DYNAMIC VERBS IN THE WORDNET OF POLISH

The paper presents patterns of co-occurrences of wordnet relations involving verb lexical units in plWordNet — a large wordnet of Polish. The discovered patterns reveal tendencies of selected synset and lexical relations to form regular circular structures of clear semantic meanings. They involve several types of relations, e.g., presupposition, cause, processuality and antonymy, do not have a necessary character (there are exceptions), but can be used in wordnet diagnostics and guidelines for wordnet editors. The analysis is illustrated with numerous positive and negative examples, as well as statistics for verb relations in plWordNet 4.0 emo. Some attempts to a more general, linguistic explanation of the observed phenomena are also made. As a background, plWordNet model of linguistic character is briefly recollected. A special attention is given to the verb part. In addition the description of dynamic verbs by relations and features is discussed in details including relation definitions and substitution tests.


Introduction
Wordnets describe words of the four basic Parts of Speech, 1 i.e. nouns, verbs, adjectives and adverbs.However, they do this in a very unbalanced way.Even in Princeton WordNet (Fellbaum, 1998) one of the largest and richest wordnets nouns are the most privileged category with rich structure of relations and good coverage, while the number of relations for adjectives is very small and the coverage is modest, not mentioning a very sparse description of adverbs.Verbs in WordNet are in the middle of the way with the significantly smaller number of relations and much more selective coverage than the one for nouns.The lower density of verb sub-networks of many wordnets, e.g.Princeton WordNet, plWordNet (Maziarz, Piasecki, Rudnicka, Szpakowicz, & Ke˛dzia, 2016) means less information.Our general aim is to introduce a richer structure of verb relations into a wordnet and make it closer to the expectations of different applications.

Dynamic verbs in the Wordnet of Polish
In the paper we will discuss an expanded model for the description of verbs in plWordNet which originates from the general model of plWordNet and consists of: features (e.g.verb class, aspect) and lexico-semantic relations.We will also strive to formulate guidelines for linguists.The discussion will be confined to the Polish language and plWordNet, but set in a perspective enabling more general conclusions.Since this topic is still very broad, even if limited to verbs in plWordNet (e.g. the guidelines for verbs in plWordNet consists of 37 pages), here we will mainly focus on dynamic verbs.

plWordNet in brief
2.1 Linguistic model The plWordNet model (cf Piasecki, Szpakowicz, &Broda, 2009 andMaziarz, Piasecki, &Szpakowicz, 2013) is based on lexical units (LUs, i.e. triples: lemma, Part of Speech, sense identifier) as basic building blocks.LUs are grouped into synsets on the basis of sharing constitutive relations and features, i.e. two LUs belong to one synset if and only if they share targets of constitutive relations and values of constitutive features.
Constitutive relations are selected lexico-semantic relations that are frequent enough and express relatively high sharing factor.They also originate from lexicography and wordnets and are relatively well-understood among linguists, e.g.hyper/hyponymy, holo/meronymy, iterativity, distributivity.
Constitutive features constrain the shape of the system of lexico-semantic relations (e.g.aspect or adjective and verb classes) or express some general pragmatic conditions on the use of lexical meanings (e.g.stylistic register ).Constitutive features are often referred to in the definitions of lexico-semantic relations, e.g.hyper/hyponymy is defined only for adjectives of the same class or inter-register synonymy links LUs of non-compatible lexical registers (cf Maziarz et al., 2013).
All plWordNet relations are precisely described guidelines for wordnet editors (i.e.lexicographers) by textual definitions, examples and substitution tests.A substitution test for a relation or its subtype consists of up to several test sentence templates that include variables to be replaced by lemmas corresponding to LUs that are examined.Each test sentence template can be positive or negative.In the former case, we expect that the template filled with the analysed word forms is a semantically well-formed sentence, in the latter case, the created sentence is expected to be semantically odd.The whole test is passed by a pair of lemmas if and only if all answers conform to the expectations.Sentence templates in substitution tests can also be compared with use examples from corpora.They connect relation definitions with language data from the corpora.Substitution tests are also presented in the user interface of WordnetLoom (Piasecki, Marcińczuk, Ramocki, &Maziarz, 2013 andNaskre˛t, Dziob, Piasecki, Saedi, &Branco, 2018) prior to adding any new relation link.They are already instantiated with the forms of the lemmas under consideration.
In short, lexico-semantic relations are the main means of description.In plWordNet they can all be traced back to language data.
Synset relations are, in fact, abbreviations of sets of constitutive relations held between the synset members.The synonymy relation is to some extent derived from constitutive relations and features that define synsets.There is no ontological difference between synset relations and lexical relations (i.e., the relations that link LUs).Both types of relations are lexico-semantic relations of strictly linguistic character between lexical meanings.The synset relations are in fact relations linking LUs belonging to the two synsets.This is in contrast to the Princeton WordNet model in which synset relations are conceptual and link lexicalised concepts represented by synsets, not LUs synset members.
In plWordNet, most of synset relations are constitutive relations.However, Dziob and Piasecki (2018) introduced non-constitutive synset relations as relations that help to expand semantic characteristics of a synset, but are not treated as the part of synset definition, because they are Dynamic verbs in the Wordnet of Polish more descriptive than discriminative, e.g.subject or circumstance relations.
plWordNet also includes also glosses (short meaning descriptions, similar to Princeton WordNet glosses) and use examples, but both are defined for LUs, not for synsets.

Construction
plWordNet is divided into four sub-databases of nouns, verbs, adjective and adverbs.However, there are many relations linking LUs of different Parts of Speech, called cross-categorial relations in EuroWordNet (Vossen, 2002).3 Verb model in plWordNet The model for the description of verbs in plWordNet 3.1 (Dziob, Piasecki, Maziarz, Wieczorek, & Dobrowolska-Pigoń, 2017) originated from the one of plWordNet 3.0 (Maziarz, Piasecki, Szpakowicz, Rabiega-Wiśniewska, & Hojka, 2011) The introduced modifications resulted in the simplification of the system of verb semantic classes and lexico-semantic relations.The changes were aimed at limiting proliferation of verb senses triggered by too fine grained semantic classes and their interactions with relation definitions.In the case of the latter, some subtypes of relations were discarded, definitions rewritten, slightly simplified and made stricter, but also a few new relations were introduced.As observed in Dziob and Piasecki (2018), the modified model resulted in the significant increase of a number of instances of verb relations, especially in the case of relations expressing associations between situations on which we will concentrate in Sec. 4.

Semantic verb classes
Seven main semantic verb classes were introduced in plWordNet 2.0 capitalising on the classification of Vendler (1957) and its modification by Laskowski (1998), namely: processes, actions, acts, accidents, activities, events and states (Maziarz et al., 2011).The classes were organised into a complex hierarchy with many subtypes represented by additional artificial synsets.This complex system of verb classification found several interesting applications, e.g.Lis and Navarretta (2014).However, as noticed in Dziob et al. (2017) and Dziob and Piasecki (2018), it posed several problems to wordnet editors.Thus, we decided to simplify the system of classes following EWN (Vossen, 2002).Only two main classes were introduced (instead of 7): static and dynamic verbs.
In this paper we will concentrate on the latter.Static verbs were defined in Dziob et al. (2017) as imperfective atelic and durative verbs that represent situations which are stable and do not include any change during their time span, i.e. we cannot distinguish any change from a state X to Y in the situation represented by a static verb, cf detailed subclasses in Dziob and Piasecki (2018).As in Vendler (1957), only imperfective verbs are static verbs.

Dynamic verbs in the Wordnet of Polish
Dynamic verbs encompass all perfective and bi-aspectual verbs.Among them, Dziob and Piasecki (2018) identified several subgroups (subclasses) focusing on differences in paraphrasing the verb meanings: 1. distributive representing situations in which something is done by many agents, in relation to many objects or affects many objects, e.g.przebadać 'to examine many people', 2. accumulative situations that have lasted or been done to such an extent that it is enough (from some point of view) e.g.ubawić sie ˛'to amuse itself', 3. perdurative doing something during a particular or limited time, e.g.przemieszkać 'to live during some period in a place', 4. delimitative expressing doing something or happening of something for some time or to some extent, e.g.pomieszkać 'to live for short time in a place', 5. action verbs a) all perfective and bi-aspectual, b) imperfective derivatives of accumulative, delimitative, perdurative, and distributive verbs (representing changing situations), c) imperfective derivatives of semelfactive verbs (i.e.representing punctual or instantaneous events), e.g.mrugać imp (from mrugna ˛ć perf ) 'to flicker', representing multiple changes, d) imperfective causative verbs (expressing occurrence of a new situation), e) processive (expressing gradual transition to a new state) (e.g.starzeć sie ˛imp , postarzeć sie ˛perf 'to becoming (become) older'), f) inchoative, representing an introductory phase of a new situation, (e.g.rozp lakiwać sie ˛imp , rozp lakać sie ˛perf 'to start crying') and g) limitative (representing ending of some situation) (e.g.wybarwiać sie ˛imp , wybarwić sie ˛perf 'to lose colour'), and also h) other imperfective verbs that represent a situation changing due to actions of entities or with respect to them, e.g.iść 'to walk imp ', biegać 'to run imp ', p lakać 'to cry imp '.Some plWordNet relations are restricted only to processive, causative and inchoative verbs, see Sec. 3.2.These relations facilitate defining verbs of these groups as verbs expressing a situation change.

Hyponymy and relationships between situations
Hyponymy and its reverse hypernymy are basic relations building the verb hierarchy in plWordNet.Hyponymy is defined with the help of the following substitution test (Maziarz et al., 2011): X inf to Y inf w specjalny sposób, jakoś 'To X is to Y in a special way, somehow' plWordNet hyponymy corresponds to troponymy relation from Princeton WordNet, described by the substitution test (Fellbaum, 1998): To V1 is to V2 in some particular manner Troponymy is also characterised as a specific kind of entailment, restricted to verbs representing temporarily coexistent situations.The same aspect of temporal co-existence is expressed in plWordNet in hyponymy/hyperonymy, meronymy/holonymy, cf Dziob and Piasecki (2018).
Meronymy for verbs in plWordNet is a relation expressing that there are two situations cooccurring in the same time in such a way that a situation represented by the verb X is the part of the one represented by Y or X is accompanying Y .
Holonymy is a reverse relation to meronymy, but it is not obligatory, i.e. not every meronymy link entails a reverse holonymy link (cf Dziob et al. 2017).The definition of meronymy is a complex issue, often discussed in literature (e.g.Murphy, 2006;Brown, 2004) and also in Dziob and Piasecki (2018), so we will omit the details in this paper.
In addition to the above four relations, in plWordNet there are several more relations that describe associations between situations.They form a rich system of verb meanings.
Presupposition and preceding both express an association of a situation represented by the source verb, with a situation that occurred earlier, represented by the target.Presupposition Dynamic verbs in the Wordnet of Polish is close to logical presupposition and informs about the necessity of an earlier occurrence of some situation, while preceding does not include such an aspect of necessity, but only expresses a possibility that one situation happens before the other one.Both relations connect a verb (perfective or imperfective) representing a situation X with the representation of an earlier situation Y characterised by a verb, noun, adjective or adverb.Neither can be established between verb synsets that include LUs associated derivationally (including pure aspectuality), possessing very close meaning and linked by inter-register synonymy or representing situations co-existing temporarily (i.e.those linked by hyponymy, meronymy or holonymy).
There are two subtypes of presupposition and preceding in plWordNet (Dziob & Piasecki, 2018): • subject identity (SI), in which the semantic subjects of both linked situations X and Y must be identical (henceforth, subject will be understood as a semantic subject), • no subject identity (NSI) without the above condition.Inchoativity express an initial phase of a situation represented by the target element: 1. variant V-V: X-ować to zaczynać/zacza ˛ć Y-ować.'X inf is to begin impinf Y inf ' (e.g.ruszyć 'to move' → poruszać sie ˛'');
Both relations cause and inchoativity express temporal sequence of situations (represented by the linked verbs) such that the later is the result of the former.However, in the case of cause there are two distinct subjects: the first one causes a new situation involving the second one, e.g.karmić 'to feed' → jeść 'to eat'.In contrast, in inchoativity, the former situation describes an initial phase of the following situation which is its continuation and includes all important aspects of the former, e.g.rozmarzyć sie ˛'to start dreaming' → marzyć 'to dream' both subjects are identical.
Inchoativity is often signalled by derivational associations, but not always, and it has been expanded to the level of synsets, beyond the derivational signal, due to its clear semantic content, see Maziarz et al. (2011), i.e. also in plWordNet it links LUs that are not derivationally associated.
Cause and processuality both express a change from an earlier situation represented by the source X to a new situation represented by the target Y , which results from or is caused by X. Cause emphasises that the resulting situation is caused/brought about the next situation of Y , Anti-causative and auto-causative verbs that render similar semantic differences are distinguished in linguistic works for English (e.g.Levin & Hovav, 1995;Koontz-Garboden, 2009;Horvath & Siloni, 2011).Anti-causative verbs express a change that happened in the object, but caused by an external agent (e.g.zatona ˛ć 'to sink'), while auto-causatives are verbs that represent such a change in which object and subject are identical (e.g.zmienić sie ˛'to change').In Polish such situations are often indistinguishable, because the information about an external agent must be overtly defined.For instance, a ship may sink because it has been sunk by pirates or because a hole appeared in her side.Thus, a synonym for zatona ˛ć 'to sink' is a reflexive verb zatopić sie ˛'≈to sink itself'.Taking this into account, similar changes in situations are described in plWordNet by means of processuality relation: Dynamic verbs in the Wordnet of Polish (e.g.zmienić sie ˛'to change itself' → inny 'different') and cause (e.g.zatopić 'to sink something' → zatona ˛ć|zatopić sie ˛'to sink/≈to sink itself') on the basis of the semantic substitution test, without referring to the valency of these verbs.
Classes of intentional and non-intentional verbs (Laskowski, 1998) were distinguished in pl-WordNet 2.0 (Maziarz et al., 2011).However, problems with the consistent recognition of the intentionality of the subject's actions caused that this distinction was abandoned in Dziob et al. (2017); Dziob and Piasecki (2018).
As mentioned, the main difference in the verb hierarchy between plWordNet and Princeton WordNet is in different hierarchical relations, respectively: hyponymy vs troponymy.Troponymy has been defined Fellbaum (1998) as a subtype of entailment: "the relation between two verbs V1 and V2 that holds when the sentence Someone V1 logically entails the sentence Someone V2" where V1 and V2 are verbs representing situations.Troponymy restricts this association to verbs representing temporarily co-existing situations.
The hyponymy substitution test of plWordNet resembles the test for troponymy in Princeton WordNet both require that the situation X is a sub-kind of the situation Y , but differ in a specific way in which the situation represented by X happens or is performed.The troponymy test adds "in some particular manner" that is covered in different aspects by several other relations of plWordNet.We can find many more such similarities in both wordnets.The comparison of relations corresponding to entailment in Princeton WordNet Fellbaum (1998) and plWordNet (Dziob & Piasecki, 2018) is presented in Table 2.The two sub-kinds of entailment are distinguished in Princeton WordNet by the criterion of temporal co-existence of the two situations represented by the linked verbs.In plWordNet, verbs representing co-existing situations are linked by hyponymy and meronymy.In the latter case, there is no requirements for subtype-like similarity.
The rich system of verb relations, especially more fine-grained forms of entailment in plWordNet allows for more elaborated description of temporal relations between situations represented by verbs.In the case of referring to situations previously occurring presupposition and preceding can be used.In the case of situations from the future, somehow resulting from the contemporary situation of X, plWordNet enables to use three synset relations: cause, processuality and inchoativity.
Processuality and cause have a lot in common both relations inform that a situation represented by the target verb Y appears as a result of the source X.However, the former is limited to cases in which there is the same subject of both situations, while cause signals that subjects of both situations are different.

Other verb relations in plWordNet
Multiplicavity is a relation which describes repetition of some state or activity, and performance an activity by many agents or on many objects.It has two subtypes: 1) distributivity restricted to perfective distributive verbs, e.g.wyt luc 'to break all things' → st luc 'to break perf ' Dynamic verbs in the Wordnet of Polish and 2) iterativity for imperfective iterative verbs, e.g.czytywać 'to read imp many times' → czytać 'to read imp '.
State is restricted to static verbs and describes an association between a verb X with a noun, an adjective or an adverb that represents a state in which the subject of X is, e.g.zielenieć 'to be green' → zielony 'green'.
Four new relations were introduced in plWordNet 3.1 (Dziob et al., 2017).Three of them are related to the verb predicate-argument structure, but only with respect to its semantic properties: 1. subject (e.g.rżeć 'to neigh' → koniowate 'equine'), 2. object (e.g.wzuwać 'to put on shoes' → but 'a shoe'), 3. circumstance linking a verb with a noun, which is an element of the prepositional phrase that can function as an adjunct to the verb (e.g.dobijać do brzegu 'to reach a shore' → brzeg 'a shore').
The fourth relation, namely manner, is motivated by a component of the substitution test for hyponymy (and a similar test for troponymy in Princeton WordNet).This component represents a manner in which the given situation proceeds (or an activity is performed).It influences the sense of a given verb in a significant way and helps to distinguish between verb senses, e.g.podsmażyć 'to fry a little' → troche ˛'a little'.
Besides synset relations, plWordNet includes many verb lexical relations.
Pure aspectuality and secondary aspectuality link verbs of different aspects, but derivationally associated.The first one links pure aspectual verb pairs, i.e. pairs in which the change of the aspect exhausts the meaning difference3 , e.g.czytać imp 'to read imp ' ↔ przeczytać perf 'to read' perf .
The secondary aspectuality relation provides information about the meaning change going beyond the aspect difference, e.g.
trzymać 'to hold imp ' ↔ potrzymać 'to hold perf a little'.Verbs linked by secondary aspectuality often belong to synsets linked by some other relation, e.g.
Proper antonymy and complementary antonymy both signal an opposition resulting from antonymic association and both link verbs of the same aspect.However, the first one links LUs X and Y , such that both X and Y cannot happen at the same time with the same subject, but if X does not take place, then it is not necessary that Y takes place e.g.zwie ˛kszać 'to increase imp ' ↔ zmniejszać 'to reduce imp '.
Complementary antonymy excludes any third possibility, if X does not take place, then Y must happen there is no other option (see also Sec. 4) e.g.zatrzymywać 'to keep imp ' ↔ oddawać 'to give imp back'.
Converseness is similar to antonymy (it is often considered to be a kind of antonymy, e.g.see Fellbaum (1998)) links verbs of the same aspect, and also expresses a kind of semantic opposition.However, it links verb LUs describing the same situation, but from two different, opposite perspectives (e.g.agent and patient, see also Sec. 4), e.g.sprzedać 'to sell perf ' ↔ kupić 'to buy perf ' In addition to the verb lexical relations discussed above, plWordNet includes other lexicosemantic relations motivated derivationally and defined on the level of LUs: role inclusion and derivationality.The former links verbs derived from nouns with their noun derivational bases.It is further subdivided into seven subtypes that express different semantic associations between a derivative and its basis (Maziarz et al., 2011;Dziob et al., 2017): subject, instrument, result, location, object, time and indefinite subtype (for less clearly identified roles).

Dynamic verbs in the Wordnet of Polish
Derivationality represents semantically less regular or frequent links signalled by a derivational association.It very often links LUs that belong to synsets participating in some other relation, e.g.uwznioślić 'to get exalted' → wznios ly 'exalted', where the synsets including these LUs are linked by the cause relation.
Statistic data concerning verb relations in plWordNet 4.0 emo are presented in Table 3.We can notice that the relations of dynamic verbs, namely: presupposition, preceding, cause and processuality are relatively frequent.Thus, they are important from the point of view of language description.

Relation
Example plWordNet is a lexico-semantic network in which every element (LUs and synsets) is linked to a vast majority of other elements:6 with some directly (by different relations), with others indirectly by chains of relation links, i.e. paths in the wordnet graph.Between many elements one can find more than one path linking them.This shows the richness of such a relational semantic dictionary as the wordnet is.In the rest of the section we will investigate examples of interconnections and co-occurrences of different relations from the linguistic point of view.

Presupposition
We can observe co-occurrences of presupposition (of both subtypes) and antonymy (to both subtypes) links.Presupposition describes a relation to the representation of an earlier situation or its element, in the latter case represented by an adjective.Thus, when presupposition co-occurs with antonymy of the members of the target synset, then the source of the presupposition link is also connected by: 1. processuality to the synsets including these antonyms in the case of the subject identity subtype, 2. and cause to such synsets in the case of the no subject identity subtype.In spite of plenty of of evidence found in plWordNet that support these relation co-occurrence patterns, we can also find cases in which these relations fail to meet each other.So, these patterns can be used only as a suggestion for a lexicographer to look for potentially missing links, not as a strict rule.

Preceding
Preceding, which is similar to presupposition in many respects, occurs in similar coincidence patterns with other relations.

Cause
Cause tells us about bringing about some other situations by the former one.It is very close to processuality with respect to the described logical construction of a cause-consequence chain.We can observe that both relations co-occur quite often among synset relations in such a way, that a verb LU which is the source of the cause link, and a verb with the processuality link, which is its target, should have the links of, respectively: cause and processuality to the same adjective or adverb.In the case of nouns, we cannot observe a similar co-occurrence of relations, For instance, in the case of gniewać 'to anger' it is linked by cause to z lość 'anger', but a non-existing processuality link would have to connect gniewać sie ˛'to become angry ref lex.(by himself)' to *stawać sie˛z lościa 'anger', that does not take place.
Example 13. {zabielać 1} 'to whiten' -cause→ {zabielać sie ˛1} 'to whiten oneself' {zabielać 1} -cause→ {bia ly 1} 'white' {zabielać sie ˛1} -processuality→ {bia ly 1} Cause also often co-occurs with both subtypes of antonymy.A target synset of the cause link: X-cause→ Y often includes also a LU z ∈ Y connected by complementary antonymy to a LU q representing a situation which is earlier than the situation represented by the source of the cause link, i.e. q is earlier than X.Because complementary antonymy introduces a necessary condition, we should add a presupposition link between the following and the earlier situation, as presupposition also excludes alternatives, i.e. q ∈ Q ←presupposition-X, where Q is the synset of q.
Example 14. {wyswobadzać 2, wyswabadzać 2, wyzwalać 1, uwalniać 4, oswobadzać 3, oswabadzać 2} 'to free [sb or sth]' -cause→ {niepodleg ly 1, niezawis ly 1} 'independent' niezawis ly 1 'independent' ←compementary antonymy→ zawis ly 1 'dependent' {podleg ly 1, zawis ly 1} ←presupposition-{wyswobadzać 2, wyswabadzać 2, wyzwalać 1, uwalniać 4, oswobadzać 3, oswabadzać 2} The relations of cause and proper antonymy also express a similar pattern of co-occurrence.If the target of cause: X-cause→ Y includes an LU z ∈ Y linked by proper antonymy to q, then q represents a situation which is earlier than X.However, in the case of proper antonymy there is no necessity aspect in the definition and we can connect X to q ∈ Q only by a weaker preceding relation which does not introduce an aspect of necessity, too.All the examples given above conform to these co-occurrence patterns for cause.However, the picture is more complicated, and these patterns can be only treated as suggestions to lexicographers to look for possible links, not as automated procedures, because we can observe exceptions.As an example we can consider other verbs linked by cause: Example 18. (negative) {urealniać 2} 'to make real' -cause→ {urealniać sie ˛1} 'to become real' urealniać sie ˛1 ←proper antonymy→ odrealniać sie ˛1 'to become unreal' It is true that if something has sie ˛urealni lo '≈made real itself', then should odrealnić sie ˛'≈become unreal' according to the pattern.In fact there is an association in the opposite direction.If something has odrealnić sie ˛'≈become unreal', then it had to earlier sie ˛urealnić 'become real', so this pair does not fulfil the test for processuality.
Cause also co-occurs with converseness for adjectives and adverbs -if a verb synset X is linked by cause to Y which includes an LU z connected by converseness to an LU q, then q represents an earlier situation and the synset q ∈ Q should be presupposed by X (i.e. to be the target of presupposition from X), as converseness due to its definition imposes the comparative or superlative grade in the case of adjectives and adverbs.
We can also notice that double cause relations: from a verb to both an adjective and a noun often co-occur with a characteristic relation between the adjective and noun.
The presented analysis has been solely performed on plWordNet and illustrated by Polish examples.plWordNet model introduces a rich system of linguistically motivated relations and Polish is an inflectional language with a complex derivational system.Thus, the analysis, especially the relation co-occurrence patterns cannot be directly mapped to the other wordnets for other languages.However, all Slavic languages have derivational systems of similar character ( Šojat & Srebačić, 2014), BulNet (Koeva, 2008) or CzechWordnet (Pala & Hlaváčková, 2007).Their wordnet models show many differences in comparison to the plWordNet's one, but such differences are not very significant, e.g. a comparison between plWordNet and BultNet presented in Piasecki and Koeva (2017).Thus, the system of verb relations proposed by us and the idea of the relation co-occurrence patterns can be applied to other Slavic wordnets, and, e.g., used in language comparison.
The application of our model to non-inflectional languages seems to be more problematic.However, it is worth-noticing that, firstly, richer systems of verb relations based on more fine grained partitioning of entailment has been considered for Princeton WordNet several times, secondly, the idea of morpho-semantic relation was proposed and applied initially to Princeton WordNet and next transferred to several other wordnets, even to Slavic ones, e.g.ButNet (Dimitrova, Tarpomanova, & Rizov, 2014).For instance, if we take a look into English glosses in Table 3, we can discover that several derivationally-motivated relations from plWordNet can be transferred to WordNet of English and this number would be much increased if zero derivation is taken into account, too.Only the minority of relations, e.g.including multiplicativity, would be very hard to be transferred, as such distinctions are expressed in many languages, including English, on the syntactic-semantic level.Moreover, all derivationally motivated relations for verbs in plWordNet have been lifted to the level of synset relation.Thus, they also link LUs that are not derivationally associated and their definitions are not based on the requirement of the existence of the derivational association.The only two problems left are verb classes and verb aspect.The first one is simple, as the system of verb classes of plWordNet originated from the systems proposed for other languages.Aspect is not lexicalised in many languages, e.g.English, contrary to the Slavic ones.However, as a result, LUs in such languages would participate in larger number of relations, or in some cases more fine grained distinctions of senses can be considered.

Applications
Co-occurrence patterns presented in Sec. 4 are more a generalisation than a complete model.It shows possibly existing correlations among groups of lexico-semantic relations.Moreover, the formulated generalisations suggest that there are missing connections in plWordNet that are predicted by them.Paradoxically, the imperfect accuracy makes the patterns a diagnostic tool: the generalisations do not determine, but predict highly likely existence of relation links.Thus, one of the results of our analysis is a set of additional markers for relations that enrich the definitions of relation and the lexicographic procedure applied.
The identified co-occurrence patterns will be used as a basis for the expansion of a diagnostic tool ?supporting linguist in maintaining high agreement between their decisions.plWordNet is built by a team of lexicographers and is not free of inconsistencies or even mistakes.Diagnostic tools utilising also language technology allows for minimising the risk of errors.However, the interpretation of a specific language phenomenon or lexical meaning is always preserved for lexicographers.Awareness of regular tendencies in relation co-occurrences may help in formalising lexicographic procedures leaving less space for intuitive interpretation, especially in relation to the identification of lexical meanings.
We have also found several cases in which it was not possible to formulate co-occurrence patterns (e.g. 8 examples for antonymy between nouns), and a higher probability of a relation is

( a )
Cause by lo Y-owo gdzie Y jest w stopniu równym lub wyższym.'it happened Y , where Y is in the positive or comparative grade.' Processuality -(b) Cause (z)robi lo sie ˛Y-owo, gdzie Y jest w stopniu równym lub wyższym.'it became Y , where Y is in the positive or comparative grade.' Processuality (z)robić coś Y-owo, gdzie Y jest w stopniu równym lub wyższym.'it did Y something, where Y is in the positive or comparative grade.' (c) Cause ktoś lub coś (z)robi l(o) coś Y-owo, gdzie Y jest w stopniu równym lub wyższym.'sb/sth did sth Y , where Y is in the positive or comparative grade.' Processuality (z)robić sie ˛Y-owo, gdzie Y jest w stopniu równym lub wyższym.'it became Y itself/by itself, where Y is in the positive or comparative grade.'

Table 1 :
Basic statistics of plWordNet 4.0 emo (http://plwordnet.pwr.edu.pl) Presupposition musia lo dziać sie ˛Y-owo, gdzie Y jest w stopniu równym.'it had to Y happen, where Y is in the positive grade (degree).'Preceding mog lo dziać sie ˛Y-owo, gdzie Y jest w stopniu równym.'he/it could happen Y , where Y is in the positive grade (degree)' (c) Presupposition ktoś lub coś musia l(o) robić coś Y-owo, gdzie Y jest w stopniu równym.'he/it had to Y do, where Y is in the positive grade (degree).'Preceding ktoś lub coś móg l/mog lo robić coś Y-owo, gdzie Y jest w stopniu równym.'he/it could Y do, where Y is in the positive grade (degree)' Presupposition musia lo być wcześniej Y-owo lub jest Y-owo do tej pory.'it/he had to be Y or it has Y been.Dynamic verbs in the Wordnet of Polish (a) Presupposition też musia lo być wcześniej Y-owo lub jest Y-owo do tej pory.'it also had to be earlier Y or it has been Y till now.' Preceding też mog lo być wcześniej Y-owo lub jest Y-owo do tej pory.'it also could earlier be Y or it has been Y ' (b) Presupposition też musia lo dziać sie ˛wcześniej Y-owo lub dzieje sie ˛Y-owo do tej pory.'also it had to earlier Y happen or it has Y happened till now.' Preceding też mog lo dziać sie ˛wcześniej Y-owo lub dzieje sie ˛Y-owo do tej pory.'it also could earlier Y happen or it has Y happened till now.' (c) Presupposition ktoś lub coś musia l(o) wcześniej też robić coś Y-owo lub robi Y-owo do tej pory.'sb/sth had to Y do or it is necessary that he/it has Y done till now.' Preceding ktoś lub coś móg l/mog lo wcześniej też robić coś Y-owo lub robi Y-owo do tej pory.'sb/sth could earlier Y do or he/it has Y done till now.' 1. variant: V-V Presupposition-SI Jeżeli stwierdzamy, że X-owa l(o), oznacza to, że musia l(o) wcześniej Y-ować.'Ifone says that sth/sb X-ed, then it means that it/he earlier had to Y ' Jeśli nie X-owa l(o), to też musia l wcześniej Y-ować lub Y-kuje do tej pory.'Ifsth/sb did not X, then it also means that it/he had to earlier Y or it/he has been Y -ed.' Prawdziwość stwierdzenia, że wcześniej Y-owa l(o) jest warunkiem koniecznym, aby sensownie stwierdzić, że X-owa l(o).'Thetruth of the statement that someone or something Y -ed earlier is a necessary condition for saying that X.' Preceding-SI: Jeśli ktoś/coś X-owa l(o), to wcześniej móg l Y-ować; istnieje przynajmniej jeszcze jeden czasownik Z taki, że może zasta ˛pić czasownik Y w tym teście.'Ifsb/sthX-ed, it means that he/it could earlier Y ; there is at least one more verb Z such that it can replace Y in this test.'Presupposition-NSIJeżelistwierdzamy, że X-owa l(o), oznacza to, że wcześniej ktoś inny lub coś innego musia l(o) Y-ować.'Ifone says that sth/sb X-ed, then it means that sth/sb else had to earlier Y ' Jeśli nie X-owa l(o), to ktoś inny lub coś innego też musia l(o) wcześniej Y-ować lub Y-kuje do tej pory.'Ifsth/sb did not X, then it also means that sth/sb else had to earlier Y or sth/sb else has been Y .' Prawdziwość stwierdzenia, że wcześniej Y-owa l(o) jest warunkiem koniecznym, aby sensownie stwierdzić, że X-owa l(o).'Thetruth of the statement that someone or something Y -ed earlier is a necessary condition for saying that X.Preceding-NSI: Jeśli ktoś/coś X-owa l, to wcześniej ktoś/coś inny/innego móg l/mog lo Y-ować; istnieje przynajmniej jeszcze jeden czasownik Z taki, że może zasta ˛pić czasownik Y w tym teście.'Ifsb/sth X-ed, it means that sb/sth else could earlier Y ; there is at least one more verb Z such that it can replace Y in this test.' 2. variant: V-N Presupposition-SI: Jeżeli stwierdzamy, że X-owa l(o), oznacza to, że musia l(o) wcześniej być Y-kiem.'Ifone says that sth/sb X-ed, it means that it/he had to earlier be Y .' Jeśli nie X-owa l(o), to też musia l(o) wcześniej być Y-kiem lub jest Y-kiem do tej pory.'Ifsth/sb did not X, it also means that it/he had to earlier be Y or it/he has been Y .' Prawdziwość stwierdzenia, że wcześniej by l(o) Y-kiem jest warunkiem koniecznym, aby sensownie stwierdzić, że X-owa l. 'The truth of the statement that sth/sb was earlier Y is a necessary condition for saying meaningfully that it/he X-ed.'Preceding-SI:Jeśli ktoś lub coś X-owa l(o), to wcześniej móg l/mog lo być Y-kiem; istnieje przynajmniej jeszcze jeden rzeczownik Z taki, że może zasta ˛pić rzeczownik Y w tym teście.Dynamic verbs in the Wordnet of Polish (b) ' Preceding mog lo być wcześniej Y-owo lub jest Y-owo do tej pory.'it could be Y or it has been Y .'(b) Presupposition musia lo dziać sie ˛Y-owo lub dzieje sie ˛Y-owo do tej pory.'it had to Y happen or it has Y happened.'Preceding mog lo dziać sie ˛wcześniej Y-owo lub dzieje sie ˛Y-owo do tej pory.'he/it could Y happen or it is possible that it has Y happened.'(c) Presupposition ktoś lub coś musia l(o) robić coś Y-owo lub robi Y-owo do tej pory.'he/it had to Y do or he/it has to have Y done' Preceding móg l/mog lo robić coś wcześniej Y-owo lub robi Y-owo do tej pory.'he/it could Y do or it is possible that he/it has Y done.' Presupposition Prawdziwość stwierdzenia, że wcześniej by l(o) Y-kowo/dzia l(o) sie ˛Y-kowo/ ktoś lub coś robi l(o) coś Y-owo jest warunkiem koniecznym, aby sensownie stwierdzić, że X-owa l. 'The truth of the statement that it was Y / Y happened, sb/sth did Y is a necessary condition for saying meaningfully that sth/sb X-ed.' Preceding Istnieje przynajmniej jeszcze jeden przys lówek Z taki, że może zasta ˛pić przys lówek Y w tym teście.'There is at least one more adverb Z such that it can replace Y in this test.'NSI Presupposition/Preceding: Jeśli ktoś lub coś X-owa l(o), to wcześniej w innych okolicznościach dotycza ˛cych kogoś lub czegoś innego 'If sb/sth X-ed, it means that earlier in some other circumstances involving somebody or something else' (a) Presupposition musia lo być Y-owo, gdzie Y jest w stopniu równym.'It was Y , where Y in the positive grade.'Preceding mog lo być Y-owo.'it could be Y ' (b) Presupposition musia lo sie ˛coś dziać Y-owo, gdzie Y jest w stopniu równym.'It had to Y happen, where Y in the positive grade.'Preceding coś mog lo dziać sie ˛Y-owo.'it could Y happen' (c) Presupposition ktoś lub coś musia l(o) robić coś Y-owo, gdzie Y jest w stopniu równym.'sb/sth had to do Y sth, where Y in the positive grade.'Preceding ktoś lub coś móg l/mog lo robić coś Y-owo.'sb/sth could Y do' Jeśli ktoś lub coś nie X-owa l(o), to w innych okolicznościach dotycza ˛cych kogoś lub czegoś innego 'If sb/sth did X, it means that earlier in some other circumstances involving somebody or something else' Presupposition Prawdziwość stwierdzenia, że wcześniej by l(o) Y-kowo/dzia l(o) sie ˛Y-kowo/ ktoś lub coś robi l(o) coś Y-owo jest warunkiem koniecznym, aby sensownie stwierdzić, że X-owa l. 'The truth of the statement that it was Y / happened Y / sb/sth did Y is a necessary condition for saying meaningfully that sth/sb X-ed.' Preceding Istnieje przynajmniej jeszcze jeden przys lówek Z taki, że może zasta ˛pić przys lówek Y w tym teście.'There is at least one more adverb Z such that it can replace Y in this test.'

Table 2 :
Relationships between Situations