WO2018192269A1 - 计算机模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台 - Google Patents

计算机模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台 Download PDF

Info

Publication number
WO2018192269A1
WO2018192269A1 PCT/CN2018/000143 CN2018000143W WO2018192269A1 WO 2018192269 A1 WO2018192269 A1 WO 2018192269A1 CN 2018000143 W CN2018000143 W CN 2018000143W WO 2018192269 A1 WO2018192269 A1 WO 2018192269A1
Authority
WO
WIPO (PCT)
Prior art keywords
program
true
library
value
semantic
Prior art date
Application number
PCT/CN2018/000143
Other languages
English (en)
French (fr)
Inventor
万继华
Original Assignee
湖南本体信息科技研究有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 湖南本体信息科技研究有限公司 filed Critical 湖南本体信息科技研究有限公司
Publication of WO2018192269A1 publication Critical patent/WO2018192269A1/zh
Priority to US16/655,550 priority Critical patent/US20200111012A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F8/00—Arrangements for software engineering
    • G06F8/20—Software design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F8/00—Arrangements for software engineering
    • G06F8/30—Creation or generation of source code
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F8/00—Arrangements for software engineering
    • G06F8/30—Creation or generation of source code
    • G06F8/33—Intelligent editors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/046—Forward inferencing; Production systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/205—Parsing
    • G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G06F40/295—Named entity recognition

Definitions

  • the invention relates to the field of human computers, in particular to a method for simulating human brain learning knowledge, a logical inference engine and a brain-like artificial intelligence service platform.
  • the ideal goal of artificial intelligence is to enable machines to learn and work in a brain-like manner.
  • Gödel's incompleteness theorem and Turing's shutdown theorem The limitations of formal systems and mechanical computing capabilities have led to the failure of such brain-like artificial intelligence techniques to date.
  • the key to solving brain-like artificial intelligence lies in solving the problem of transforming natural language into computer programming language and how to simulate human brain logic judgment reasoning in programming language.
  • the method of translating natural language into computer language proposes that "computer uses the lexicon and word segmentation rules to extract the grammatical components, logical conjunctions and logical semantics in natural language words, and translates the grammatical components based on the information management library.
  • translating natural language into computer language means that the machine can work according to the instructions given by people with natural sentences (propositions), which means that the machine can be natural.
  • Language interacts with humans and can simulate the logical thinking of the human brain in a judgmental reasoning manner. It also means that the goal of brain-like artificial intelligence can be achieved.
  • the present invention provides a method for computer simulation of human brain learning knowledge, which includes the following steps:
  • a class library for storing a basic element of a class corresponding to a syntax component of a natural language statement and a true property corresponding to the two logical units of the main and the predicate composed of the basic element of the class, wherein: the semantic property of the object element corresponding to the subject is The affirmative property is true, represented by binary code 1.
  • the semantic property of the function element corresponding to the predicate is positive or negative, and the semantic property of the function corresponding to each natural statement predicate can only be true in one of the affirmative and negative properties. , represented by binary code 1 or 0, 1 means definitely true, 0 means negation is true;
  • a resource library for storing information resources of the above scene or event, and corresponding to the basic elements of the class library and the properties of the object and function elements;
  • An intelligent information management library for storing a judgment inference algorithm program similar to human brain management thinking and a smart application for managing behavior and a correspondence between the class library, the resource library, and the thesaurus;
  • the computer reads or adds the words and part of speech in the natural language sentence indicating the grammatical components to the lexicon, and then calls the semantic analyzer to create and store the basic elements and semantic properties of the class generated by the natural language single sentence in the class.
  • the library configures and stores the scene corresponding to the basic elements and semantic properties of the class in the resource library, wherein the semantic properties of the object and the function are consistent with the properties of the corresponding scene of the resource library; the nature of the corresponding scene of the object is affirmative 1 True, the nature of the function corresponding to the scene is affirmative 1 or negative 0.
  • the function corresponding to the semantic property of 1 is a scene with a positive truth, and the corresponding property of the semantic property is 0. a logical knowledge element corresponding to the object, the functional unit, and the functional unit;
  • the computer calls the semantic analyzer for the intelligent application requirements, and generates a smart application by a natural language program that satisfies the application requirements by a natural language sentence, a complex sentence or a statement set, and stores it in the intelligence.
  • Information Management Library
  • the thesaurus is divided into a system vocabulary, a private vocabulary and a public lexicon, the system vocabulary is used for storing logical conjunction words and generating negative semantic words, and the private vocabulary is used for storing user-defined special words.
  • the common vocabulary is used to store public words of the part of speech.
  • said class library comprises an ontology heterogeneous function, which is a method of simultaneously or defining a representation of a different word or term referring to the same scene or scene element as the same word or term.
  • the present invention also provides a logical inference engine constructed according to a computer simulation method for human brain learning knowledge, which adopts software or hardware form, and includes a knowledge information acquisition module, a judgment inference calculation module, an operation program generation module, and an operation program execution module;
  • the knowledge information obtaining module is configured to obtain an application program in the intelligent information management library and an algorithm program for judging the inference calculation application program;
  • the judgment inference calculation module is configured to perform judgment and inference calculation on the application program by using the obtained axiom algorithm program;
  • the operating program generating module is configured to generate an operable judgment inference conclusion program according to the result of the logic calculation;
  • the operation program execution module is configured to execute the judgment inference conclusion program.
  • the judgment reasoning algorithm program comprises a judgment algorithm program and an inference algorithm program, wherein the judgment algorithm program is a binary code of a semantic property of a function in a single sentence program and a binary code in a class library indicating that the corresponding function is a semantic property.
  • the reasoning algorithm program is based on the logical relationship between the return value of the single sentence in the determining program and the complex sentence formed by the single sentence and the single sentence and the single sentence. Calculation program; the logical relationship between statements is divided into sufficient, necessary, necessary and sufficient conditional relations and or three types of relations.
  • the calculating program calculates the sufficient condition by: determining that the return value of the front piece is true, then the value of the post-initiating piece is true, and determining that the return value of the front piece is false, the value of the post-initiating piece is true and determining If the return value of the latter is true, then the value of the predecessor is true, and the return value of the post is determined to be false.
  • the value of the predecessor is false; the calculation program calculates the necessary conditional condition: If the return value is true, the value of the post-release piece is true, the return value of the pre-event is judged to be false, then the value of the post-release piece must be false, and the return value of the post-return piece is determined to be true.
  • the calculation program calculates the condition for the subdivision: if the return value of the predecessor is true, the value of the post is pushed, and the predecessor is determined. If the return value is false, the value of the post-release piece must be false. If the return value of the post-receipt is true, the value of the pre-extension piece must be true. If the return value of the post-receipt is false, the value of the pre-extension piece must be false.
  • the method for calculating or calculating the calculation program is: determining that the return value of one of the single sentences is true, and then introducing another single sentence must be false; determining that the return value of one of the single sentences is false, then another single sentence is true.
  • the calculating program calculates the relationship relationship by: determining that the return value of each single sentence in the precondition is true, the conclusion of the post-release piece is true, and the return value of the single sentence in the precondition is false. The conclusion of the post-release is false.
  • the operation program generating module is configured to generate an operation program according to the judgment inference conclusion obtained by the calculation program, and the method for generating the judgment operation program is: a program that returns a true value, and is defined in the acquisition application.
  • the sequence control structure generates the operation program by the method for generating the judgment operation program; and the inference operation program is generated
  • the method is as follows: an application of an inference relationship is divided into an analysis program statement and an operation program statement; an analysis program is a program that introduces a conclusion statement, an operation program is a statement program that is introduced as a conclusion; and a conclusion statement program is introduced as a true
  • the semantic properties defined in the function defined in the application corresponding to the library objects and The function and the scene generation operation program configured in the resource library
  • the conclusion is a fake statement program, according to the semantic property of the function defined in the acquisition application, the corresponding class library object and function and the configuration in the resource sequence Scene generation operation program.
  • the conclusion is that the real statement program can generate the operation program corresponding to the library object function and the scene configured in the resource library according to the original value or the reverse of the semantic property defined in the
  • the invention further provides a brain-like artificial intelligence service platform formed by using a logical inference engine, which comprises:
  • the artificial intelligence product developer registers the landing platform, downloads the SDK toolkit, and uses the learning knowledge method and calls the semantic analyzer to create a brain-like knowledge base that satisfies the requirements of the development application product;
  • the platform listens or reads the natural language of the input and calls the semantic analyzer to automatically generate an application request or instruction designed by the artificial intelligence product developer;
  • the present invention first provides a method for computer simulation of human brain learning knowledge, which enables a computer to learn knowledge in a brain-like manner; and then provides a logical inference engine that can perform judgment inference calculation; the present invention uses artificial
  • the method uses the cognitive model of the human brain to intelligently calculate and judge the objective things and the intelligent mechanism based on the cognitive model to simulate the logical mechanism to the computer system, realizes the machine to simulate the intellectual function learning knowledge of the human brain, and forms a brain-like artificial Intelligent service platform.
  • Figure 1 is a schematic illustration of a method of machine simulation human brain learning in accordance with the present invention.
  • FIG. 2 is a schematic diagram showing the structure and calculation flow of the logical inference engine of the present invention.
  • FIG. 3 is a schematic diagram showing the workflow of the brain-like artificial intelligence service platform of the present invention.
  • FIG. 1, FIG. 2, and FIG. 3, are intended to be illustrative of the present invention and are not intended to limit the invention.
  • a method of computer simulation of human brain learning knowledge comprising the steps of:
  • a resource library for storing information resources of the above scene or event, and corresponding to the semantic nature of the subject and the concept in the natural statement;
  • a class library for storing a basic element of a class corresponding to a syntax component of a natural language sentence and a logical property corresponding to the two logical units of the main body and the predicate composed of the basic element of the class, wherein: the logical property of the object element corresponding to the subject is affirmative The property is true and is represented by binary code 1.
  • the logical property of the function element corresponding to the predicate is true or positive, and can only be true if one of the properties is true, represented by binary code 1 or 0, and 1 is affirmative. True, 0 means the negative nature is true;
  • An intelligent information management library for storing a judgment inference algorithm program similar to human brain management thinking and a smart application for managing behavior and a correspondence between the class library, the resource library, and the thesaurus;
  • the computer inputs or adds the words and part of speech in the natural language expressions to the lexicon, and then calls the semantic analyzer to create and store the basic elements and semantic properties of the class generated by the natural language statement in a class.
  • the library configures and stores the scene corresponding to the basic elements and semantic properties of the class in the resource library, wherein the semantic properties of the object and the function are consistent with the logical properties of the corresponding scene of the resource library; the logical properties of the object and the corresponding scene are A positive 1 is true, the logical property of the function and the corresponding scene is affirmative 1 or negative 0 is true; where the function with a semantic property of 1 corresponds to a logical property whose affirmative scene is true and the semantic property is 0, the corresponding logical property is negative.
  • the scene is true, thereby forming a logical knowledge element with the main object, the concept corresponding object, and the function unit; this is a method for the computer to learn the concept element or morpheme in the form of brain-like logic
  • the computer calls the semantic analyzer for the intelligent application requirements, and automatically generates the intelligent application by the natural language program that satisfies the application requirements by the natural language single sentence, complex sentence or statement set, and stores it in the intelligent language program.
  • Intelligent information management library This is the way computers learn serial or systematic knowledge in the form of brain-like logic.
  • the ontology brain knowledge base simulates the methods of human brain memory and learning, which can be summarized into the following three levels:
  • the first is to learn the concept of words and the objective things they refer to, that is, to learn the words that represent the scene or event.
  • the relationship between the "sun" and the physical sun in the sky is the knowledge of this level. That is, a string of words is used as a token to correspond to the scene or event it represents.
  • the program saves the knowledge to the thesaurus, the resource library and the class library, wherein the knowledge of the words stored in the thesaurus, the scene or the event is saved to the resource library (database), and the operation and recognition knowledge of the scene or event is saved to the class. Library, and one-to-one correspondence. Storing this knowledge in the machine is equivalent to having the machine learn these knowledge about words.
  • the second is to learn the logical recognition and judgment knowledge formed by the main and predicate concepts of sentences (propositions) as the basic logical structure, that is, to know "what is what, what is not what, what can be, what can not be done” and so on.
  • the ontology-like brain knowledge base expresses the knowledge of this level to the class library, and corresponding various syntactic components in the sentence to the basic elements of the class, wherein the subject is represented as an object, a predicate representation method or attribute, and other components are represented as parameters, and It corresponds to words and scenes in the thesaurus and the repository.
  • the subject and the predicate part are divided into two logical units, and the logical semantic properties of each logical unit are corresponding to the logical nature of the scene state represented by it, so that the machine can understand and judge the knowledge in the brain-like form. Storing this knowledge in the machine is equivalent to having the machine learn these knowledge about the statement.
  • the third is to learn the knowledge of a set of sentences that form a single sentence or multiple repetitions in a logically connected relationship.
  • the present invention translates this knowledge into a computer language method to form an application, and stores the knowledge in an information management library, and associates with knowledge in a class library, a resource library, and a thesaurus, and through the semantic analysis.
  • the analytical translation of the device and the judgment and reasoning of the logical inference engine form an intelligent knowledge system. That is, various smart applications.
  • the specific translation method is described in the patent ZL201310657042.8.
  • the translation of grammatical components, logical conjunctions, and logical semantics in natural language texts is performed by translating the string codes of the grammatical components into string codes representing the names of the basic elements in the computer object-oriented language, and translating the logical conjunctions.
  • the program-directed program transfers the instruction code, translates the logical semantics into binary codes representing affirmative and negative, and then stitches the three codes into an application.
  • the logical inference engine generates a preset executable intelligent program to form a system intelligent knowledge management system. Storing this knowledge by the machine is equivalent to letting the machine learn the intelligent knowledge of using natural language as a program to perform various tasks. This approach can also be extended to the active learning level in which a machine can form knowledge by reading in or listening to natural language.
  • the vocabulary in the learning knowledge method distinguishes system lexicon, private lexicon and public lexicon.
  • the system lexicon is used to store logical conjunctions and generate negative semantics.
  • the private vocabulary is used to store user-defined special words corresponding to their special fields or block libraries and resource libraries.
  • the public vocabulary is used for storage. Public words of the word norm. This method is conducive to the formation of knowledge block, changing the current machine deep learning relies on the high resource consumption of big data analysis and the big data monopoly situation, and also facilitates the connection of knowledge in different fields in the form of blockchain.
  • the intelligent knowledge support and application enables the knowledge of the network system to truly surpass the human brain.
  • the class library in the learning knowledge method includes an ontology heterogeneous function, which refers to a representation of the same scene or scene element in different natural language words or terms, and points to or defines the same word or term, for implementation.
  • the natural language method to establish a human-computer interaction service platform and its application provides great convenience, effectively ensuring that end users can perform human-computer interaction based on artificial intelligence platform in an easy and free way.
  • Computer simulation of human brain learning refers to the simulation of human cognitive models and intelligent mechanisms into computer systems, so that the machine has brain-like intelligence functions.
  • the establishment of the above-mentioned ontology brain knowledge base enables the machine to have a brain-like learning style and memory.
  • the logic inference engine of the present invention adopts a software or hardware form, and includes a knowledge information acquisition module, a judgment inference calculation module, an operation program generation module, and an operation program execution module;
  • the knowledge information acquisition module is configured to acquire an application in the intelligent information management library.
  • an algorithm program for calculating an application by the judgment reasoning is configured to perform a judgment inference calculation on the application program by using the acquired axiom algorithm program;
  • the operation program generation module is configured to generate an operable according to the result of the logic calculation Determining an inference conclusion program;
  • the operation program execution module is configured to execute a judgment inference conclusion program.
  • the logic inference engine of the invention implements the logical judgment inference calculation of the main predicate program unit by using four modules, so that the programming language including the main and predicate program units converted from the natural language can generate the reasoning thinking and the human brain.
  • An executable intelligent program for artificial intelligence applications that provides an effective technical solution for brain-like artificial intelligence applications.
  • the judgment reasoning algorithm program includes a judgment algorithm program, and the judgment algorithm program performs the same or different comparison with the binary code of the semantic property of the function in the single sentence program and the binary code of the class library indicating the logical property of the corresponding function. Calculate, if the two are the same as 11 or 00, it is judged that the single-sentence program is true and returns a judgment value of 1; the single-sentence program with a return value of 1 corresponds to the scene resource definition according to the original semantic property and the object and function name in the application.
  • the program if the difference between the two is 10 or 01, the judgment is false and the judgment value is 0; the single-sentence program with the return value of 0 is changed according to the original semantic property in the application, and the object and the function name are corresponding to the program.
  • the scene resource defines the program.
  • the intelligent program generation module generates a corresponding executable intelligent program according to the value that is determined to be true, and intuitively guarantees the logical reliability of the program. Specifically:
  • h represents propositional variables
  • j represents semantic and logical properties
  • jz, jw represent principal predicate or object and function unit and their semantic properties respectively
  • jT z and jT w represent the real scene and logic of the corresponding subject and predicate respectively True property
  • ⁇ represents the same or operator
  • 0 represents the mapping operator.
  • Axiom 1 and axiom 2 are algorithms for calculating affirmative propositions as true and false
  • the judgment reasoning algorithm program includes an inference algorithm program, which is based on a return sentence of a single sentence in a judgment program, a calculation procedure for a single sentence and a single sentence, and a logical relationship after a sentence sentence is composed of a single sentence, and the logical relationship between the sentences is divided into Full, necessary, necessary and sufficient conditions, and or three types of relations. Because the sufficient condition and the necessary condition have the complementary relationship between the front and the back, the calculation program calculates the sufficient conditional relationship as follows: determining that the return value of the previous sentence is true, then the value of the single sentence is true, and the previous sentence is determined. If the return value is false, the value of the single sentence after the push is true, and the return value of the single sentence is true.
  • the value of the single sentence is true, and the return value of the single sentence is false.
  • the method of calculating the necessary conditional condition is as follows: determining that the return value of the previous sentence is true, then the value of the single sentence is true, and the return value of the previous sentence is false, and the single sentence is pushed. The value must be false, and the return value of the single sentence is determined to be true. The value of the original single sentence must be true, and the return value of the subsequent single sentence is false. The value of the previous sentence can be true; the calculation program calculates the necessary points.
  • the conditional method is as follows: if the return value of the previous sentence is true, then the value of the single sentence after the launch is true, and the return value of the previous sentence is false, then the value of the single sentence must be false, and the return value of the subsequent single sentence is determined. True The value of the pre-sentence single sentence must be true. If the return value of the subsequent single sentence is judged to be false, the value of the pre-sentence single sentence must be false.
  • the method for calculating or calculating the calculation program is as follows: determining that the return value of one of the single sentences is true, then introducing another single sentence must be false; and determining that the return value of one of the single sentences is false, then another single sentence must be introduced.
  • the method for calculating the relationship between the calculation program is: determining that the return value of each single sentence in the precondition is true, then the conclusion of the post-release is true, and the return condition of the single sentence in the precondition is false, then after the launch The conclusion of the piece is false.
  • the operation program generating module of the present invention is configured to generate an operation program according to the judgment inference conclusion obtained by the calculation program, wherein the method for generating the judgment operation program is: a program whose return value is true, in the original application program The semantic property defined on the function, the corresponding object and function and the scene generation operation program configured in the resource library; the program whose return value is falsely defined, after the semantic property of the function defined in the original application is reversed, The corresponding object and function and the scene generated in the resource library generate an operation program; the method for generating the inference operation program is: the reasoning is a true conclusion program, according to the semantic properties of the function defined in the original application, the corresponding object and function and The scene generation operation program in the configuration resource library; the reasoning is a false conclusion program, according to the semantic property of the function defined in the original application, the corresponding object and function and the scene generation operation program configured in the resource library .
  • the reasoning is that the program can be really concluded. After the semantic properties or reversals defined in the original application conclusions,
  • Reasoning is the highest form of human brain thinking. It is a logical thinking method that is known to introduce unknowns. The so-called known is to solve for a problem, the value of the true and false nature of the preconditions is known. With this known condition, the latter can be concluded with sufficient justification based on the sufficient conditional relationship. This is the highest and most basic intelligent way of thinking that the human brain uses logical reasoning to solve problems. In addition, individuals can experience or see the logic of this form and the reliability and validity of its reasoning. Specifically:
  • Jh1 ⁇ jh2 ⁇ (1(jh1) ⁇ 1(jh2)) ⁇ 1(jh1 ⁇ jh2) ⁇ (0(jh1) ⁇ (1(jh2)) ⁇ 0(jh1 ⁇ jh2))
  • the above reasoning method can be used to calculate the reasoning of various connection relations that may occur between propositions.
  • the computer By symbolizing and informatizing the thinking form of the human brain to judge and reason, the computer learns and works in a smart way with brain-like judgment and reasoning.
  • the machine can obtain the human-computer interaction result that meets the application requirements, and ensure the logical correctness of the output.
  • the specific implementation is as follows:
  • Example 1 Trump is an American.
  • JHAction jha new JHAction()
  • Semanteme semanteme jha.JHSemanteme(say);
  • Boolean fal LanguagComparisonEreality(z1j,w1j,RLV);
  • Example 2 Trump is Chinese.
  • JHAction jha new JHAction()
  • Semanteme semanteme jha.JHSemanteme(say);
  • Boolean fal LanguagComparisonEreality(z1j,w1j,RLV);
  • Example 3 If Trump is an American, then Trump is not Chinese.
  • JHAction jha new JHAction()
  • Semanteme semanteme jha.JHSemanteme(say);
  • Example 4 Only Trump is not an American, Trump may be Chinese.
  • JHAction jha new JHAction()
  • Semanteme semanteme jha.JHSemanteme(say);
  • Example 3 (sufficient and necessary condition): The cup has a caliber of 5 cm, which is equal to (if and only) when the cup is qualified.
  • JHAction jha new JHAction()
  • Semanteme semanteme jha.JHSemanteme(say);
  • the correct conclusion that is introduced here is that the cup is unqualified. That is, if the predecessor of the necessary and sufficient conditions is false, the latter part must be false. If it is false, it will be true if the nature of the predicate is reversed.
  • the logical inference engine of the present invention can derive real information conforming to the scene or event, and ensure that the generated executable program simulates the judgment intelligence of the person. And the statements of the logical relationship can be linked to form a more complicated logical judgment and reasoning calculation, and the machine simulation human brain thinking can be realized.
  • the brain-like artificial intelligence service platform of the invention comprises:
  • the artificial intelligence product developer registers the landing platform, downloads the SDK toolkit, and uses the learning knowledge method and calls the semantic analyzer to create a brain-like knowledge base that satisfies the requirements of the development application product;
  • the platform listens or reads the natural language of the input and calls the semantic analyzer to automatically generate an application request or instruction designed by the artificial intelligence product developer;
  • the present invention can utilize the network cloud platform to integrate the semantic analyzer, the learning method and the logical inference engine into a universal brain intelligence tool, thereby realizing the wide application of brain-like artificial intelligence.
  • All kinds of artificial intelligence developers can access the brain intelligence tools formed by semantic analyzers and logical inference machines through the human-computer interaction cloud platform portal, develop ontology brain knowledge bases and related intelligent products, and end users of various artificial intelligence products. Both can work or interact with the machine through the human-computer interaction cloud platform portal.
  • a brain-like artificial intelligence system based on the human-computer interaction cloud service platform is formed.
  • the computer learning knowledge method of the present invention simulates a human brain cognitive model, a logical inference engine simulates a human brain intelligence mechanism, and an artificial intelligence platform provides an intellectual function that simulates information exchange between people.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Machine Translation (AREA)

Abstract

计算机模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台，其中计算机模拟人脑学习知识的方法包括：建立计算机类脑知识库，包括词库、类库、资源库、智能信息管理库；计算机将调用语义分析器将由自然语言语句单句生成的类基本元素和语义性质以类的方法创建并存储于类库；计算机基于类库中的智能知识元素，针对智能应用需求调用语义分析器生成智能应用程序，并将其存储于智能信息管理库。本方法以人工的方法将人脑以智能计算和判断来认识客观事物的认知模型和基于认知模型进行逻辑推理的智能机制模拟到计算机系统，实现机器模拟人脑的智能功能进行学习和工作，形成了类脑人工智能服务平台。

Description

计算机模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台 技术领域
本发明涉及人计算机领域，具体涉及模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台。
背景技术
人工智能的理想目标，是使机器能以类脑的方式学习和工作。实现这个目标，需要以人工的方法将人的认知模型和智能机制模拟到计算机，即以类脑方法处理计算机中的数据信息，由于哥德尔不完全性定理和图灵停机定理所证明的逻辑形式系统和机械计算能力的局限性，导致这种类脑人工智能的技术至今未能实现。
所谓哥德尔定理和图灵停机定理证明的逻辑局限性，通俗地说，就是其定理证明，以“肯定”和“否定”或“真”和“假”的形式对命题(自然语句)进行逻辑判断的计算是算法上无解的。而人脑认知的基本功能就是逻辑判断和推理，并且推理是依赖于判断之后进行计算的。因为计算机是算法机器，即是在算法驱动下运行物理装置。如果判断无算法，则说明人脑基于判断的认知思维是不可计算的；如果不将认知思维形式实现为可计算的逻辑算法，就无法将人的认知模型和智能机制模拟到计算机。因而，解决类脑人工智能的关键，就在于解决将自然语言转换计算机程序语言和如何以程序语言模拟人脑逻辑判断推理的算法问题。
针对上述困难，万继华于2006年在《自然辩证法研究》发表关于《基因语义性质的哲学和实证考察》的论文，提出以本质唯一性解决语义二义性的本体信息哲学思想，同时提出了通过抽取命题的语义性质为信息实体对命题进行逻辑判定，以消除在哥德尔不完全性定理中形成语义悖论的基因量子算法。创建了以四组基因量子语义信号编码的真值表判定方法。之后，出版专著《本体逻辑原理与应用》(广东科技出版社2008 年出版)，系统地论述了计算机如何理解人类自然语言的本体哲学理论和逻辑计算技术。并于2008年10月在全国第四届逻辑系统、智能科学与信息科学学术会议上作了《基于哲学本体论的真值演算系统——实现计算机理解自然语言的逻辑方法》的学术报告，其报告以论文编入《逻辑学及其应用研究》的学术论文集(贵州民族出版社2009年出版)。并已于2016年10月获得《将自然语言翻译为计算机语言的方法、语义处理器和人机对话系统》的发明专利权(专利号：ZL201310657042.8)。
上述发明专利中将自然语言翻译为计算机语言的方法，提出了“计算机利用词库和分词规则识别提取自然语言文字中的语法成分、逻辑联结词和逻辑语义，并基于信息管理库将语法成分翻译成表示计算机面向对象语言中的基本元素名称的字符串代码、将逻辑联结词翻译成表示程序控制的程序转移指令代码、将逻辑语义翻译成表示肯定和否定的二进制代码，并由这些代码拼接成为程序语言”的智能技术方案。因为语言是表达人脑认知和思维的信息工具及载体，将自然语言翻译为计算机语言就意味着机器可以按照人以自然语句(命题)给出的指令进行工作，也意味着机器能以自然语言与人进行交互并能以判断推理方式模拟人脑的逻辑思维，也意味着类脑人工智能的目标可以实现。
但该专利方案并未给出计算机如何学习知识和如何以判断推理的计算方法智能进行工作和执行程序，由此导致该专利没有完整和系统地解决人工智能需要机器以类脑方式进行学习和工作的技术问题。最近由于证明和完善了逻辑判定推理的公理化算法，基于该算法可形成机器判断推理和模拟人脑学习及工作的人工智能系统。
发明内容
针对上述技术问题，本发明提供一种计算机模拟人脑学习知识的方法，其包括以下步骤：
(1)建立计算机类脑知识库，包括词库、类库、资源库、智能信息管理库、其中：
词库，用于存储以自然语言表示场景或事件的词语及与词语对应的词性；
类库，用于存储与自然语言语句的语法成分对应的类基本元素和由类基本元素组成的与主、谓两个逻辑单元对应的真的性质，其中：对应主语的对象元素的语义性质以肯定性质为真，用二进制代码1表示，对应谓语的函数元素的语义性质以肯定或否定为真，每一个自然语句谓语对应的函数的语义性质只能以肯定和否定中的一种性质为真，用二进制代码1或0表示，1表示肯定为真，0表示否定为真；
资源库，用于存储上述场景或事件的信息资源，并与类库中的类基本元素及对象和函数元素为真的性质对应；
智能信息管理库，用于存储类似人脑管理思维的判断推理算法程序和管理行为的智能应用程序以及所述类库、资源库、词库三者之间的对应关系；
(2)计算机将自然语言语句中表示语法成分的词语及词性读入或添加到词库，然后调用语义分析器将由自然语言单句生成的类基本元素和语义性质以类的方法创建并存储于类库，同时将与类基本元素和语义性质对应的场景进行配置并存储于资源库，其中对象、函数的语义性质与资源库对应场景的性质保持一致为真；对象对应场景的性质以肯定1为真，函数对应场景的性质以肯定1或否定0为真，其中语义性质为1的函数对应性质以肯定为真的场景、语义性质为0的函数对应性质以否定为真的场景，由此形成以主、谓概念对应对象、函数单元的逻辑知识元素；
(3)计算机基于类库中的智能知识元素，针对智能应用需求调用语 义分析器，将由自然语言单句、复句或语句集来满足应用需求的自然语言程序生成智能应用程序，并将其存储于智能信息管理库。
作为优选，所述词库分为系统词库、私人词库和公共词库，系统词库用于存储逻辑联结词和生成语义性质的否定词，私人词库用于存储用户自定义的专用词语对应其专用领域或区块性的类库和资源库，公共词库用于存储词性规范的公共词语。
作为优选，所述类库包括本体异构函数，所述本体异构函数是将不同词语或术语指称同一场景或场景元素的表述同时对应或定义为同一词语或术语的方法。
本发明还提供一种根据计算机模拟人脑学习知识的方法构造的逻辑推理机，采用软件或硬件形式，包括知识信息获取模块、判断推理计算模块、操作程序生成模块、操作程序执行模块；所述知识信息获取模块获用于获取智能信息管理库中的应用程序和以判断推理计算应用程序的算法程序；所述判断推理计算模块用于以获取的公理算法程序对应用程序进行判断推理计算；所述操作程序生成模块用于根据逻辑计算的结果生成可操作的判断推理结论程序；所述操作程序执行模块用于执行判断推理结论程序。
作为优选，所述判断推理算法程序包括判断算法程序和推理算法程序，判断算法程序是以单句程序中的函数语义性质的二进制代码与类库中表示其对应函数为真的语义性质的二进制代码进行相同或不同的比较计算，如果二者相同即为11或00，则判断单句程序为真并返回判定值为1；返回值为1的单句程序，按获取应用程序中的语义性质及对象和函数名对应其场景资源判定该程序；如果二者不同即为10或01，则判断为假并返回判断值为0；返回值为0的单句程序，按获取应用程序中的语义性质变反后，以与对象和函数相符的语义性质对应其场景资源判定该程序； 所述推理算法程序是基于单句在判定程序中的返回值和由单句及单句与单句形成的复句之间的逻辑关系推出结论的计算程序；语句间的逻辑关系分为充分、必要、充要条件关系和或、与三类关系。
作为优选，所述计算程序计算充分条件的方法为：判定前件的返回值为真则推出后件的值必真、判定前件的返回值为假则推出后件的值为可真、判定后件的返回值为真则推出前件的值为可真、判定后件的返回值为假则推出前件的值必假；所述计算程序计算必要分条件的方法为：判定前件的返回值为真则推出后件的值为可真、判定前件的返回值为假则推出后件的值必假、判定后件的返回值为真则推出前件的值必真、判定后件单句的返回值为假则推出前件的值为可真；所述计算程序计算充要分条件的方法为：判定前件的返回值为真则推出后件的值必真、判定前件的返回值为假则推出后件的值必假、判定后件的返回值为真则推出前件的值必真，判定后件的返回值为假则推出前件的值必假。
作为优选，所述计算程序计算或关系的方法为：判定其中一个单句的返回值为真，则推出另一个单句必假；判定其中一个单句的返回值为假，则推出另一个单句必真。
作为优选，所述计算程序计算与关系的方法为：判定前提条件中的每一个单句的返回值为真，则推出后件的结论为真，判定前提条件中有一个单句的返回值为假，则推出后件的结论为假。
作为优选，所述操作程序生成模块用于根据所述计算程序得出的判断推理结论生成操作程序，生成判断操作程序的方法是：以返回值为真判定的程序，按获取应用程序中定义在函数上的语义性质，对应类库中的对象和函数以及配置在资源库中的场景生成操作程序；以返回值为假判定的程序，将获取应用程序中定义在函数上的语义性质变反后，对应类库中的对象和函数及配置在资源库中的场景生成操作程序；由多个单 句组成的程序，按顺序控制结构以所述生成判断操作程序的方法生成操作程序；生成推理操作程序的方法是：把一个推理关系的应用程序，分为分析程序语句和操作程序语句；分析程序是推出结论语句的程序，操作程序是被推出作为结论的语句程序；推出为真的结论语句程序，按获取应用程序中定义在函数上的语义性质，对应类库对象和函数及配置在资源库中的场景生成操作程序；推出结论为假的语句程序，按获取应用程序中定义在函数上的语义性质变反后，对应类库对象和函数及配置在资源序中的场景生成操作程序。推出结论为可真的语句程序，按获取应用程序结论语句中定义在函数上的语义性质原值或变反后，对应类库对象函数及配置在资源库中的场景生成操作程序。
本发明再提供一种利用逻辑推理机形成的类脑人工智能服务平台，其包括：
(1)将所述语义分析器、类脑学习知识的方法和逻辑推理机集成为通用智力工具，创建网络人工智能服务台平，形成网络共享类脑服务功能；
(2)人工智能产品开发方注册登陆平台，下载SDK工具包，利用学习知识方法并调用语义分析器创建满足开发应用产品需求的类脑知识库；
(3)终端用户基于人工智能产品开发方的应用产品以自然语言发出应用请求或指令传输到服务平台；
(4)平台监听或读取输入的自然语言并调用语义分析器自动生成人工智能产品开发方设计的应用请求或指令；
(5)调用逻辑推理机计算和执行人工智能产品开发方提供的满足终端用户应用需求的智能知识或工作程序，完成人机交互及其工作任务。
从以上技术方案可知，本发明首先提供一种计算机模拟人脑学习知识的方法，可使计算机以类脑方式学习知识；然后提供一种逻辑推理机，其可进行判断推理计算；本发明以人工的方法将人脑以智能计算和判断 来认识客观事物的认知模型和基于认知模型进行逻辑推理的智能机制模拟到计算机系统，实现机器模拟人脑的智力功能学习知识，形成了类脑人工智能服务平台。
附图说明
图1是本发明的机器模拟人脑学习的方法的示意图。
图2是本发明的逻辑推理机的结构及计算流程示意图。
图3是本发明的类脑人工智能服务平台的工作流程示意图。
具体实施方式
下面结合图1、图2和图3详细介绍本发明，在此本发明的示意性实施例以及说明用来解释本发明，但并不作为对本发明的限定。
一种计算机模拟人脑学习知识的方法，其包括以下步骤：
(1)建立计算机类脑知识库，包括词库、类库、资源库、智能信息管理库，其中：
词库，用于存储以自然语言表示场景或事件的词语及与词语对应的词性；
资源库，用于存储上述场景或事件的信息资源，并与自然语句中的主、谓概念的语义性质对应；
类库，用于存储与自然语言语句的语法成分对应的类基本元素和由类基本元素组成的与主、谓两个逻辑单元对应的逻辑性质，其中：对应主语的对象元素的逻辑性质以肯定性质为真，用二进制代码1表示，对应谓语的函数元素的逻辑性质以肯定或否定性质为真，且只能以其中的一种性质为真，用二进制代码1或0表示，1表示肯定性质为真，0表示否定性质为真；
智能信息管理库，用于存储类似人脑管理思维的判断推理算法程序和管理行为的智能应用程序以及所述类库、资源库、词库三者之间的对应 关系；
(2)计算机将自然语言词句中表示语法成分的词语及词性输入或添加到词库，然后调用语义分析器将由自然语言语句单句生成的类基本元素和语义性质以类的方法创建并存储于类库，同时将与类基本元素和语义性质对应的场景进行配置并存储于资源库，其中对象、函数的语义性质与资源库对应场景的逻辑性质保持一致为真；对象与对应场景的逻辑性质以肯定1为真，函数与对应场景的逻辑性质以肯定1或否定0为真；其中语义性质为1的函数对应逻辑性质为肯定的场景为真、语义性质为0的函数对应逻辑性质为否定的场景为真，由此形成以主、谓概念对应对象、函数单元的逻辑知识元素；这是计算机以类脑逻辑形式学习概念为单位的知识元素或语素的方法。
(3)计算机基于类库中的智能知识元素，针对智能应用需求调用语义分析器，将由自然语言单句、复句或语句集来满足应用需求的自然语言程序自动生成智能应用程序，并将其存储于智能信息管理库。这是计算机以类脑逻辑形式学习序列性或系统性知识的方法。
从以上可知，本体类脑知识库模拟了人脑记忆和学习的方法，可概括为以下三个层次：
一是学习词语概念与其指称的客观事物，即学习以词语表示场景或事件的知识。如”太阳”这个和天空中的实体太阳对应的关系就是这个层面的知识。即把一个词语的字符串作为记号对应到它所表示的场景或事件。本方案将这些知识保存到词库、资源库及类库，其中词语知识保存到词库、场景或事件的素材知识保存到资源库(数据库)、对场景或事件的操作和识别知识保存到类库，并一一对应。将这些知识存储在机器中就相当于使机器学到了这些关于词语的知识。
二是学习以语句(命题)的主、谓概念为基本逻辑结构形成的逻辑认 知和判断知识，即认识“什么是什么、什么不是什么、什么能怎么样、什么不能怎么样”等等。本体类脑知识库将这个层面的知识表示到类库，并将语句中的各种语法成分对应为类基本元素，其中主语表示为对象、谓语表示方法或属性、其他成分表示为参数，同时将其与词库和资源库中的词语及场景对应。并且以主语和谓语部分将其分为两个逻辑单元，同时将各个逻辑单元的逻辑语义性质与其所表示的场景状态的逻辑性质相对应，使机器能以类脑形式理解和判断知识。将这些知识存储在机器中就相当于使机器学到了这些关于语句的知识。
三是学习以逻辑联结关系将单句组成复句或多重复句的语句集的知识。本发明将这些知识翻译为计算机语言的方法使之形成应用程序，并将这些知识存储到信息管理库中，同时与类库、资源库和词库中的知识关联起来，并通过所述语义分析器的解析翻译和逻辑推理机的判断推理计算形成智能知识体系。即各种智能应用程序。具体的翻译方法为专利ZL201310657042.8所述。即将自然语言文字中的语法成分、逻辑联结词和逻辑语义对应翻译，具体是分别将语法成分的字符串代码翻译成表示计算机面向对象语言中的基本元素名称的字符串代码、将逻辑联结词翻译成表示程序控制的程序转移指令代码、将逻辑语义翻译成表示肯定和否定的二进制代码，再将该三种代码拼接成应用程序。同时通过所述逻辑推理机生成预设的可执行智能程序，形成系统的智能知识管理体系。机器将这些知识存储下来就相当于使机器学到了按自然语言作为程序去完成各种工作任务的智能知识。这种方法还可拓展到机器通过读入或听入自然语言就能形成知识的主动学习层次。
所述学习知识方法中的词库区分了系统词库、私人词库和公共词库。系统词库用于存储逻辑联结词和生成语义性质的否定词，私人词库用于存储用户自定义的专用词语对应其专用领域或区块性的类库和资源库， 公共词库用于存储词性规范的公共词语。这种方法有利于形成知识区块化，改变当前机器深度学习依赖大数据分析的高资源消耗和大数据垄断局面，也有利将各个不同领域的知识以区块链的方式联结，形式强大而广泛的智能知识支持及应用，使网络系统的知识真正超越人脑。
所述学习知识方法中的类库包括了本体异构函数，这种将以不同自然语言词语或术语指称同一场景或场景元素的表述，同时指向或定义为同一词语或术语的方法，为实现以自然语言方式建立人机交互服务平台及其应用提供极大方便，有效保障了终端用户能以轻松自如的方式基于人工智能平台进行人机交互。
计算机模拟人脑学习，是指将人的认知模型和智能机制模拟到计算机系统而使机器具备类脑智力功能，上述本体类脑知识库的建立，使得机器具备了类脑的学习方式和记忆功能。
本发明的逻辑推理机采用软件或硬件形式，包括知识信息获取模块、判断推理计算模块、操作程序生成模块、操作程序执行模块；所述知识信息获取模块用于获取智能信息管理库中的应用程序和以判断推理计算应用程序的算法程序；所述判断推理计算模块用于以获取的公理算法程序对应用程序进行判断推理计算；所述操作程序生成模块用于根据逻辑计算的结果生成可操作的判断推理结论程序；所述操作程序执行模块用于执行判断推理结论程序。本发明的逻辑推理机采用四大模块实现对主谓语程序单元的逻辑判断推理计算，从而使由自然语言转换而来的包括主、谓语程序单元的程序语言，能够生成符合人脑判断推理思维和人工智能应用需求的可执行智能程序，为类脑人工智能的应用提供有效的技术解决方案。
具体来说，所述判断推理算法程序包括判断算法程序，判断算法程序是以单句程序中的函数语义性质的二进制代码与类库中表示其对应函 数的逻辑性质的二进制代码进行相同或不同的比较计算，如果二者相同为11或00，则判断单句程序为真并返回判定值为1；返回值为1的单句程序，按应用程序中原有的语义性质及对象和函数名对应其场景资源定义该程序；如果二者不同为10或01，则判断为假并返回判断值为0；返回值为0的单句程序，按应用程序中原有的语义性质变反后，以对象和函数名对应其场景资源定义该程序。
上述计算过程的逻辑可靠性和普遍有效性是直观可证的。因为以程序语言中的主、谓程序单元的语义性质值和类库中描述真实场景的主、谓逻辑单元的性质值进行映射计算，等于以主、谓概念与真的场景对应映射。如果比较二者的性质相同则判断为真，不同则判断为假，这是直觉上自然为真的。如果被判定为真则值一定与被判定程序的值相同，如果被判定为假，则与被判定程序中的值不同。因为逻辑性质只有真假二种，二者不同则必为假，将为假的值変反则又必为真。由此无论被判定程序是真还是假，都能得出判定为真的计算结果。智能程序生成模块根据该判定为必真的值生成对应的可执行智能程序，直观地保障了程序的逻辑可靠性。具体来说：
(1)将主谓语程序单元中对应主语、谓语部分的字符串代码分别与类库中对应的两个逻辑单元的对象和函数的字符串代码进行比较，比较结果完全相同，则产生与所述主谓语程序单元中对应主语、谓语部分的语义性质值相同的二进制比较值，即二进制代码；比较不相同，则产生与所述主谓语程序单元中对应主语、谓语部分的对象和函数的语义性质值相反的二进制值。例如，“奥巴马是前任美国总统”，转换后的主语、谓语部分的语义性质值为11，且对应主语、谓语部分的字符串代码与类库中的对应的两个逻辑单元的字符串代码完全相同，从而得到的比较值为11；再如，“奥巴马不是前任美国总统”，转换后的主语、谓语部分 的语义性质值为10，且对应主语、谓语部分的字符串代码与类库中的对应的两个逻辑单元的字符串代码不同，从而得到的比较值为11。
(2)将比较值与所述主谓语程序单元中对应主语、谓语部分的语义性质值进行对应映射计算，得到为真或为假的判定值；例如，“奥巴马是前任美国总统”，得到的比较值为11，将其与主谓语程序单元中对应主语、谓语部分的语义性质值11进行映射，得到判定值为真，并返回判判定值为1。这意味着该主谓语程序单元表示的语句是与真实场景相符合的；再如，“奥巴马不是前任美国总统”，得到的比较值为10，将其与类库中对应主语、谓语部分的对象、函数语义性质值11进行映射计算，得到判定值为假，并返回判定值为0。这意味着该主谓语程序单元表示的语句与真实场景或事实不相符。
(3)再如果二者相同为11或00，且返回判定值为1的，则按应用程序中原有的语义性质及对象和函数名对应其场景资源定义该程序；如果二者不同为10或01，且返回判断值为0的；则按应用程序中原有的语义性质变反后，以对象和函数名对应其场景资源定义该程序。从而得到与“奥巴马是不前任美国总统”的谓语语义性质相反的值，生成“奥巴马是前任美国总统”的可执行程序。上述计算模型可表示公式：
Figure PCTCN2018000143-appb-000001
其中：h表示命题变量；j表示语义和逻辑性质変量；jz，jw分别表示主谓语或对象和函数单元及其语义性质；jT z、jT w分别表示对应主语、谓语部分的真实场景及逻辑为真性质；⊙表示同或算符；0表示映射算符。上述判断计算的编码模型只有4种，这4种编码模型穷尽了人脑用于命题或语句判断的所有计算模型。在这4种判断计算模型中任何一个为真或为假的命题，经运用该编码模型计算后，都能得到判定为真的正确结论。该编码计算的过程使机器模拟了人脑的判断思维过程。以下为4种 编码的计算公理模型(符号＝＞表示判定)：
公理1：
Figure PCTCN2018000143-appb-000002
公理2：
Figure PCTCN2018000143-appb-000003
公理3：
Figure PCTCN2018000143-appb-000004
公理4：：
Figure PCTCN2018000143-appb-000005
公理1和公理2是计算肯定命题为真和假的算法，公理3和公理4是计算否定命题为真和假的算法。因为每一个命题中的主语和谓语总是以其字符串为记号而自然对应于具有唯一性质的事实或场景的，多个记号对于同一场景的已被本体异构函数定义同义词。并且每个命题都只能有肯定或否定和为真或假的语义和逻辑性质，如公理1中的1(1h)＝1z⊙1w表示肯定命题判定的返回值为真则命题的值不变、公理2中的0(1h)＝1z⊙0w表示肯定命题判定的返回值为假则命题的值通过谓语变反而成真、公理3中的1(0h)＝1z⊙0w表示否定命题判定的返回值为真则命题的值不变、公理4中的0(0h)＝1z⊙1w表示否定命题判定的返回值为假则命题的值通过谓语变反而成真。所以，能够直观看出公理1至4的算法在逻辑上满足了任一命题的判定计算。由于命题中的主、谓成分与程序中对象、函数对应，所以公理1至4满足所有程序语句的逻辑判断算法需求。
判断推理算法程序包括推理算法程序，该推理算法程序是基于单句在判定程序中的返回值和单句与单句以及由单句组成语句集之后的逻辑关系推出结论的计算程序，语句间的逻辑关系分为充分、必要、充要条件关系和或、与三类关系。因为充分条件和必要条件存在前后件互补的关系，所以所述计算程序计算充分条件关系的方法为：判定前件单句的返回值为真则推出后件单句的值必真、判定前件单句的返回值为假则推出 后件单句的值为可真、判定后件单句的返回值为真则推出前件单句的值为可真、判定后件单句的返回值为假则前件单句的值必假；所述计算程序计算必要分条件的方法为：判定前件单句的返回值为真则推出后件单句的值为可真、判定前件单句的返回值为假则推出后件单句的值必假、判定后件单句的返回值为真则推出前件单句的值必真、判定后件单句的返回值为假则推出前件单句的值可真；所述计算程序计算充要分条件的方法为：判定前件单句的返回值为真则推出后件单句的值必真、判定前件单句的返回值为假则推出后件单句的值必假、判定后件单句的返回值为真则推出前件单句的值必真，判定后件单句的返回值为假则推出前件单句的值必假。所述计算程序计算或关系的方法为：判定其中一个单句的返回值为真，则推出另一个单句必假；判定其中一个单句的返回值为假，则推出另一个单句必真。所述计算程序计算与关系的方法为：判定前提条件中的每一个单句的返回值为真，则推出后件的结论为真，判定前提条件中有一个单句的返回值为假，则推出后件的结论为假。
本发明的所述操作程序生成模块用于根据所述计算程序得出的判断推理结论生成操作程序，其中生成判断操作程序的方法是：以返回值为真定义的程序，按原有应用程序中定义在函数上的语义性质，对应对象和函数及配置在资源库中的场景生成操作程序；以返回值为假定义的程序，将原有应用程序中定义在函数上的语义性质变反后，对应对象和函数及配置在资源库中的场景生成操作程序；生成推理操作程序的方法是：推理为真的结论程序，按原有应用程序中定义在函数上的语义性质，对应对象和函数及在配置资源库中的场景生成操作程序；推理为假的结论程序，按原有应用程序中定义在函数上的语义性质变反后，对应对象和函数及配置在资源库中的场景生成操作程序。推理为可真的结论程序，按原有应用程序结论中定义在函数上的语义性质或变反后，对应对象和 函数及配置在资源库中的场景生成操作程序。
推理是人脑思维的最高形式。是一种由已知推出未知的逻辑思维方法。所谓已知就是针对一个问题求解时，对前提条件的真假性质值要是已知的。有了这个已知条件，则可根据充分条件关系，以充分理由推出后件即结论。这是人脑以逻辑推理方法，实现问题求解的最高也是最基本的智能思维方式。并旦毎个人都能在直觉中体验或看出这种形式的逻辑规律及其推理的可靠性和有效性。具体来说：
(1)当一个充分条件前提的命题h 1被计算为真时，则后件h 2自然为真，当其前提中的命题h 1被计算为假时，则后件的h 2自然为真或假。相反，当后件中的命题h 2被计算为真时，则前件的h 1自然为真或假，当后件中的命题h 2被计算为假时，则前件的h 1自然为假。其算法可用公式表达为：
(jh1→jh2)→(1(jh1)→1(jh2))∨(0(jh1)←0∨1(jh2))∨(1(jh2)←0∨1(jh1))∧(0(jh2)→0(jh1))
(2)当一个必要条件前提的命题h 1被计算为真时，则后件h 2自然为真或假，当其前提中的命题h 1被计算为假时，则后件的h 2自然为假。相反，当后件中的命题h 2被计算为真时，则前件的h 1自然为真，当后件中的命题h 2被计算为假时，则前件的h 1自然为真或假。其算法可用公式表达为：
(jh1←jh2)→(1(jh1)←0∨1(jh2))∨(0(jh1)→0(jh2))∨(1(jh2)→1(jh1))∨0((jh2)←0∨1(jh1))
(3)当一个充要条件前的命题h 1被计算为真时，则后件h 2自然为真，当其前提中的命题h 1被计算为假时，则后件的h 2自然为假。相反，当后件中的命题h 2被计算为真时，则前件的h 1自然为真，当后件中的命题h 2被计算为假时，则前件的h 1自然为假。其算法可用公式表达为：
Figure PCTCN2018000143-appb-000006
Figure PCTCN2018000143-appb-000007
(4)当一个或多个与关系语句为充分条件前提的命题时，则需毎个命题都真后件才为真，当其中有一个为假时，则后件自然为假。其算法可用公式表达为：
jh1∧jh2→(1(jh1)∧1(jh2))→1(jh1∧jh2)∨(0(jh1)∧(1(jh2))→0(jh1∧jh2))
(5)当一个或(异或)关系语句的命题出现时，则会自然形式彼真则假，且反之亦然。其算法可用公式表达为：
(jh1∨jh2)→(1(jh1)→0(jh2))∨(0(jh1)→1(jh2))∨(1(jh2)→0(jh1))∨(0(jh2)→1(jh1))
因为人脑推理思维中的逻辑关系只有上述5种，所以通过上述推理方法能计算命题之间可能出现的各种联结关系的推理。通过将人脑进行判断推理的思维形式符号化和信息化，从而实现计算机以类脑判断推理的智能方式学习和工作。机器通过上述可执行程序，可得到符合应用需求的人机交互结果，确保输出的逻辑正确性。具体实施例如下：
示例1：特朗普是美国人。
String say＝”特朗普是美国人。”；
listener.MatchListener(say)；//JH平台监听；
JHAction jha＝new JHAction()；
Semanteme semanteme＝jha.JHSemanteme(say)；
//接收句子的语法成分；
int z1j＝semanteme.getZ1j()；//获取成分中主语逻辑值，
这里的值为1；
int w1j＝semanteme.getW1j()；//获取成分中谓语逻辑值，
这里的值为1；
Int[]RLV＝jha.getComparison(say)；//获取实际比对逻辑值，这里的值为实际主逻辑值1，实际谓逻辑值1；
Boolean fal＝LanguagComparisonEreality(z1j，w1j，RLV)；
//获得语句程序和场景对比值得出这句话的真值即返回值为1。这里以肯定为true：判定”特朗普是美国人”。
示例2：特朗普是中国人。
String say＝”特朗普是中国人。”；
listener.MatchListener(say)；//JH平台监听；
JHAction jha＝new JHAction()；
Semanteme semanteme＝jha.JHSemanteme(say)；
//接收句子语法成分；
int z1j＝semanteme.getZ1j()；//获取成分中主语逻辑值，
这里的值为1；
int w1j＝semanteme.getW1j().//获取成分中谓语逻辑值，
这里的值为1；
lnt[]RLV＝jha.getComparison(say)；//获取实际比对逻辑值，这里的值为实际主逻辑值1，实际谓逻辑值0
Boolean fal＝LanguagComparisonEreality(z1j，w1j，RLV)；
//获得语言和场景的对比值得出这句话的真值即返回值为0。这里以谓语性质变反后：判定”特朗普不是中国人”。
示例3(必要条件)：如果特朗普是美国人，那么特朗普不是中国人。
String say＝”如果特朗普是美国人，那么特朗普不是中国人。”；
JHAction jha＝new JHAction()；
Semanteme semanteme＝jha.JHSemanteme(say)；
String JudgeConditions＝semanteme.getqj()；//获取成分中前件判断 条件“特朗普是美国人”；
通过示例1我们得到了一个返回值为1：那么后件一定成立即为真。即充分条件的前件为真则后件必真。这里推出的正确结论为：因为特朗普是美国人，所以特朗普不是中国人。；
示例4(必要条件)：只有特朗普不是美国人，特朗普才可能是中国人。
String say＝”只有特朗普不是美国人，特朗普才可能是中国人。”；
JHAction jha＝new JHAction()；
Semanteme semanteme＝jha.JHSemanteme(say)；
String JudgeConditions＝semanteme.getqj()；//获取成分中前件判断条件“特朗普不是美国人”；
通过示例1我们得到了一个判断返回值为0，那么后件不成立的即必假。即必要条件的前件为假则后件必假。为假的判定结论语句的谓词要变后才为真。这里推出的的正确结论为：因为特朗普是美国人，所以特朗普不是中国人。
示例3(充要条件)：杯子的口径为5厘米，等于(当且仅)当杯子合格。
String say＝”杯子的口径为5厘米，等于杯子合格。”；
JHAction jha＝new JHAction()；
Semanteme semanteme＝jha.JHSemanteme(say)；
String JudgeConditions＝semanteme.getqj()；//获取成分中前件判断条件“杯子的口径为5厘米”；
通过示例1和2，我们在加工杯子获取杯子的实际口径数时是不定的，可能等于5厘米也可能不等于5厘米，所以我们得到的实际口径数与标准口径数是不确定的，那么这里有两种可能的返回值，一种为1，一种为0。当返回值为1时，这里推出的的正确结论为：杯子合格。即充要条件的前件为真则后件必真。为真则要将谓词的性质不变。
当返回值为0时，这里推出的的正确结论为：杯子不合格。即充要条件的前件为假则后件必假。为假则要将谓词的性质変反后才真。
从上可知，本发明的逻辑推理机可以推导出符合场景或事件的真实信息，保证生成的可执行程序模拟了人的判断推理智能。并且可将所述吝种逻辑关系的语句联结起来，形成更为复杂的逻辑判断推理计算，真正实现机器模拟人脑思維。
本发明的类脑人工智能服务平台包括：
(1)将所述语义分析器、类脑学习知识方法和逻辑推理机集成为通用智力工具(SDK)，创建网络人工智能服务台平，形成网络共享类脑服务功能；
(2)人工智能产品开发方注册登陆平台，下载SDK工具包，利用学习知识方法并调用语义分析器创建满足开发应用产品需求的类脑知识库；
(3)终端用户基于人工智能产品开发方的应用产品以自然语言发出应用请求或指令传输到服务平台；
(4)平台监听或读取输入的自然语言并调用语义分析器自动生成人工智能产品开发方设计的应用请求或指令；
(5)调用逻辑推理机计算和执行人工智能产品开发方提供的满足终端用户应用需求的智能知识或工作程序，完成人机交互及其工作任务。
由此可见，本发明可利用网络云平台，将所述语义分析器、学习方法和逻辑推理机集成为通用类脑智力工具，由此实现类脑人工智能广泛应用。各类人工智能开发商都能通过人机交互云平台入口，获取由语义分析器和逻辑推理机形成的类脑智力工具，开发本体类脑知识库和相关智能产品，各个人工智能产品的终端用户也都能通过人机交互云平台入口指挥机器为其工作或进行交互。由此就形成了以人机交互云服务平台为智能技术基础的类脑人工智能系统。本发明的计算机学习知识方法模 拟了人脑认知模型、逻辑推理机模拟了人脑智能机制、人工智能平台提供模拟了在人与人之间进行信息交流的智力功能。
上述实施方式仅供说明本发明之用，而并非是对本发明的限制，有关技术领域的普通技术人员，在不脱离本发明精神和范围的情况下，还可以作出各种变化和变型，因此所有等同的技术方案也应属于本发明的范畴。

Claims (10)

  1. 计算机模拟人脑学习知识的方法，其包括以下步骤：
    (1)建立计算机类脑知识库，包括词库、类库、资源库、智能信息管理库，其中：
    词库，用于存储以自然语言表示场景或事件的词语及与词语对应的词性；
    类库，用于存储与自然语言语句的语法成分对应的类基本元素和由类基本元素组成的与主、谓两个逻辑单元对应的真的性质，其中：对应主语的对象元素的语义性质以肯定性质为真，用二进制代码1表示，对应谓语的函数元素的语义性质以肯定或否定为真，每一个自然语句谓语对应的函数的语义性质只能以肯定和否定中的一种性质为真，用二进制代码1或0表示，1表示肯定为真，0表示否定为真；
    资源库，用于存储上述场景或事件的信息资源，并与类库中的类基本元素及对象和函数元素为真的性质对应；
    智能信息管理库，用于存储类似人脑管理思维的判断推理算法程序和管理行为的智能应用程序以及所述类库、资源库、词库三者之间的对应关系；
    (2)计算机将自然语言语句中表示语法成分的词语及词性读入或添加到词库，然后调用语义分析器将由自然语言单句生成的类基本元素和语义性质以类的方法创建并存储于类库，同时将与类基本元素和语义性质对应的场景进行配置并存储于资源库，其中对象、函数的语义性质与资源库对应场景的性质保持一致为真；对象对应场景的性质以肯定1为真，函数对应场景的性质以肯定1或否定0为真，其中语义性质为1的函数对应性质以肯定为真的场景、语义性质为0的函数对应性质以否定为真的场景，由此形成以主、谓概念对应对象、函数单元的逻辑知识元素；
    (3)计算机基于类库中的智能知识元素，针对智能应用需求调用语义分析器，将由自然语言单句、复句或语句集来满足应用需求的自然语言程序生成智能应用程序，并将其存储于智能信息管理库。
  2. 根据权利要求1所述计算机模拟人脑学习知识的方法，其特征在于：所述词库分为系统词库、私人词库和公共词库，系统词库用于存储逻辑联结 词和生成语义性质的否定词，私人词库用于存储用户自定义的专用词语对应其专用领域或区块性的类库和资源库，公共词库用于存储词性规范的公共词语。
  3. 根据权利要求1所述计算机模拟人脑学习知识的方法，其特征在于：所述类库包括本体异构函数，所述本体异构函数是将不同词语或术语指称同一场景或场景元素的表述同时对应或定义为同一词语或术语的方法。
  4. 根据权利要求1或2或3所述计算机模拟人脑学习知识的方法构造的逻辑推理机，采用软件或硬件形式，其特征在于：包括知识信息获取模块、判断推理计算模块、操作程序生成模块、操作程序执行模块；所述知识信息获取模块获用于获取智能信息管理库中的应用程序和以判断推理计算应用程序的算法程序；所述判断推理计算模块用于以获取的公理算法程序对应用程序进行判断推理计算；所述操作程序生成模块用于根据逻辑计算的结果生成可操作的判断推理结论程序；所述操作程序执行模块用于执行判断推理结论程序。
  5. 根据权利要求4所述逻辑推理机，其特征在于：所述判断推理算法程序包括判断算法程序和推理算法程序，判断算法程序是以单句程序中的函数语义性质的二进制代码与类库中表示其对应函数为真的语义性质的二进制代码进行相同或不同的比较计算，如果二者相同即为11或00，则判断单句程序为真并返回判定值为1；返回值为1的单句程序，按获取应用程序中的语义性质及对象和函数名对应其场景资源判定该程序；如果二者不同即为10或01，则判断为假并返回判断值为0；返回值为0的单句程序，按获取应用程序中的语义性质变反后，以与对象和函数相符的语义性质对应其场景资源判定该程序：所述推理算法程序是基于单句在判定程序中的返回值和由单句及单句与单句形成的复句之间的逻辑关系推出结论的计算程序；语句间的逻辑关系分为充分、必要、充要条件关系和或、与三类关系。
  6. 根据权利要求5所述逻辑推理机，其特征在于：所述计算程序计算充分条件的方法为：判定前件的返回值为真则推出后件的值必真、判定前件的返 回值为假则推出后件的值为可真、判定后件的返回值为真则推出前件的值为可真、判定后件的返回值为假则推出前件的值必假；所述计算程序计算必要分条件的方法为：判定前件的返回值为真则推出后件的值为可真、判定前件的返回值为假则推出后件的值必假、判定后件的返回值为真则推出前件的值必真、判定后件单句的返回值为假则推出前件的值为可真；所述计算程序计算充要分条件的方法为：判定前件的返回值为真则推出后件的值必真、判定前件的返回值为假则推出后件的值必假、判定后件的返回值为真则推出前件的值必真，判定后件的返回值为假则推出前件的值必假。
  7. 根据权利要求5所述逻辑推理机，其特征在于：所述计算程序计算或关系的方法为：判定其中一个单句的返回值为真，则推出另一个单句必假；判定其中一个单句的返回值为假，则推出另一个单句必真。
  8. 根据权利要求5所述逻辑推理机，其特征在于：所述计算程序计算与关系的方法为：判定前提条件中的每一个单句的返回值为真，则推出后件的结论为真，判定前提条件中有一个单句的返回值为假，则推出后件的结论为假。
  9. 根据权利要求4所述逻辑推理机，其特征在于：所述操作程序生成模块用于根据所述计算程序得出的判断推理结论生成操作程序，生成判断操作程序的方法是：以返回值为真判定的程序，接获取应用程序中定义在函数上的语义性质，对应类库中的对象和函数以及配置在资源库中的场景生成操作程序；以返回值为假判定的程序，将获取应用程序中定义在函数上的语义性质变反后，对应类库中的对象和函数及配置在资源库中的场景生成操作程序；由多个单句组成的程序，按顺序控制结构以所述生成判断操作程序的方法生成操作程序；生成推理操作程序的方法是：把一个推理关系的应用程序，分为分析程序语句和操作程序语句；分析程序是推出结论语句的程序，操作程序是被推出作为结论的语句程序；推出为真的结论语句程序，按获取应用程序中定义在函数上的语义性质，对应类库对象和函数及配置在资源库中的场景生成操作程序；推出结论为假的语句程序，按获取应用程序中定义在函数上的语义性质变反后，对应类库对象和函数及配置在资源库中的场景生成操 作程序。推出结论为可真的语句程序，按获取应用程序结论语句中定义在函数上的语义性质原值或变反后，对应类库对象函数及配置在资源库中的场景生成操作程序。
  10. 利用权利要求4至9中任意一项所述逻辑推理机形成的类脑人工智能服务平台，其包括：
    (1)将所述语义分析器、类脑学习知识的方法和逻辑推理机集成为通用智力工具，创建网络人工智能服务台平，形成网络共享类脑服务功能；
    (2)人工智能产品开发方注册登陆平台，下载SDK工具包，利用学习知识方法并调用语义分析器创建满足开发应用产品需求的类脑知识库；
    (3)终端用户基于人工智能产品开发方的应用产品以自然语言发出应用请求或指令传输到服务平台；
    (4)平台监听或读取输入的自然语言并调用语义分析器自动生成人工智能产品开发方设计的应用请求或指令；
    (5)调用逻辑推理机计算和执行人工智能产品开发方提供的满足终端用户应用需求的智能知识或工作程序，完成人机交互及其工作任务。
PCT/CN2018/000143 2017-04-17 2018-04-17 计算机模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台 WO2018192269A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/655,550 US20200111012A1 (en) 2017-04-17 2019-10-17 Method for computer simulation of human brain learning, logical reasoning apparatus and brain-like artificial intelligence service platform

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710247831.2 2017-04-17
CN201710247831.2A CN107169569A (zh) 2017-04-17 2017-04-17 一种逻辑推理机、机器模拟人脑学习和工作的方法及人工智能系统

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/655,550 Continuation US20200111012A1 (en) 2017-04-17 2019-10-17 Method for computer simulation of human brain learning, logical reasoning apparatus and brain-like artificial intelligence service platform

Publications (1)

Publication Number Publication Date
WO2018192269A1 true WO2018192269A1 (zh) 2018-10-25

Family

ID=59849671

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/000143 WO2018192269A1 (zh) 2017-04-17 2018-04-17 计算机模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台

Country Status (3)

Country Link
US (1) US20200111012A1 (zh)
CN (2) CN107169569A (zh)
WO (1) WO2018192269A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111859990A (zh) * 2020-07-30 2020-10-30 威海微法信息科技有限责任公司 基于语义逻辑唯一性判断的处理问答数据的方法及系统

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107169569A (zh) * 2017-04-17 2017-09-15 湖南本体信息科技研究有限公司 一种逻辑推理机、机器模拟人脑学习和工作的方法及人工智能系统
CN110472723A (zh) * 2018-05-09 2019-11-19 郑州科技学院 一种机器模拟人脑学习和工作的人工智能方法
JP6600398B1 (ja) * 2018-09-15 2019-10-30 株式会社ブロードリーフ Ａｉ創作物の検証装置
KR102620904B1 (ko) * 2019-01-10 2024-01-03 브레인 코그니티브스 피티이. 엘티디. 자연 솔루션 언어
CN109872244B (zh) * 2019-01-29 2023-03-10 汕头大学 一种任务指导型智慧农业种植专家系统
US10983761B2 (en) * 2019-02-02 2021-04-20 Microsoft Technology Licensing, Llc Deep learning enhanced code completion system
WO2020190897A1 (en) * 2019-03-15 2020-09-24 General Electric Company Using graph patterns to augment integration of models into a semantic framework
US11334767B2 (en) * 2019-09-24 2022-05-17 Hrl Laboratories, Llc System and method of perception error evaluation and correction by solving optimization problems under the probabilistic signal temporal logic based constraints
US11941870B1 (en) 2019-09-24 2024-03-26 Hrl Laboratories, Llc System for action recognition error detection and correction using probabilistic signal temporal logic
US11350039B2 (en) * 2019-09-24 2022-05-31 Hrl Laboratories, Llc Contrast and entropy based perception adaptation using probabilistic signal temporal logic based optimization
US11854252B1 (en) 2019-09-24 2023-12-26 Hrl Laboratories, Llc Automated probabilistic axiom generation and incremental updates
CN112819166A (zh) * 2019-11-15 2021-05-18 万继华 一种基于量子信息计算、存储和通信的编码方法
US11775772B2 (en) * 2019-12-05 2023-10-03 Oracle International Corporation Chatbot providing a defeating reply
CN111260075A (zh) * 2020-01-10 2020-06-09 厦门驿全智能科技有限公司 机器模拟待模拟目标学习和工作的人工智能方法及系统
CN111775158B (zh) * 2020-06-08 2022-04-01 华南师范大学 人工智能伦理规则实现方法、专家系统和机器人
CN111772629B (zh) * 2020-06-08 2023-03-24 北京航天自动控制研究所 一种脑认知技能移植的方法
CN111881665B (zh) * 2020-09-27 2021-01-05 华南师范大学 词嵌入表示方法、装置及设备
CN113139657B (zh) * 2021-04-08 2024-03-29 北京泰豪智能工程有限公司 一种机器思维实现方法及装置
US11604626B1 (en) * 2021-06-24 2023-03-14 Amazon Technologies, Inc. Analyzing code according to natural language descriptions of coding practices

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101763342A (zh) * 2009-12-31 2010-06-30 中兴通讯股份有限公司 生成计算机代码的方法及自然语言解释中心和应用控制端
CN103488625A (zh) * 2012-06-12 2014-01-01 国际商业机器公司 自然语言处理的本体驱动词典生成和含糊解决系统和方法
CN103617159A (zh) * 2012-12-07 2014-03-05 万继华 将自然语言翻译成计算机语言的方法、语义分析器及人机对话系统
US20160154631A1 (en) * 2013-07-12 2016-06-02 Bryant G. CRUSE Method and system for machine comprehension
CN107169569A (zh) * 2017-04-17 2017-09-15 湖南本体信息科技研究有限公司 一种逻辑推理机、机器模拟人脑学习和工作的方法及人工智能系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101344936A (zh) * 2007-06-21 2009-01-14 上海北控智能科技有限公司 一种基于行为科学的计算机辅助识别的方法
CN104809501B (zh) * 2014-01-24 2018-05-01 清华大学 一种基于类脑协处理器的计算机系统
US9588961B2 (en) * 2014-10-06 2017-03-07 International Business Machines Corporation Natural language processing utilizing propagation of knowledge through logical parse tree structures
CN106383835A (zh) * 2016-08-29 2017-02-08 华东师范大学 一种基于形式语义推理和深度学习的自然语言知识挖掘系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101763342A (zh) * 2009-12-31 2010-06-30 中兴通讯股份有限公司 生成计算机代码的方法及自然语言解释中心和应用控制端
CN103488625A (zh) * 2012-06-12 2014-01-01 国际商业机器公司 自然语言处理的本体驱动词典生成和含糊解决系统和方法
CN103617159A (zh) * 2012-12-07 2014-03-05 万继华 将自然语言翻译成计算机语言的方法、语义分析器及人机对话系统
US20160154631A1 (en) * 2013-07-12 2016-06-02 Bryant G. CRUSE Method and system for machine comprehension
CN107169569A (zh) * 2017-04-17 2017-09-15 湖南本体信息科技研究有限公司 一种逻辑推理机、机器模拟人脑学习和工作的方法及人工智能系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111859990A (zh) * 2020-07-30 2020-10-30 威海微法信息科技有限责任公司 基于语义逻辑唯一性判断的处理问答数据的方法及系统
CN111859990B (zh) * 2020-07-30 2023-11-21 韩朝晖 基于语义逻辑唯一性判断的处理问答数据的方法及系统

Also Published As

Publication number Publication date
CN107169569A (zh) 2017-09-15
CN108874380A (zh) 2018-11-23
US20200111012A1 (en) 2020-04-09
CN108874380B (zh) 2021-06-25

Similar Documents

Publication Publication Date Title
WO2018192269A1 (zh) 计算机模拟人脑学习知识的方法、逻辑推理机及类脑人工智能服务平台
Zhu et al. Knowledge-based question answering by tree-to-sequence learning
Yang et al. A survey of knowledge enhanced pre-trained models
CN112288075A (zh) 一种数据处理方法及相关设备
US10706356B1 (en) System and method for understanding human level meaning using a 9-dimensional hypercube of cognitive frames
Liu et al. Visual question answering via attention-based syntactic structure tree-LSTM
CN111194401B (zh) 意图识别的抽象和可移植性
Gao et al. A review on cyber security named entity recognition
Chen et al. Formalisation of product requirements: from natural language descriptions to formal specifications
Xu et al. Building a natural language query and control interface for IoT platforms
CN110781666A (zh) 基于生成式对抗网络的自然语言处理文本建模
Cheng et al. Ontology-based semantic classification of unstructured documents
Ahmed et al. Developing an ontology of concepts in the Qur'an
Bitter et al. Natural language processing: a prolog perspective
Longo et al. A Reactive Cognitive Architecture based on Natural Language Processing for the task of Decision-Making using a Rich Semantic.
CN115345153A (zh) 一种基于概念网络的自然语言生成方法
Shen et al. Knowledge-based reasoning network for relation detection
WO2016055895A1 (en) Natural language processing utilizing logical tree structures and propagation of knowledge through logical parse tree structures
Xue et al. Constructing Controlled English for Both Human Usage and Machine Processing.
CN116991980B (zh) 文本筛选模型训练方法及相关方法、装置、介质及设备
Pan et al. Providing context for free text interpretation
Piad-Morffis et al. A neural network component for knowledge-based semantic representations of text
Kubricht et al. Towards an Automated Language Acquisition System for Grounded Agency
Mukherjee et al. A comparative analysis of permutation combination based and grammatical rule based knowledge provider system
Mulwad et al. Automated Creation and Human-assisted Curation of Computable Scientific Models from Code and Text

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18788281

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18788281

Country of ref document: EP

Kind code of ref document: A1