Non-verbal complements of modal verbs: The case of directional adverbs in Czech

This paper deals with structures where Czech modal verbs ( muset ‘must’, moci ‘can’, smět ‘be allowed’) combine, at surface, with an adverbial complement and which involve an event of movement to the place denoted by this complement. Since modal verbs normally select a VP complement, the question arises whether these structures contain an elided or a null verb GO, or whether modal verbs here directly select a directional adverbial, whose motion interpretation supplies a ‘missing’ verb of movement. We show in this paper that there is not enough evidence to posit a null lexical verb GO in the structures under discussion. We then argue that these structures are licensed by modality like non-finite or non-sentential wh-clauses that may also contain a directional adverbial without an overt verb of movement. However, in declarative clauses, which require a verbal head to bear tense and agreement feature and to support the negative prefix ne- expressing sentential negation, the modality must be overtly realized by a modal verb.


Introduction
This paper deals with structures where modal verbs in Czech combine, at surface, with an adverbial complement (mostly PP) like in (1) and which involve an event of movement to the place denoted by this complement. We focus on strict modals muset (must), moci (can) and smět (be allowed), in comparison with the volitional verb chtít (want), 1 which allows for the same directional complements, see (2). 2 (1) Czech a. Musím do Prahy/ k doktorovi. must.1SG to Prague.GEN to doctor.DAT 'I have to go to Prague'/ 'I have to go to the doctor.' b. Po tom infarktu může jenom na zahradu. after that heart attack can.3SG only on garden.ACC 'After his heart attack, he can only go out in the garden.' c. Po tom infarktu nesmí ani na zahradu. after that heart attack NEG.can.3SG even on garden.ACC 'After his heart attack, he is not allowed to even go out in the garden.' In contrast, this structure is impossible with other lexical verbs that may combine at surface with motion verbs and their directional complements, like zkusit 'to try, akceptovat 'to accept', rozhodnout 'to decide': 3 1 When relevant, the differences between the strict modals and the verb want are explicitly put forward in the paper.
2 Abbreviations in glosses: ACC: accusative case, DAT: dative case, GEN: genitive case, IMF: imperfective, PF: perfective, FUT: future, REFL: reflexive, SG: singular, PL: plural, NEUT: neuter, NEG: negation, POSS: possessive. 3 These verbs do not allow for NCA in Czech: (i) Zítra zkusím jít do kina. / Zítra *(to) zkusim. tomorrow try.1SG go to cinema.GEN tomorrow it try.1SG 'Tomorrow I'll try to go to the cinema./Tomorrow I'll try (it).' (ii) Akceptoval jít dovnitř. / Akceptoval *(to). accepted go inside accepted it 'He accepted to go inside.' / 'He accepted (it).' The structures in (1) and (2) raise thus a question with respect to the selectional properties of modal verbs: since modal verbs normally select a VP complement, should we assume that the structures in (1) and (2) contain an elided or a null verb GO,4 or some null copula verb? Or should we better account for these structure by assuming that modal verbs may directly select a directional adverbial, whose motion interpretation supplies a 'missing' verb of movement?
Though we are not against the idea of a null verb GO in the grammar, we show in this paper that there is not enough evidence to posit such a null lexical verb in structures with strict modal verbs and directional adverbials in Czech. Rather, we argue that these structures are licensed by modality like nonfinite or non-sentential wh-clauses. However, in declarative clauses, which require a verbal head to bear tense and agreement feature and to support the negative prefix ne-expressing sentential negation, the modality must be overtly realized by a modal verb.
The paper is organized as follows. In section 2, we show basic arguments against ellipsis analysis. In section 3, we present arguments for a null verb GO that have been put forward by van Riemsdijk for Germanic languages and by Marušič and Žaucer (2005) for Slovenian. We then argue that there is no straightforward evidence for positing a null verb GO in Czech. In section 4, we propose an alternative GO-less analysis of our structure.

Against an ellipsis account
Modal verbs in Czech allow for ellipsis of their VP-complement, as we can see in (4a). Structures involving ellipsis differ however from structures involving adverbials in at least two aspects. First, elided elements need a linguistic antecedent 5 in order to be licensed and interpreted, compare (4a) with the VPantecedent and (4b) without antecedent.
(4) a. Jan si mohl vzít dovolenou v sprnu, ale jeho Jan REFL could take vacation in August but his kolega nemohl. colleague NEG.could 'Jan was allowed to take his vacation in August but his colleague was not.' b. * Janův kolega nemohl. 6 Jan.POSS colleague NEG.could 'John's colleague could not.' Second, VP-ellipsis requires a contrastive remnant element. Typically, this contrastive element will be the subject, as in (4), but it can also be the complement of the non-finite lexical verb, that had been topicalized and moved out of the VP before ellipsis took place, as shown in (5). In the same way, the contrast may involve adverbial elements, like in (6). Note that in absence of the VP antecedent, the sequence 'modal verb + locative adverbial' in (6b) is ungrammatical. Importantly, no antecedent is necessary to license and interpret the sentences in (1) and (2) 7 above. Likewise, the directional PP is not necessarily contrastively focused with another PP in the context. We conclude thus that ellipsis of a lexical verb of movement can be thus be easily rejected. In the next section, we turn to another analysis that considers that the structures we are dealing with contain a null verb GO.

Arguments in favour of a null verb GO
Structures like in (1) exist in other languages. Van Riemsdijk (2002) argues for a null verb GO in Germanic languages (except for English) like in Swiss German in (7a). His main argument comes from the contrast between (7a) and (7b) with respect to the position of the adverbial element häi (home). When a lexical verb of motion is present in the sentence, the adverbi-7 Gruet-Skrabalova (2019) also shows that even epistemic modal verbs allow for ellipsis: (i) Může to být pravda, ale nemusí.
can it be true but NEG.must 'It might be true, but it doesn't have to.' In contrast, modal verbs combining with directional adverbials have always deontic reading, see section 3.1. al cannot occur in clause-final position, see (7b). The fact the adverbial in (7a) is acceptable in clause-final surface position suggests that it is followed by a null verb GO. Since Czech is not a V2 language, this argument cannot be applied to our data.  Marušič and Žaucer (2005) argue for the existence of a null verb GO in Slovenian, whose distribution would be however larger than that of a null GO in Germanic. In the next subsections, we discuss the main arguments they present in favor of their claim: the presence of contradictory temporal adverbials, the use of purpose PPs, VP conjunction and covert modality. We argue that these arguments are not really conclusive, at least for Czech. Marušič and Žaucer (2005) argue that the possibility to have two contradictory temporal adverbs in (8b), but not in (8a), indicates that the sentence (8b) contains two temporally independent events and thus a syntactic structure with two VPs. The second adverb in (8b) would thus be dependent of the VP involving the null verb GO. In Czech, the simultaneous presence of 'yesterday' and 'today' in (9) is infelicitous even when the modal verb is followed by the overt verb jít 'go'. Note however that the verb in (9a) has deontic reading (i.e. he had to go to the doctor at a moment x). The example (10a), where the modal verb has epistemic reading (i.e. he thought it necessary (yesterday) to go to the doctor (today)) is acceptable. It is thus the obligation of 'going somewhere' that cannot be situated at another moment that the event of 'going somewhere' itself. The fact that the verb 'go' cannot be omitted in both (9b) and (10b) implies that the structures where modals combine with a directional PP only have deontic reading. This is actually the case in (1) and (2) above.

Temporal adverbials
(9) Czech a. * Včera musel jít k doktorovi dneska. yesterday must.PR.3SG.M go to doctor.DAT today '(Intended:) Yesterday he had to go to the doctor today.' b. * Včera musel k doktorovi dneska. Since the structures with strict modal verbs and directional adverbials only involve one event, we conclude that they do not require the presence of a null verb GO. Even if we admit that a single node T could be compatible with two V nodes, 9 we consider that the data like in (9) and (10) do not establish a strong piece of evidence in favor of a null V head.

Purpose PPs
A modal verb in Slovenian can combine not only with a directional PP, like in (12a), but also with a non-directional PP with the 'purpose' preposition po, like in (12b), which implies 'movement with a purpose'. Since po cannot occur with other than motion verbs, Marušič and Žaucer (2005) claim that in (12b), which is read as 'he must go and get bread', a motion verb must actually present but is not pronounced.
Peter must to store for bread 'Peter must go (to the store) and get some bread.' In Czech, the PP after the modal verb can also be introduced by the purpose preposition pro (for) or na (on/for), as shown in (13). Such a purpose PP is not by itself directional but it implies a place where we have to go in order to get the DP introduced by pro or na. This place can be stated explicitly by a directional PP, as shown in (14). But usually, the directional PP is not necessary, because it can be inferred from the purpose PP itself: the croissants are bought in a store, the mushrooms grow in the forest, and the children have to be picked up from school. We claim thus that a purpose PP do not require an overt verb of movement. Rather, we consider that an event of movement implies both direction and goal (see section 4 for more details), which has as consequence that purpose PPs appear in the same modal structures that directional PPs. We conclude that the data discussed here do not necessarily imply the presence of a null verb GO in structures where modal verbs combine with a purpose PP.

Coordination
In Slovenian, modal verbs can have scope over conjunction.
Assuming that conjuncts must be identical, Marušič and Žaucer (2005)  However, that there has been shown in the literature (e.g. Bayer 1996) that categorial identity of conjuncts is not obligatory. The conjuncts need to be semantically compatible and able to appear alone in the position of the coordinate phrase, as we can see in (16). It is thus not surprising that we can conjoin directional adverbials after modals with an overt VP, as in (17), since both may function as a predicative phrase. These examples can simply be analyzed as involving two conjoined PredPs. We thus conclude that the conjoined structures in (17) do not necessarily imply the presence of two VPs, and therefore that of a null verb GO in the conjunct containing the directional PP.

Covert modality
Finally, Marušič and Žaucer (2005) show for Slovenian that infinitival wh-clauses, that get some sort of modal interpretation (cf. Bhaat 2000), can also occur with no overt verb and a directional or purpose PP. Assuming that a clause should not exist without a verb, Marušič and Žaucer (2005)  In Czech, both directional and purpose PPs like na nádraží ('to station') and pro chleba ('for bread') respectively may also appear in embedded interrogative clauses without an overt motion verb: Moreover, these PPs may also appear in independent nonsentential interrogative fragments like in (20). In contrast, these contexts do not license static PPs that require the presence of the copula být ('be') both in embedded wh-clause in (21a) and in non-sentential interrogative fragments in (21b). Assuming thus that wh-contexts contain some covert modality, we can suppose that this is precisely this covert modality that licenses directional and by extension purpose PP (but not the static PPs), and not a motion verb. We thus conclude that these contexts do not necessarily require the presence of a null verb GO.

Summary
In this section, we presented the main arguments Marušič and Žaucer (2005) give in favor of a null verb GO in structures with directional adverbials. However, we do not think them very convincing at least for Czech. The impossibility to have two contradictory temporal adverbials implies that the structures under discussion involve only one event, which does not require the presence of a null V of movement. The fact that purpose PPs behave like directional PPs is not surprising because the purpose PPs imply a place where we have to go in order to get something. The data only show that directional and purpose PPs may occur in verbless contexts provided these contexts are in some way modal. Finally, the coordination of a PP and a VP after he modal verbs does not imply VPconjunction either, because coordination does not require strict categorial identity. VP and directional PP can be conjoined because they are semantically predicative phrases. In the next section, we propose another line of reasoning that allows for a GO-less analysis of the structures we deal with.

For a GO-less analysis
We have seen above that directional and purpose PPs may appear in wh-contexts, which contain some sort of covert modality. The same observation can be made for exclamative contexts, which are also associated with modality (cf. Le Querler 1996). As shown in (23), exclamatives allow for directional but not for static PPs: We will henceforth assume that PPs with motion and purpose 11 interpretation can be licensed by modality. The question arises how these PPs can be licensed in declarative clauses which are not by themselves modal and require an overt verbal element to bear agreement and tense features 12 . This is the reason why we have to use the copula být 'be' with nominal, adjectival or adverbial predicates; the copula 'be' however licenses only adverbials with non-motion interpretation, i.e. static PPs or source PPs like in (24). We claim that the PPs under discussion can be inserted into the syntactic structure in two ways. They can first be selected by lexical verbs of movement which do not by themselves express the direction nor the goal of the movement. The motion verbs in Germanic and Slavic languages actually express the manner but not the direction (cf. Talmy 1991), see the verbs in (25). The directional or goal PPs function thus as part of a complex predicate whose meaning is 'to move in some manner x to some place y in order to get z'. The predicate expressing a motion event contains thus three variables: x, y and z. The variables y and z can be easily let unexpressed because leaving out the place or the goal variable allows still to obtain a clause with a verbal predicate. In contrast, if we leave out the manner variable, which is expressed on the lexical verb itself, we end up with a verbless predicate, and the sentence will be ruled out by the grammar. We propose however that we may insert these PPs into syntax without a verb of movement, i.e. as predicates, provided that they are supported by some verbal element. Since directional/goal PPs can be semantically licensed by modality, the verbal element required as a support for these PPs would be a modal verb. The modal verb would thus function as a verbal support whose role is to establish the predicative relation between the subject and the non-verbal predicate, to bear tense and agreement features, and to bear negative prefix ne-expressing sentential negation. This proposal implies that modal verbs always combine with a predicative phrase, which could be verbal or nonverbal, the latter one being limited to directional and goal PPs, see (26). 13 The verbal predicate would not be limited to the verbs of movement. Thus, in (27a), the structure containing an overt verb of movement denotes a movement event in which a human subject has to go in some manner to the place where Adam actual-ly in order to achieve the goal of picking him up from that place. The manner is explicitly given by the verb: to walk, to drive, to run. In (27b), the structure containing only the goal PP implies that there is a movement event in which a human subject has as goal to pick up Adam from the place where Adam actually is. The hearer can however infer the manner of achieving this goal from the situation or from his informational background (e.g. usual situation).

Conclusion
In this paper, we have argued that structures where strict modal verbs combine at surface with an adverbial complement do not require to postulate the presence of a null verb GO. We have shown that these structures refer to a single semantic event, denoted by a single predicate, which can contain verbal or non-verbal materiel. We have also shown that covert modality licenses directional and goal PPs in wh-clauses, nonsentential wh-fragments and exclamative clauses. We have therefore argued that modality may license such PPs also in declarative sentences provided that there is an overt verbal element able to bear functional features. In our proposal, we have put forward that directional and goal PPs refer to place and goal variables that are parts of a movement event. We have proposed that these PPs can be inserted into the syntactic structure either as complements of a lexical verb of movement, or as non-verbal predicates. The latter insertion requires that two conditions be met within the declarative clause: presence of modality and presence of a verbal head. These two conditions are successfully met in sentences with modal verbs. Modal verbs are functional heads that express semantic modality ant that are able to bear tense and agreement features, and also to support the negative prefix ne-expressing sentential negation. The adverbial predicate in these structures implies a movement event, whose manner variable can be inferred from the subject, the situation or the hearer's knowledge.