Keywords

1 Introduction

One of the long-standing interests of phraseology lies in the study of compositionality, i.e., to what extent the syntactic and semantic properties of multi-word units can be deduced from the properties of their respective parts. Indeed, compositionality or the lack thereof has a wide range of consequences both in theoretical linguistics and in NLP applications (e.g., information retrieval and machine translation). In this paper, we focus on the syntactic formation of nominal structures of Czech deverbal nouns and verbal structures in which they take part, on light verb constructions representing multi-word units. We show that the deep and surface syntactic structures of light verb constructions can be treated as compositional, being composed by applying simple syntactic rules to the structure of deverbal nouns and the light verbs. Although we only investigate examples in Czech (taken mostly from the Prague Dependency Treebank), we believe the same mechanisms operate in other languages as well.

Czech deverbal nouns, being derived from verbs either by productive suffixes -ní/tí, e.g., podporování ‘supporting’, or by non-productive suffixes or the zero suffix, e.g., podpora ‘support’, can denote processes (in a broad sense, covering actions and states) as their base verbs (1), or abstract (2) or concrete (3) results of the processes denoted by their base verbs.

figure a

Deverbal nouns are typically characterized by valency, i.e., by the ability to open a certain number of valency positions for other dependent units, representing their valency complementations. The description of valency structure of deverbal nouns is of great importance as deverbal nouns referring to processes (1) and to abstract results of processes denoted by their base verbs (2) – along with deadjectival nouns (e.g., trpělivost ‘patience’) or some primary nouns (e.g., láska ‘love’) – form not only nominal structures but verbal structures as well, the latter as parts of light verb constructions. In contrast to constructions with full verbs, the syntactic structure of light verb constructions is not determined solely by the verb but also by the deverbal noun with which the verb combines. Although the participation of light verbs and deverbal nouns in the syntactic structure formation of light verb constructions has attracted much attention of theoretical linguistics, see esp. [1, 3, 4, 13], many aspects of this process still remain unclear.

In this paper, we demonstrate a close interplay between valency structures of light verbs and deverbal nouns in the syntactic formation of Czech light verb constructions. Czech, an inflectional language encoding syntactic relations via morphological forms, provides an excellent basis for studying this mechanism because the forms clearly indicate whether a complementation is syntactically structured as a complementation of the light verb, or as a complementation of the deverbal noun. Our theoretical findings are applied in an extensive annotation of light verbs and deverbal nouns in the valency lexicons VALLEX and NomVallex, respectively. We show that the information provided by these lexicons is sufficient for generating well-formed light verb constructions.

The paper is structured as follows. First, we lay out the main tenets of the description of the valency structure of deverbal nouns (Sect. 2), with an emphasis on differences between the expression of valency complementations of deverbal nouns in nominal structures (Sect. 3) and in verbal structures (Sect. 4). Second, a lexicographic representation of light verbs and deverbal nouns is proposed (Sect. 5), which, supplemented with grammatical rules, allows for generation of well-formed light verb constructions.

2 Valency of Deverbal Nouns in FGD

In the description of the valency structure of deverbal nouns, we make use of the valency theory formulated within the Functional Generative Description (FGD), a dependency oriented framework which takes a stratificational approach to language description, see esp. [14]. In FGD, valency belongs to the tectogrammatical layer, i.e., a layer of linguistically structured meaning corresponding to the deep syntactic layer, see esp. [10]. The valency theory has been first elaborated for the description of verbs. Corresponding to the argument-adjunct distinction in other linguistic theories, two kinds of valency complementations of verbs are distinguished in FGD – inner participants (actants) and free modifications (adjuncts) [12]. As actants, the valency complementations corresponding to the subject and the direct and indirect object are classified; their morphemic form is typically determined by the verb. Actants occur with a verb only once regardless of coordination and apposition (e.g., šli do kina. ‘(John and Mary)\(_{\,\mathsf{ACT}}\) went to the cinema.’ provides only a single occurrence of the relevant complementation). Both obligatoryFootnote 1 and optional actants characterize a verb in a unique way and in this sense, they have to be listed in its valency frame, see below. Five types of actants are distinguished mostly on syntactic criteria: ‘Actor’ (ACT), ‘Patient’ (PAT), ‘Addressee’ (ADDR), ‘Effect’ (EFF), and ‘Origin’ (ORIG) (e.g., ročně. ‘The government\(_{\,\mathsf{ACT}}\) restricted uranium mining\(_{\,\mathsf{PAT}}\) from the current 950 tonnes\(_{\,\mathsf{ORIG}}\) to 500 tonnes\(_{\,\mathsf{EFF}}\) per year’). Among free modifications fall the valency complementations corresponding to adverbials; in contrast to actants, their morphemic form is not determined by the verb. Free modifications can typically occur more than once with a verb (e.g., oznámila britská vláda světu,... ‘Not before the morning\(_{\,\mathsf{TWHEN}}\) on Sunday\(_{\,\mathsf{TWHEN}}\) September\(_{\,\mathsf{TWHEN}}\) the \(3^{rd}\) at 11 o’clock\(_{\,\mathsf{TWHEN}}\) did the British government announce...’). Just as actants, free modifications are either obligatory, or optional. However, unlike actants, only obligatory free modifications characterize a verb in a unique way (e.g., Petr přijel dom \(\mathring{u}\) \(_{\,\mathsf{DIR3}}\). ‘Peter arrived home\(_{\,\mathsf{DIR3}}\).’ and chovaly. ‘Children behaved well\(_{\,\mathsf{MANN}}\).’) and thus they have to be listed in the valency frame.

The same inventory of valency complementations is applied in the description of valency characteristics of deverbal nouns denoting processes and those that refer to abstract results of the processes. The inventory of valency complementations of deverbal nouns referring to concrete entities, especially to concrete results of processes, is broader, comprising other specific complementations such as the actant MATerial (e.g., ‘one package of butter\(_{\,\mathsf{MAT}}\)’) and the free modifications APPurtenance (e.g., ‘sales department\(_{\,\mathsf{APP}}\)’) and AUTHor (e.g., výzdoba ‘decoration by Michelangelo\(_{\,\mathsf{AUTH}}\)’), see [11].

The valency characteristics of both verbs and deverbal nouns are described in the form of valency frames. A valency frame is modeled as a sequence of valency slots. Each slot stands for one valency complementation and it comprises a functor (labeling the syntactic-semantic relation of the given complementation to its governing word) and the information on its obligatoriness; this information is supplemented with a list of possible morphemic forms which determine the expression of the complementation in the surface syntactic structure.

3 Nominal Structures

Czech deverbal nouns express the meaning ranging from a process via an abstract result of the process to its concrete result. Those deverbal nouns that are derived from verbs by productive suffixes -ní/-tí typically preserve the meaning of the verbs from which they are derived; as a result, they mostly express processes as their base verbs (i). For example, the deverbal noun plánování expresses the process of planning just as its base verb plánovat ‘to plan’. The deverbal nouns derived by non-productive suffixes or the zero suffix express processes only rarely. Typically, they either denote an abstract result of the process expressed by their base verbs (ii), or they refer to a concrete result of the process (iii). For example, the deverbal noun plán ‘plan’ derived by the zero suffix expresses both the abstract result of the planning process (e.g., Jan \(\mathring{u}\) v plán studovat ‘John’s plan to study’) and its concrete result (e.g., plán zahrady ‘a plan of the garden’).

(i) The deverbal nouns referring to processes typically inherit the valency structure from their base verbs, i.e., the number and type of valency complementations and their obligatoriness remain the same. Only morphemic forms of those valency complementations that are expressed as the nominative subject and the accusative direct object of the base verb undergo systemic changes:

figure b

These changes are regular enough to be described in terms of rules [11]. For example, the valency frame of the deverbal noun plánování ‘planning’ denoting the process of planning, see (5), is identical with the valency frame of its base verb plánovat ‘to plan’, see (4), except for the morphemic forms of individual valency complementations. These morphemic forms exhibit systemic shifts; the form of the ACT is changed from nominative to one of the following forms: genitive, instrumental and a possessive adjective or pronoun, and the form of the PAT changes from accusative to genitive or a possessive adjective or pronoun while other forms remain unchanged.Footnote 2

figure c

(ii) The deverbal nouns denoting abstract results of the processes expressed by their base verbs behave similarly to the deverbal nouns expressing processes in that they typically exhibit the same valency structure as their base verbs in terms of the number and type of valency complementations (i.e., the functors and obligatoriness of the complementations). However, the morphemic forms of valency complementations of these deverbal nouns are – in addition to systemic shifts – often subject to non-systemic shifts as well [7]. For example, as in case of the deverbal noun plánování ‘planning’, the valency frame of the deverbal noun plán ‘plan’, see (6), referring to the abstract result of the process of planning, consists of the same number of valency complementations as the valency frame of its base verb plánovat ‘to plan’, see above (4). Further, their functors and their obligatoriness are preserved as well. However, in contrast to the deverbal noun plánování ‘planning’, the morphemic forms of the valency complementations of this deverbal noun undergo non-systemic shifts besides systemic ones, see the prepositional groups expressing the PAT na+accusative, o+locative and pro+accusative.

figure d

(iii) Valency frames of the deverbal nouns that refer to the concrete results of the processes denoted by their base verbs can substantially differ from the valency frames of the verbs. The changes in their valency frames can concern the number of valency complementations, their type with respect to both functors and obligatoriness, as well as possible morphemic forms. For example, the valency frame of the deverbal noun plán ‘plan’ with the concrete meaning (7) comprises an optional free modification AUTH (marked as typ) and an obligatory PAT.

figure e

4 Verbal Structures

In addition to nominal structures, the deverbal nouns denoting processes (type (i) in Sect. 3) and the deverbal nouns referring to abstract results of processes (type (ii) in Sect. 3) can be employed in verbal structures as well. In this case, the deverbal noun selects a particular light (semantically impoverished) verb to form a complex predicate. This complex predicate exhibits a discrepancy between semantic and syntactic behavior. It is the deverbal noun that represents the semantic core of the given predicate, contributing most if not all of the predicate’s semantic participants. However, it is the light verb that syntactically governs the deverbal noun. In Czech, complex predicates of the given type form (idiomatic) light verb constructions.

Complex predicates in which the deverbal noun is expressed as the direct object of the light verb (i.e., by the prepositionless accusative) represent the central type of Czech complex predicates, e.g., mít plán ‘to have a plan’. Rarely, the deverbal noun can be expressed as the indirect object by a prepositionless case other than accusative, e.g., dojít újmy \(_{gen}\) ‘to come to harm’, podrobit zkoumání \(_{dat}\) ‘to put under scrutiny’, zahrnout výčitkami \(_{instr}\) ‘to shower with reproaches’, see esp. [9]. These predicates form light verb constructions (LVCs); the syntactic structure formation of LVCs is regular to a great extent (Sects. 4.1 and 4.2).

Less frequently, complex predicates in which the deverbal noun is expressed as an adverbial of the light verb occur in Czech, e.g., mít v plánu ‘to intend, lit. to have in plan’. These predicates form the so-called idiomatic light verb constructions representing the borderline between LVCs and idioms due to a greater degree of irregularities in their syntactic formation (Sect. 4.4), see [2].

4.1 The Deep Syntactic Structure

The deep structure of LVCs is formed by both valency complementations of the light verb and complementations of the deverbal noun within the complex predicate. For example, the deep structure of LVCs with the complex predicate mít plán ‘have a plan’ consists of both valency complementations of the deverbal noun plán ‘plan’, see the valency frame in (6) in Sect. 3, and complementations of the light verb mít ‘have’, see the frame in (9).

The valency frame of the deverbal noun corresponds to the usage of the noun in nominal structures (Sect. 3). Individual valency complementations of deverbal nouns are semantically saturated by semantic participants. For example, the ACT and PAT of the noun plán ‘plan’ are mapped onto ‘Agent’ and ‘Goal’, respectively.

The valency frame of the light verb is typically identical with the frame of its full verb counterpart, see the frame of the full verb mít ‘have’ in (8) and the frame of the light verb mít ‘have’ in (9) and examples (10) and (11), respectively. They only differ in that the valency position reserved for the deverbal noun, representing the nominal component of the complex predicate with the light verb, is marked by the functor CPHR.

figure f

While valency complementations of full verbs are mapped onto semantic participants, complementations of light verbs do not correspond to any participants; the only exception is represented by light verbs with causative meaning that provide the semantic participant ‘Causator’, see below. For example, the ACT and PAT of the full verb mít ‘have’ are mapped onto ‘Possessor’ and ‘Possession’, respectively. In contrast, the ACT of the light verb mít ‘have’ (the only valency complementation in its valency frame except for CPHR) is not semantically saturated by any participant of the verb.

Although valency complementations of the light verb are not semantically specified by any semantic participants of the verb, they do not remain semantically unsaturated. To acquire semantic capacity, valency complementations of the light verb enter in coreference with complementations of the deverbal noun. For example, the ACT of the light verb mít ‘have’ within the complex predicate mít plán ‘to have a plan’ corefers with the ACT of the noun plán ‘plan’; both these ACTors thus refer to the ‘Agent’, see the scheme of the mapping of the semantic participants onto valency complementations in the complex predicate mít plán ‘to have a plan’ provided in (12) (\(\Rightarrow \) indicates the mapping of participants, \(\rightarrow \) shows coreference). Figure 1 displays the deep syntactic structure of the complex predicate mít plán ‘have a plan’.

figure g
Fig. 1.
figure 1

The deep and surface syntactic structure of the complex predicate mít plán ‘to have a plan’.

In case of causative light verbs, the semantic participant ‘Causator’ is provided by the light verb, being mapped onto one of its valency complementations. For example, within the complex predicate přinést radost ‘to bring joy’ two semantic participants – ‘Experiencer’ and ‘Stimulus’ – are provided by the noun radost ‘happiness, joy’. These participants correspond to the ACT and PAT of the noun, respectively, see the frame of the noun radost ‘happiness, joy’ in (13). In addition, the causative light verb přinést ‘to bring’ contributes the ‘Causator’, mapped onto its ACT, to the complex predicate přinést radost ‘to bring joy’, see the frame in (14). The ADDR of the light verb – the only remaining semantically unsaturated valency complementation in its valency frame – does not correspond to any participant. To be semantically saturated, the ADDR enters in coreference with the ACT of the deverbal noun radost ‘happiness, joy’, see the scheme of the mapping of the semantic participants within the complex predicate přinést radost ‘to bring joy’ in (15).

figure h

4.2 The Surface Syntactic Structure

The surface structure of LVCs can be composed of both valency complementations of the light verb and complementations of the deverbal noun. Each semantic participant is typically expressed in the surface structure only once. Despite being primarily contributed to LVCs by deverbal nouns, semantic participants tend to be expressed on the surface as valency complementations of light verbs. As corpus evidence shows, the surface structure formation of Czech LVCs is governed by the following principles [5]:

  • From the valency frame of the light verb, all valency complementations are expressed, namely:

    • - the valency complementation with the functor CPHR, indicating the deverbal noun,

    • - the valency complementation corresponding to the ‘Causator’ (if present),

    • - all valency complementations which corefer with complementations of the deverbal noun.

  • From the valency frame of the deverbal noun, those valency complementations are expressed on the surface that are not in coreference with any complementation of the light verb.

For example, from the valency frame of the light verb mít ‘to have’, besides the CPHR reserved for the predicative noun, the ACT coreferring with the ACT of the deverbal noun plán ‘plan’ is expressed in the surface structure of the LVC with the complex predicate mít plán ‘to have a plan’. From the valency frame of the noun plán ‘plan’, only the PAT, not coreferring with any complementation of the verb, is expressed on the surface; see the valency frames (6) and (9) and the scheme (12). Example (16) illustrates the LVC with the complex predicate mít plán ‘to have a plan’; see also its surface syntactic structure in Fig. 1.

figure i

The surface structure of the LVC with the complex predicate přinést radost ‘to bring joy’ with the causative light verb přinést ‘to bring’ consists of all valency complementations of the light verb přinést ‘to bring’: the ACT corresponding to the ‘Causator’, the ADDR coreferring with the ACT of the deverbal noun radost ‘happiness’ (thus expressing the ‘Experiencer’) and the CPHR standing for the deverbal noun. From the valency frame of the deverbal noun, only the PAT that does not corefer with any complementation of the verb is expressed on the surface, see the valency frames (13) and (14), scheme (15) and example (17).

figure j

4.3 A Quantitative Comparison of Nominal and Verbal Structures in Corpus Data

Prague Dependency Treebank (PDT),Footnote 3 containing Czech texts with complex and interlinked morphological, surface syntactic and deep syntactic annotation, provides an excellent basis for the study of syntactic behavior of deverbal nouns in nominal and verbal structures. We have divided the instances of nouns in PDT into two groups: the occurrences in which the noun has the CPHR functor,Footnote 4 representing verbal structures, and all other occurrences of nouns, representing their nominal structures. We have limited our investigation to the nominal lemmas that appear at least once in a verbal structure; Table 1 gives basic statistics of investigated lemmas and instances.

The graph in Fig. 2 summarizes the frequency of different types of valency complementations of deverbal nouns expressed on the surface relative to all their occurrences; the remaining space to 100% is filled by the occurrences of that functor that are not expressed on the surface.

The figure shows that nominal ACT and ADDR are expressed much more rarely in verbal structures (where they typically corefer with verbal complementations) than in nominal structures. On the other hand, the PAT of productively derived nouns is expressed more often in verbal structures; a preliminary investigation of a sample of the VALLEX lexicon (Sect. 5.2) shows that a nominal PAT enters a coreference relation with a complementation of the light verb about three times less than a nominal ACT.

Table 1. Investigated PDT data
Fig. 2.
figure 2

For each functor, the percentage of its occurrences in which it is expressed in the surface structure of the sentence is given. The missing columns correspond to actants that did not appear with any noun instance in the investigated sample.

4.4 Remarks on Idiomatic Light Verb Constructions

By idiomatic light verb constructions (ILVCs) we understand the LVCs that are formed by the complex predicates within which the deverbal noun is expressed as an adverbial of the light verb (i.e., by various prepositional groups, e.g., dát do pořádku ‘to put in order’, mít pod kontrolou ‘to have under control’, mít na starosti ‘to be responsible, lit. to have on care’, mít v úmyslu ‘to intend, lit. to have in intention’).

In ILVCs, semantic participants are primarily provided – as in LVCs – by deverbal nouns that represent their semantic core. The only exception is represented by the complex predicates within which light verbs fulfill the causative function (e.g., dostat pod kontrolu ‘to get under control’, uvést do chodu ‘to give a start, lit. lead into operation’, dát do ochrany ‘to give under protection, lit. to give in protection’), which contribute the participant ‘Causator’ initiating the event expressed by the deverbal noun.

The syntactic structure formation of ILVCs is governed by the same principles as that of LVCs:

  • the deep syntactic structure of ILVCs results from an interplay between the valency frame of the deverbal noun and the frame of the light verb (Sect. 4.1) and it is characterized by coreference between valency complementations of the deverbal noun and complementations of the light verb;

  • the surface syntactic structure of ILVCs is guided by the same principles as that of LVCs (Sect. 4.2).

For example, the deep structure of the ILVC formed by the complex predicate mít v plánu ‘to intend, lit. to have in plan’ is underlain by the valency frame of the light verb mít ‘to have’ (18) and the frame of the deverbal noun plán ‘plan’ (6) repeated here in (19). This predicate exhibits coreference between the nominal ACT and the verbal ACT and at the same time between the nominal PAT and the verbal PAT, expressing the ‘Agent’ and ‘Goal’ provided by the noun, respectively, see scheme (20) and the deep structure of this ILVC in Fig. 3.

figure k

In the surface structure of the ILVC with the complex predicate mít v plánu ‘to intend, lit. to have in plan’, all valency complementations from the valency frame of the light verb (18) are expressed, namely ACT and PAT coreferring with the nominal ACT and PAT, and the CPHR occupied by the deverbal noun. No valency complementations from the valency frame of the deverbal noun, namely neither the ACT nor the PAT, are expressed on the surface as they corefer with the verbal complementations, see the principles of the surface structure formation in Sect. 4.2 and example (22). The surface structure of the ILVC is displayed in Fig. 3.

Fig. 3.
figure 3

The deep and surface syntactic structure of the complex predicate mít v plánu ‘to have a plan, lit. to have in plan’.

However, ILVCs exhibit a lesser degree of regularity in morphemic expressions of valency complementations than LVCs. The irregularities concern morphemic forms of those valency complementations that semantically correspond to participants of propositional character. Corpus evidence shows that information on their morphemic forms cannot be easily inferred from the valency frame of the light verb containing the valency complementation referring to the given participant.

For example, the semantic participant ‘Goal’ characterizing the complex predicates mít plán ‘to have a plan’ and mít v plánu ‘to intend, lit. to have in plan’ represents a proposition. In LVCs formed by the complex predicate mít plán ‘to have a plan’, the ‘Goal’ is realized on the surface as an attribute of the deverbal noun plán ‘plan’, i.e., as its PAT (see principles in Sect. 4.2 and Fig. 1). The PAT can be expressed in LVCs by any of the morphemic forms prescribed in the valency frame of this noun, see the valency frame (19) and example (21).

In contrast, in ILVCs formed by the complex predicate mít v plánu ‘to intend, lit. to have in plan’, the Goal is expressed on the surface as the direct object of the light verb mít ‘to have’, i.e., as its PAT. In the valency frame of the light verb mít ‘to have’ (18) – underlying also this verb in the complex predicates without any propositional participant, e.g., mít v užívání ‘to make use, lit. to have in use’, mít v oblibě ‘to have a liking, lit. to have in liking’, only the form of a prepositionless accusative is prescribed for the given PAT. However, the PAT corresponding to the propositional participant ‘Goal’ in ILVCs with the complex predicate mít v plánu ‘to intend, lit. to have in plan’ can have the form of the infinitive and the dependent content clause as well, see (22).

figure l

To capture irregularities in the expression of propositional participants in ILVCs, we establish the following principle:

  • The valency complementation of the light verb which in the surface structure of ILVCs occupies the direct object may be expressed, additionally to the forms listed in the valency frame of the light verb, by an infinitive or a dependent content clause on condition that these forms are listed as a morphemic expression of the coreferring nominal complementation.

5 Lexicographic Representation of Deverbal Nouns

In this section, we introduce NomVallex and VALLEX, valency lexicons which provide information on the basis of which nominal (Sect. 5.1) and verbal structures containing deverbal nouns (Sect. 5.2) can be generated.

5.1 Nominal Structures in a Lexicon

The valency lexicon NomVallex provides all the information necessary for producing nominal structures governed by Czech deverbal nouns, together with additional syntactic information [6]. This lexicon has adopted the design of the valency lexicon of Czech verbs, VALLEX, see Sect. 5.2.

In NomVallex, each deverbal noun is described by a lexeme, an abstract twofold unit associating all lexical forms of the noun with lexical units (individual senses of the noun). Each nominal lexeme is represented by a set of lemma(s); it is formed by a set of lexical unit(s) corresponding to the individual senses. The lexical units are specified with respect to their types, i.e., whether they denote processes, abstract results or concrete results of processes.

Key information on the valency structure of each lexical unit is provided by the valency frame, see Sect. 2. Besides the valency frame, each lexical unit is accompanied with a gloss describing the given meaning and an example providing its corpus evidence. Further, each lexical unit can be assigned additional syntactic and semantic information, e.g., on reciprocity, control, and semantic class membership.

Each deverbal noun forming complex predicates with light verbs has the attribute lvc providing references to individual valency frames of light verbs with which the noun combines; the particular light verbs are stored in the VALLEX lexicon, see Sect. 5.2. On the basis of these references, individual complex predicates can be obtained. See Fig. 4 displaying a simplified entry of the deverbal noun plán ‘plan’.

Fig. 4.
figure 4

Simplified entry of the noun plán ‘plan’.

5.2 Verbal Structures in a Lexicon

The derivation of verbal structures with deverbal nouns (LVCs and ILVCs, see Sect. 4) requires a close cooperation between the lexicographic representation of deverbal nouns (see Sect. 5.1) and light verbs on the one hand, and grammatical rules, on the other (see Sect. 4). The representation of LVCs is proposed here for the valency lexicon of Czech verbs VALLEX, see [8]. The structure of this lexicon is the same as that of NomVallex (Sect. 5.1).

Light verbs in VALLEX are treated as specific senses of verbs. Each light verb is assigned its respective valency frame. The valency position reserved in the valency frame for deverbal nouns is labeled by the CPHR functor. Each valency frame of the light verb is assigned three special attributes: lvc, instig, and map. The attribute map provides a list of pairs of valency complementations of the deverbal noun and the light verb that are in coreference. The attribute lvc provides references to relevant deverbal nouns. If relevant, the attribute instig introduces that valency complementation of the light verb onto which the semantic participant ‘Causator’ is mapped. See Fig. 5 displaying the simplified entry of the verb mít ‘to have’.

The lexicon thus economically captures recurring patterns of complex predicates with the given light verb, and the grammatical rules enable users to generate well-formed LVCs and ILVCs.

Fig. 5.
figure 5

Simplified entry of the verb mít ‘to have’, showing the valency frames for the core meaning of the full verb and its corresponding light verbs.

6 Conclusion

In this paper, we have proposed a lexicographic representation of nominal and verbal structures of Czech deverbal nouns, focusing primarily on the verbal ones, i.e., light verb constructions and idiomatic light verb constructions. We have demonstrated that their well-formed syntactic structure can be derived by a complex process requiring a close interplay of information provided by lexicons and grammatical rules, making it possible to better understand the syntactic compositionality of these types of multi-word units.