1. Introduction
Kiparsky & Kiparsky (Reference Kiparsky, Kiparsky, Bierwisch and Heidolph1970) proposed that factive clauses are embedded in a complex NP headed by a silent fact. In a recently updated version of this proposal, Kastner (Reference Kastner2015) extends this analysis to all kinds of presuppositional clauses, including sentential subjects and topics (cf. Davies & Dubinsky Reference Davies, Dubinsky, Gerdts, Moore and Polinsky2009, Takahashi Reference Takahashi2010) and also complements of so-called response-stance verbs like ‘agree’ and ‘deny’. In addition, Kastner assumes that presuppositional clauses can be nominal not only by virtue of being embedded in a complex NP, but also by being introduced by a (possibly silent) definite D (
$ \Delta $
), which, he argues, happens in English, as illustrated in (1a)–(1b), cf. non-presuppositional clauses in (1c).Footnote
2
In its general form, Kastner’s (Reference Kastner2015) proposal (referred to as the DP requirement), can be summarized as in (2). The proposal receives crosslinguistic support from languages like Hebrew and Greek where clauses can be introduced by an overt definite D (a DP shell), which appears to systematically correlate with the presuppositionality of the clause.
In this paper, I discuss an apparent problem with this proposal posed by Russian data.Footnote 3 I show that in Russian the correlation between presuppositionality and the presence of the DP shell (to) is only partial. Specifically, while in sentential subjects and topics the DP shell is indeed obligatory (or strongly preferred), in complements of presuppositional verbs the DP shell is generally optional. In addition, clauses with DP shells in Russian are not always presuppositional.
I propose a solution to this problem based on the idea that lexical items can realize sequences of heads in a head-complement relation, i.e. spans (Svenonius Reference Svenonius2012, Ramchand Reference Ramchand2018). Specifically, I suggest that presuppositional verbs in Russian can lexicalize the DP shell (DΔ) by virtue of having spanning lexical entries that correspond to V-DΔ (or V-P-DΔ) spans. I further suggest that in English the DP shell can also be lexicalized by the complementizer (that), corresponding to a DΔ-C (or a P-DΔ-C) span. I argue that the spanning analysis is superior to an alternative analysis in terms of a (separate) null DΔ, which cannot account for the distribution of to in a satisfactory way.
Section 2 discusses the distribution of to in Russian indicative declarative clauses and introduces the problem for the DP requirement. Section 3 presents the spanning account and discusses arguments for it and against the null DΔ alternative. Section 4 extends the spanning analysis to C-drop. Section 5 concludes.
2. The problem: subject–complement asymmetry
In Russian, argument clauses can be nominalized by the (case-marked) n.sg. form of the demonstrative ‘that’. I assume that such clauses have the structure in (3), with to as the definite D head (overtly realizing DΔ) which takes the CP as its complement (Hartman Reference Hartman2012, Knyazev Reference Knyazev2016).Footnote 4
At first glance, the distribution of nominalized clauses (see e.g. Hartman Reference Hartman2012, Knyazev Reference Knyazev2016) seems to comply with the DP requirement in (2). Thus, sentential subjects (corresponding to external arguments), as in (3), which are taken to be presuppositional, almost obligatorily require to (cf. (6c)).Footnote 5 Conversely, complements of non-presuppositional verbs, as in (4), normally disallow to. Footnote 6
However, the correspondence between presuppositionality and the presence of to is not perfect: complements of presuppositional verbs, as in (5a), as well as postverbal subjects (corresponding to internal arguments), as in (5b), allow but crucially do not require to, in violation of (2a).Footnote 7
It might be assumed that presuppositional clauses without to in examples like (5) contain a null variant of DΔ, as in (6a–b). However, this analysis is problematic because then null DΔ should also be possible in sentential subjects, as in (3). Yet, overt to is required in this case, as shown in (6c). The problem can be characterized as a subject-complement asymmetry.Footnote 8
An additional problem for the DP requirement in (2) is that clauses in object-of-P/oblique positions are generally optional, regardless of whether the predicate is non-presuppositional, as in (7a), or presuppositional, as in (7b).Footnote 9
The presence of to in examples like (7a) could be accounted for by an independent requirement for complements of P, as well as complements assigned oblique case, to be nominalized so that in such cases (2b) would not apply. Yet, the absence of to in examples like (7b) would still violate (2a) unless they contain null DΔ, which runs into the problem of constraining the distribution of null DΔ.Footnote 10
3. A spanning account
3.1. The account of the subject–complement asymmetry
As shown in (6), a null DΔ leads to overgeneration in the case of sentential subjects. I wish to argue that this problem can be avoided if morphemes are allowed to spell-out multiple terminals. I adopt the spanning implementation of this idea, which uses the notion of a span, i.e. ‘a contiguous sequence of heads in a complementation relation’ (Ramchand Reference Ramchand2018: 27). On this view, if a lexical item corresponds to a span of category features (e.g. <X, Y, Z>), it can realize any contiguous subspan of it in a given structure (e.g. X, X-Y, X-Y-Z, etc. but not e.g. X-Z or W-X). This is referred to as the Superset Principle. It is also commonly assumed that every syntactic head must be lexicalized and that zero morphemes are by default avoided (see e.g. Caha Reference Caha, Baunaz, Haegeman, Clercq and Lander2018).
The subject–complement asymmetry in (6) can be straightforwardly explained in the spanning framework. Suppose that presuppositional verbs in Russian like otricat’ ‘deny’ correspond to a span that includes DΔ, as illustrated in (8a). Suppose further that to corresponds to DΔ, as I assumed earlier, and čto to (indicative) C, as shown in (8b–c). Given these (simplified) lexical entries, the optionality of to in presuppositional complements (cf. (5a)) follows from the fact that, due to the Superset Principle, otricat’ ‘deny’ can lexicalize either the V-DΔ span, as in (9), or just its V subspan (in which case DΔ is lexicalized by to), as in (10).Footnote 11 Postverbal (internal) sentential subjects of presuppositional verbs (cf. (5b)) are accounted for in a similar way.
By contrast, the obligatoriness of to in preverbal (external) sentential subjects (cf. (3), (6c)), follows from the fact that they are not in the complement position and thus, by definition of a span, DΔ cannot form a span with V, precluding lexicalization by a spanning verb. As a result, DΔ is lexicalized by to, assuming that zero morphemes are disallowed. Thus, the overgeneration problem in (6c) does not arise because null DΔ is not postulated in the first place.
The optionality of to in presuppositional clauses in object-of-P/oblique position with verbs like sožalet’ ‘regret’ (cf. (7b)) can also be easily captured on the spanning view if we allow such verbs to include not just DΔ but also P (or the head that licenses/realizes the oblique case) in their span, as in (11). Presuppositional clauses without to would be observed when the verb lexicalizes the whole V-P-DΔ span, whereas clauses with to would correspond to the lexicalization of just the V subspan (with P and DΔ lexicalized by designated morphemes).Footnote 12
An alternative analysis with a null DΔ requires an additional stipulation: it would need to posit a null P (PΔ) that only selects for null DΔ (as contrasted with verbs like otricat’ ‘deny’, which would select either for overt or null DΔ), given that verbs like sožalet’ ‘regret’ do not take overt to or ordinary DPs (without overt P), as shown in (12). This is problematic as we expect null and overt DΔ (to) to differ only phonologically, but not in syntactic selectional features.
The optionality of to in non-presuppositional clauses in object-of-P/oblique position with verbs like nadejat’sja ‘hope’ (cf. (7a)) can be taken to follow from the fact that such verbs (disjunctively) select not only for a PP/oblique complement (in which case the clause is obligatorily nominalized), but also for a CP. The latter option is available because non-presuppositional clauses do not require DΔ.Footnote 13 By contrast, in the respective case of presuppositional clauses an analysis with a disjunctive selection is unavailable because a (bare) CP complement would violate the DP requirement.Footnote 14
To summarize, the spanning analysis of presuppositional verbs provides a simple and straightforward account of the distribution of the DP shell in Russian, particularly of the subject–complement asymmetry in (6). Below, I present two further arguments in favor of this analysis over the null DΔ alternative.
3.2. Further arguments for the spanning account
A similar overgeneration problem as in the case of sentential subjects (cf. (6c)) arises with sentential topics, which are also presuppositional (Kastner Reference Kastner2015) and hence subject to the DP requirement in (2a). While not as strongly as in the former case, sentential topics without to are disfavored, as shown in (13a–b).Footnote 15
Yet, a null DΔ/PΔ account, illustrated in (14a–b), wrongly predicts that such examples should be fully acceptable. By contrast, on the spanning account the structures in (14a–b) are ruled out since DΔ (and PΔ) cannot form a span with V (assuming that spell-out happens after topicalization), resulting in the lexicalization of DΔ (and PΔ) by separate morphemes.Footnote 16 , Footnote 17
More importantly, on the spanning account we expect that presuppositional verbs will vary as to whether they require overt to. Footnote 18 This is because whether to may be dropped depends on whether a given verb has a spanning entry (that includes DΔ). We do indeed find such variability. Although many presuppositional predicates alternate between complements with and without to, some, especially less common ones, e.g. obuslovlen (ins) ‘driven (by)’, as in (15a), otreagirovat’ (na) ‘react (on)’, as in (15b), stojat’ (na) ‘stick (to)’ and others require overt to. Footnote 19 The obligatoriness of to with such verbs would follow from the fact that they simply do not have spanning entries, unlike otricat’ ‘deny’ and sožalet’ ‘regret’.
By contrast, such verb-by-verb variability is not predicted by the null DΔ account. Rather, we expect language-by-language variability because whether null DΔ is available (as a separate morpheme) depends on the lexicon of a particular language.Footnote 20 While the difference between otreagirovat’ ‘react’ and sožalet’ ‘regret’ can still be captured on the null DΔ/PΔ account, e.g. by taking sožalet’ ‘regret’ to select either for overt P or null PΔ (see (12) and the surrounding discussion) and otreagirovat’ ‘react’ to select only for overt P, overall the spanning account provides a simpler and more principled explanation of the relevant facts.
3.3. Comparison with English
I argued that in Russian bare CPs (without to) are disallowed as sentential subjects and topics because their DP shell cannot be lexicalized by a spanning verb (since DΔ is not in the complement position). This raises the question as to why English that-clauses are allowed in the corresponding examples. I wish to suggest that in English DΔ can be lexicalized by a spanning complementizer corresponding to a DΔ-C span, as illustrated in (16a–b), cf. a non-spanning entry for Russian čto in (8c).Footnote 21 As a result, that-clauses, unlike Russian čto-clauses, can occur in non-complement positions such as subjects and topics. Note that that can still lexicalize just a C head (due to the Superset Principle).
In fact, there is evidence that that actually lexicalizes a larger span that includes P, as in (17a). First, that-clauses can occur with verbs that select for PPs and do not take DPs such as agree, as in (17b). Moreover, this is possible when the clause is extraposed to the right, when P cannot remain in the complement position and thus cannot be lexicalized by the verb (given that rightward P-stranding is prohibited), as in (17c). Second, the entry in (17a) may provide an explanation for why that-clauses, unlike Russian clauses with to (cf. (7)), are disallowed as objects of P, as shown in (17d). Specifically, we could argue that the lexicalization of the P-DΔ-C span directly by that is preferred to and thereby blocks the lexicalization of the same span with a separate P plus that (due to a blocking principle favoring fewer exponents when possible (Svenonius Reference Svenonius2012)).Footnote 22
Space precludes a detailed examination of the distribution of English that-clauses (see e.g. Bruening & Al Khalaf Reference Bruening and Al Khalaf2020 for a recent discussion). Instead, I now turn to evidence for spanning verbs in a different domain.
4. The spanning analysis of complementizer drop
The proposed account of the distribution of the DP shell (DΔ) in Russian in terms of its lexicalization by a spanning verb can be extended to the distribution of zero complementizer (vs. that), or ‘C-drop’. C-drop is in some respects similar to ‘to-drop’ in Russian. For example, it is possible in complements, as in (18a), but disallowed in sentential subjects and topics, as shown in (18b–c).
The spanning analysis provides a straightforward account of the asymmetry in (18). Verbs that allow C-drop like think can be analyzed as having <V, C> entries, with C-drop corresponding to the lexicalization of the whole V-C span by the verb, as in (19), whereas overt C (that) would correspond to the lexicalization of just the V subspan. In turn, the ungrammaticality of C-drop in sentential subjects and topics would follow from the fact that the CP is not in the complement position and thus V cannot form a span with C, precluding a spanning lexicalization.Footnote 23
A further argument for the spanning account comes from the well-known fact that not all verbs allow C-drop and, more importantly, that there are some idiosyncratic restrictions. For example, C-drop is disallowed with manner-of-speaking verbs (Stowell Reference Stowell1981) and, most relevantly, presuppositional verbs including response-stance (Hegarty Reference Hegarty1992) and emotive factive verbs, as in (21a). At the same time, C-drop is allowed with some presuppositional verbs, including cognitive factives, as in (21b), as well as with various exceptions among emotive factives and response-stance verbs, as in (21c–d).
On the spanning analysis, the variability among presuppositional predicates in (21) is expected and can be captured by analyzing presuppositional predicates that allow C-drop like doubt/deny as having spanning entries that include not just C but also DΔ in their span, i.e. as <V, DΔ, C>, as in (20), whereas verbs that disallow C-drop like regret would be analyzed as lacking <V, DΔ, C> entries (cf. a parallel argument in Section 3.2).Footnote 24
By contrast, the observed verb-by-verb variability is not explained by the traditional analysis of C-drop in terms of null C and its licensing conditions (Bošković & Lasnik Reference Bošković and Lasnik2003). In a system with a null DΔ (Kastner Reference Kastner2015), one may try to capture the distribution of C-drop in terms of selection, e.g. by taking DΔ to select only for overt, but not null, C. However, this is problematic since overt and null C are not expected to differ in syntactic features (cf. Section 3.2). Moreover, it is unclear how this account can deal with exceptions among presuppositional verbs. Should there be another DΔ that selects for either null or overt C?
A selection-based analysis where DΔ selects only for overt, but not null, C may be more plausible for Russian, where C-drop is consistently disallowed with presuppositional verbs, as shown in (22a), and where overt to is incompatible with null C, as shown in (22b). But this account still invokes an unmotivated featural difference between overt and null C. On the spanning view, by contrast, the relevant data can be easily captured by saying that Russian simply lacks verbs that include both C and DΔ in their span (and only has verbs of the <V, C> class), whereas the restriction in (22b) would follow from the contiguity condition on spans (the verb would have to skip P and DΔ to lexicalize C).Footnote 25
To summarize, the spanning analysis provides a straightforward account of the distribution C-drop, specifically of the subject/topic–complement asymmetry, as well as of its partly idiosyncratic character.Footnote 26
5. Conclusion
In this paper, I presented an account of the (non)-realization of the overt DP shell (to) in Russian, in particular, its obligatoriness in sentential subjects/topics and its optionality in complements of presuppositional verbs. The account was based on the DP requirement, namely, that presuppositonal clauses are introduced by the definite determiner (DΔ) (Kastner Reference Kastner2015), and, crucially, on the proposal that D is lexicalized by verbs with spanning lexical entries (that include DΔ). I also suggested that in English DΔ can be lexicalized by a spanning complementizer (that), an option unavailable in Russian. Finally, I showed that the spanning account can be extended to C-drop (by taking verbs that allow C-drop to include C in their span), capturing its intriguing similarities with to-drop.
Many questions remain: What is the exact mechanism of spell-out in the spanning framework? How does it interact with the timing of the derivation? Whether and how do spanning and non-spanning lexicalizations compete? Regarding the latter, it is sometimes assumed that the grammar contains a general blocking principle that (under certain conditions) favors a single spanning exponent over multiple exponents lexicalizing the same span (as in the case of du vs. de l(e) in French (Svenonius Reference Svenonius2012)). An alternative view is that blocking is not a principle of grammar but rather is a result of routinization of particular structures during the learning process and thereby is subject to frequency effects and other factors outside grammar proper.Footnote 27 These and other important issues must be addressed in future work. For now we may conclude that spanning provides a promising approach to the study of clausal complementation.
