Natural language processing

1 . A computer-implemented method comprising:

receiving first input audio data representing a first spoken input;

performing automatic speech recognition (ASR) processing using the first input audio data to generate a first transcript of the first spoken input;

receiving first context data associated with the first spoken input, wherein the first context data is determined by a first component, and includes a first value;

based at least in part on the first context data, determining a first component description associated with the first component;

determining the first component description indicates the first value is to be modified prior to being processed by a first language model;

based at least in part on determining the first component description indicates the first value is to be modified, generating a first modified value;

determining second context data corresponding to the first context data, wherein the second context data includes the first modified value, instead of the first value;

determining first data associating the first value and the first modified value;

processing the first transcript and the second context data to generate a first prompt including the first transcript and the second context data;

processing, using the first language model, the first prompt to generate first model output data indicating a first action to be performed by a second component, wherein the first model output data includes the first modified value;

based at least in part on the first data, processing the first model output data to generate a first request for the second component to perform the first action, wherein the first request includes the first value, instead of the first modified value;

based at least in part on the first request, causing the second component to perform the first action;

receiving, from the second component, first response data associated with performance of the first action;

generating, based at least in part on the first response data, first output data responsive to the first spoken input; and

causing presentation of the first output data.

2 . The computer-implemented method of claim 1 , wherein the first value is associated with a first identifier type and profile data associated with the first spoken input, and the method further comprises:

using the profile data, determining a name associated with the first value;

based at least in part on the first identifier type, determining a first mapping representing an association between a previous value and a previous modified value;

based at least in part on the name and the previous modified value, determining the first modified value, wherein the first modified value is a natural language description of the first value; and

generating a second mapping representing an association between the first value and the first modified value.

3 . The computer-implemented method of claim 1 , wherein the first response data includes a second value and the method further comprises:

based at least in part on the first response data, determining a second component description associated with the second component;

determining the second component description indicates the second value is to be modified;

based at least in part on determining the second component description indicates the second value is to be modified, generating a second modified value;

determining second response data corresponding to the first response data, wherein the second response data includes the second modified value, instead of the second value; and

generating a mapping representing the first value and the first modified value are associated, wherein generating, based at least in part on the first response data, the first output data responsive to the first spoken input further comprises processing the second response data.

4 . The computer-implemented method of claim 1 , wherein the first value is associated with a first identifier type and the first context data includes a component identifier associated with a third component that generated the first context data, and the method further comprises:

determining the component identifier corresponds to the first component description;

based at least in part on the component identifier corresponding to the first component description, retrieve the first component description from a storage; and

processing the first value and the first component description to determine the first component description includes the first identifier type, wherein determining the first component description indicates the first value is to be modified is based at least in part on determining the first component description includes the first identifier type.

5 . A computer-implemented method comprising:

receiving first input data;

receiving first context data associated with the first input data, wherein the first context data is determined by a first component, the first context data including a first value;

determining a component description associated with the first component;

determining the component description indicates that the first value is to be modified prior to processing by a first language model;

determining second context data corresponding to the first context data, wherein the second context data includes a first modified value instead of the first value;

processing, using the first language model, the first input data and the second context data to determine first data indicating a first action to be performed by a second component, wherein the first action is to be performed using the first value;

based at least in part on the first data, determining second data representing the first action, wherein the second data includes the first value;

based at least in part on the second data, receiving, from the second component, first response data associated with performance of the first action;

generating, based at least in part on the first response data, first output data corresponding to the first input data; and

causing presentation of the first output data.

6 . The computer-implemented method of claim 5 , wherein the first value is associated with a first identifier type and profile data associated with the first input data, and the method further comprises:

using the profile data, determining natural language data associated with the first value;

based at least in part on the first identifier type, determining a first mapping representing an association between a previous value and a previous modified value;

based at least in part on the natural language data and the previous modified value, determining the first modified value, wherein the first modified value is a natural language description of the first value; and

generating a second mapping representing an association between the first value and the first modified value.

7 . The computer-implemented method of claim 5 , further comprising:

processing the second data to generate a first executable application programming interface (API) call usable to cause the first component to perform the first action, the first executable API call including the first value, wherein receiving the first response data associated with performance of the first action based at least in part on the second data comprises processing the first executable API call.

8 . The computer-implemented method of claim 5 , wherein the first response data includes a second value and the method further comprises:

based at least in part on the first response data, determining third data associated with the first component, the third data indicating the second value is to be modified;

based at least in part on the third data indicating the second value is to be modified, generating a second modified value;

determining second response data corresponding to the first response data, wherein the second response data includes the second modified value, instead of the second value; and

generating a mapping representing the second value and the second modified value are associated, wherein generating, based at least in part on the first response data, the first output data corresponding to the first input data comprises processing the second response data.

9 . The computer-implemented method of claim 8 , wherein the first input data includes a first user input, and the method further comprises:

determining the second response data is responsive to the first user input, wherein generating the first output data is based at least in part on determining the second response data is responsive to the first user input;

based at least in part on determining the second response data is responsive to the first user input, using the third data to determine the second value;

generating third response data corresponding to the second response data, wherein the third response data includes the second value, instead of the second modified value; and

based at least in part on the third response data, causing the first component to perform the first action.

10 . The computer-implemented method of claim 5 , wherein the first value is associated with a first identifier type, and the method further comprises:

determining third data including one or more identifier types associated with values to be modified; and

processing the first value and the third data to determine the third data includes the first identifier type, wherein determining that the first value is to be modified prior to processing by the first language model is based at least in part on determining the third data includes the first identifier type.

11 . The computer-implemented method of claim 5 , wherein the first value is associated with a first identifier type and the first context data includes a component identifier associated with the first component, and the method further comprises:

determining the component identifier corresponds to the component description;

based at least in part on the component identifier corresponding to the component description, retrieving the component description from a storage;

processing the first value and the component description to determine the component description includes the first identifier type; and

based at least in part on determining the component description includes the first identifier type, determining the component description indicates the first value is to be modified.

12 . The computer-implemented method of claim 5 , wherein the first language model is a large language model (LLM), and the method further comprises:

processing the second context data to generate a first prompt including the second context data, wherein the first prompt is a first instruction for the LLM to determine a component to process with respect to the first input data; and

processing, using the LLM, the first prompt to generate the first data.

13 . A computing system comprising:

at least one processor; and

at least one memory including instructions, that when executed by the at least one processor, cause the computing system to:

receive first input data;

receive first context data associated with the first input data, wherein the first context data is determined by a first component, the first context data including a first value;

determine a component description associated with the first component;

determine the component description indicates that the first value is to be modified prior to processing by a first language model;

determine second context data representing the first context data, wherein the second context data includes a first modified value instead of the first value;

process, using the first language model, the first input data and the second context data to determine first data indicating a first action to be performed by a second component, wherein the first action is to be performed using the first value;

based at least in part on the first data, determine second data representing the first action, wherein the second data includes the first value;

based at least in part on the second data, receive, from the second component, first response data associated with performance of the first action;

generate, based at least in part on the first response data, first output data corresponding to the first input data; and

cause presentation of the first output data.

14 . The computing system of claim 13 , wherein the first value is associated with a first identifier type and profile data associated with the first input data, and wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

using the profile data, determine natural language data associated with the first value;

based at least in part on the first identifier type, determine a first mapping representing an association between a previous value and a previous modified value;

based at least in part on the natural language data and the previous modified value, determine the first modified value, wherein the first modified value is a natural language description of the first value; and

generate a second mapping representing an association between the first value and the first modified value.

15 . The computing system of claim 13 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

process the second data to generate a first executable application programming interface (API) call usable to cause the first component to perform the first action, the first executable API call including the first value, wherein receiving the first response data associated with performance of the first action based at least in part on the second data comprises processing the first executable API call.

16 . The computing system of claim 13 , wherein the first response data includes a second value and wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

based at least in part on the first response data, determine third data associated with the first component, the third data indicating the second value is to be modified;

based at least in part on the second data indicating the second value is to be modified, generate a second modified value;

determine second response data corresponding to the first response data, wherein the second response data includes the second modified value, instead of the second value; and

generate a mapping representing the second value and the second modified value are associated, wherein generating, based at least in part on the first response data, the first output data corresponding to the first input data comprises processing the second response data.

17 . The computing system of claim 16 , wherein the first input data includes a first user input, wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

determine the second response data is responsive to the first user input, wherein generating the first output data is based at least in part on determining the second response data is responsive to the first user input;

based at least in part on determining the second response data is responsive to the first user input, use the third data to determine the second value;

generate third response data corresponding to the second response data, wherein the third response data includes the second value, instead of the second modified value; and

based at least in part on the third response data, cause the first component to perform the first action.

18 . The computing system of claim 13 , wherein the first value is associated with a first identifier type, and wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

determine third data including one or more identifier types associated with values to be modified; and

process the first value and the third data to determine the third data includes the first identifier type, wherein determining that the first value is to be modified prior to processing by the first language model is based at least in part on determining the third data includes the first identifier type.

19 . The computing system of claim 13 , wherein the first value is associated with a first identifier type and the first context data includes a component identifier associated with the first component, and wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

determine the component identifier corresponds to the component description;

based at least in part on the component identifier corresponding to the component description, retrieve the component description from a storage;

process the first value and the component description to determine the component description includes the first identifier type; and

based at least in part on determining the component description includes the first identifier type, determine the component description indicates the first value is to be modified.

20 . The computing system of claim 13 , wherein the first language model is a large language model (LLM), and wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

process the second context data to generate a first prompt including the second context data, wherein the first prompt is a first instruction for the LLM to determine a component to process with respect to the first input data; and

process, using the LLM, the first prompt to generate the first data.