RussNet: тезаурус  русского языка
English version

Отношения в RussNet-е

The list of semantic relations presented in RussNet is based mostly on Princeton WordNet Lexical and Conceptual Relations, and EuroWordNet Language-Internal Relations. But there are some differences in treatment of synonymy, hyponymy, antonymy, and role_realtions. Also a set of new relations that combine semantic features with grammatical or derivational ones was introduced, such as relations between members of aspect pairs, between neutral words and their expressive derivatives etc. Mostly these are relations between members of synsets (‘literals’) rather than synsets themselves.

Relations between synsets:

Relations between literals:


Synonymy is a fundamental relation, on which all wordnets, and RussNet in particular, are based. Sets of synonyms (so called synsets) are regarded as the main structural units of wordnets.

There are two different ways to define synonymy:
  • in terms of substitution;
  • in terms of semantic similarity.
    In Princeton WN and EWN the first way is adopted: “two expressions are synonymous in a linguistic context C if the substitution of one for the other in C does not alter the truth value” [Miller et al., 1990]. In RussNet we combine the substitution method with that of semantic similarity, since there are
  • many contexts in which words are interchangeable, because of sharing the same categorical meaning, but still denoting different concepts;
  • many words in Russian which have similar meanings but are hardly interchangeable in any context due to the syntactic, stylistic, expressive or other differences.

This complex approach to the synonymy definition is supported by Russian lexicographic tradition [Apresjan 1999, Evgenjeva 1970, and Babenko 1999].

Thus, in RussNet the term SYNONYMY is generally used to refer to the relation between words, compound words or collocations
  • denoting the same concept;
  • belonging to the same POS;
  • and (optionally) interchangeable in some contexts.

Synonymy is a symmetric relation, i.e. if A is synonymous to B, B is synonymous to A.

In practice, the relation of synonymy may be set up as a hypothesis on the basis of lexicographer’s intuitions or some dictionary of synonyms, then it should be verified with the synonymic definitions from the explanatory dictionary (see the Definition Analysis for details), and finally it should pass the standard implicative tests of the following type [Cruse 1986, Vossen 1998 ]:

Word1 in Contextc entails and is entailed by Word2 in Contextc

e.g.: Word1 = гиппопотам
Word2 = бегемот
Это - гиппопотам, следовательно, это - бегемот. (true)
Это - бегемот, следовательно, это - гиппопотам. (true)
Therefore, гиппопотам and бегемот are synonyms.

e.g.: Word1 = роза
Word2 = цветок
Мне подарили розы, следовательно, мне подарили цветы. (true)
* Мне подарили цветы, следовательно, мне подарили розы. (false)
Therefore, роза and цветок are not synonyms.

The list of traditional relations that link synset members in the frames of RussNet includes:
  • Absolute Synonymy
    Synonyms of this type are interchangeable in any context. These are words with different roots, usually of different origin and having different derivational structure. Such synonyms are extremely rare. They are obviously included into the same synset (e.g. бегемот - гиппопотам; языкознание - языковедение - лингвистика; космонавт - астронавт).
  • Duplicates (Variants).
    Such synonyms are rarely interchangeable in a context due to their stylistic colouring or domain preferences (though a native speaker might be unable to explain the distinction between them). They are words with the same root and different sets of affixes. They are included into the same synset (e.g. камешек - камушек, синонимичный - синонимический, федеративный - федеральный, семья - семейство, волочь - волочить, выкорм - выкормка - выкармливание, and лиса - лисица).
  • Stylistic Synonymy.
    These words are not interchangeable in a context due to the literary genre distinction. They are words with different roots and origin. They are included into the synset with special stylistic attributes: ARCHAIC or (устар.), INFORMAL or (прост.), SPOKEN or (разг.)(e.g. глаз - око, сообщение - донесение).
  • Expressive Synonymy.
    These words are not interchangeable in a context due to the meliorative and pejorative emotional connotation. Such words have different roots, and are included into the synset with special expressive attributes: X_EXPRESSES_POSITIVE_EMOTIONS or (мелиор.), X_EX­PRESSES_NEGATIVE_EMOTIONS or (пейор.)(e.g. спать - кимарить).
  • Derivational_Synonymy.

For better structuring we set order into the synset and regard one of its members as a representative of the whole synset, its dominant literal. Usually it is a literal without (or with minimum) stylistic and expressive colouring. In synset structuring we rely upon the frequency data: being a neutral term, dominant is expected to occur in texts more often then other members of the corresponding synset. We foresee the correlation between two modes of synset organization: from frequent to infrequent, and from the dominant through its absolute synonyms and duplicates to the stylistic and expressive synonyms.

e.g.: глаз1 - has relative frequency about 1329 ipm,
око (высок.) - 20.08 ipm,
зенки (прост.) - 2.26 ipm,
глазик1 (мелиор.) - 1.68 ipm,

Thus the synset is structured as follows: {глаз1, око (высок.), зенки (прост.), глазик1 (мелиор.), глазок2 (мелиор.)}

See examples


Hyponymy or IS_A relation plays a crucial role in constructing of lexical databases and ontologies, because it allows to organize data in hierarchical structures (trees), within which the top-down inheritance of semantic features are shown. The meaning of the superordinate term (‘parent’, ‘hyperonym’), is included into the meanings of all its subordinates (‘children’, ‘hyponyms’), e. g., words like chair, sofa, bench etc. contain as a part of their meanings a pointer to the meaning of seat.

Hyponymy is a transitive relation: e.g. being a kind of {цветок}, {роза} has inherited not only all semantic features of {цветок}, but also that of its superordinates: {растение}, {живой организм}, etc.

Within RussNet we use the term HYPONYMY to refer to the relation of semantic inclusion:
  • existing between synsets that belong to the same POS (nouns, verbs or adjectives);
  • linking subordinates to their direct superordinates.

The general implicative test sentence for hyponymy is unidirectional:

Word1 in Contextc entails but is not entailed by Word2 in Contextc

e.g.: Word1 = идти
Word2 = двигаться
Петр идет, следовательно, Петр движется. (true)
Петр движется, следовательно, Петр идет. (false)

What concerns hyponymy, in RussNet we allow the existence of multi-parent relations, i.e. one synset may have several hyperonyms. E.g., actress “female actor” is subordinated to both actor and woman. Though it is rather difficult to deal with multiple inheritances within databases, we believe that it helps us to reflect the real structure of semantic relations within language.

We suppose that in future the existence of multi-parent relations accompanied with introduction of special relation attributes may help us to distinguish between following types of HYPONYMY:
  • IS_A vs. INSTANCE_OF relations: e.g., королевство is a kind of государство, while Куба is an instance of государство.
  • ROLE_OF vs. TYPE_OF relations: e.g. убийца is a role of человек, while мужчина is a type of человек [Gangemi, Guarino et al, 2002].

See examples


Although in Princeton WN antonymy is regarded as a relation between words rather than synsets, in RussNet antonymy is considered to be one of the semantic relations between synsets.

Yet we by no means are to reject the differentiation of direct and indirect antonymy. We suppose that setting order into a synset (for details see SYNONYMY chapter) helps us to manage this problem adequately. As Word Association Norms show, in Russian it is usually synset representatives (‘dominant literals’) that are related by antonymy directly, all other members of synsets are opposed through this pair, i.e. indirectly. E.g., большой is strongly associated with маленький, маленький is associated with большой, while небольшой is associated first of all with маленький, its association with большой is rather weak. But there still is a possibility that several pairs of direct antonyms may appear in the frame of two synsets, like in English large <-> small, big <-> little. However, our data proves this phenomenon is not that characteristic for Russian.

Antonymy is a symmetric relation, i.e. if A is antonymous to B, B is antonymous to A.

For verification of antonymy we apply bidirectional implicative test sentences with negation:

Word1 in Contextc entails not-Word2 in Contextc

Word2 in Contextc entails not-Word1 in Contextc

e.g.: Word1 = подтверждать,
Word2 = опровергать
Он подтвердил мои опасения, следовательно, он не опроверг мои опасения (true);
Он опроверг мои опасения, следовательно, он не подтвердил мои опасения (true).
Therefore, we conclude that synsets {подтверждать, доказывать} and {опровергать} are antonyms.

Thus, in RussNet the term ANTONYMY is generally used to refer to the relation between synsets that
  • belong to the same POS;
  • have some common semantic component(s): belong to the same semantic field, share the same hypernym;
  • are opposed by some essential semantic component(s);
  • are interchangeable in a context with negation.
It is generally accepted that semantics of antonymy is quite complex [Cruse 1986,Miller 1993], and although native speakers have little problems with recognizing it, among linguists there is no agreement on ways antonymy should be defined. Here is the variety of semantic opposites that are treated as antonyms in the frames of RussNet:
  • contraries: большой - маленький;
  • complementaries: живой - мертвый;
  • reversives: e.g. прийти - уйти.

See examples


We believe it is reasonable to differentiate such oppositions as, for example, покупать - продавать, муж - жена from antonymous ones. We treat CONVERSION as a separate relation that differs from other types of opposition in that:
  • converses refer to the same situation, but from the viewpoint of different participants;
  • unlike antonyms, converses are opposites that share no common superordinate term.

E.g., покупать and продавать are converse terms because they describe the same situation of exchanging goods for money or its equivalent, покупать - from the viewpoint of buyer, while продавать - from the viewpoint of seller. {Покупать} is a hyponym of {получать, приобретать}, while {продавать} is subordinated to {отдавать, передавать2}.

See examples


Meronymy or HAS_A relation links synsets denoting wholes with that denoting their parts.

Though meronyms can sometimes be arranged in hierarchical structures (e.g., body parts), usually they are rather incorporated into a net-like structures than trees (e. g., point may be a part of knife, as well as a part of pencil, pin etc.)

Unlike the hyponymy, transitivity of meronymy is quite limited: e.g., ручка - часть двери, дверь - часть дома are normal, but it is quite odd to say ручка - часть дома.

Meronymy is an asymmetric relation, it may not be always reversible to holonymy: e.g., whereas a forest is not a forest unless it consists of trees, a tree does not necessarily grow in a forest (it may be a street or a desert) [Cruse 1986].

In RussNet following types of MERONYMY are distinguished:
  • MEMBER_OF, e.g.: дерево - лес;
  • PART_OF, e.g.: ветка - дерево;
  • SUBSTANCE_OF, e.g.: алюминий - самолет;
  • PORTION_OF, e.g.: кусок - пирог;
  • LOCATION_OF, e.g.: Москва - Россия;

See examples


ENTAILMENT, based on the logical relation of strict implication, is one of the prominent relations specific mostly for verbs and their derivatives. According to [Fellbaum, 1998], “the different relations that organize the verbs can be cast in terms of one overarching principle, lexical entailment”. Two basic kinds of lexical entailment can be distinguished: one involves ‘temporal inclusion’ (the two situations referred to by the verbs in the relation partially or totally overlap); the other involves ‘temporal exclusion’ (the two situations are variously temporally disjoint).

These temporal relationships between verbs are taken as a basis for a further distinction of four kinds of entailment:
  • a. + Temporal Inclusion
    • a.1 co-extensiveness (e. g., to limp - to walk)
    • a.2 proper inclusion (e.g., to snore - to sleep)
  • b. - Temporal Exclusion
    • b.1 backward presupposition (e.g., to succeed - to try)
    • b.2 causation (e.g., to give - to have)

Within Princeton WN (a1) is referred to as TROPONYMY,
(a2) and (b1) are referred to as ENTAILMENT,
(b2) is referred to as CAUSATION.

In EuroWordNet data related to the WN 1.5 entailment relation are encoded in a different a way [Alonge, 1996]:
(a1) is referred to as HYPONYMY,
(a2) is referred to as SUBEVENT/IS_SUBEVENT,
(b1) is referred to as CAUSATION with non-factitive label,
(b2) is referred to as CAUSATION.

In RussNet we encode data related to the entailment principle in a following way:
(a1) is referred to as HYPONYMY
(a2) is referred to as SUBEVENT/IS_SUBEVENT
(b1) is referred to as BACKWARD_PRESUPPOSITION
(b2) is referred to as CAUSATION


Thus, within RussNet we apply the term BACKWARD PRESUPPOSITION to the entailment relations between synsets that:
  • belong to the same POS (verbs, verbal nouns or verbal adjectives);
  • refer to temporally disjoint situations;
  • situations referred may have different agents;
  • meaning of one synset is embedded as a semantic component into the meaning of the other synset.

E.g.: on the one hand, отвечать temporally succeeds спрашивать, on the other hand it presupposes nothing but спрашивать as its necessary starting condition.
E.g.: переспрашивать presupposes понять+ negation

See examples


Within RussNet we apply the term SUBEVENT / IS_SUBEVENT to the entailment relations between synsets that:
  • belong to the same POS (verbs, verbal nouns or verbal adjectives);
  • refer to two temporally conjoint situations;
  • situations referred have the same agent
  • refer to the complex activity and its simple part.

For verifying SUBEVENT / IS_SUBEVENTrelations we use test sentences of the following type:

Word1 in Contextc, but not Word2 in Contextc (false)
e.g.: Word1 = приносить,
Word2 = приходить.
*Он принес, но не пришел (false).
Thus, {приносить} has subevent {приходить}

See examples


The term CAUSATION is used to indicate the entailment relation between synsets that:
  • belong to the same POS (verbs);
  • refer to temporally disjoint and conjoint situations;
  • one of the synsets refers to an event causing another event, process or state referred to by the second synset
  • events referred have different agents
    E.g.: kill -> die, give -> have.

See examples

All kinds of ENTAILMENT are unidirectionalrelations in that if Word1 entails Word2 it does not mean that Word2 necessarily entails Word1. Death is not necessarily a result of killing (убить CAUSE умереть, but not *умереть IS_CAUSED убить), while paying is always a part of buying (платить IS_SUBEVENT покупать, and покупать SUBEVENT платить).


‘INCHOATIVE'(BEGIN) is a semantic relation between verb synsets denoting some state, event, or action and the beginning of this state, event or action, e.g.: {полюбить, возлюбить (устар.)} BEGIN {любить1}. Its counterpart, ‘TERMINATIVE'(END) is a semantic relation opposing some state, event, or action to its ending, e.g.: {разлюбить, охладеть2} END {бить1}

BEGIN/END are complex relations, bordering upon antonymy, backward presupposition and aspect opposition. Firstly, having the opposite direction, inchoative term may be opposed to terminative one as its reversive, e.g.: {полюбить, возлюбить (устар.)} <-> {разлюбить, охладеть2}. Secondly, both inchoative and terminative terms presuppose being in the state, e.g. {полюбить, возлюбить (устар.)} => {любить1}, {разлюбить, охладеть2} => {любить1}, though being in some state may assume no inherent beginning or ending ?<= {любить1} =>?. Thirdly, ending of some state may resemble its culminating point, thus being associated with perfect aspect category.

See examples


INVOLVED (and its counterpart ROLE) is a set of relations linking verb synsets to other PoS: nouns, adjectives and adverbs, including

As opposed to EuroWordNet we use this terms not only to encode the data concerning sense of other words strongly involved into a verb sense [Alonge, 1993], but also to demonstrate probable argument structures of verbs. We generalize EWN treatment of these relations so that INVOLVED/ROLE refer to verb-to-class relations, with verb-to-word relations being its extreme case, when class contains only 1 member.

E.g.: {беседовать2, говорить2} INVOLVED_AGENT + {человек, лицо3};

{гукать2} INVOLVED_ AGENT {младенец}.

{Младенец} is linked to a single synset {гукать2}, while + {человек, лицо3} refers not only to synset itself, but also to all its hyponyms, thus covering a class of synsets being probable agent arguments of {беседовать2, говорить2}.

INVOLVED relations play crucial role in valency frames description (for details see Verbs chapter).

See examples


It is a relation between neutral words and their expressive derivatives. They denote the same concept, but contrast in the speaker’s attitude to the concept. As derivatives and corresponding stem words differ in style, they are not interchangeable in context, e. g. старик (old man) - старикан, старикашка (impolite address to an old man), дом (house) - домик1 (house to which the speaker has positive emotions). Derivatives are included into the same synset as corresponding stem words, e.g. {старик, ста­рикан, старикашка}, but with special expressive attributes: e.g. домик will have (мелиор.) or X_EX­PRES­SES_ POSITIVE_EMOTIONS, while старикан and старикашка will be marked by (пейор.) or X_EX­PRESSES_ NEGATIVE_EMOTIONS.

See examples


Like Derivational_Synonymy, DERIVATIONAL_HYPONYMY is a relation between cognate literals (nouns or adjectives). But in this case a derivative and corresponding stem-word denote the concepts that slightly differ from each other, e.g., дом (house) - домик2 (small house), дом (house) - домина (big house). The clear sense component is added by derivational affix to the stem-word meaning, so that the resulting word couldn't be regarded as its expressive synonym. That is why we treat such pairs as derivational hyperonym - derivational hyponym, and include into different synsets.

According to the meaning component added by derivational affix we divide DERIVATIONAL_HYPONYMY into several types:
  • DH_a - augmentative (X_IS_BIG), e.g. дом (house) - домина (big house),
    • DH_am - maximization (X_IS_MAXIMAL);
  • DH_d - diminutive (X_IS_SMALL), e.g., дом (house) - домик2 (small house),
    • DH_dm - minimization (X_IS_MINIMAL), e.g., (последний - распоследний)

See examples


DERIVATIONAL_INVOLVED_RELATIONS are established to link a verb to its derivatives, designating action participants. These are relations between literals, that implies strong semantic incorporation supported by derivational motivation:

e.g.: сеять DER_INVOLVED_OBJECT сеянец,
сеять DER_INVOLVED_AGENT сеятель,

These are unidirectional relations.

Introduction of Derivational_INVOLVED_Relations allows us to present the inheritance of the argument structure and selectional restrictions of stem verbs by their derivatives,

e.g.: бороться за правое дело => борьба за правое дело => борец за правое дело.


Aspect is a category specific for verbal derivatives in highly inflectional languages, like Russian and other Slavonic. Aspect pairs interact in a complex manner: on the one hand, they look like very close synonyms, though on the other hand, they realise a very important semantic oppositions, concerning ‘Aktionarten’ differentiation [Vendler 1967], such as activity - action.

Thus, in RussNet IMPERFECT/PERFECT are relations that have semantic-grammatical nature and links verb literals denoting temporarily unbounded event, action, or state characterized by multiple or unspecified recurrence, on the on hand, with cognate literals, denoting temporarily bounded event, action, or state characterized by accomplishment of inherent culminating point, on the other hand.

E.g.: делать PERFECT сделать,
помочь IMPERFECT помогать.

Aspect opposition has obligatory formal representation: derivational affixes -ыва-, с-, -ну- etc.

See examples

  Ваши комментарии и замечания пишите на  Последнее обновление 14 июня 2005