one bidirectional encoder representations from transformers-based model.

4 . The computer-implemented method of claim 1 , wherein generating at least one natural language answer comprises processing the subset of the set of passages using at least one sequence-to-sequence generation model.

5 . The computer-implemented method of claim 4 , wherein the at least one sequence-to-sequence generation model comprises at least one bidirectional auto-regressive transformer.

6 . The computer-implemented method of claim 1 , wherein performing one or more automated actions comprises automatically training the first set of multiple artificial intelligence techniques based at least in part on one or more of the ranking of the at least a portion of the set of passages.

7 . The computer-implemented method of claim 1 , wherein retrieving the set of passages comprises processing at least a portion of the input query using the first set of multiple artificial intelligence techniques in conjunction with at least one approximate nearest neighbors index.

8 . The computer-implemented method of claim 1 , wherein retrieving the set of passages comprises:

encoding at least a portion of the input query into one or more vectors;

identifying each vector from the one or more vectors having an inner product-related value above a given threshold value; and

retrieving a passage, corresponding to each identified vector, from the at least one knowledge base.

9 . The computer-implemented method of claim 1 , wherein performing one or more automated actions comprises outputting the at least one generated natural language answer to at least one user device.

10 . The computer-implemented method of claim 1 , wherein the first set of multiple artificial intelligence techniques and the second set of one or more artificial intelligence techniques each comprises at least one identical artificial intelligence technique.

11 . The computer-implemented method of claim 1 , wherein software implementing the method is provided as a service in a cloud environment.

12 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computing device to cause the computing device to:

retrieve, in response to an input query, a set of passages from at least one knowledge base by processing at least a portion of the input query using at least a first set of multiple artificial intelligence techniques, wherein retrieving the set of passages comprises:

identifying at least a given number of the passages, represented in vector form, having at least a designated amount of similarity to one or more parts of the input query represented in vector form by processing at least a portion of the input query using at least one dense passage retrieval model, at least one contextualized late interaction over bidirectional encoder representations from transformers (ColBERT) model, and at least one artificial intelligence-based keyword search model; and

combining, to create the set of passages, one or more portions of the at least a given number of passages identified using each of the at least one dense passage retrieval model, the at least one ColBERT model, and the at least one artificial intelligence-based keyword search model;

rank at least a portion of the set of passages by (i) generating scores for the at least a portion of the set of passages using at least a second set of one or more artificial intelligence techniques, wherein the scores each comprise a probability of relevance value, trained with at least one cross-entropy loss, for a given passage from the at least a portion of the set of passages, and (ii) ranking the at least a portion of the set of passages in accordance with the scores;

generate at least one natural language answer, in response to the input query, by processing a subset of the set of passages in connection with one or more automated knowledge distillation techniques based at least in part on the ranking of the at least a portion of the set of passages; and

perform one or more automated actions based at least in part on one or more of the ranking of the at least a portion of the set of passages and the at least one generated natural language answer.

13 . The computer program product of claim 12 , wherein ranking at least a portion of the set of passages comprises processing the set of passages using at least one bidirectional encoder representations from transformers-based model.

14 . The computer program product of claim 12 , wherein generating at least one natural language answer comprises processing the subset of the set of passages using at least one sequence-to-sequence generation model.

15 . The computer program product of claim 14 , wherein the at least one sequence-to-sequence generation model comprises at least one bidirectional auto-regressive transformer.

16 . A system comprising:

a memory configured to store program instructions; and

a processor operatively coupled to the memory to execute the program instructions to:

retrieve, in response to an input query, a set of passages from at least one knowledge base by processing at least a portion of the input query using at least a first set of multiple artificial intelligence techniques, wherein retrieving the set of passages comprises:

identifying at least a given number of the passages, represented in vector form, having at least a designated amount of similarity to one or more parts of the input query represented in vector form by processing at least a portion of the input query using at least one dense passage retrieval model, at least one contextualized late interaction over bidirectional encoder representations from transformers (ColBERT) model, and at least one artificial intelligence-based keyword search model; and

combining, to create the set of passages, one or more portions of the at least a given number of passages identified using each of the at least one dense passage retrieval model, the at least one ColBERT model, and the at least one artificial intelligence-based keyword search model;

rank at least a portion of the set of passages by (i) generating scores for the at least a portion of the set of passages using at least a second set of one or more artificial intelligence techniques, wherein the scores each comprise a probability of relevance value, trained with at least one cross-entropy loss, for a given passage from the at least a portion of the set of passages, and (ii) ranking the at least a portion of the set of passages in accordance with the scores;

generate at least one natural language answer, in response to the input query, by processing a subset of the set of passages in connection with one or more automated knowledge distillation techniques based at least in part on the ranking of the at least a portion of the set of passages; and

perform one or more automated actions based at least in part on one or more of the ranking of the at least a portion of the set of passages and the at least one generated natural language answer.

17 . The system of claim 16 , wherein ranking at least a portion of the set of passages comprises processing the set of passages using at least one bidirectional encoder representations from transformers-based model.

18 . The system of claim 16 , wherein generating at least one natural language answer comprises processing the subset of the set of passages using at least one sequence-to-sequence generation model.

19 . The system of claim 18 , wherein the at least one sequence-to-sequence generation model comprises at least one bidirectional auto-regressive transformer.

20 . The system of claim 16 , wherein retrieving the set of passages comprises processing at least a portion of the input query using the first set of multiple artificial intelligence techniques in conjunction with at least one approximate nearest neighbors index.