3.1. Distributional differences

We show below that, unlike (24), sentences such as (25) and (26) contain the case-bearing to. For convenience, we refer to the structurally reduced variant in (24) – analysed below as an nP – as to_n and to the KP variant in (25)–(26) as to _k .

At first glance, (24) and (25) differ only in the case of the predicate (nom vs. ins) and in agreement: the right-hand KP obligatorily controls agreement in (24), whereas the copula in (25) takes the third-person singular neuter form, also observed in sentences with to being the subject of lexical verbs (see (26)). However, there are several important differences between to in (24) on the one hand, and to in (25) and (26) on the other.

First, only to _k is distributionally equivalent to KPs headed by lexical nouns. That is, one can replace to with such a KP in (25) and (26), but – regardless of agreement – not in (24), as shown in (27). (To save space, we do not show examples of to occurring with lexical verbs, as it behaves identically to to with instrumental predicates.)

Second, in contrast to to _k , to_n cannot be modified by adjectives, see (28).

Third, only to _k can act as a nominaliser of a complementiser phrase (CP), see (29).

Finally, when to _k is used anaphorically, its antecedent must be a $ \phi $ -less expression, for instance, a clause. In contrast, to_n can refer to any antecedent, including a lexical KP denoting a human; see (30).

The differences are summarised in Table 1.

Table 1.

Differences between to_n and to _k .

Table 1. Long description

The table consists of four columns. The first column lists the property being compared, the second column describes to sub n, the third column describes to sub K, and the fourth column provides example references.

* Row 1: Property is (to plus N O M predicate) for to sub n and (e.g., to plus I N S predicate) for to sub K.

* Row 2: Distribution of K P. to sub n is marked with a minus sign. to sub K is marked with a plus sign. Example (27).

* Row 3: Modifiability by adjectives. to sub n is marked with a minus sign. to sub K is marked with a plus sign. Example (28).

* Row 4: Can nominalise a C P. to sub n is marked with a minus sign. to sub K is marked with a plus sign. Example (29).

* Row 5: Anaphoric possibilities. to sub n allows any antecedent. to sub K allows phi-less antecedents. Example (30).

In brief, to _k behaves like a full KP: it is distributionally equivalent to lexical KPs, it can be modified by adjectives and it can nominalise a CP – similarly to certain nouns taking CP dependents (e.g., fakt ‘fact’, myśl ‘thought’). Its distributional equivalence to a KP also explains its behaviour with respect to what antecedents it can refer to. Namely, to _k competes with other pronominals having KP distribution, that is, personal pronouns. However, in contrast to to _k , such pronouns have number and gender features (for instance, on ‘pers.sg.m’, ono ‘pers.sg.n’), and therefore outcompete to _k whenever the antecedent bears the corresponding number and gender features. Consequently, to _k is an elsewhere option, occurring only when personal pronouns cannot be used, i.e., when the antecedent is $ \phi $ -less – for instance, when it is expressed by a CP or an infinitival TP. The other to, to_n, is special: it is a strictly referential item, but it does not behave like a KP – it cannot be replaced by KPs, it cannot be modified by adjectives, nor can it nominalise a CP. As it is not distributionally equivalent to KPs (such as personal pronouns), it does not compete with them, so it can refer to any antecedent.

Given that to_n and to _k share the same morphological form and the same primarily demonstrative function, they should not be analysed as unrelated homonyms. We propose that they are closely related and share a referential feature, namely, the root of the feature geometry (Harley & Ritter Reference Harley and Ritter2002), called here $ \pi $ . This root is exponed in Polish as t-, as witnessed in various demonstrative forms, such as the adverb t-ak ‘in this manner’ and the numeral t-yle ‘this much’. Apart from the $ \pi $ feature, to_n and to _k also share the nominal categoriser, n, exponed as -o; hence the common form to. (We will have more to say on this in the following subsection.) However, distributional differences discussed above indicate that only to _k projects to a full KP. Overall, we postulate the following structures for to_n and to _k , with the dashed line signalling our agnosticism about projections between nP and KP:

(31)

This analysis, treating to _k as a full KP projection, predicts that to _k inflects for case. This is borne out, as shown in (32). Crucially, the case-inflected to matches all the characteristics of to _k summarised in Table 1.

To summarise, assuming that to_n is the nominalised version of the referential feature, which can be embedded in a case projection (forming to _k ), explains (i) why the two – to _k and to_n – are spelled out as the same form, (ii) why to _k competes with other case-bearing pronominals while to_n does not and (iii) why only to _k behaves syntactically like a complete nominal projection: it inflects, combines with adjectives and may nominalise CPs.

3.2. Feature geometry, $ \boldsymbol{\pi} $ and $ \boldsymbol{\phi} $

Following Harley & Ritter (Reference Harley and Ritter2002) and much subsequent work (see, for instance, McGinnis Reference McGinnis2005, Béjar & Řezáč Reference Béjar and Řezáč2009), we assume that features relevant for agreement are organised within a hierarchy (a feature geometry). The hierarchy we assume for Polish is presented in (33). Although simplified in some respects that are irrelevant here,Footnote ¹⁸ it explicitly represents the distinction between the feature $ \pi $ – responsible for referentiality – and the feature $ \phi $ , a strictly grammatical feature present on third-person singular neuter items.

(33)

Distinguishing between [ $ \pi $ ] and [ $ \pi $ , $ \phi $ ] makes it possible to capture the difference between $ \phi $ -less referential expressions on the one hand (such as infinitival and clausal phrases in argumental positions, as well as to in both its incarnations: to_n and to _k ) and third-person singular neuter items on the other (such as the personal pronoun ono ‘sg.n’ and neuter gender nouns such as okno ‘window.sg.n’ or krzesło ‘chair.sg.n’).Footnote ¹⁹

There are at least two arguments for the necessity of this distinction in Polish. First, coordination of two singular neuter ([ $ \pi $ , $ \phi $ ]) items results either in closest conjunct agreement or in plural agreement, as shown in (34). By contrast, coordination of two infinitival phrases or two to _k s in the position of the subject of a lexical verb obligatorily triggers the so-called ‘default agreement’ (Dziwirek Reference Dziwirek, Dziwirek, Farrell and Mejías-Bikandi1990, Preminger Reference Preminger2014), i.e., a 3sg.n form, which signals the lack of an appropriate agreement controller; see (35)–(36). Note that coordinating two anaphoric uses of to _k is difficult, as it is then difficult to determine which to refers to which antecedent. However, it is easy to coordinate two deictic uses of to _k (see (36a)) or two CPs nominalised by to _k (see (36b)).Footnote ²⁰

(36)	a.	{Zaskoczyło	/ *Zaskoczyły}	mnie	to→	i	to→.
		surprised.3sg.n	/ surprised.3pl	I.acc	this	and	this
		‘This and this surprised me.’ (‘ $ \to $ ’ = pointing gesture)

	b.	Skandalem	{było	/ *były}	[to,	że	Polska	przegrała],	i
		scandal.sg.m.ins	was.sg.n	/ were.pl	this	comp	Poland	lost	and
		[to,	że	kibice	się	pobili].
		this	comp	fans	refl	brawled
		‘[That Poland lost]nominalised and [that fans got into a brawl]nominalised was a scandal.’

This syntactic difference between truly neuter nouns in (34) and $ \phi $ -less constituents in (35)–(36) is accounted for by assuming that infinitival phrases and the word to (whether to_n or to _k ) are not singular neuter, that is, they do not have even the minimal set of $ \phi $ features (the feature [ $ \phi $ ] itself), in contrast to true singular neuter (pro)nouns. Given this lack of $ \phi $ features on the conjuncts, the usual feature resolution responsible for the plural in (34) cannot take place, so not only particular conjuncts but the whole coordinations in (35)–(36) remain $ \phi $ -less.

The second phenomenon pointing to the same difference – between genuine third-person neuter ([ $ \pi $ , $ \phi $ ]) and $ \phi $ -less ([ $ \pi $ ]) items – is based on possible antecedents of pronouns. Polish personal pronouns used anaphorically must agree with their antecedents: they presuppose that their antecedents bear corresponding $ \phi $ features. In this context, note that the singular neuter pronoun ono cannot refer to an instance of to _k , as shown in (37).

The personal pronoun ono ‘pers.sg.n’ expects its antecedent to have the corresponding features (crucially, [ $ \phi $ ]), in the same way the pronoun on ‘pers.sg.m’ requires its antecedent to have the feature [masc]. The word to _k in A’s utterance lacks $ \phi $ , and thus ono cannot take it as an antecedent. The elsewhere form – to _k itself – is used in B’s utterance.

This behaviour of infinitival TPs and to _k may be contrasted with that of Polish gerund forms, which – just like infinitival TPs – refer to event types, but may be shown to have [ $ \pi $ , $ \phi $ ]. First, they can trigger plural agreement when coordinated; compare (38) below with (36) above. Second, they may be referred to with the pronoun ono ‘sg.n’; compare (39) with (37).

Another difference between $ \phi $ -bearing items and to is that to does not easily bind reflexive pronouns:

While (40a) is fully acceptable, (40b) is considerably worse. This pattern is expected if the reflexive requires a minimal set of $ \phi $ features on its binder: ono in (40a) bears the feature [ $ \phi $ ], while to in (40b) lacks it.Footnote ²¹

Let us finish this section with some speculations on the $ \phi $ -lessness of to_n and to _k . The existence of such $ \phi $ -less pronouns should not be particularly surprising. As argued above, personal (on-based) pronouns presuppose a specific set of $ \phi $ features on their antecedents – minimally the feature [ $ \phi $ ] (the singular neuter pronoun ono). However, $ \phi $ -less CPs and infinitival TPs also introduce (abstract) objects into the discourse. Accordingly, the existence of $ \phi $ -less pronouns is required to refer to such entities.

This is in line both with Déchaine & Wiltschko (Reference Déchaine and Wiltschko2002), where pronouns are argued to come in different sizes, and with Collins (Reference Collins2024), where implicit arguments (e.g., agents in short passives) are argued to be realised by three kinds of caseless pronouns in English – pro_gen (generic), pro_def (definite) and pro_un (existential) – the last of them also lacking $ \phi $ features. On this approach, the existence of featurally deficient pronouns ‘is an immediate consequence of a UG that allows for pronouns defined by various sets of features’ (Collins Reference Collins2024: 58). In fact, as shown in the table in (41), to _k fills a gap in the taxonomy of pronouns implicit in Collins (Reference Collins2024).

(41)

In summary, there are two variants of the Polish pronoun to: a maximal but $ \phi $ -less nominal projection (KP) and a reduced structure (nP). We consider this result to be in the spirit of Déchaine & Wiltschko (Reference Déchaine and Wiltschko2002), which proposes that ‘it is necessary to recognize (at least) three pronoun types: pro-DP, pro- $ \phi $ P, and pro-NP’ (p. 409). But while to _k might be an instance of Déchaine & Wiltschko’s (Reference Déchaine and Wiltschko2002) pro-DP (given that for them maximal nominal projections are DPs rather than KPs), to_n instantiates a new overt pronoun type, featurally similar to Collins’s (Reference Collins2024) phonetically empty pro_un.

In any case, regardless of its appropriate place in the taxonomy and typology of pronouns, to_n is not a nominative KP, so it is not an accessible agreement controller, which explains why agreement is controlled by the nominative KP predicate lower in the tree.

4.1. Intervention

The construction at hand does not allow topicalisation and wh-movement, as shown in (45)–(46). (The word to in parentheses in (45) is an optional topic marker, synchronically unrelated to the demonstrative; the two can co-occur, as in (45b).) In the (a) examples, the right-hand predicate is fronted over KP₁: in (45a), for information-structural purposes, and in (46a), as an instance of wh-movement. Both examples are clearly ungrammatical. Compare these with examples in (b), which differ from (a) only in lacking KP₁. Their full acceptability demonstrates that the intervention effect is triggered precisely by the presence of KP₁.Footnote ²³ To provide a complete picture, examples in (c) are canonical predicational constructions with instrumental predicates and regular subject-verb agreement. These again permit wh-movement and topicalisation, since the phrase that the predicate moves over is simply the Spec,TP subject.

Crucially, in studies specifically aiming at distinguishing between Spec,TP and positions above Spec,TP, intervening in A′-movement is regarded as an important diagnostic. For instance, Landau (Reference Landau2009), in refuting the ‘broad subject’ analysis of certain phrases in Hebrew – which posits two specifier positions in TP – appeals to this diagnostic and considers it ‘a lethal problem’ (Landau Reference Landau2009: 97) for a TP-internal analysis.Footnote ²⁴ The same reasoning also applies to the Polish construction at hand: KP₁ intervenes in A′-movement because it is base-generated in an A′ position itself.Footnote ²⁵

4.2. Refuting counterarguments

Works rejecting the dislocated analysis of Polish KP₁ refer to three phenomena suggesting a TP-internal status: (i) KP₁ may be realised as a quantifier phrase (Citko Reference Citko2008), (ii) KP₁ can control a reflexive pronoun and (iii) KP₁ can control the subject of an adversative adjunct clause (Bondaruk Reference Bondaruk2013). We argue that these diagnostics do not provide evidence for a TP-internal status of KP₁, because, as will be shown, (i) cross-linguistically quantifiers are attested in dislocated positions, (ii) Polish reflexive pronouns can be controlled from outside the TP in which they are placed and (iii) subjects of adversative adjunct clauses can be bound by items other than Spec,TP subjects.

Firstly, Citko (Reference Citko2008: 263, fn. 1) dismisses the dislocated analysis on the basis of the observation that quantificational constituents can occur as KP₁ (see (47)), although they are ‘impossible in left dislocated positions’ (see an attempted hanging topic construction in (48)):

However, quantificational phrases are attested in A′ positions, as in clitic left dislocation constructions best known from Romance languages. Consider the Italian example (49), which contains a left-peripheral phrase co-occurring with a coreferential clitic in the main clause. Similarly, (50) is a Lebanese Arabic sentence containing a left-peripheral quantificational phrase.

Hence, (48) is ungrammatical not because A′ constituents cannot be quantifiers, they can, but because hanging topics must be referential rather than quantificational. Crucially, being in a dislocated position is not equal to being a hanging topic. The latter are best analysed as syntactically unintegrated phrases, simply juxtaposed with the main clause and interpreted as connected to this clause only at the discourse level (Shaer & Frey Reference Shaer and Frey2004, Shaer Reference Shaer, Shaer, Cook, Frey and Maienborn2009, Villa-García Reference Villa-García2023). We argue that KP₁ occupies a dedicated position in the left periphery, not that it is a hanging topic.

Secondly, Bondaruk (Reference Bondaruk2013, Reference Bondaruk2015) shows that KP₁ can control subject-oriented reflexive pronouns, see (51).

Importantly, this cannot be attributed to to binding the reflexive and – being coreferential with KP₁ – acting as a proxy. As shown in Section 3, such constructions lacking KP₁ are degraded; see (40b), repeated as (52). Therefore, we must conclude that KP₁ directly binds the reflexive.

However, in Section 4.1, we presented syntactic evidence that the left-hand KP behaves as a dislocated phrase, in that it gives rise to intervention effects. How, then, can this be reconciled with the ability of KP₁ to bind reflexives?

We argue that TP does not constitute a locality domain for binding reflexives, and consequently, a TP-external KP₁ can bind a reflexive within that TP. Independent support for the availability of TP-external binders comes from constructions like (53). In this example, the reflexive can be bound not only by pro – controlled by the matrix object Piotrowi ‘Piotr.dat’ – but also by the matrix subject, which clearly lacks any syntactic representation within the infinitival TP.

The correct generalisation, then, is that the binder of the reflexive must have at least a minimal set of $ \phi $ -features (accounting for the degraded acceptability of (52)) and be located within the same CP – rather than the same TP – as the reflexive (which explains the availability of both interpretations in (53)).Footnote ²⁶

Thirdly, Bondaruk (Reference Bondaruk2013) argues that the dislocated analysis of KP₁ is wrong because KP₁ can control the subject of an adversative adjunct clause (aac) introduced by mimo ‘despite’:

The argument rests on the assumption that only Spec,TP subjects can control the pro in aac. This assumption, however, is immediately falsified by attested examples such as (55)–(56). In (55), the subject of the aac is controlled by a semantically prominent individual, not syntactically represented in the matrix clause at all. In (56), the controller likewise is not represented syntactically (if the precious metals were provided by, say, the ruler’s servants) or alternatively, it is realised as the dative complement of the impersonal form zwrócono ‘returned.imps’ (if the ruler himself supplied the metals).

In summary, the assumption that KP₁ is base-generated above the TP agreement domain immediately explains intervention effects. Together with the status of to_n discussed in the previous section, this assumption also explains why the nominative predicate is the highest accessible phrase and, hence, controls verbal agreement. Known arguments against this assumption either fail entirely (as in the case of quantificational phrases and adversative adjunct clauses) or can be reconciled with the TP-external analysis of KP₁ by granting it mixed A/A′ status or by adopting an analysis of binding insensitive to this distinction (as in the case of reflexive binding).

5.1. PredP(s) and nominative KP predicates

We adopt the currently standard view on the structure of copular clauses, which holds that there is a functional head mediating nonverbal predication, Pred (see, for instance, Bowers Reference Bowers1993, Reference Bowers, Baltin and Collins2001, Baker Reference Baker2003: Chapter 2, den Dikken Reference den Dikken2006 for general discussion and Citko Reference Citko2008, Bondaruk Reference Bondaruk2013 for its application to Polish copular clauses).Footnote ²⁹ Its maximal projection, PredP, is selected by the copula – whether it is a lexical verb (V), a light verb (v) or simply the Tense head. This last minimal approach, on which the copula (być ‘be’) is an exponent of T in the context of PredP, is also adopted in this work (following, for instance, Baker Reference Baker2003 and Roy Reference Roy2013). In information-structurally neutral contexts, the constituent closest to T, i.e., the subject of predication in Spec,PredP, moves to Spec,TP to satisfy epp. The schematic structure of such copular clauses is presented in (57).

(57)

The first question to be addressed is why predicates in constructions under investigation appear in the nominative case. From the perspective of Polish grammar, this is unexpected. Nominal predicates in canonical clauses – when the subject of predication is a nominative KP – typically take the instrumental case, as shown in (7), repeated below as (58).

Relatedly, why is it the case that non-canonical subjects occurring in the constructions analysed in this paper – PP[z] and to_n – cannot occur with instrumental predicates? It is straightforward to show that they cannot in the case of PP[z], see (59). As for to_n, see Section 3, where we show that the item occurring with instrumental predicates is to _k , i.e., a canonical KP subject, rather than to_n.

At its core, our proposal follows Citko (Reference Citko2008) and Bondaruk (Reference Bondaruk2013): we assume there are two Pred heads – a ‘canonical’ one, which assigns the instrumental case to its complement (Pred_INS), and a ‘defective’ one, which does not assign case (Pred $ {}_{\varnothing } $ ).Footnote ³⁰ The two differ with regard to selectional restrictions, as shown in (60). The canonical Pred_INS takes an instrumental predicate and combines with the same kind of specifiers as lexical verbs – typically KPs. By contrast, Pred $ {}_{\varnothing } $ is ‘defective’ in the sense that it places no selectional restrictions on its complement and specifier.Footnote ³¹

(60)	head	specifier	complement
	PredINS	KP*	KPINS
	Pred $ {}_{\varnothing } $	—	—

KP* in the table refers to ‘canonical’ subjects, i.e., ones that lexical verbs normally combine with. Besides nominative nominal phrases, these include infinitival phrases and CPs, as shown in the context of Pred_INS in (61).

(61)	a.	[Wprowadzić	ją	w	błąd] $ {}_{\mathrm{InfP}} $	było	rzeczą	bardzo	trudną.
		lead.inf	she.acc	in	mistake	was.sg.n	thing.ins	very	difficult
		‘To mislead her was a very difficult thing (to do).’ (NCP).
	b.	[Że	się	od razu	nie	pozabijali] $ {}_{\mathrm{CP}} $	było	zasługą
		comp	refl	immediately	neg	killed	was.sg.n	achievement.ins
		matki…
		mother.gen
		‘That they didn’t kill each other right away was thanks to the mother (lit. the mother’s achievement).’ (NCP)

Given that Pred $ {}_{\varnothing } $ has no restrictions on its complement and specifier, the next question is why only non-canonical subjects – PP[z] and to_n – can occur with it; for example, why is the Pred $ {}_{\varnothing } $ version of (58), with the nominative predicate, unacceptable with the canonical subject? We propose that this is straightforwardly explained in terms of Pāṇinian blocking: since KP subjects have the dedicated Pred_INS selecting for such items, Pred $ {}_{\varnothing } $ (the ‘elsewhere’ Pred) is blocked, as it is less specified for such subjects.

Interestingly, canonical subjects can occasionally occur with nominative predicates, but this structure is rare and requires an expressive predicate, as in (62)–(63).

Paradoxically, this phenomenon confirms the analysis in terms of Pred_INS and Pred $ {}_{\varnothing } $ . It has been reported that violating economy conditions does not necessarily lead to ungrammaticality if it induces a pragmatic effect (Schlenker Reference Schlenker2005, Patel-Grosz & Grosz Reference Patel-Grosz and Grosz2017). This is exactly the case here: using Pred $ {}_{\varnothing } $ with a nominative subject, instead of the better match, Pred_INS, is only licensed when the expression is intended to convey a special interpretive effect, namely, marking the speaker’s expressive attitude towards the assertion.

Importantly, (62) also shows the workings of agreement when two nominative phrases are present in the agreement domain: the highest one controls agreement, precisely as predicted by Bobaljik’s generalisation.

Before we proceed to the formal analysis of agreement, let us show how our minimal inventory of Preds interacts with control facts mentioned in Section 2; see (17), repeated as (64).

The predicate niezła paczka ‘a nice bunch’ is fine with both to and PP in the finite matrix structure (as illustrated for PP in (9)). However, a control structure cannot be formed. This follows naturally under the assumption that pro bears case (see Chomsky & Lasnik Reference Chomsky, Lasnik, Jacobs, von Stechow, Sternefeld and Vennemann1993, Sigurðsson Reference Sigurðsson2008). Given that case-bearing items have a more specialised Pred available, the defective Pred $ {}_{\varnothing } $ is blocked.

Crucially, as shown in (65), strongly expressive nominative predicates can occur in control structures. This directly follows from the analysis proposed for constructions like (62)–(63): once the predicate is sufficiently expressive, a violation of the economy conditions described above is pragmatically licensed. As a result, pro can occur with Pred $ {}_{\varnothing } $ despite the availability of the more specified alternative, Pred_INS.

Having laid out our assumptions regarding the basic structure of copular clauses, we now turn to the formal implementation of the agreement mechanism.

5.2. Agreement

Bobaljik (Reference Bobaljik2008) does not provide a formal derivational analysis of case-dependent Agree. To implement Bobaljik’s generalisation formally, we require a mechanism involving (i) case-discriminating agreement probes which (ii) copy the $ \phi $ features of their goal. Under a certain set of assumptions, these can be achieved by any model of Agree, including the classic formulation in Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky, Kenstowicz and Hale2001). A central feature of Chomsky’s approach, however, is the claim that agreement and case-valuation proceed in tandem, and the eventually nominative phrase must be ‘active’ – bear an unvalued case (i.e., the [ucase: ] feature) – in order to participate in agreement. This assumption, in turn, runs counter to one of the core insights of Bobaljik (Reference Bobaljik2008): that agreement is dependent on case (see also Preminger Reference Preminger2014).Footnote ³⁵

For concreteness, we adopt the interaction/satisfaction model of Agree (Deal Reference Deal, Bui and Özyldz2015, Reference Deal2024, Clem & Deal Reference Clem and Deal2026, a.o.), as it allows us to present the analysis in the most perspicuous way. We modify this approach in a way directly implementing case-dependent Agree. As specified below, the agreement probe postulated here searches for the nominative case. In this respect, we depart from Clem & Deal (Reference Clem and Deal2026), where case marking is treated as a by-product of agreement.

In the interaction/satisfaction model, agreement probes are specified by two parameters: interaction (which features the probe interacts with) and satisfaction (which features, if any, cause the probe to stop probing further). We propose that Polish T both interacts with and is satisfied by nominative KPs, thus implementing Bobaljik’s generalisation, rather than by some set of $ \phi $ features, as is more frequently assumed. That is, the probe on T is specified as [int: nom, sat: nom].

Given this, we adopt the ‘coarse’ approach to feature copying (see, for instance, Grishin Reference Grishin2024, Deal & Royer Reference Deal and Royer2025), which assumes that ‘all features of the goal are copied onto the probe, not just those that entail the interaction feature’ (Grishin Reference Grishin2024: 19).Footnote ³⁶ That is, by saying that the probe T interacts with nom, we imply that it copies all the features present on a phrase bearing nom, including the $ \phi $ features, which results in observed agreement.

Let us illustrate this analysis with a structure involving to_n (but not KP₁), such as (4), repeated again as (66).

The right-hand KP predicate occurs in the nominative case and obligatorily controls agreement. The word to is a reduced demonstrative lacking a case projection, which makes it invisible to the probe, see (67).Footnote ³⁷ The word to moves to Spec,TP for epp reasons (crucially, detached from agreement, as is argued for Slavic in, for instance, Lavine & Freidin Reference Lavine and Freidin2002, Bailyn Reference Bailyn2004).

(67)

Constructions with PP subjects of predication are fully analogous, with PP[z] in place of to_n in (67) (cf. (71) in Section 5.4).

Moreover, (67) also illustrates how agreement operates in structures such as (5), repeated below as (68). KP₁ is base-generated above TP (and, hence, not shown in (67)) and is therefore fully irrelevant for the T probe.

Note that under the view that PredP is a phase (see Citko Reference Citko2014: §5.1), a locality issue arises, since the probe would have to enter the phase, thereby violating the Phase Impenetrability Constraint (PIC; Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000). However, issues concerning the interaction between agreement and PIC are not unique to our analysis and have prompted a range of proposals addressing how such configurations should be handled. One possibility is to weaken the PIC, as proposed in Chomsky (Reference Chomsky, Kenstowicz and Hale2001), who argues that a phase is sent to Spell-Out only after another phase head has been merged. On this view, the structures analysed here do not involve a PIC violation: T probes into PredP immediately, before the next phase head, C, is merged. Other possibilities are that agreement is not restricted by phases (only movement is), as argued in Bošković (Reference Bošković2007), or that Pred (at least Pred $ {}_{\varnothing } $ ) is a ‘defective’ phase head – i.e., one lacking a full set of uninterpretable $ \phi $ features – and is therefore transparent for agreement (as proposed in Richards Reference Richards, Alexiadou, Kiss and Müller2012: 151 to account for apparent PIC-violating agreement).

Due to space constraints, we do not attempt to adjudicate between these approaches here, focusing instead on additional data in the following two subsections that confirm the analysis postulated in this paper.

5.3. KP₁ outside agreement domain

Given the interaction/satisfaction model, especially as presented in Clem & Deal (Reference Clem and Deal2026), one could claim that KP₁ is within the agreement domain after all, but the ‘insatiable’ probe interacts with both KP₁ and the KP₂ predicate, and only expones the features of the last – ‘outermost’ (Grishin Reference Grishin2024, Przepiórkowski Reference Przepiórkowski2025) – item it interacts with, i.e., the features of KP₂. This would be compatible with an analysis on which KP₁ is base-generated within TP, but not with the analysis proposed above. Dually, if it could be shown that the probe does not interact with KP₁ at all, this would be a confirmation of the analysis proposed here.

In this subsection, we demonstrate that KP₁ indeed does not interact with the probe: KP₁’s features fail to be exponed on the copula even when it is the only nominative phrase equipped with $ \phi $ features in the sentence. Consider (69), where the predicate is a $ \phi $ -less adverbial phrase.Footnote ³⁸

There are no nominative KPs in these sentences other than KP₁,Footnote ³⁹ so – on the ‘insatiable’ Agree account sketched above – KP₁s should control agreement here. But it is clear that they do not: if they did, the copula would expone the feminine gender, which it does not. Instead, only the singular neuter form of the verb is allowed here, resulting from ‘default agreement’ (Dziwirek Reference Dziwirek, Dziwirek, Farrell and Mejías-Bikandi1990, Preminger Reference Preminger2014), i.e., the lack of an actual agreement controller. This confirms that KP₁ is base-generated outside the agreement domain.

5.4. Long-distance agreement

Consider (70), with copular constructions embedded under raising verbs.

In these sentences, the finite modal verb expresses the features of the only nominative phrase in (70a) (canonical copular clause) and (70b) (non-canonical copular clause with a prepositional subject), or of the right-hand nominative phrase in (70c) (non-canonical copular clause with to_n and KP₁), as might be expected. However, unlike all previous examples, (70b) and (70c) are instances of long-distance agreement, a phenomenon in which a matrix verb exhibits agreement with an argument of an embedded verb. The literature related to such phenomena is abundant,Footnote ⁴⁰ with various theoretical proposals: from those arguing that such phenomena involve a sequence of cyclic, strictly local agreement processes, with the effect that the non-locality is only apparent (e.g. Haug & Nikitina Reference Haug and Nikitina2016), to those which suggest that agreement is not limited by virtually any locality constraints (Bošković Reference Bošković2007). Although a detailed analysis of how the phenomenon presented in (70b)–(70c) can be understood within the broader theoretical research on long-distance agreement is beyond the scope of the paper, we would like to note that the account offered above captures these patterns immediately.

From the perspective of Bobaljik’s generalisation adopted here, there is nothing special about agreement in (70b)–(70c): the highest accessible phrase controls agreement with the copula, see (71) corresponding to (70b) above.Footnote ⁴¹ Once the embedded TP is built and selected by the raising modal verb, the main sentential T merges and probes for a nominative KP, finding it at the very bottom of the tree.

(71)

As expected, control verbs – such as chcieć ‘want’ – cannot occur with such non-canonical structures, although they can occur with a canonical copular clause with an instrumental predicate:

The crucial factor here is that control verbs project their own subject (in Spec,vP), which – as in the case of all other lexical verbs – cannot be ‘non-canonical’ (nP, PP[z]). As postulated in Section 5.1, lexical verbs select for KP $ {}^{\ast } $ – the shorthand for ‘canonical’ subjects. By contrast, raising verbs do not project their Spec,vP and raise their (possibly non-canonical) subjects from the embedded TP.

Another relevant piece of evidence pointing in the same direction is related to available interpretations of modal verbs. Note that the canonical copular sentences embedded under móc ‘can’ allow for both epistemic and non-epistemic interpretations (cf. Kratzer Reference Kratzer, von Stechow and Wunderlich1991). That is, on the one hand, (73a) can be interpreted so that Janek has the ability or is permitted to be ‘a terrible pig’ (non-epistemic). On the other, it can be interpreted epistemically, i.e., as saying that Janek being ‘a terrible pig’ is compatible with the evidence. In contrast, the non-canonically agreeing sentence (73b) can only be interpreted epistemically: the sentence simply asserts that it is compatible with the evidence that Janek was a terrible pig. For example, it cannot be interpreted deontically, as saying that somebody permitted Janek to misbehave. (Similar considerations apply to non-canonical copular constructions with to_n.) This is in line with the common view that epistemic modals are expressed by raising verbs, while non-epistemic modals are typically expressed by control verbs (e.g., Jackendoff Reference Jackendoff1972, Kratzer Reference Kratzer, von Stechow and Wunderlich1991, Brennan Reference Brennan1993).

In summary, when non-canonical copular constructions discussed in this paper are complements of epistemic raising verbs, the nominative KP predicate still controls agreement, but this time it is long-distance agreement between a verbal head (the epistemic modal) and a phrase embedded within its (infinitival) TP complement. That is, what the finite verb agrees with in such constructions cannot reasonably be analysed as this verb’s subject.

This is apparently the first report of long-distance agreement in Slavic in the literature; it strikingly confirms the view of agreement expressed in Bobaljik’s generalisation (6) and disconfirms the view common in traditional Slavic linguistics that verbs only agree with their subjects.Footnote ⁴²