# US 12,602,209 B1 · Constrained generative framework application development environment > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,602,209 B1 - Document ID: US12602209B1 - Status: Granted - Kind: B1 - Application number: 18/478,034 - Filed: 2023-09-29 - Granted: 2026-04-14 - Earliest claimed: 2023-09-29 - Assignees: Amazon Technologies, Inc. - Applicants: Amazon Technologies, Inc. - Inventors: James Gung; Arshit Gupta; John Baker; Yi Zhang; Saab Mansour; Santosh Kumar Ameti; Ruhaab Markas; Ganesh Kumar Gella; Katrin Kirchhoff - Web page: https://patentgrep.com/patent/US12602209B1 ## CPC - G06F 8/38: for implementing user interfaces - G06N 20/00: Machine learning - G06F 40/30: Semantic analysis - G06F 40/35: Discourse or dialogue representation - G06F 40/56: Natural language generation - G06N 3/02: Neural networks - G06F 9/453: {Help systems} - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Systems and methods provide for modular, iterative framework to resolve inputs provided to an interaction environment. The input may be decomposed into different component parts and then relevant actions for each of the component parts may be predicted. An action may be selected and relevant parameters may be populated based on the input. If parameters remain unresolved, additional queries may be presented in order to resolve the remaining parameters. Multiple actions may be executed and then prepared to generate a combined response responsive to the input. Actions for a given interaction environment may be domain-specific and also may be developer-defined for a given goal or task to restrict one or more underlying language systems. ## Claims ### Claim 1 A computer-implemented method, executed by one or more processors of a computing system comprising a dialogue management framework, the method comprising: receiving, by a dialogue manager, an input to an interaction environment; determining one or more sub-components of the input associated with independent actions for resolution; determining, by a state manager, one or more actions from a set of actions, the one or more actions associated with the one or more sub-components of the input; selecting an action for a selected sub-component of the one or more sub-components; determining, from the input, one or more values corresponding to one or more parameters of the action; determining, one or more additional values corresponding to one or more additional parameters remain unresolved; generating, by an interaction service, a response to the input associated with the one or more additional values; receiving, by the dialogue manager, a reply to the response; determining, from the reply, the one or more additional values; determining each value of the one or more parameters of the action is resolved; and executing, by an action executor of the dialogue management framework, the action using the one or more parameters. ### Claim 2 (depends on 1) The computer-implemented method of claim 1, further comprising: generating a first partial response using the action; selecting a second action for the selected sub-component; executing the second action to generate a second partial response; and generating a full response combining the first partial response and the second partial response. ### Claim 3 (depends on 1) The computer-implemented method of claim 1, further comprising: determining, from the reply, a revised sub-component associated with a second action; halting execution of the action; determining one or more second action values for one or more section action parameters; and executing the second action. ### Claim 4 (depends on 1) The computer-implemented method of claim 1, further comprising: providing a builder environment to create the interaction environment; receiving a developer input including a domain and the set of actions; and generating the interaction environment using the set of actions. ### Claim 5 (depends on 1) The computer-implemented method of claim 1, wherein the set of actions are developer-defined actions to restrict one or more outputs of an underlying generative language model. ### Claim 6 A computer-implemented method, executed by one or more processors of a computing system comprising a dialogue management framework, the method comprising: determining, by a dialogue manager, a first portion and a second portion of an input; predicting, by a state manager, a first relevant action for the first portion, the first relevant action being included within a set of actions; selecting a meta-action, from a pre-defined set of meta-actions, based at least on the first relevant action; causing, by an action executor, the meta-action to execute using one or more parameters of the first relevant action; determining at least some first relevant action values for a set of first action parameters are unresolvable from the input; and generating, by an interaction service, a responsive query directed toward the at least some first relevant action values. ### Claim 7 (depends on 6) The computer-implemented method of claim 6, further comprising: determining, from a second input, a set of unresolved first relevant action values; and executing the first relevant action. ### Claim 8 (depends on 7) The computer-implemented method of claim 7, further comprising: predicting a second relevant action for the second portion, the second relevant action being included within the set of actions; selecting a second meta-action, from the pre-defined set of meta-action, based at least on the second relevant action; determining, from at least one of the input and the second input, a set of unresolved second relevant action values; executing the second relevant action; and generating a response to the input based, at least, on respective outputs of the execution of the first relevant action and the second relevant action. ### Claim 9 (depends on 6) The computer-implemented method of claim 6, wherein at least some actions of the set of actions are developer-defined to constraint an underlying generative language model. ### Claim 10 (depends on 6) The computer-implemented method of claim 6, further comprising: determining an initial first relevant action value from the input; adding the initial first relevant action value to an initial first relevant action parameter; determining a next first relevant action value is unresolved by the input; generating a request for information associated with the next first relevant action value; determining the next first relevant action value after receiving an answer to the request; adding the next first relevant action value to a next first relevant action parameter; and determining a following first relevant action value from the input after adding the next first relevant action value. ### Claim 11 (depends on 6) The computer-implemented method of claim 6, further comprising: predicting a second relevant action for the second portion, the second relevant action being included within the set of actions; resolving a portion of the one or more parameters of the first relevant action; resolving a portion of second relevant action parameters; and resolving remaining first relevant action parameters after resolving the portion of the second relevant action parameters. ### Claim 12 (depends on 6) The computer-implemented method of claim 6, wherein the input is at least one of an auditory input or a textual input. ### Claim 13 (depends on 6) The computer-implemented method of claim 6, wherein the input is received within an interaction environment, the interaction environment comprising a task-oriented interaction environment within a defined domain. ### Claim 14 (depends on 6) The computer-implemented method of claim 6, further comprising: provide a builder environment to create an interaction environment; receive a developer input including a domain and the set of actions; and generate the interaction environment using the set of actions. ### Claim 15 A system, comprising: at least one processor; and memory including instructions that, when executed by the at least one processor, cause the system to: determine a first portion and a second portion of an input; predict a first relevant action for the first portion, the first relevant action being included within a set of actions; select a meta-action, from a pre-defined set of meta-actions, based at least on the first relevant action; execute an action executor to cause the meta-action to execute using one or more parameters of the first relevant action; determine at least some first relevant action values for a set of first action parameters are unresolvable from the input; and generate a responsive query directed toward the at least some first relevant action values. ### Claim 16 (depends on 15) The system of claim 15, wherein the instructions when executed further cause the system to: determine, from a second input, a set of unresolved first relevant action values; and execute the first relevant action. ### Claim 17 (depends on 16) The system of claim 16, wherein the instructions when executed further cause the system to: predict a second relevant action for the second portion, the second relevant action being included within the set of actions; select a second meta-action, from the pre-defined set of meta-actions, based at least on the second relevant action; determine, from at least one of the input and the second input, a set of unresolved second relevant action values; execute the second relevant action; and generate a response to the input based, at least, on respective outputs of the execution of the first relevant action and the second relevant action. ### Claim 18 (depends on 15) The system of claim 15, wherein at least some actions of the set of actions are developer-defined to constraint an underlying generative language model. ### Claim 19 (depends on 15) The system of claim 15, wherein the instructions when executed further cause the system to: determine an initial first relevant action value from the input; add the initial first relevant action value to an initial first relevant action parameter; determine a next first relevant action value is unresolved by the input; generate a request for information associated with the next first relevant action value; determine the next first relevant action value after receiving an answer to the request; add the next first relevant action value to a next first relevant action parameter; and determine at least one following first relevant action value from the input after adding the next first relevant action value. ### Claim 20 (depends on 15) The system of claim 15, wherein the instructions when executed further cause the system to: predict a second relevant action for the second portion, the second relevant action being included within the set of actions; resolve a portion of the one or more parameters of the first relevant action; resolve a portion of the second relevant action by using one or more parameters associated with the second relevant action; and resolve remaining first relevant action parameters after resolving the portion of the one or more parameters associated with the second relevant action.