| CPC G10L 15/1815 (2013.01) [G10L 15/063 (2013.01); G10L 2015/0631 (2013.01)] | 20 Claims |
|
1. A computing system, comprising:
at least one memory configured to store a pre-trained autoregressive generative language model without fine-tuning, an encoder language model defining intents and trained to classify which of the intents are expressed in received natural language utterances, and an intent classification dataset having initial intent samples for each of the intents of the encoder language model; and
at least one processor configured to execute stored instructions to cause the computing system to perform actions comprising:
selecting a set of initial intent samples from the intent classification dataset, wherein the set of initial intent samples is associated with at least one intent of the encoder language model;
providing the set of initial intent samples as input to the pre-trained autoregressive generative language model;
receiving a set of generated intent samples associated with the at least one intent as output from the pre-trained autoregressive generative language model;
adding at least a portion of the set of generated intent samples associated with the at least one intent to the intent classification dataset to yield an augmented intent classification dataset; and
fine-tuning the encoder language model using the augmented intent classification dataset.
|
|
12. A method, comprising:
selecting a set of initial intent samples from an intent classification dataset that includes initial intent samples for each of intent of a encoder language model, wherein the encoder language model defines intents and is trained to classify which of the intents are expressed in received natural language utterances, and wherein the set of initial intent samples is associated with at least one intent of the encoder language model;
providing the set of initial intent samples as input to a pre-trained autoregressive generative language model that lacks fine-tuning;
receiving a set of generated intent samples associated with the at least one intent as output from the pre-trained autoregressive generative language model;
adding at least a portion of the set of generated intent samples associated with the at least one intent to the intent classification dataset to yield an augmented intent classification dataset; and
fine-tuning the encoder language model using the augmented intent classification dataset.
|
|
17. A non-transitory, computer-readable medium storing instructions executable by a computer processor, the instructions comprising instructions to:
select a set of initial intent samples from an intent classification dataset that includes initial intent samples for each of intent of a encoder language model, wherein the encoder language model defines intents and is trained to classify which of the intents are expressed in received natural language utterances, and wherein the set of initial intent samples is associated with at least one intent of the encoder language model;
provide the set of initial intent samples as input to a pre-trained autoregressive generative language model that lacks fine-tuning;
receive a set of generated intent samples associated with the at least one intent as output from the pre-trained autoregressive generative language model;
add at least a portion of the set of generated intent samples associated with the at least one intent to the intent classification dataset to yield an augmented intent classification dataset; and
fine-tune the encoder language model using the augmented intent classification dataset.
|