Intelligent control with hierarchical stacked neural networks

• US 10,510,000 B1
• Filed: 06/08/2015
• Issued: 12/17/2019
• Est. Priority Date: 10/26/2010
• Status: Active Grant

• Current Assignee: Michael Lamport Commons
• Original Assignee: Michael Lamport Commons
• Inventors: Commons, Michael Lamport
• Primary Examiner(s): Woolwine, Shane D
• Application Number:
  US14/733,472
• Time in Patent Office: 1,653 Days
• Field of Search: 706 20
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]
• CPC Class Codes: G10L 15/16 using artificial neural networks

Abstract

A system and method of detecting an aberrant message is provided. An ordered set of words within the message is detected. The set of words found within the message is linked to a corresponding set of expected words, the set of expected words having semantic attributes. A set of grammatical structures represented in the message is detected, based on the ordered set of words and the semantic attributes of the corresponding set of expected words. A cognitive noise vector comprising a quantitative measure of a deviation between grammatical structures represented in the message and an expected measure of grammatical structures for a message of the type is then determined. The cognitive noise vector may be processed by higher levels of the neural network and/or an external processor.

Claims

1. A method for analyzing grammar in a natural language message, comprising:

providing an artificial neural network having an input layer, a hidden layer, and an output layer, each comprising a plurality of neurons, and together being trained to produce an artificial neural network output from a natural language neural network input dependent on training according to a natural language grammar;

receiving a message having a type;

detecting an ordered set of words within the message;

linking the set of words found within the message to a corresponding set of expected words, the set of expected words having semantic attributes;

detecting a set of grammatical structures represented in the message, based on the type of the received message, the ordered set of words and the semantic attributes of the corresponding set of expected words;

determining, with the artificial neural network, a degree of consistency of the set of grammatical structures represented in the message with a natural language grammar, dependent on the semantic attributes of the set of expected words according to the type of the message, and being dependent on training according to the natural language grammar, to produce a vector output of the artificial neural network representing at least a type of grammatical deviation of the set of grammatical structures represented in the message from the natural language grammar;

at least one of storing and outputting a vector based on the output of the artificial neural network.

2. The method of claim 1, wherein the message comprises text created by a word processing program.

3. The method of claim 1, wherein the message is an email message.

4. The method of claim 1, wherein the message is a search query.

5. The method of claim 1, wherein the message is a voice message.

6. The method of claim 5, wherein the semantic attributes correspond to proper spelling, grammar, and word use in the spoken or written language.

7. The method of claim 1, wherein the set of expected words is represented in a dictionary.

8. The method of claim 1, wherein the vector is produced dependent on a degree to which the set of grammatical structures represented in the message deviates from the semantic attributes of the set of expected words and comprises an indication of the existence of a grammatical error.

9. The method of claim 1, wherein the vector is produced dependent on a degree to which the set of grammatical structures represented in the message deviates from the semantic attributes of the set of expected words, the method further comprising outputting a suggested correction for a grammatical error.

10. A method of processing language, comprising:

providing at least one artificial neural network, each artificial neural network comprising;

at least one input layer, receiving inputs to a plurality of input neurons and producing a plurality of input neuron responses;

at least one hidden layer, receiving the plurality of input neuron responses to a plurality of hidden layer neurons and producing a plurality of hidden layer responses;

at least one output layer, receiving the plurality of hidden layer responses to a plurality of output layer neurons and producing at least one output;

the input neuron responses, the hidden layer responses and the output layer responses being defined according to neural network training, wherein the at least one artificial neural network is trained to abstract information present in the received inputs, wherein the at least one output represents an abstraction of the received inputs according to the training;

receiving inputs comprising human language messages;

processing the human language messages with the at least one artificial neural network according to a human language grammar represented in the training; and

the at least one output comprising a vector signal representing a type of grammatical deviance of the inputs from the human language grammar represented in the training.

11. The method according to claim 10, further comprising:

determining a type of the communication represented in the inputs;

detecting an ordered set of words within the inputs;

linking the set of words found, to a corresponding set of expected words, the set of expected words having semantic attributes; and

said processing comprises;

detecting, with the at least one artificial neural network, a set of grammatical structures represented in the inputs, based on type, the ordered set of words and the semantic attributes of the corresponding set of expected words; and

determining a consistency of the set of grammatical structures with the semantic attributes of the set of expected words according to the type.

12. The method according to claim 10, wherein the inputs comprise a search query for a semantically-searchable database comprising text.

13. The method according to claim 10, wherein the at least one artificial neural network comprises a plurality of artificial neural networks arranged in a hierarchical stack, each artificial neural network within the hierarchical stack being trained according to a respective level of cognitive development, and having a plurality of outputs representing a non-arbitrary organization of actions.

14. The method according to claim 13, further comprising generating an additional output representing a noise content of the inputs, representing a quantitative representation of a portion of the data content in the received inputs which does not represent the non-arbitrary organization of actions.

15. The method according to claim 10, wherein the at least one artificial neural network has a training which is adaptively updated.

16. A method of processing language, comprising:

providing at least one artificial neural network, each artificial neural network comprising;

at least one input layer, receiving inputs to a plurality of input neurons and producing a plurality of input neuron responses;

at least one hidden layer, receiving the plurality of input neuron responses to a plurality of hidden layer neurons and producing a plurality of hidden layer responses;

at least one output layer, receiving the plurality of hidden layer responses to a plurality of output layer neurons and producing at least one output;

the input neuron responses, the hidden layer responses and the output layer responses being defined according to human semantic communication grammar training, wherein the at least one output represents a vector artificial neural network output corresponding to a compliance of the received inputs with a predetermined grammar of the human semantic communication grammar training;

receiving the inputs comprising human semantic communications;

processing the human semantic communications with the at least one artificial neural network according to the human semantic communication grammar training; and

the at least one output comprising a vector signal representing a type of grammatical difference of the processed human semantic communications of the received inputs from the predetermined grammar of the human semantic communication grammar training.

17. The method according to claim 16, further comprising:

determining a type of the communication represented in a respective received input human semantic communication;

detecting an ordered set of words within the respective received input human semantic communication;

linking the set of words found to a corresponding set of expected words, the set of expected words having semantic attributes; and

said processing comprising;

detecting, with the at least one artificial neural network, a set of grammatical structures represented in the received input semantic communication, based on the determined type, the detected ordered set of words, and the semantic attributes of the corresponding set of expected words; and

determining a consistency of the set of grammatical structures with the semantic attributes of the set of expected words according to the determined type.

18. The method according to claim 16, wherein the respective received input human semantic communication comprises a search query for a semantically-searchable database.

19. The method according to claim 16, wherein the at least one artificial neural network comprises a plurality of artificial neural networks arranged in a hierarchical stack, each artificial neural network within the hierarchical stack being trained according to a respective level of cognitive development, and having a plurality of outputs representing a non-arbitrary organization of actions.

20. The method according to claim 19, further comprising generating an additional vector output representing a noise content of the inputs, comprising a quantitative representation of a portion of the data content in the received inputs which does not represent the non-arbitrary organization of actions.