Semantic thesis.

Semantic thesis - a structure for knowledge representation in the form of nodes connected by arcs. The earliest semantic thesis have been developed as an intermediary language for machine translation, and many modern versions are still similar in its characteristics with the natural language. However, recent versions of semantic thesis have become more powerful and flexible to compete frame systems, logic programming and other languages of submission.
Beginning in the late 50th years were created and implemented dozens of versions of semantic thesis. Although the terminology and structure is different, there are similarities inherent in virtually all semantic thesis:
1. semantic thesis nodes represent concepts of objects, events, states;
2. different nodes of a concept are of different values if they are not marked, that they belong to the same concept;
3. arc semantic thesis create relationships between nodes-concepts (labels on arcs indicate the type of relationship);
4. Some relationships between the concepts represent the linguistic case, such as agent, object, recipient, and an instrument (other mean time, spatial, logical relations and relations between individual proposals;
5. concepts are organized by level according to the degree of generalization because, for example, nature, animal, animal, carnivorous,;
But there are differences: the concept of value in terms of philosophy, methods and the commonality of the quantifiers and logical operators, networks and ways of manipulating the rules of the withdrawal, the terminology. This varies from author to author. Despite some differences, networks are convenient for reading and processing machine, as well as powerful enough to represent the semantics of natural language.


Read also:
Knowledge based jobs
Graphs of quadratic equations
Academic proposal format
Grammer parsing

Knowledge based jobs.

Knowledge based jobs introduced the logical formulas in the form of trees, but that little resemble the modern semantic network. Another pioneer was the Charles Sanderz Ac, who used the image recording in organic chemistry.

He formulated the rules of the findings using ekzistentsionalnyh graphs.
In psychological influence Зельц used the space for the submission of certain hereditary characteristics in the hierarchy of concepts. Research of knowledge based jobs had a tremendous impact on the study of tactics in chess, which in turn influenced such thinkers as Simon and Newell.

With respect to linguistics, the first scientist involved in the development of graphic descriptions became Tener. He used a graphic record of its grammatical dependencies. Tener had a great influence on the development of linguistics in Europe.

For the first time, knowledge based jobs have been used in machine translation systems in the late 50's - early 60's. The first such system that has created Masterman, includes 100 primitive concepts such as, for example, the people, a thing, DO NOT. Using these concepts, she described the level of 15000 units of the dictionary, which also had a mechanism for transfer characteristics with gipertipa to subtype. Some systems of machine translation based on the correlation networks Tsekkato, representing a set of 56 different relationships, some of which - declensional relations, subtype relations, member and part of the whole. He used a network consisting of concepts and relationships to guide the actions of parsing and resolving ambiguity.

In knowledge based jobs systems, semantic networks are used to answer various questions, explore learning, remembering and reasoning. In the late 70's networks have been widely disseminated. In the 80 boundaries between networks, frame structures and linear forms of writing are gradually disappearing. Expressive power is no longer a decisive argument for choosing networks or linear forms of writing because the ideas recorded on one form of records can be easily transferred to another. Conversely, particularly important were secondary factors such as readability, efficiency, neiskusstvennost and theoretical elegance, also takes into account for knowledge based jobs the ease of introduction into the computer, editing and printing.
Read also:
Semantic thesis
Graphs of quadratic equations
Academic proposal format
Grammer parsing

Graphs of quadratic equations.

The most simple networks, which are used in artificial intelligence systems - relational graphs of quadratic equations. They consist of nodes connected by arcs. Each node represents a concept, and each arc - the relationship between different concepts. Figure 1 presents the sentence "The dog eagerly glozhet bone. Four rectangles represent the notion of a dog, a process glozheniya, bones, and such characteristics as greed. Inscriptions on the arcs indicate that the dog is the agent glozheniya, bone is the object of glozheniya and greed - this way glozheniya.

The terminology used in this area varies. To achieve some uniformity, nodes connected by arcs, commonly called a graphs of quadratic equations, a structure where there is a whole nest of the nodes, or where there are relationships between the different order of columns is called a network. In addition to the terminology used to explain, as different methods of image. Some use mugs instead of rectangles, some types of writing directly on the arcs, not enclosing them in ovals, and some use abbreviations, such as O or A to denote the agent or object, and some use different types of arrows. Figure 2 shows a graphs of quadratic equations of the conceptual dependency Shenk. <=> Means the agent. INGEST (absorb) - one of the primitives Shenk: YES - to absorb a solid object; drink - the object to absorb liquid; Breathing - absorb gaseous object. Additional windows on the left shows that the transition from bone to the dog an unspecified location.

Since it is difficult to put into the computer, some charts and they occupy much space in the press, many scholars write their entry in the more compact form. For example, the same sentence Owl invited to write in a linear form, using some elements of Figure 1
Read also:
Semantic thesis
Knowledge based jobs
Academic proposal format
Grammer parsing

Academic proposal format.

BOX with the Center in the verb.

Verbs connected with a group of nouns with case relations. For example, the sentence "Mary gave a book to Fred", Mary agent of giving, book an object of this process, and Fred recipient of the verb "give". In addition to case relations in a sentence in natural language also has the means to link the individual academic proposal format. Such relationships are needed for the following:

Unions. The easiest way to combine offers - is to put the union between them. Some unions, such as "and", or "if" denote a logical connection, and some, such as "after", "when", "long", "since as" and "because "express the temporal relations and the cause.
Verbs requiring a subordinate sentence. Declensional frames many verbs require a subordinate academic proposal format, which is usually a direct complement. To this type belong the verbs "talk," "believe," "believe," "know", "be convinced", "danger", "seek", etc.

Determinants related to the whole academic proposal format. Many adverbs and propozitsionnye phrases refer only to the verb, but some define the whole sentence. Such adverbs like "usually" or "might", in most cases placed at the beginning of the academic proposal format. And for example, the word "one" determines the whole story, the next after him.


Modal verbs and times. Such verbs as "may", "can", "must", "should", "would" and "could" have a modal value and relate to the entire academic proposal format, where they occur. Temporal relation can be expressed as the past tense of verbs, and the circumstances "now", "tomorrow" or "once" and others.

Linked discourse. In addition to the relationship expressed in one sentence, there are also higher-order relationships between the various academic proposal formats of the story or any other story. Many of them are not expressed explicitly: temporal relationships and the arguments may be, for example, implicitly expressed by the order of the academic proposal formats one after another in the text.

Precisely because the verb is given such an important role in the academic proposal format, many of the theory make it their central link. This approach takes its origin from the Indo-European language family, where the modality and temporal relations are expressed by a change in verbal form. Consider the following example: "While a dog was eating a bone, a cat passed by unnoticed". In this our proposal format, reported that when the sentence "While a dog was eating a bone" is true, the second sentence "A cat passed unnoticed" is also true. Figure 3 shows a graph with the center of the verb. Union "while" (WHL) connects node PASS-BY to the site EAT. Figure 3 shows that the dog is the agent nezamechaniya (not noticing).


Graphs with the center of the verb - a relational graphs, where the verb is central to any our proposal format. Markers of time and relationships are written right next to the concepts, which are verbs. Counts of conceptual dependency Roger Shenk also use this approach.
Despite the fact that the graphs with the center of the verb rather flexible in its structure, they have several limitations. One of them is that they do not distinguish between the determinants, which relate only to the verb, and the determinants related to the our proposal entirely. Consider the following examples:

The dog greedily ate the bone.
Greedily, the dog ate the bone.

These graphs are also poorly cope with the academic proposal formats located within the other our proposals.
When working with relational graphs having a problem with the whole variety of temporal relations and relations modality. Despite the fact that many scientists use these graphs to solve complex problems, so they are still not developed a general method for solving them. In the above example should mark PAST applies to the entire our proposal, which says that a dog eats a bone, not just to the verb EAT, it is clear that the bone was later eaten by a dog as a whole. It should also be pointed out that the process of passing a cat and a process has not comments her dog took place in one and the same time.



PROPOZITSIONNYE NETWORK.

In propozitsionnyh networks nodes represent entire sentences. These nodes are the points of contact for the relationship between the various our proposals related text. On the other hand they determine the time and modality for the entire context. The following examples illustrate the relationship, for which records are needed propozitsionnye sites:

Sue thinks that Bob believes that a dog is eating a bone.
If a dog is eating a bone, it is unwise to try to take it away from him.

In the first sentence for the verb "think" and "believe" the whole academic proposal format is an addition: Bob believes that "A dog is eating a bone", what he thinks Sue is a more complex sentence-"Bob believes that a dog is eating a bone ". Such proposals nesting within the other our proposals can be repeated arbitrarily many times. To portray such a proposal consultant, you must use propozitsionnye sites that contain nesting boxes. Figure 4 shows propozitsionnaya network for this proposal. Note that (EXP) - experiencer, ie one who feels, connects with Sue THINK and BELIEVE with Bob, but EAT DOG and interconnected elliptical ratio (AGNT). The reason for different types of relations is the fact that think and believe is a condition experienced by people, but eating is the action performed by the agent.


In the second example, presented two proposals are in respect of conditions. Antecedent is the sentence "A dog is eating a bone", and the consequent sentence "It is unwise to try to take it away from him". Infinitives "to try" and "to take" point to the other, nesting proposals. At nesting proposal also indicates turnover "it is unwise". This proposal must also contain correspondence between the "it", "him" and "bone" and "dog". Links match marked by a dotted line. For the formal record of this proposal also uses the universal quantifier and the existence and some elements of logic.
All relational graphs and graphs with the center of the verb have much in common. But among them there are also differences:
1. The inclusion of context, or merely its symbol with a reference to the scheme.
2. Strict breeding: the same concept may or may not occur in two different contexts, none of which do not nest in another.
3. Showing the relationship of conformity. In the context of intersecting, that is when they are one and the same concept is found in two different contexts, these bonds are not listed.
However, this is just stylistic differences, which do not significantly affect the logic of the construction.


Hierarchy.

The hierarchy of types and subtypes is a standard feature of semantic networks. The hierarchy may include essentially: TAXI DOG carnivorous Animals Living Being physical object ESSENCE. They may also include events: Sacrifice ANY ACT event or condition: Extasy HAPPINESS Emotional Condition STATUS. Hierarchy of Aristotle included 10 major categories: substance, quantity, quality, relation, place, time, status, activity and passivity. Some scholars have supplemented it with their classes.
The symbol between more general and more specific symbol is read as: "Do H-tip/podtip."
The term "hierarchy" usually refers to a partial order, where some types are more common than others. The ordering is partial, because many types simply can not be compared with each other. Compare HOUSE DOG and DOG HOUSE meaningless if they compare, but the word is a subtype DOGHOUSE HOUSE, but not DOG. Consider some types of graphs:

Acyclic graph. Any partial ordering can be represented as a graph without cycles. Such a graph has branches that diverge and come together again, which allows some sites have multiple nodes parents. Sometimes this type of graph called confusing.

Trees. The most common type of hierarchy is a graph with one vertex. In such graphs imposed restrictions on acyclic graphs: the top of the graph represents one common type, and every other type of X has only one parent U.

Lattice. Unlike tree nodes in the lattice can have multiple parent nodes. However, there are imposed other restrictions: any pair of type X and Y as a minimum should have a common gipertip Khieu and subtype HiliU. Because of this limitation grille looks like a tree, which has the main peak at each end. Instead of just one vertex array has a single peak, which is gipertipom all categories, and another peak, which is a subtype of all types.
Read also:
Knowledge based jobs
Graphs of quadratic equations
Semantic thesis
Grammer parsing

Grammer parsing.

The main property of the hierarchy is the ability to inherit subtypes qualities gipertipov: all characteristics that are inherent in animals, and mammalian, fish and birds. The basis of the theory of succession is the theory of syllogisms of Aristotle: If A - characteristic of B and B - x-ka C, then A Har-ka of all S.
The advantages of hierarchy and inheritance:
The type hierarchy is an excellent structure for indexing the knowledge base and its efficient organization.
Adherence to a branch with the help of the hierarchy is much faster.

SYNTACTIC ANALYSIS OF LANGUAGE AND ITS generation.

Semantic networks can help the grammer parsing to resolve the semantic ambiguity. Without this kind of representation the whole burden falls on the analysis of language syntax rules and semantic tests. The structure of the semantic network shows clearly how individual concepts are interconnected. When the grammer parsing encounters any ambiguity, it can use the semantic network in order to choose one or the other option. When you work with semantic networks used different techniques grammer parsing.

Parsing, which is based on the syntax. Work is controlled by the grammar grammer parser immediate constituents and operators of building structures and testing them. At that time, as the input data are analyzed, operators for constructing structures create a semantic network, and operators are testing check constraints on a partially constructed network. If no restrictions are not found, then used with the grammatical rule is rejected and the grammer parsing checks for another opportunity. This is the most common approach.

Parser using semantics. The grammer parsing operates using semantics as well as a parser, which is based on the syntax. However, it operates not with the syntactic categories such as the subject group and a group of predicate, and with the high-level concepts such as SHIP and transport.

Conceptual grammer parsing. Semantic network predicts the possible constraints that may arise in the relationship between words, as well as to predict the words that can meet later in the sentence. For example, the verb to give requires an animate agent, but also predicts the possibility of a recipient and an object, which will be given. Schenck was one of the most active supporters of the concept grammer parsing.

Parsing based on the examination of words. Due to the existence of a large number of irregular structures in natural language, many people, instead of going to any universal generalizations, use special dictionaries, a collection of several independent procedures, which are called by the experts of words. Analysis of the proposals considered as a process, carried out jointly by the various word-expert. The main proponent of this approach was Small.

Arguments for and against various grammer parsing techniques are often not based on specific data, and more on an already established opinion. And only one project in practice to compare several types of parsing - the language of semantic representations, a project developed at the University of Berlin. For several years they have created four different types of grammer parsers for the analysis of the German language and its writing into the language of semantic representations, which represents a network.

The first grammer parser was created in the likeness of a conceptual parser Shenk. It was noted that although adding to his vocabulary of new words was pretty easy, but the analysis could be conducted only in simple sentences, and only in relative clauses. Extend the syntactic processing of the parser was difficult.

Second semantically oriented grammer parser was extended network transition. It was easier to compile the syntax, but the apparatus syntax is slower than the first review of the parser.
Then the work was carried out with the word expert grammer parsing. Here you can easily were processed in special cases, but the dispersion between the individual components of the grammar makes it almost impossible its common understanding, support and modification.
Parser, which was created relatively recently - it is syntactically oriented parser, based on common grammar phrasal structure. He most The systematic and generalized and relatively quickly.

These results are basically consistent with the view of other linguists: a syntax-oriented grammer parsers most holistic, but they need a certain set of network operators for a smooth interaction between the grammar and semantic networks.
Generation of the language on a semantic network represents the inverse parsing. Instead of grammer parsing a chain with a view to generating network generator produces grammer parse language network for some chains. There are two ways of generating the language of the semantic network.

1. Generator of the language just to be on the network, turning concepts into words, and relations listed next to the arcs in the relationship of natural language. This method has many limitations.

2. Approaches targeted to control the generation of syntax language with the help of grammatical rules, which use the network in order to determine what the following rule should apply.

However, in practice, both methods have many similarities: for example, the first method is a sequence of nodes that are processed by the generator of language-oriented syntax.

Machine learning.

Graphs and networks are simple concepts for programs that are exploring new structures. Their advantage in learning is the ease of adding and removing, as well as the comparison of arcs and nodes. Below are the programs that are used to study semantic networks.

Winston used relational graphs to describe structures such as arches and towers. The machines are offered examples of correct and incorrect descriptions of these structures, and programs to create graphs, which pointed out all the necessary conditions to ensure that this structure is an arch or tower.
Salveter used graphs with the center of the verb for the submission of case relations, which require different verbs. His program MORAN for each verb displays declensional frame, comparing the same situation before and after their description using this verb.

Schenk developed a theory Memory-Organization Packets for an explanation of how people learn new information from the specific life situations. This MOP-it is a generalized abstract structure, which are not related to any particular situation individually.

Apply in practice.

Semantic networks can be written on almost any programming language on any machine. Most popular in this respect, languages LISP and PROLOG. However, many versions have been created and FORTRAN, PASCALe, C and other programming languages. To store all the nodes and arcs need more memory, although the first system were performed in 60-ies on the machines, which were much smaller and slower than today's computers.

One of the most common languages, designed for recording natural language in the form of networks - is PLNLP (Programming Language for Natural Language Processing) programming language for natural language processing, created Haydernom. This language is used for large grammars with extensive coverage. PLNLP works with two kinds of rules:

1. using the rules of decoding is grammer parsing the language of the linear chain and construct the network.

2. using the encoding rules network is generated by the scan chain, or other language transformed network.

In addition to special language for semantic networks was also developed special hardware. On conventional computers can be successfully carried out operations with language parsing and scanning operation of networks. However, for large knowledge bases to find the necessary rules or access to predznaniyam may take a very long time. To allow different processes take place simultaneously search Falman developed a system NETL, which is a semantic network, which can be used with parallel hardware. So he wanted to create a model of the human brain, which signals can move through different channels simultaneously. Other scientists have developed a parallel software for searching the most likely interpretation of ambiguous sentences of natural language.
Read also:
Knowledge based jobs
Graphs of quadratic equations
Academic proposal format
Semantic thesis

Organs on left side of human body.

Since the late 40-ies scientists have an increasing number of university and
industrial research labs headed for an audacious goal:
construction of computers operating in such a way that the results
their work would be impossible to distinguish from the organs on left side of human body mind.
Patiently move forward in their hard work, researchers
Whether working in the field of artificial intelligence (AI), found
that came to grips with a very intricate, far-going
schimi beyond the traditional computer science. It turned out that first
need to understand the mechanisms of the learning process, the nature of language and chuvs -
idents of perception. It turned out that the creation of machines that simulate
work of the organs on left side of human body brain is required to understand how to operate
its billions of interconnected neurons. And then many researchers
came to the conclusion that perhaps the most difficult problem facing
modern science - knowledge about the functioning of the organs on left side of human body mind, and not merely an imitation of his work. What directly touches upon -
Valo fundamental theoretical problems of psychology. In
Indeed, academics find it difficult even to come to a common point of view relative
PRE of the subject of their investigations - intelligence. Here, as in
parable of the blind men trying to describe an elephant, trying to keep
a long-cherished definition.
Some believe that intelligence - the ability to solve complex problems;
others view it as a learning ability, generalization and ana -
logiyam still others - like the opportunity to interact with the outside world through
communication, perception and awareness of the perceived. Nevertheless, many ISS -
ledovateli AI tend to take a test of machine intelligence proposed
in the early 50-ies the outstanding English mathematician and
in computer science Alan Turing. The computer can be considered
reasonable - Turing argued - if he can make us behav -
rit, that we are not dealing with a machine, but a man.
Mechanical approach.
The idea of thinking machines "human-like", which seemed -
elk would think, move, hear, speak, and generally behave as
living people is rooted in the deep past. The ancient Egyptians and
Romans felt awed before the cult statues, Coto
rye gesticulating and spake prophecies (of course not without the help
priests). Medieval chronicles are full of stories about the machines, how -
GOVERNMENTAL walk and move almost as well as their owners - the people. In the Middle
century and even later there were rumors that someone from the sages have
homunculus (little man-fellows) - real live, ACT -
feel its potential merits. Prominent Swiss physician and estestvois -
pytatel XVI in Theophrastus Bombast von Gogengeym (better known
name Paracelsus) left the leadership in manufacturing homunculus, in co -
torus is described by a strange process, starting with instillation of lo -
shadiny manure tightly corked organs on left side of human body sperm. "We
as gods, - proclaimed Paracelsus. - We repeat the greatest of miracles
Lord's - the creation of man "(4)
Knowledge based jobs
Graphs of quadratic equations
Academic proposal format
Semantic thesis
Grammer parsing