# US 2021/0334474 A1 · Natural Language System and Methods

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0334474 A1
- Document ID: US20210334474A1
- Status: Published application
- Kind: A1
- Application number: 17/302,260
- Filed: 2021-04-28
- Published: 2021-10-28
- Earliest claimed: 2020-04-28
- Assignees: Leela Al, Inc.
- Applicants: Leela Al, Inc.
- Inventors: Cyrus Shaoul; Henry Bowdoin Minsky; Milan Singh Minsky
- Web page: https://patentgrep.com/patent/US20210334474A1

## CPC

- G06F 40/35: Discourse or dialogue representation
- G06F 11/3075: {the data filtering being achieved in order to maintain consistency among the monitored data, e.g. ensuring that the monitored data belong to the same timeframe, to the same system or component}
- G06F 16/2237: {Vectors, bitmaps or matrices}
- G06F 16/2477: {Temporal data queries}
- G06F 16/3329: Natural language query formulation
- G06F 18/211: Selection of the most significant subset of features
- G06F 18/29: {Graphical models, e.g. Bayesian networks}
- G06F 40/279: Recognition of textual entities
- G06N 3/006: based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/042: Knowledge-based neural networks; Logical representations of neural networks
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/08: Learning methods
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06N 3/09: Supervised learning
- G06N 3/091: Active learning
- G06N 3/092: Reinforcement learning
- G06N 5/02: Knowledge representation; Symbolic representation
- G06N 5/025: Extracting rules from data
- G06N 5/042: {Backward inferencing}
- G06V 10/70: using pattern recognition or machine learning (optical pattern recognition or electronic computations therefor G06V10/88)
- G06V 10/82: using neural networks
- G06V 20/41: {Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items (segmenting video sequences G06V20/49)}
- G06V 20/44: {Event detection}
- G06F 40/30: Semantic analysis
- G06T 2207/20081: Training; Learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

A natural language understanding (NLU) system utilizes a knowledge network having interconnected actor perceiver predictor (APP) nodes associated with context, action, and result. For a received utterance, an utterance type is determined, and an input message is generated based on the utterance type. The input message includes word-concept groupings, which include words of the received utterance and concept tags associated with the words. An action type is determined based on the utterance type. The knowledge network is searched to locate a subset of the APP nodes that exceed a threshold connection weight and thereby discriminate the word-concept groupings from other word-concept groupings represented by nodes of the network. Metadata for the word-concept groupings is retrieved from the subset of the APP nodes, and a response to the received utterance is generated by incorporating the metadata into a response template.

## Claims

### Claim 1

A system for natural language understanding, comprising:
an input module configured to 1) determine an utterance type for a received utterance and 2) generate an input message based on the utterance type, the input message including word-concept groupings, each of the word-concept groupings including at least one word of the received utterance and at least one concept tag associated with the word;
a network comprising a plurality of interconnected actor perceiver predictor (APP) nodes, each APP node of the plurality of APP nodes associated with a context, action, and result;
a concept discrimination module (CDM) configured to 1) determine an action type based on the utterance type, 2) search the network to locate a subset of the APP nodes that exceed a threshold connection weight and thereby discriminate the word-concept groupings from other word-concept groupings represented by APP nodes of the network, and 3) retrieve metadata for the word-concept groupings from the subset of the APP nodes;
an output module configured to 1) determine a response template based on the action type and 2) generate a response to the received utterance by incorporating the metadata into the response template; and
a user interface configured to convey the response to a user.

### Claim 2 (depends on 1)

The system of claim 1, further comprising a conversation status module (CSM) configured to determine an update to a conversation status based on the utterance type and a previous conversation status, the CDM being configured to determine the action type based on the conversation status.

### Claim 3 (depends on 2)

The system of claim 2, wherein the conversation status indicates whether a user is awaiting a response from the system, the system is awaiting a response from a user, or neither the user nor the system is awaiting a response.

### Claim 4 (depends on 1)

The system of claim 1, wherein the input module is further configured to determine the utterance type based on a selected use case, the selected use case being associated with a dialog context associated with the utterance and indicating a set of concept tags for inclusion in the input message.

### Claim 5 (depends on 1)

The system of claim 1, wherein the utterance type is selected from a plurality of utterance types including at least one of question, command, statement, and exclamation.

### Claim 6 (depends on 1)

The system of claim 1, wherein the input module is further configured determine an utterance sub-type for the received utterance, the input module determining the utterance sub-type based on the utterance type and at least one word of the received utterance associated with the utterance type.

### Claim 7 (depends on 1)

The system of claim 1, wherein the CDM is further configured to modify connection weights between concepts and entities stored at the network based on the received utterance and a present state of the network.

### Claim 8 (depends on 1)

The system of claim 1, wherein the CDM is further configured to select a subset of the metadata based on at least one of a dialog history and external data associated with the received utterance.

### Claim 9 (depends on 1)

The system of claim 1, wherein the network is further configured to generate a request to the output module to generate the response, the request indicating the action type and a plurality of response word-concept groupings incorporating the metadata.

### Claim 10 (depends on 1)

The system of claim 1, wherein the network is further configured to update a property of at least one of the APP nodes based on a relation between two of the word-concept groupings.

### Claim 11 (depends on 1)

The system of claim 1, wherein the network is further configured to generate a new APP node from one of the word-concept groupings, the new APP node being connected to at least one other APP node based on a relation between two of the word-concept groupings.

### Claim 12 (depends on 1)

The system of claim 1, wherein the action type is a first action type, and wherein the network is further configured to determine a second action type based on a state of the network.

### Claim 13 (depends on 1)

The system of claim 1, wherein, prior to generating the response, the network is configured to generate an inquiry for information associated with the received utterance, the output module generating a message based on the inquiry for conveyance to the user via the user interface.

### Claim 14 (depends on 1)

The system of claim 1, wherein the CDM is further configured to 1) search the network to locate neighbor nodes to the subset of the APP nodes based on connection weight to the subset of the APP nodes, and 2) retrieve secondary metadata for the word-concept groupings from the neighbor nodes.

### Claim 15 (depends on 1)

The system of claim 1, wherein the CDM is further configured to locate the subset of the APP nodes as a function of a set of concept tags of the word-concept groupings.

### Claim 16

A method of natural language understanding, comprising:
determining an utterance type for a received utterance;
generating an input message based on the utterance type, the input message including word-concept groupings, each of the word-concept groupings including at least one word of the received utterance and at least one concept tag associated with the word;
determining an action type based on the utterance type;
searching a network to locate a subset of the APP nodes that exceed a threshold connection weight and thereby discriminate the word-concept groupings from other word-concept groupings represented by APP nodes of the network, the network comprising a plurality of interconnected actor perceiver predictor (APP) nodes, each APP node of the plurality of APP nodes associated with a context, action, and result;
retrieving metadata for the word-concept groupings from the subset of the APP nodes;
determining a response template based on the action type;
generating a response to the received utterance by incorporating the metadata into the response template; and
conveying the response to a user.

### Claim 17 (depends on 16)

The method of claim 16, further comprising:
determining an update to a conversation status based on the utterance type and a previous conversation status; and
determining the action type based on the conversation status.

### Claim 18 (depends on 17)

The method of claim 17, wherein the conversation status indicates whether a user is awaiting a response from a system, the system is awaiting a response from a user, or neither the user nor the system is awaiting a response.

### Claim 19 (depends on 16)

The method of claim 16, further comprising determining the utterance type based on a selected use case, the selected use case being associated with a dialog context associated with the utterance and indicating a set of concept tags for inclusion in the input message.

### Claim 20 (depends on 16)

The method of claim 16, wherein the utterance type is selected from a plurality of utterance types including at least one of question, command, statement, and exclamation.

### Claim 21 (depends on 16)

The method of claim 16, further comprising:
determining an utterance sub-type for the received utterance; and
determining the utterance sub-type based on the utterance type and at least one word of the received utterance associated with the utterance type.

### Claim 22 (depends on 16)

The method of claim 16, further comprising modifying connection weights between concepts and entities stored at the network based on the received utterance and a present state of the network.

### Claim 23 (depends on 16)

The method of claim 16, further comprising selecting a subset of the metadata based on at least one of a dialog history and external data associated with the received utterance.

### Claim 24 (depends on 16)

The method of claim 16, further comprising generating a request to the output module to generate the response, the request indicating the action type and a plurality of response word-concept groupings incorporating the metadata.

### Claim 25 (depends on 16)

The method of claim 16, further comprising updating a property of at least one of the APP nodes based on a relation between two of the word-concept groupings.

### Claim 26 (depends on 16)

The method of claim 16, further comprising generating a new APP node from one of the word-concept groupings, the new APP node being connected to at least one other APP node based on a relation between two of the word-concept groupings.

### Claim 27 (depends on 16)

The method of claim 16, wherein the action type is a first action type, and further comprising determining a second action type based on a state of the network.

### Claim 28 (depends on 16)

The method of claim 16, further comprising, prior to generating the response:
generating an inquiry for information associated with the received utterance; and
generating a message based on the inquiry for conveyance to the user via a user interface.

### Claim 29 (depends on 16)

The method of claim 16, further comprising:
searching the network to locate neighbor nodes to the subset of the APP nodes based on connection weight to the subset of the APP nodes; and
retrieving secondary metadata for the word-concept groupings from the neighbor nodes.

### Claim 30 (depends on 16)

The method of claim 16, further comprising locating the subset of the APP nodes as a function of a set of concept tags of the word-concept groupings.
