Virtual assistant humor management
In accordance with one disclosed method, first data representing first text may be processed using a first machine learning model configured to generate second data representing an explanation as to why the first text is humorous. The second data may be processed, together with the first data, by a second machine learning model configured to generate a value indicating that the first data and the second data correspond to a pun. A device may be caused to take at least a first action based at least in part on the value.
1 . A computer-implemented method, comprising:
receiving, from a first device, first audio data representing an utterance corresponding to a dialog between the first device and a system;
performing speech processing using the first audio data to determine first data representing first text corresponding to the utterance;
sending, to a dialog management component, the first data;
processing the first data using a text-to-text transfer transformer (T5) model configured to generate output based on input such that the output represents an explanation as to why the input is humorous;
receiving, from the T5 model, second data representing an explanation as to why the first text is humorous;
processing the first data and the second data using a bidirectional encoder representations from transformers (BERT) model configured to classify input text as corresponding to a pun;
determining that the BERT model classified the first data and the second data as corresponding to a pun;
sending an indication of the pun to the dialog management component; and
determining, by the dialog management component, and based at least in part on the first data and the indication of the pun, output data responsive to the utterance.
2 . The computer-implemented method of claim 1 , further comprising, prior to receiving the first audio data:
training the T5 model using training data including at least third data representing a first text prefix corresponding to generation of pun explanations, fourth data representing a first pun, and fifth data representing a first explanation of the first pun, with the fifth data being used to supervise the training.
3 . The computer-implemented method of claim 1 , further comprising, prior to receiving the first audio data:
training the BERT model using training data including at least third data representing second text, fourth data representing an explanation of the second text, and a first value indicating that the second text represents a pun, with the first value being used to supervise the training.
4 . The computer-implemented method of claim 1 , further comprising:
receiving, from the dialog management component, dialog data representing at least one user input and at least one system response corresponding to the dialog;
wherein the output data is determined further based at least in part on the dialog data.
5 . A computer-implemented method, comprising:
receiving first data representing first text of a sentence including a pun;
processing the first data using a first machine learning model to generate second data representing an explanation as to why the sentence includes a pun, wherein the first machine learning model is configured using training data including at least third data representing a first text prefix corresponding to generation of pun explanations, fourth data representing a first pun, and fifth data representing a first explanation of the first pun; and
causing a first device to take at least a first action based at least in part on the second data.
6 . The computer-implemented method of claim 5 , wherein the first machine learning model comprises a text-to-text transfer transformer model.
7 . The computer-implemented method of claim 6 , further comprising:
training the text-to-text transfer transformer model using the training data, wherein the fifth data is used to supervise the training.
8 . The computer-implemented method of claim 5 , wherein:
the first machine learning model is configured to generate output text based on input text such that the output text represents an explanation as to why the input text includes a pun.
9 . A computer-implemented method, comprising:
receiving first data representing first text of a sentence including a pun;
processing the first data using a first machine learning model to generate second data representing an explanation as to why the sentence includes a pun;
processing the first data and the second data using a second machine learning model configured to generate an output value based on input text such that the output value indicates whether the input text correspond to a pun;
receiving, from the second machine learning model, a first value indicating that the first data and the second data correspond to a pun; and
causing a first device to take at least a first action based at least in part on the first value.
10 . The computer-implemented method of claim 9 , wherein the second machine learning model comprises a bidirectional encoder representations from transformers (BERT) model configured to classify the input text as corresponding to a pun or not corresponding to a pun.
11 . The computer-implemented method of claim 10 , wherein the BERT model is configured using training data including at least third data representing second text, fourth data representing an explanation of the second text, and a first value indicating that the second text represents a pun.
12 . The computer-implemented method of claim 11 , further comprising:
training the BERT model using the training data, wherein the first value is used to supervise the training.
13 . The computer-implemented method of claim 9 , further comprising:
receiving, by a dialog component, a first communication corresponding to the first text;
determining the first data based on the first communication;
generating, by the dialog component, a response recognizing that the first communication corresponded to a pun; and
sending a second communication corresponding to the response to the first device to cause the first device to output a representation of the response.
14 . The computer-implemented method of claim 9 , wherein the first machine learning model is configured using training data including at least third data representing a first text prefix corresponding to generation of pun explanations, fourth data representing a first pun, and fifth data representing a first explanation of the first pun.
15 . A system, comprising:
at least one processor; and
at least one memory comprising instructions that, when executed by the at least one processor, cause the system to:
receive first data representing first text of a sentence including a pun;
process the first data using a first machine learning model including a neural network configured to generate second data representing an explanation as to why the sentence includes a pun, wherein the first machine learning model is configured using training data including at least third data representing a first text prefix corresponding to generation of pun explanations, fourth data representing a first pun, and fifth data representing a first explanation of the first pun; and
cause a first device to take at least a first action based at least in part on the second data.
16 . The system of claim 15 , wherein the first machine learning model is configured to generate output text based on input text such that the output text represents an explanation as to why the input text includes a pun.
17 . A system, comprising:
at least one processor; and
at least one memory comprising instructions that, when executed by the at least one processor, cause the system to:
receive first data representing first text of a sentence including a pun;
process the first data using a first machine learning model including a neural network configured to generate second data representing an explanation as to why the sentence includes a pun;
process the first data and the second data using a second machine learning model configured to generate an output value based on input text such that the output value indicates whether the input text correspond to a pun;
receive, from the second machine learning model, a first value indicating that the first data and the second data correspond to a pun; and
cause a first device to take at least a first action based at least in part on the first value.
18 . The system of claim 17 , wherein the second machine learning model comprises a bidirectional encoder representations from transformers (BERT) model configured to classify the input text as corresponding to a pun or not corresponding to a pun.
19 . The system of claim 18 , wherein the BERT model is configured using training data including at least third data representing second text, fourth data representing an explanation of the second text, and a first value indicating that the second text represents a pun.
20 . The system of claim 19 , wherein the at least one memory further comprises additional instructions that, when executed by the at least one processor, cause the system to:
train the BERT model using the training data, wherein the first value is used to supervise the training.
21 . The system of claim 17 , wherein the at least one memory further comprises additional instructions that, when executed by the at least one processor, cause the system to:
receive, by a dialog component, a first communication corresponding to the first text;
determine the first data based on the first communication;
generate, by the dialog component, a response recognizing that the first communication corresponded to a pun; and
send a second communication corresponding to the response to the first device to cause the first device to output a representation of the response.
22 . The system of claim 17 , wherein the first machine learning model is configured using training data including at least third data representing a first text prefix corresponding to generation of pun explanations, fourth data representing a first pun, and fifth data representing a first explanation of the first pun.
23 . The system of claim 22 , wherein the first machine learning model comprises a text-to-text transfer transformer model.
24 . The system of claim 23 , wherein the at least one memory further comprises additional instructions that, when executed by the at least one processor, cause the system to:
train the text-to-text transfer transformer model using the training data, wherein the fifth data is used to supervise the training.