Semantic modeling. Semantic network model

Logical model of knowledge.

A logical model is a formal system - some logical calculus. All knowledge about the subject area is described in the form of formulas of this calculus or rules of inference. Description in the form of formulas makes it possible to represent declarative knowledge, and inference rules provide procedural knowledge. Knowledge representation languages ​​of the logical type were widely used in the early stages of the development of intelligent systems, but were soon supplanted (or, in any case, strongly supplanted) by languages ​​of other types. This is explained by the cumbersomeness of the records based on classical logical calculus. When creating such records, it is easy to make mistakes, and searching for them is very difficult. The lack of visibility and readability (especially for those whose activities are not related to the exact sciences) made it difficult to spread languages ​​of this type.

Frame model of knowledge.

Frame(English frame - frame or frame) proposed by M. Minsky in the 1970s. as a knowledge structure for the perception of spatial scenes. This model has a deep psychological basis. A frame is an abstract image or situation. A frame is also a formalized model for displaying an image. Distinguish sample frames, or prototypes, stored in the knowledge base, and instance frames, which are created to display real situations based on incoming data. The frame model is quite universal, since it allows you to display the entire diversity of knowledge about the world through: - frame-structures, to designate objects and concepts (loan, pledge, bill); - frame-roles(manager, cashier, client); - script frames(bankruptcy, shareholders meeting); - situation frames(alarm, accident, device operating mode), etc. The main advantage of frames as a model of knowledge representation is the ability to reflect the conceptual basis of the organization of human memory, as well as its flexibility and clarity. Special knowledge representation languages ​​in frame networks FRL (Frame Representation Language) and others make it possible to effectively build industrial ES. Frame-oriented expert systems such as ANALYST and MODIS are widely known.

Term semantic means “semantic”, and semantics itself is a science that establishes relationships between symbols and the objects that they designate, i.e. the science that determines the meaning of signs. A semantic network is a directed graph whose vertices are concepts and whose arcs are relationships between them. A characteristic feature of semantic networks is the mandatory presence of three types of relationships: - class - class element; - property - value; - an example of a class element. The problem of finding a solution in a knowledge base such as a semantic network comes down to the problem of finding a fragment of the network corresponding to a certain subnet corresponding to the question posed. The main advantage of this model is that it is in accordance with modern ideas about the organization of human long-term memory. The disadvantage of the model is the difficulty of searching for output on the semantic network. To implement semantic networks, there are special network languages, for example NET, etc. Expert systems that use semantic networks as a knowledge representation language are widely known - PROSPECTOR, CASNET, TORUS.

According to the form of description, knowledge is divided into:

Declarative (facts)- this is knowledge of the form “A is A”. Declarative knowledge is divided into objects, object classes and relations. An object is a fact that is given by its meaning. Object class is the name under which a specific collection of fact objects is united. Relationship- determine connections between classes of objects and individual objects that have arisen within the subject area.

Procedural- this is knowledge of the form “If A, then B.” Procedural knowledge includes sets of rules that show how to derive new distinctive features of classes or relationships for objects. The rules use all types of declarative knowledge, as well as logical connectives. When processing rules, it should be noted that the analysis of relationships is recursive, i.e. one rule causes a deep search of all possible variants of knowledge base objects.

The boundary between declarative and procedural knowledge is very fluid, i.e. the designer can describe the same thing as a relation or as a rule.

For modeling domains, relational DBMSs have become widespread. Their use in a wide variety of applications shows that the relational data model is quite universal. However, designing a relational database in terms of relationships is often a very complex and inconvenient process for the designer.

At the same time, the limitations of the relational data model are manifested in the following aspects:

· The model does not provide sufficient means of representing the semantic content of the data. The semantics of the real subject area must be mapped into the designer's mind in a model-independent manner. In particular, this relates to the problem of representing integrity constraints;

· For many applications, it is difficult to model the domain using flat tables. In some cases, at the very initial design stage, the developer has to describe the subject area in the form of a single (possibly even non-normalized) table.

Although the entire design process takes place on the basis of taking into account dependencies, the relational model does not provide any means for representing these dependencies.

Although the design process begins with identifying some application-relevant domain objects ("entities") and identifying relationships between these entities, the relational data model does not offer any apparatus for separating entities and relationships.

The needs of database designers for more convenient and powerful domain modeling tools are realized when using

= semantic data models =

Any developed semantic data model, like the relational model, includes structural, manipulative And holistic parts.

The main purpose of semantic models is to provide the ability to express the semantics of data.

Most often in practice, semantic modeling is used at the first stage of database design. In this case, a conceptual database schema is produced in terms of the semantic model, which is then manually converted to a relational (or some other) schema. This process is carried out under the control of methods in which all stages of such transformation are quite clearly specified.

Less often automated compilation of the conceptual diagram into a relational model is implemented.

There are two known approaches:

· based on an explicit representation of the conceptual scheme as input information for the compiler;

· construction of integrated design systems with automated creation of a conceptual diagram based on interviews with subject matter experts.

Third possibility, which is just beyond the scope of research and experimental projects, is working with databases in a semantic model, that is, DBMSs based on semantic data models.

Again, two options are considered:

· providing a user interface based on a semantic data model with automatic mapping of constructs into a relational data model (this is a task of approximately the same level of complexity as automatic compilation of a conceptual database schema into a relational schema);

· direct implementation of a DBMS based on some semantic data model.

The closest to the second approach are modern object-oriented DBMSs, whose data models are close to semantic models in many respects (although in some aspects they are more powerful and in others weaker).

Basic concepts of the “Entity - Relationship” model.

Semantic data model Entity - Relationship– model " Entity - Connections"(ER-model).

Most modern approaches to database design (mainly relational) are based on the use of varieties of the ER model. The model was proposed by Chen in 1976. Domain modeling is based on the use of graphical diagrams that include a small number of heterogeneous components.

Due to the clarity of presentation of conceptual database diagrams, ER models are widely used in CASE systems that support automated design of relational databases. Among the many varieties of ER models, the most developed model is used in the CASE system (from ORACLE).

The main concepts of the ER model are:

“essence – connection – attribute”.

Essence is a real or imaginable object, information about which must be stored and accessible. In ER model diagrams, an entity is represented as a rectangle containing the name of the entity. In this case, the name of the entity is the name of the type, and not of some specific instance of this type. The “AIRPORT” entity, with example objects “Sheremetyevo” and “Heathrow”, is shown in Fig. 2.24.

For greater expressiveness and better understanding, the entity name can be accompanied by examples of specific objects of this type.

Each instance of an entity must be distinguishable from every other instance of the same entity (this requirement is somewhat analogous to the requirement that there are no duplicate tuples in relational tables).

Connection is a graphically depicted association established between two entities. This association is always binary and can exist between two different entities or between an entity and itself ( recursive connection).

In any connection, two ends are identified (in accordance with the existing pair of connected entities), each of which indicates end name, end degree(how many instances of this entity are associated), mandatory communication(i.e., whether any instance of a given entity must participate in a given relationship).

A connection is represented as a line connecting two entities or leading from an entity to itself. In this case, at the point of “docking”, connections with the entity are used three-point entry into the entity rectangle if that entity in the relationship can be used a lot of(many) entity instances, and single point entry, if only one entity instance.

Mandatory end connection is depicted solid, A optional –intermittent line.

Like an entity, a relationship is a generic concept; all instances of both pairs of related entities are subject to the rules of association.

In the example shown in Fig. 2.25, the relationship between the entities TICKET and PASSENGER connects tickets and passengers. In this case, the end of the entity with the name “for” allows you to associate more than one ticket with one passenger, and each ticket must be associated with some passenger. The end of the entity named "has" means that each ticket can only be owned by one passenger, and the passenger is not required to have at least one ticket.

End of connection with name "son" determines the fact that one father can have more than one son. End of connection with name "father" means that not every person can have sons.

Attribute An entity is any detail that serves to clarify, identify, classify, numerically characterize, or express the state of an entity. Attribute names are entered in a rectangle representing the entity, under the entity name, and are depicted in small letters, possibly with examples.

Normal forms of ER circuits.

As in relational database schemes, the concept of normal forms is introduced in ER schemes. Note that the formulations of normal forms of ER-schemas make the meaning of normalization of relational schemes clearer.

Here are brief and informal definitions of the first three normal forms:

· duplicate attributes or groups of attributes are eliminated, i.e. implicit entities “disguised” as attributes are identified;

· attributes that depend only on part of the unique identifier are eliminated; this part of the unique identifier defines a separate entity.

· attributes that depend on attributes that are not included in the unique identifier are eliminated. These attributes are the basis of a single entity.

More complex elements of the model include the following:

· Subtypes and supertypes of entities. As in programming languages ​​with developed typical systems (for example, in object-oriented programming languages), the possibility of inheriting an entity type based on one or more supertypes is introduced. Interesting nuances are associated with the need for a graphic representation of this mechanism.

· Many-to-many connections. Sometimes it is necessary to link entities in such a way that there may be multiple instances of the entity at both ends of the relationship ( for example, all members of a cooperative jointly own the property of the cooperative). To do this, a type of connection “many – with –” is introduced.

many."

· Specified degrees of connection. Sometimes it is useful to determine the possible number of instances of an entity participating in a given relationship ( for example, an employee is allowed to participate in no more than three projects at a time). To express this semantic constraint, it is allowed to indicate at the end of the connection its maximum or mandatory degree.

· Cascade deletions of entity instances. Some relationships are so strong (in the case of a one-to-many relationship) that when you delete the reference entity instance (corresponding to the one end of the relationship), you must also delete all entity instances corresponding to the many end of the relationship. The corresponding requirement for "cascading deletion" can be formulated when defining an entity.

· Domains . As with the relational data model, it is sometimes useful to be able to define a potentially valid set of values ​​for an entity (domain) attribute.

These and other more complex elements of the Entity-Relationship data model make it significantly more powerful, but at the same time make it somewhat more difficult to use. When actually using ER diagrams for database design, you need to become familiar with all the possibilities.

Let's look at one of the mentioned elements - entity subtype .

Essence may be divided into two or more mutually exclusive subtypes. Each of them includes common attributes and/or relationships. These common attributes and/or relationships are explicitly defined once at a higher level. Subtypes can define their own attributes and/or relationships. Basically "subtyping" may continue at lower levels, but experience shows that in most cases two or three levels are sufficient.

Essence, on the basis of which subtypes are determined, is called supertype. Subtypes must form a complete set, that is, any instance of a supertype must belong to some subtype. Sometimes for completeness it is necessary to define an additional subtype OTHER.

Example - with upertype AIRCRAFT– shown in Fig. 2.27.

Sometimes it is convenient to have two or more different subtypes of an entity.

For example, the essence of PERSON can be divided into subtypes based on professional characteristics (PROGRAMMER, MILKmaid, etc.), or maybe -

by gender (MAN, WOMAN).

In Fig. Figure 8 shows an example of a semantic network.

The problem of finding a solution in a knowledge base such as a semantic network comes down to the task of finding a fragment of a network or subnet that corresponds to the question posed.

Semantic connection (SS) reflects the relationship of concepts in a conceptual system. In the lexicon, they correspond to lexemes of any kind, including those representing the predicators “less”, “equal”, “if, then”, etc.
Extra-lexical properties of SS are expressed through:

• Rf - reflexivity;
• Nrf - non-reflexivity;
• Arf - anti-reflexivity (not a single reflection);
• Sm - symmetry;
• Ns - asymmetry;
• Ans - antisymmetry (no symmetry);
• As - asymmetry (contextual property - reversing a connection gives another connection from the list);
• Tg - transitivity;

Ntr - intransitivity.

Extra-lexical properties of semantic connections in judgments are checked as follows.

Regarding the combination of the listed properties, SS are divided into types presented in (Table 2.1.).
1. Reflexivity is determined by the substitution criterion:

instead of object A, object B is substituted (AgB -> BgB) and one of the following answers is selected:
quite possible (tautology) ~» Rf;
not excluded -> Nrf;
impossible -> Arf.
Example. Autonomic disorders are accompanied by autonomic disorders. Answer 1 for Com.

2. Symmetry is determined by the permutation criterion:
objects A and B are swapped (AgB - BgA) and the validity of the received proposal is determined. If the answer is affirmative, the statement is assigned the property Sm, otherwise - the property Ns.

Example. Headache is always accompanied by autonomic disorders, and Autonomic disorders are always accompanied by headache. The answer is "No" for Com. This corresponds to the Ns property.
The Ns property is refined into stronger properties: Ans and As. The first occurs for any examples of the analyzed connection. For example, for the connection Com the Ans property holds.

Plus of the model: Easy to implement.

Disadvantage of the model: Poorly structured - with a large number of elements you can get confused, and with an increase in the amount of information, a combinatorial explosion can occur. When creating any thing, any product, any work, a person faces the need for an inevitable choice among a huge number of possible options. Let us see what a simple enumeration of these options can lead to in the following phenomenon. This phenomenon is known in cybernetics as a combinatorial explosion. What kind of “beast” this is is easy to demonstrate with a simple example. Let's say that there is a certain alphabet consisting of only 10 characters (letters). ...
From such an alphabet you can compose 10^^100 texts of 100 letters each. A hypothetical computer capable of processing 10^^18 such texts per second would spend 10^^74 years on the overall analysis of all texts. For comparison, according to modern cosmogonic concepts, ~10^^10 years have passed since the Big Bang of the explored part of the Universe.

A semantic network is an information model of a subject area, in the form of a directed graph, the vertices of which correspond to objects of the subject area, and arcs (edges) define the relationships between them. Objects can be concepts, events, properties, processes. Thus, the semantic network is one of the ways to represent knowledge. The name combines terms from two sciences: semantics in linguistics studies the meaning of language units, and a network in mathematics is a type of graph - a set of vertices connected by arcs (edges). In a semantic network, the role of vertices is played by the concepts of the knowledge base, and the arcs (and directed ones) define the relationships between them. Thus, the semantic network reflects the semantics of the subject area in the form of concepts and relationships.

The idea of ​​systematization based on some semantic relations was proposed by early scientists. An example of this is the biological classification of Carl Linnaeus of 1735. If we consider it as a semantic network, then this classification uses a subset relation, modern AKO.

The ancestors of modern semantic networks can be considered existential graphs, proposed by Charles Peirce in 1909. They were used to represent logical statements in the form of special diagrams. Peirce called this method “the logic of the future.”

The work of the German psychologist Otto Seltz in 1913 and 1922 had an important start in the study of networks. In them, he used graphs and semantic relationships to organize the structures of concepts and associations, as well as study methods of inheriting properties. Researchers J. Anderson (1973), D. Norman (1975) and others used this work to model human memory and intellectual properties.

Computer semantic networks were developed in detail by Richard Richens in 1956 as part of the Cambridge Language Center project on machine translation. The machine translation process is divided into 2 parts: translation of the source text into an intermediate form of representation, and then this intermediate form is translated into the desired language. Semantic networks were just such an intermediate form.

Mathematics allows us to describe most phenomena in the world around us in the form of logical statements. Semantic networks emerged as an attempt to visualize mathematical formulas. The main representation for a semantic network is a graph. However, we should not forget that behind the graphic image there is certainly a strict mathematical notation, and that both of these forms are not competing, but complementary.

The main form of representation of a semantic network is a graph. The concepts of the semantic network are written in ovals or rectangles and are connected by arrows with signatures - arcs. This is the most convenient form for humans to perceive. Its shortcomings appear when we begin to build more complex networks or try to take into account the features of natural language.

Classification of semantic networks:

For all semantic networks, the division by arity and number of relationship types is valid.

According to the number of types, networks can be homogeneous or heterogeneous. Homogeneous networks have only one type of relationship (arrow), for example, the above-mentioned classification of biological species (with a single relationship AKO). In heterogeneous networks, the number of relationship types is more than two. Classic illustrations of this model of knowledge representation represent precisely such networks. Heterogeneous networks are of greater interest for practical purposes, but also more difficult to study.

By arity, networks with binary relations (connecting exactly two concepts) are typical. Binary relationships are indeed very simple and conveniently look like an arrow between two concepts on a graph. In addition, they play an exceptional role in mathematics. In practice, however, you may need relationships that connect more than two objects—N-ary ones. In this case, a difficulty arises - how to depict such a connection on a graph so as not to get confused. Conceptual graphs alleviate this difficulty by representing each relationship as a separate node.

In addition to conceptual graphs, there are other modifications of semantic networks; this is another basis for classification (by implementation).

§ 500. The problem of defining structural-semantic models.

This problem is related to the definition of typical situations (processes, relationships) that can be described by sentences. Despite the infinite variety of specific situations, they can be reduced to certain classes. The situation that is denoted by the sentence is determined by the nature of the process (event) and the roles of the participants in this process. Therefore, in linguistics there have been two ways of defining types of situations: actant-role and predicate-verbal. In the first case, attention is paid to the function (role) of the participants in the process (actants), in the second - to the nature of the process itself. In this case, the types of situations are correlated with the semantic types of predicates and verbs. Since the type of process and the role of the actants of the phenomenon are interdependent, the most effective solution to the problem is to simultaneously take into account the type of process and the functions of the actants.

The beginning of the development of an actant-role representation of the situation is associated with the name of L. Tenier. As noted in § 368, Tenier distinguished three types of actants: first, second and third (prime, second, tiers actant), to which he gave, however, a contradictory interpretation. On the one hand, he considered them as semantic entities (the 1st actant performs the action, the 2nd undergoes it, the 3rd is the actant for whose benefit or to the detriment of which the action is carried out): Alfred (1) donne le livre (2 ) à Charles (3). On the other hand, when identifying them, he proceeded from syntactic forms, so that the 1st actant is identified with the subject, the 2nd with the direct object, and the 3rd with the indirect. In the phase Alfred a été frappé par Charles, the word Charles is considered as a “second actant of the passive”, since it is introduced by a preposition, although it indicates a real agent.

Greimas made an important step in the development of semantic syntax. He made a distinction between semantic actants, which stand out in the situation described, and syntactic ones (see the example of Adam, Eve and the apple in § 368). He expanded the number of actants, presenting them in the form of an opposition of three pairs: subject/object, sender/addressee, assistant/opponent. Greimas's merit is also his attempt to connect semantic actants with literary facts. The fact is that it is literature (from folklore and anecdotes to great novels) that describes all the situations in which a person and the world around him may find himself. A structural analysis of Russian fairy tales by the Soviet researcher V. Propp and the dramatic works of the French literary critic Souriot showed that in works of various contents constant typical functions are distinguished: hero (subject), object, helper, counteractor, etc. This confirmed that in semantic actants, identified by linguists, reflect the categories of real phenomena of the objective world.

Subsequent research took the path of clarifying the nomenclature of actants and striving to connect them with the type of the process itself. If actants are defined as any substance affected by the process, then their number should include not only the subject, object, addressee, but also the instrument and the substance associated with another by relations of belonging. Actants should also include temporal and local specifiers, since they are designated primarily by nouns (à la maison, l "année passée); their adverbial designations (ici, alors) represent deictic replacements. These actants correspond to seven substantive CPs in the sentence structure: subject , direct and indirect addition, addition of a tool, addition to a name, circumstances of place and time.

Somewhat later, the grammar of “cases” began to develop, in which the term “actant” gave way to the term “case,” denoting the verb part in a certain role. In this grammar, functional roles (cases) began to be linked to the nature of the process itself. Fillmore identified six cases: agentive (animate initiator of the process), instrumental, objective, locative, dative, counter-agent (agent with a passive verb). In a phrase Pierre ouvre la porte the subject expresses the agentive “case”, in La clé ouvre la porte - gun.

Chafe more closely linked the type of case with the type of process, identifying six types of process: state, process itself, action, process-action, state and environmental change, and seven cases: agent (Pierre a cassé une assiette), patient (Pierre dort), experiencer (Pierre veut boire), beneficiary (Pierre a une voiture), instrument, location, addition (Marie chante une chanson). We see the division of the subject into a number of entities depending on the type of process (agent in active action, patient in a state, experiencer in a relationship, testing something).

Pothier distinguishes two types of processes, which he calls voix: attributive (a message about the characteristics of an object): Le chat dort, le chat est noir, le chat est un animal and active (a message about the impact of one object on another): Marie coupe la tarte. The center of the utterance is the verb, around which three zones of actants (cases) are organized:

Zone I includes the main cases (i.e. functions): nominative (subject of the attributive relation): La lune est ronde, ergative (subject of active action): La luneéclaire la campagne, accusative (addition). Zone II a covers cases denoting the source of an event: causative (reason), instrumental (tool), agentive (object of a passive verb); zone III – cases denoting elements following the main relation: dative (à Jean), benefactive (pour Jean), final (goal). Zone III combines circonstants - spatial, temporal and qualitative characteristics.

The positive side of the theories under consideration is the desire to distinguish between the syntactic and semantic structure of a sentence, which is necessary due to the asymmetry of form and content in syntax ( § 360). The fact that the number and nomenclature of functions (actants, cases) do not coincide indicates objective complexity: in the real world, as in language, there are no clear dividing lines.

Rigel and others, distinguishing between syntactic relations and “deep cases” (“semantic roles”), define 9 “roles” for the French language, stipulating, however, that depending on different factors and approaches, their nomenclature may be different. These cases are as follows: 1) agent (agent, acteur) – initiator and “controller” of the process: L"enfant caress le chien; 2) object (patient) - a substance that is directly affected by the process: Le chien a mordu l"enfant; 3) beneficiary (dative, bénéficiaire) – an animate being affected by the process: Le tribunal a retire son permis de conduire au chauffard; //// a reçu la Légion d'honneur; 4) the focus (siège) of a physical or mental state: Les vitres tremblent; Jean est content; 5) tool (instrument) – Pierre ouvre la porte avec cette clef; // Cette clef ouvre la porte; 6) locative – spatial concretizer: Les clef sont dans le tiroir; // Le tiroir contenait des clés; 7) goal (but) – the entity towards which the action is directed: Pierre lance la balle à Jean; // Le tireur a raté le cible; 8) result (résultatif) – a consequence of the process: Jean a écrit plusieurs romances; 9) source from which another entity originates or moves away: Venus son de l"onde; // Il s"est separé de sa femme.

Identification of predicate types.

The classification of predicates (types of processes) is also extremely complex, because some types gradually transform into others. The complexity of solving this issue is evidenced by the experience of Potier, who first identified five fundamental semantic types of sentences, then expanded them to 10, then reduced them to 6, highlighting “voices”: existential: Il y a A; identifying: A est president; situational: A se trouve à B; descriptive: A est bien, A rit; possessive: A a B; subjective: A voit Q. Some types have subtypes, all of them have three “statuses” - static: A se trouve à B, dynamic: A entre à B, causative: C fait entrer Aà B, which gives from 18 to 32 basic types of sentences; but this scheme also has vulnerabilities.

Already in this list we find an attempt to identify types of situations (predicates) based on intersecting features. In the most general form, one can distinguish, on the one hand, processes contained in the subject itself, and processes covering the subject and object, and on the other, static and dynamic processes. The intersection of these two features gives four fundamental types of process, which are symbolized by common verbs:

objectless: static (être) – dynamic (aller),

object: static (avoir) – dynamic (faire).

T. V. Bulygina conducts a detailed component analysis of predicates (and therefore types of situations), including oppositions: timelessness/episodicity; dynamic/non-dynamic; duration/non-duration; promising/unpromising; gradual/non-gradual; controllability/uncontrollability, etc. Combinations of these features make it possible to objectively distinguish between phenomena, processes, events, incidents, etc. Experiences in classifying predicates (and at the same time semantic types of sentences) were undertaken by Russian linguists using material from different languages ​​[see. 3, 25 (2), 35, 41, 48 (2)]. In many ways, these classifications (despite the difference in terminology) coincide - the four main classes noted above are distinguished everywhere. But there are also differences, partially explained by the specifics of the languages. Thus, in the Russian language, subject quantification is distinguished as a separate type (There are many wolves or; There were five of them). In French, this semantic type does not have a specific form and can be considered as a semantic variant of the qualifying predicate (Ils étaient cinq). When describing a specific language, it is important to take into account linguistic forms when highlighting situations.

In objective reality, we can distinguish primarily the following fundamental types of situations:

1. Situation of existence: Les Martiens existent-ils? Displayed by the existential predicate: exister, être, il y a.

2. Definition of an object, its characteristics, qualities: Pierre est étudiant; represented by a nominal predicate.

3. Subjectless process: Il pleut; is framed using an impersonal sentence.

4. Subjective-objectless process, concentrated in the subject itself: Pierre dort. Expressed with an intransitive verb.

5. Subject-object process, the manifestation of the subject in relation to another object. Here the differences are: a) localization of the subject: Pierre va à Paris; b) the relationship of the subject to the object: Pierre aime Marie; c) the action of the subject in relation to the object: Pierre lit un livre. These situations are described using Vt or Vi with a mandatory addition or circumstance.

Many models can have options - static and dynamic, and within the latter inchoative (beginning, formation of a phenomenon) and finite (cessation of action), as well as causative. We will consider the relationship between structural-semantic models (surface level) and semantic types of sentences (deep level) in semasiological and onomasiological aspects (from forms to content and from content to forms).