# US 12,001,805 B2 · Explainable natural language understanding platform

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

- Number: US 12,001,805 B2
- Document ID: US12001805B2
- Status: Granted
- Kind: B2
- Application number: 18/306,479
- Filed: 2023-04-25
- Published: 2023-11-09
- Granted: 2024-06-04
- Earliest claimed: 2022-04-25
- Assignees: Gyan Inc.
- Applicants: Venkatesan Srinivasan; Vishaal Jatav
- Inventors: Venkatesan Srinivasan; Vishaal Jatav
- Web page: https://patentgrep.com/patent/US12001805B2

## CPC

- G06F 40/30: Semantic analysis
- G06F 40/35: Discourse or dialogue representation
- G06F 40/205: Parsing
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A system and method for a real time, human-analogous natural language understanding engine (NLU Platform). The Platform understands natural language content in its full compositional context, is completely explainable and is effective with no or little training data. The invention utilizes a machine representation of natural language to enable human-analogous understanding by machines. The invention combines knowledge-based linguistics, discourse models, compositionality and the rhetorical structure in language. It can incorporate global knowledge about a concept to improve its representation and understanding. The platform does not rely on statistically derived distributional semantics and can be used for the complete range of natural language understanding tasks in the world. In one exemplary application, the NLU Platform can be used for automated knowledge acquisition or as an automated research engine for any topic or set of topics from one or more document repositories like the internet.

## Claims

### Claim 1

A natural language understanding system comprising at least one processor configured to automatically carry out a task of human-analogous natural language understanding of natural language content from one or more text documents, the system comprising:
a computer system hosting a natural language understanding engine which creates a machine representation of the meaning of a document using a compositional language model;
the engine creating its machine meaning representation for a document by decomposing the entire document into its compositional structure to reflect multiple layers of meaning, and doing so in an invertible manner;
the engine decomposing the document into its compositional structure by parsing the document according to a discourse model identified for the document, identifying the main idea of the document, and identifying the start of new ideas in the document,
decomposing the document into sub-documents by ideas, and creating meaning representation for the entire document by decomposing the subdocuments into their compositional elements; and
the computing system comprising one or more virtual or dedicated servers or similar computing devices and programmed with executable instructions.

### Claim 2 (depends on 1)

The system of claim 1, where the compositional language model does not utilize statistical machine learning or statistically derived distributional semantics like word embeddings, for creating its meaning representation graph.

### Claim 3 (depends on 1)

The system of claim 1, where the natural language understanding engine utilizes computational linguistics, compositionality and rhetorical structure in language and processes one or more documents without any training or training data.

### Claim 4 (depends on 1)

The system of claim 1, where the natural language understanding engine works entirely in natural language and does not convert any part of the document into real valued vector encodings.

### Claim 5 (depends on 1)

The system of claim 1, where a new discourse model is created and integrated without any programming.

### Claim 6 (depends on 1)

The system of claim 1, where the main idea of the document is determined based on concept importance using a combination of linguistic and rhetorical structure, attributes and discourse model.

### Claim 7 (depends on 1)

The system of claim 1, where the start of a new idea is determined based on determining the nature of relationship between each successive compositional element including a paragraph, a sentence, and a clause based on linguistic attributes and nature of rhetorical expressions linking the compositional elements including a plurality of linguistic rules, rhetorical relations and co-referential relationships.

### Claim 8 (depends on 7)

The system of claim 7, where the co-referential relationships between concepts in two sentences is the use of the exact same concept or a pronomical or a pragmatic reference to the concept contained in an external global knowledge store.

### Claim 9 (depends on 1)

The system of claim 1, where decomposing the sub-documents includes at the first level, identifying sentence level compositional elements and the relationships amongst them that make up the sub-documents using linguistic attributes and the nature of the linguistic relationship of the compositional elements to the primary sentence of the sub-document and each other.

### Claim 10 (depends on 9)

The system of claim 9, where decomposing the sub-documents at the next level, includes identifying within sentence level compositional elements and the nature of the rhetorical and linguistic relationships amongst them using linguistic and rhetorical attributes and rules.

### Claim 11 (depends on 10)

The system of claim 10, where the decomposing the sub-documents continues until every single compositional unit including at the word level is part of a meaning representation graph created for the entire document.

### Claim 12 (depends on 9)

The system of claim 9, where new rules are added to detect relationships between and among compositional elements without any programming.

### Claim 13 (depends on 1)

The system of claim 1, where the meaning representation is in the form of a meaning representation graph comprising the decomposed compositional elements which is queried by concept, surface form of the rhetorical expression or abstract rhetorical relationship types.

### Claim 14 (depends on 13)

The system of claim 13, where the meaning representation graph is invertible back into the complete document.

### Claim 15 (depends on 13)

The system of claim 13, where surface form of the rhetorical expression or abstract rhetorical relationship types is a map between rhetorical expressions and abstract rhetorical relations that is configurable and is a function of cue words, cue phrases and linguistic attributes of the discourse including parts of speech, sentence type, and co-referential relationships.

### Claim 16 (depends on 1)

The system of claim 1, where the decomposing the compositional structure includes establishing co-referential relationships between words and the detection of such co-referential relationships are improved by external knowledge stored in a global knowledge store.

### Claim 17 (depends on 16)

The system of claim 16, where the global knowledge store stores knowledge at a sense level for each concept and continuously improved without any programming.

### Claim 18 (depends on 16)

The system of claim 16, where the global knowledge store stores knowledge at a sense level for each concept.

### Claim 19 (depends on 1)

The system of claim 1, where the document is a text document of data formats comprising text, html, Docx, PDF, Pptx, eBook and audio and video data which are transcribed into text.

### Claim 20 (depends on 19)

The system of claim 19, where the document is not grammatically complete including cryptic social media messages.

### Claim 21 (depends on 1)

The system of claim 1, where the engine identifies rhetorical phrases between sentences and within each sentence among compositional units including between subject and object, and also maps such identified rhetorical phrases to a set of normalized rhetorical relations.

### Claim 22 (depends on 1)

The system of claim 1, wherein the engine creates an expandable multi-level summary of a document corresponding to a multi-level meaning representation graph.

### Claim 23 (depends on 1)

The system of claim 1, wherein the engine creates an expandable multi-level summary of a document according to user preferences expressed as rules operating on rhetorical relations identified in the document.

### Claim 24 (depends on 1)

The system of claim 1, wherein the engine processes one or a plurality of documents without requiring any training or training document set.

### Claim 25 (depends on 24)

The system of claim 24, where the engine creates a combined meaning representation graph for an entire collection of documents by aggregating meaning representation graphs for each of the documents after removing redundant concepts across meaning representation graphs at the document level.

### Claim 26 (depends on 24)

The system of claim 24, where the engine creates an aggregate summary for an entire collection of documents by aggregating summaries for each of the documents and removing redundant content.

### Claim 27 (depends on 1)

A system for creating a knowledge collection of relevant documents using the meaning representation of system of claim 1.

### Claim 28 (depends on 27)

The system of claim 27, where the knowledge collection is based on a query using internet search engines, private document collections and sources or a combination thereof.

### Claim 29 (depends on 27)

The system of claim 27, where the engine organizes the collection of relevant documents into a pre-determined set of sub-topics by determining the most relevant match for each document; and
where a pre-determined set of sub-topics is not available, the engine surfaces an intrinsic set of topics from the collection of relevant documents using the meaning representation of the system for each of the documents in the collection.

### Claim 30 (depends on 29)

The system of claim 29, where the pre-determined set of sub-topics is an intrinsic set of sub-topics that is determined as the most representative ideas across the collection of documents by aggregating the ideas in each of the documents.

### Claim 31 (depends on 30)

The system of claim 30, where the process of determining the most representative ideas across the collection of documents integrates knowledge from the global knowledge store to enable the engine to recognize relationships across ideas not explicitly present in the documents, uses linguistic characteristics including its role in the documents it is found in.

### Claim 32 (depends on 29)

The system of claim 29, where a degree of relevance of each article in the collection of documents to each of the pre-defined sub-topics is determined using a combination of concept importance, sentence importance and context strength.

### Claim 33 (depends on 29)

The system of claim 29, where the engine creates an aggregate summary for each of the sub-topics by aggregating summaries of the documents classified in that sub-topic.

### Claim 34

A method of automatically performing the function of human-analogous natural language understanding of natural language content from one or more text documents, the method comprising:
generating a machine representation of the meaning of a document using a compositional language model;
generating the machine meaning representation for a document by decomposing the entire document into its compositional structure to reflect multiple layers of meaning, and doing so in an invertible manner; and
decomposing the document into its compositional structure by parsing the document according to a discourse model identified for the document, identifying the main idea of the document, and identifying the start of new ideas in the document, decomposing the document into sub-documents by ideas, and creating meaning representation for the entire document by decomposing the subdocuments into their compositional elements.

### Claim 35 (depends on 34)

The method of claim 34, where the compositional language model does not utilize statistical machine learning or statistically derived distributional semantics like word embeddings, for creating its a meaning representation graph.

### Claim 36 (depends on 34)

The method of claim 34, where the compositional language model utilizes computational linguistics and rhetorical structure in language and processes one or more documents without any training or training data.

### Claim 37 (depends on 34)

The method of claim 34, where the method works entirely in natural language and does not convert any part of the document into real valued vector encodings.

### Claim 38 (depends on 34)

The method of claim 34, where a new discourse model is created and integrated without any programming.

### Claim 39 (depends on 34)

The method of claim 34, where the main idea of the document is determined based on concept importance using a combination of linguistic and rhetorical structure, attributes and discourse model.

### Claim 40 (depends on 34)

The method of claim 34, where the start of a new idea is determined based on determining the nature of relationship between each successive compositional element including a paragraph, a sentence, and a clause based on linguistic attributes and nature of rhetorical expressions linking the compositional elements including a plurality of linguistic rules, rhetorical relations and co-referential relationships.

### Claim 41 (depends on 34)

The method of claim 34, where decomposing the sub-documents includes at the first level, identifying sentence level compositional elements and the relationships amongst them that make up the sub-documents using linguistic attributes and the nature of the linguistic relationship of the compositional elements to the primary sentence of the sub-documents and each other.

### Claim 42 (depends on 41)

The method of claim 41, where decomposing the sub-documents at the next level, includes identifying within sentence level compositional elements and the nature of the rhetorical and linguistic relationships amongst them using linguistic and rhetorical attributes and rules.

### Claim 43 (depends on 42)

The method of claim 42, where the decomposition of the sub-documents continues until every single compositional unit including at the word level is part of the meaning representation graph.

### Claim 44 (depends on 41)

The method of claim 41, where new rules can be added to detect relationships between and among compositional elements without any programming.

### Claim 45 (depends on 34)

The method of claim 34, where the meaning representation is in the form of a multi-layered graph comprising the decomposed compositional elements which can be queried by concept, surface form of the rhetorical expression or abstract rhetorical relationship types.

### Claim 46 (depends on 45)

The method of claim 45, where the surface form of the rhetorical expression or abstract rhetorical relationship types is a map between rhetorical expressions and abstract rhetorical relations are configurable and can be a function of cue words, cue phrases and linguistic attributes of the discourse including parts of speech, sentence type, and co-referential relationships.

### Claim 47 (depends on 34)

The method of claim 34, where decomposing the compositional structure includes establishing co-referential relationships between words and the detection of such co-referential relationships are improved by external knowledge stored in a global knowledge store.

### Claim 48 (depends on 47)

The method of claim 47, where the global knowledge store can store knowledge at a sense level for each concept and continuously improved without any programming.

### Claim 49 (depends on 34)

The method of claim 34, where a document is not grammatically complete including cryptic social media messages.

### Claim 50 (depends on 34)

The method of claim 34, wherein the method creates an expandable multi-level summary of a document corresponding to a multi-level meaning representation graph.

### Claim 51 (depends on 50)

The method of claim 50, wherein the method creates an expandable multi-level summary of a document according to user preferences expressed as rules operating on the rhetorical relations identified in the document.

### Claim 52 (depends on 34)

The method of claim 34, wherein the method processes one or a plurality of documents without requiring any training or training document set.

### Claim 53 (depends on 34)

A method for creating a knowledge collection of relevant documents using the meaning representation of the method of claim 34.

### Claim 54 (depends on 53)

The method of claim 53, where the knowledge collection is based on a query using internet search engines, private document collections and sources or a combination.

### Claim 55 (depends on 53)

The method of claim 53, where the method organizes the knowledge collection of relevant documents into a pre-determined set of sub-topics by determining the most relevant match for each document; and where a pre-determined set of sub-topics is not available, the method surfacing an intrinsic set of topics from the collection of relevant documents for each of the documents in the collection.

### Claim 56 (depends on 55)

The method of claim 55, where the intrinsic set of sub-topics is determined as the most representative ideas across the collection of documents by aggregating the ideas in each of the document.

### Claim 57 (depends on 56)

The method of claim 56, where the process of determining the most representative ideas across the document collection also integrates knowledge from the global knowledge store to enable the method to recognize relationships across ideas not explicitly present in the documents.

### Claim 58 (depends on 55)

The method of claim 55, where the degree of relevance of each document in the document collection to each of the pre-defined sub-topics is determined using a combination of concept importance, sentence importance and context strength.

### Claim 59 (depends on 34)

The method of claim 34, where the method creates a combined meaning representation graph for an entire collection of documents by aggregating meaning representation graphs for each of the documents after removing redundant concepts across meaning representation graphs at the document level.

### Claim 60 (depends on 34)

The method of claim 34, where the method creates an aggregate summary for an entire collection of documents by aggregating summaries for each of the documents and removing redundant content.