Bottleneck of AI and response to ontology semantics (Zhao zelin, Gao Xinmin)

Intention was once one of the hottest topics in the 19-20th century. Coincidentally, in the new century's turning point, it was once again favored by people. What is different is that it is no longer just an academic question, but also an engineering nature. Today's philosophy of mind and other specific science that cares about intelligent issues, such as artificial intelligence, computer science, and cognitive science, despite their different ways of thinking, however, in the end, it is found that intention is the unique feature and essential condition of smart phenomena. However, as the crystallization of modern science and technology, computers show the so-called intelligence. Although it is far better than human intelligence in many aspects, it can only convert the form according to the formal rules, instead of actively and consciously associating with external events like human intelligence, that is, there is no meaning involved, or there is no semantics or intention. Therefore, in essence, it is just a syntactic machine, not a semantic machine like a human. Some even believe that the existing machine intelligence is not intelligent at all. Therefore, a bottleneck in AI research is to study how to make intelligent machines intentational, and how to transform syntactic machines into semantic machines. There have been many new solutions around this topic, such as the explanatory semantics of kakhan, the "mind architecture" of mcking, and the non-representative intelligence of Brooks, ontology semantics is one of the dazzling wonders.

Motivation, method and basic category of ontology Semantics

Neenberger (S. nirenburg) and Raskin (v. raskin) said: "Ontology semantics is a theory about natural language meaning and a solution about natural language processing, it uses the constructed world model or Ontology as the basic framework for extracting and expressing the meaning of natural language texts as the prerequisite for introducing knowledge from texts. This scheme also aims to form natural language text based on the meaning of natural language ." [1] This means that ontology semantics has a dual motive. The first is the motivation at the application or engineering level, and the second is the motivation at the basic theoretical level. The former is more urgent in terms of actual needs. Neenberger and others realized that the biggest problem with the existing machine intelligence is that it can only complete syntax processing or symbol Conversion, which determines that even if it is fast, convenient, and "versatile ", and cannot change its tool role. Because it is critical to human intelligence, that is, it has no intention. The so-called "intention" refers to the fact that there is an external situation concerning nature or directionality, that is, there is an awareness of its things, and there is a transcendence of itself, instead of purely symbolic form conversion. From the perspective of semantics, intention is semantic. Semantics is the symbolic meaning, reference, and true value conditions involved in human intelligence. Obviously, the intention, semantics, meaning, and other words are essentially the same. Because of this, today's intention theories, meaning theories, and semantics have a tendency to converge. However, no machine has shown the above properties so far. Famous American philosopher and cognitive scientist John R. searle) pointed out that the so-called intelligence implemented by computers itself is just a form of symbol processing, and they "have no intention; they are completely meaningless. ...... In the line of linguistics, they are just syntaxes and meaningless. It seems to be the intention of a computer, but it only exists in the hearts of those who program and use the computer for the computer, and those who send the input and interpretation of the output ." [2] If you understand information from the perspective of meaning, you cannot even say that a computer has the function of processing information. He said, "What computers do is not 'information processing', but Processing Form symbols. Program The compiler and the computer output interpreter use symbols to replace objects in reality. This fact is completely out of the computer scope ." [3] nilunberger and others not only recognized this point, but also stressed that "meaning is a key factor in the future high-end natural language processing, without this ability to utilize the meaning of text, it is impossible for people to make real breakthroughs in natural language processing ,...... In the past, most of the work in this field was not noticed ." [4] The purpose of their research on Ontology semantics is to change this situation, that is, to study how computers use and process the meaning of texts at the technical level, and how to make machine intelligence more intentional.

To accomplish the above tasks, an essential task is to study human intelligence and its operating mechanism, to study the basis and conditions of human intention, especially to reveal the processing mechanism of human natural language, explains its fundamental principles and methods to build a basic model of human semantic processing. To simulate such intelligence, as well as its meaning acceptance, understanding, completion, and output mechanisms, it is necessary to enter the context in which the Speaker and the listener or language producers and consumers communicate with each other, this article discusses how to systematize our concepts of language description, how to systematize the concepts of computing program processing significance, and how to form a representative theory that is more practical and applicable. Determined by this task, ontology semantics puts forward its own methodology principles. Since it is going to accomplish application tasks, it will certainly try to form such a hypothesis, that is, to reconstruct the capabilities of human processing languages and the knowledge and processes they need, that is to find out how natural language processing is possible. To this end, it has such a theoretical preset, that is, its commitment to the intelligent view of weak people, rather than the intelligent view of strong workers. The latter believes that computer programs should not only simulate the human brain in terms of functions, but also simulate the process and details of structural and physical execution. The former advocates that only functional simulation is required to simulate the semantic capabilities of the human brain. Whether the simulation is successful depends on whether the semantic processing capability of the machine is functionally equivalent to that of the human being. Secondly, the unique methodology of ontology semantics also emphasizes that to make machine processing of natural language semantic, it must be based on ontology. Because humans can understand and produce meaning, the fundamental condition is that humans have an ontology schema. With this kind of ontology, once a symbolic word enters the human field of view, it will be classified into a specific semantic domain to obtain a specific Semantic Value. However, the ontology mentioned here has its unique meaning.

Neenberger and others noted that the word "ontology" is ambiguous. Although "ontology" is widely used, it can be divided into two categories: one is pure philosophical usage, and the other is specific scientific and engineering usage. Nilunberger agrees with guarino (N. guarino)'s Viewpoint on "ontology". He calls the previous use "upper-case ontology" and the latter "lower-case ontology ". The lower-case ontology has two forms: Formal Ontology and engineering ontology. Guarillo pointed out: the so-called "form ontology ...... It is a theory about prior Division, such as the reality in the world (physical objects, events, regions, amount of material ......) In the meta-level category used to simulate the world (concept, attribute, quality, status, function, part ......) ". [5] the ontology of engineering is very different from the ontology in philosophy. It does not care about the meaning of the "yes" of metaphysics, nor does it care about the actual ontology classification. It is concerned with the integration factors in the information system and also involves the ontology determination of the results of conceptual analysis. Therefore, it is a veritable engineering ontology. [6] In ontology semantics, "ontology" is different from form ontology and Philosophical Ontology, But it learns useful things from them. Neenberger and others said: their "Ontology construction attempts to get help from the form ontology and Philosophical Ontology" [7] based on their reference, they put forward a new understanding of the "ontology, it establishes an extremely personalized ontology. "The words in each language in ontology semantics use the same ontology to describe the meaning, because it must contain all the meanings in that ontology ." [8] nilunberger and others believe that "to recognize the possibility of expression and processing, one must find such specific significance factors, which are an alternative to the reality of the external world. In ontology semantics, ontology is the most appropriate thing that can direct the external world. It is actually a model of the world. It is constructed based on this. "[9] In short, the ontology mentioned in ontology semantics is just a conceptual framework in the language processing system, its role is to perform ontology positioning on the input words and create conditions for their semantics.

Conception of semantic processing system based on binary ontology Semantics

To answer the question of how machine semantic processing is possible, and to simulate machine processing of human natural language, the first question that must be solved is: how is human natural language processing possible? According to the study of ontology semantics, the possible condition is that humans have the ability to associate it with language, other skills, irrational aspects such as emotion and Will, because the meaning given to words is often emotional. In addition, it is the purpose, plan and procedure of the activity, and finally various knowledge resources.

Ontology semantics holds that the most important condition for human beings to understand and produce meaning is that they have an ontology schema. With the help of it, any language has its own ownership as soon as it enters the mind, and is placed into the category it belongs to. If you hear the word "red, people now have this classification: it refers to attributes, which belong to the same category as "green" and "blue". It is the basic concept and category of an object, so it is not an object. Neenberger and others said: "Ontology semantics is intended to explore the use of these concepts in introspection and reflection. People often talk about attributes. Fictional realities (unicorns or Hermes) and abstract things regard them as existent. However, for us, deciding to put them in the ontology is not rooted in the fact that these are actually referred in natural language, but because we believe: because people have these concepts in their universe, language calls them." [10] Therefore, in the semantic machine model, we must first establish this kind of ontology schema. According to their views, "ontology provides the original language required to describe the meaning of a language's vocabulary unit, and the original language required to describe the meaning of encoding in natural language representation. To provide these things, the ontology must contain definitions of these concepts, which can be understood as a reflection of the world's things and event categories. In terms of structure, ontology is a series of architectures, or a series of ordered attributes-value pairs ." [11] It is an ontology positioning for the meaning of the word term to be characterized, that is, to describe what kind of existence it belongs to, and what its characteristics, properties, and boundary conditions are. For example, when a word "pay" is input, it must first go through the ontology step. In other words, the word must first be represented as an ontology concept and put into the ontology concept system, once this is done, its attributes and values are defined. With the concept framework of ontology, dynamic knowledge resources with meaning representation can be generated continuously in this static knowledge resource. Dynamic Knowledge resources are based on the tasks and requirements of applications.

With a clear and quantified understanding of the conditions required for human processing of natural languages, it is possible for computers to obtain such conditions through the establishment of appropriate networks, in this way, machines are sensitive to meaning and ultimately have semantic processing capabilities. Ontology semantics believes that this is not impossible, at least with great development prospects. In this regard, ontology semantics is a bold attempt to construct a typical semantic processing model. The specific operation is: First let the processor have a static and dynamic knowledge source, and then let it have the corresponding processing capabilities. Based on practice, neenberger and others explain in detail the basic principle and process of machine semantic processing by analyzing the recognized stratified model of natural language processing.

in the View of neenberger and others, intelligent subjects must process at least six basic links to understand the meaning of texts. The first step is text analysis, that is, to generate a formal expression that represents the meaning of the input text. Determined by this task, it must have a analyzer and a generator. In the text analysis process, to input text into the system, you must first retag the text after "pre-processing" and analyze different texts in different languages, genres, and styles, so that the text can be analyzed by the system. The second step is to use static knowledge resources of ecology, morphology, grammar, and NLP for morphological analysis of the labeled text to form a reference form for distinguishing text words. For example, when we encounter the input of the word "book", morphological analysis will analyze it like this: "book, noun, plural", "book, verb, present time, third person, singular. The third step will send them to the analyzer and activate the entrance of this analyzer. This entry contains many types of knowledge and information, such as syntaxes and semantics. It is used to check and purify the results of morphological analysis. For example, the English text may contain words in French, German, and Italian languages, and some ambiguous words. What's more troublesome is that some words have not appeared in the vocabulary analyzer, therefore, it cannot be checked. In these cases, it is necessary to check and identify unfamiliar words, such as the steps and methods to deal with them. Step 4 is syntactic analysis. The fifth step is to determine the Basic Semantic subordination. For example, to establish the proposition structure of future meaning representation, determine which factors will become the topic of these propositions, and determine the attribute location of the proposition.

On this basis, ontology semantics puts forward the complete conception of semantic processing machine. Nilunberger believes that machines must have processors and static knowledge resources to complete text meaning representation. The first step is to analyze the input text with the help of static knowledge resources (ecology, syntax, morphology, NLP, word source and ontology, and fact materials, then, the knowledge resources are used to generate text meaning representation. Both the analysis module and the semantic generator cannot be separated from static knowledge resources. How can we get knowledge resources? It depends on learning. "Ontology semantics must involve Learning: the more they work, the more knowledge they store about the world, and the better the expected results ." [12] in addition to static knowledge, computers must also have dynamic knowledge to complete semantic representation. They are procedural knowledge about meaning representation and knowledge about reasoning types. In addition, the processor must have the dynamic ability to dynamically extract the stored knowledge and apply it to Knowledge Representation. "In ontology semantics, these goals are achieved by associating the representation, vocabulary, and ontology of text meanings," nilenberger and others said ." [13] "our solution to characterize the meaning of text uses two methods: An Example of the concept of ontology and an example of parameters unrelated to ontology. The former provides abstract and non-indexed propositions that are consistent with any possible textual meaning representation examples. These examples are obtained in this way, that is, basic ontology statements are provided. They have specific situations and contain parameter values, such as aspects, methods, and co-pointers ." [14] Here, the concept of ontology is abstract but necessary mainly because it provides the classification of existence and words, for example, for the meaning to be characterized, it first needs to determine which of the objects, attributes, aspects, methods, processes, activities, and quantities it belongs to with the help of this ontology category. In short, for the meaning or meaning of a word, we must first use the ontology to determine the category of existence that it should be included in. On this basis, we use non-ontology parameters to analyze its specific and situational values.

Three features and problems

Ontology semantics has its own distinctive characteristics compared with other artificial intelligence theories and natural language processing systems. First, it emphasizes that the processing of meaning does not require syntactic analysis, at least not through syntactic analysis. In its view, the machine's acceptance, characterization, processing, generation, and output of meaning, or making the machine's syntactic processing semantic or intentional, rely mainly on not the original keyword matching, syntax conversion relies on comprehensive simulation of human intelligence. Second, ontology semantics recognizes that human mental states are intentional and natural languages are semantic dependent on complex factors, and forms a comprehensive solution of research significance on the basis of this understanding. In specific engineering practices, it pays attention to many factors in meaning processing, that is, not only knowledge factors, but also non-knowledge factors hidden in human intelligence, and "internalizes" them into the artificial intelligence system they build in a specific way. Third, ontology semantics attaches great importance to the role of ontology schema in the intentization of human mental states and the semantic nature of natural language, and transforms this cognitive achievement into the field of engineering technology, in this way, the natural language processing system has been boldly explored in terms of the hub and Mechanism of semantic generation, and has achieved enlightening initial results. Fourth, ontology semantics does have important practical significance and broad application prospects. The most important application value is that it can produce representation of text meaning. Because its semantic processing system can use static knowledge resources to analyze input texts, dynamically extract the stored knowledge by means of the dynamic capabilities of the processor, and apply it to knowledge representation, then, the knowledge resources are used to generate text meaning representation, and the man-machine interaction at the meaning communication level is completed by specific output devices.

It can be said that ontology semantics is facing the bottleneck of the current development of artificial intelligence. Based on engineering practice and philosophical reflection, it has not only made great value explorations in building realistic semantic machines, it also provides reference and further reflection for revealing the basic principles of natural language processing. Therefore, ontology semantics is an indispensable choice for the development of artificial intelligence. However, this does not mean that ontology semantics is a complete and impeccable AI theory, although it is a solution proposed in response to the serl Chinese house argument and other complaints about artificial intelligence, it is still subject to such criticism: the meaning and generated meaning processed by the natural language system seem to be inseparable from the interpretation of the design operator. If so, at best, it only has the intention and semantics derived. Second, the biggest problem is that the system established by this theory still lacks initiative, consciousness, consciousness, and purpose, and these properties are exactly the inherent characteristics of human intention. Therefore, it seems that there is still a long way to go to turn machines into intentional autonomous systems like humans.

Note

[1] [4] [6] [7] [8] [9] [10] [11] [12] [13] [14] S. nuremburg and V. raskin, ontological semantics, Cambridge, MA: The MIT Press, 2004, XIII, XIII, pp.138-139, P.154, p.111, p.88, p.135, p191, p160, p160, p174.

[2] [3] sel: the mind, the brain, and the program are contained in Margaret BOLDEN: the philosophy of artificial intelligence, Shanghai Translation Press, 2001, 113rd pages, 116th pages.

[5] n. guarino, "Formal Ontology", in N. guarino et al (eds .), special issue, the role of formal ontology in the information technology, International Journal of Human and computer, 1995 (43) 5-6.