1. Introduction
This article addresses an argument, made by Chomsky (Reference Chomsky2013) and widely echoed, that DP raising to Spec-T, which in earlier versions of Minimalism was motivated by a property of T called “EPP”, could be forced by independently motivated labelling considerations: a DP raises not because it is attracted by a property of T (in what Moro and Roberts Reference Moro and Roberts2023 call a pull chain) but because an unlabellable symmetry point must be dissolved (a push chain in Moro and Roberts’s terms).Footnote 1I will evaluate this argument in detail, and argue that Spanish copular constructions pose a challenge to the reduction of the classical EPP effects to labelling. In this context, we need to define two theoretical notions: what the EPP is and how labelling works. We start by defining the EPP in current generative theory.
Classically, the EPP would be defined as the requirement that Spec-IP/TP be filled, implemented, for example by means of a D feature in T (Chomsky Reference Chomsky2015a). Chomsky (Reference Chomsky1982: 10) defines EPP as a requirement that “clauses must have subjects”, but the notion of “subject” in transformational grammar, to the extent to which it has any role in the theory, is clearly configurational (it has been since Aspects), as opposed to a primitive as in Relational Grammar or Lexical Functional Grammar (see also Chomsky Reference Chomsky1981). The Spec-TP position can be occupied either by movement (i.e., Internal Merge, IM) or by lexical insertion (e.g., External Merge, EM of an expletive).
In contemporary Minimalist theory, since structure building has been reduced to unordered set formation without projection, there is stricto sensu no such thing as Spec-TP (or, more generally, no X-bar theoretic specifiers, heads, or complements) in any formally meaningful way, and thus a redefinition is necessary. The EPP has always been strange: we start from a desired output configuration, and reverse engineer a way to get to it. The motivations for such output configuration will not be reviewed here; I will assume for the sake of the argument that there are independent reasons to want a DP in Spec-T (using traditional X-bar theoretic terms for expository purposes only), including, for example, constituent order (see section 3). Needless to say, if these reasons were not valid, the whole discussion would become moot.
At the core of current Minimalist theory is the operation Merge, which, applied to syntactic objects X and Y in a workspace WS, delivers the set containing X and Y, removing the elements X and Y from WS and leaving only the set {X, Y} accessible for further operations (Collins Reference Collins, Bauke and Blümel2017; Chomsky Reference Chomsky2020a, Reference Chomsky, Franco and Lorusso2020b, Reference Chomsky2021; Epstein et al. Reference Samuel, Kitahara, Seely, Epstein, Kitahara and Seely2015a, Reference Samuel, Kitahara, Seely, Epstein, Kitahara and Seely2015b, Reference Samuel, Kitahara and Seely2022; Chomsky et al. Reference Chomsky, Seely, Berwick, Sandiway Fong, Kitahara, McInnerney and Sugimoto2023):
Merge(X, Y) = {X, Y} (“Simplest Merge” henceforth)
Regardless of whether “removal” of X and Y is a separate operation (Chomsky Reference Chomsky2019) or whether its effects are delivered by a condition Minimal Yield (Chomsky Reference Chomsky2021), the essence of structure building in contemporary Minimalism is free (untriggered, unmotivated) unordered set formation. All syntactic conditions, therefore, must be formulated exclusively in terms of possible and impossible set-theoretic objects and mappings between these.
The article is structured as follows: Section 1.1 defines the properties of Simplest Merge and of the objects it generates. Sections 1.2 and 1.3 are devoted to a discussion of EPP and labelling, respectively, assuming Simplest Merge. Section 2 presents the empirical data used to ground the theoretical discussion: Spanish copular constructions, with adjectival and nominal predicates. I conclude that the data analyzed here argues against a reduction of classical EPP effects to labelling, contra Chomsky (Reference Chomsky2013 et seq.) and much related work.
1.1 Structure building: on “Simplest Merge”
Following recent Minimalist literature, this article defines Simplest Merge as an operation of set formation. Two additional considerations become relevant at this point. First, strictly speaking, even though Merge always delivers sets, these are not always unordered. Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Belletti2004, Reference Chomsky2020a, Reference Chomsky2021), Langendoen (Reference Langendoen, Carnie, Willie and Harley2003), Epstein et al. (Reference Samuel, Kitahara and Seely2022), and others assume in addition to Simplest Merge (also known as Set Merge) an operation, Pair-Merge, which “adjoins α to β to form <α, β>” (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000: 133). Despite <α, β>, as an ordered pair, being set-theoretically equivalent to {{α}, {α, β}} (Krivine Reference Krivine1971: 3), Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000: 133) says “[…] it is natural to conclude that the adjoined element α leaves the category type unchanged: the target β projects.”Footnote 2It seems more natural to see this object as {β, {α, β}} (where β is the label) than as {α, {α, β}} (where α is the label), an observation also made by Langendoen (Reference Langendoen, Carnie, Willie and Harley2003). This ordered-set-formation operation has been invoked by Chomsky in the analysis of adjuncts, the introduction of which does not change the category of the input to adjunction (see also Dowty Reference Dowty, Lang, Maienborn and Fabricius-Hansen2003).Footnote 3Specifically, Pair-Merge has been mentioned by Chomsky in connection to experiencers in raising constructions (Chomsky Reference Chomsky2021: 35), adjectival modification in NPs (2020a: 49–50), and “adjuncts” more generally (Reference Chomsky and Belletti2004: 117), as well as maintaining that at CI it is supposed to deliver “predicate composition” (Reference Chomsky and Belletti2004: 118). Pair-Merge would involve syntactic objects derived in parallel (assuming some kind of multidimensional workspace or set thereof, cf. Chomsky Reference Chomsky2020a, Reference Chomsky2021), and after its application the adjoined object becomes invisible for both labelling and sub-extraction (Chomsky Reference Chomsky2020a).
The second point pertains to the properties of the syntactic objects being Merged: while Chomsky and others emphasize that Merge is essentially Form Set (Collins Reference Collins, Bauke and Blümel2017, Epstein et al. Reference Samuel, Kitahara and Seely2022, Chomsky et al. Reference Chomsky, Seely, Berwick, Sandiway Fong, Kitahara, McInnerney and Sugimoto2023; also Svenonius Reference Svenonius, Allott, Lohndal and Rey2021, with some reservations), the specific axiomatization of set theory that underpins Merge is not always clear (see also Gärtner Reference Gärtner2022). In their discussion of chain formation, Chomsky et al. (Reference Chomsky2019: 246) state that a separate operation COPY is unnecessary given IM on “standard set-theoretic assumptions”, and Svenonius (Reference Svenonius, Allott, Lohndal and Rey2021: 143) cites Russell and Whitehead’s Principia Mathematica as a source of the idea that “sets are mathematically fundamental”. However, even accepting Svenonius’s appeal to the “fundamental” character of sets as an argument in favour of sets being an appropriate formalization for the theory of syntax, which specific version of axiomatic set theory is assumed does make a difference. For a theory of natural language syntax, I contend, the set-theoretic underpinnings of structure building must allow for urelements: objects which are not sets but which may be members of a set (see e.g. Barwise Reference Barwise1975). Importantly, perhaps the most widely used version of axiomatic set theory, Zermelo-Fraenkel set theory, has sets as primitives, with no urelements (Krivine Reference Krivine1971). Syntactic urelements can be features, morphemes and roots, or lexical items (depending on the level of granularity adopted: lexicalist theories and strongly decompositional theories differ greatly in this point; cf., e.g., Williams Reference Williams2011 and Borer Reference Borer2005 respectively), but syntactic theory needs some irreducible non-set object to work with. This is a somewhat controversial point that deserves some commentary, as it will become relevant. For example, Chomsky (Reference Chomsky2020a: 37) says:
We want to say that [X], the workspace which is a set containing X is distinct from X.
[X] ≠ X
We don’t want to identify a singleton set with its member. If we did, the workspace itself would be accessible to MERGE. However, in the case of the elements produced by MERGE, we want to say the opposite.
{X} = X Footnote 4
We want to identify singleton sets with their members.
This may lead to contradictions, depending on the set-theoretic framework adopted (see Krivochen Reference Krivochen2023 for further discussion). For example, in Zermelo-Fraenkel set theory (ZF henceforth) {Ø} is a subset of every set (Kuratowski and Mostowski Reference Kuratowski and Mostowski1968: 9), but Ø is not an element of every set (importantly, Ø ≠ {Ø}: informally, the empty set contains “nothing”, but is not itself nothing). Note, also, that in the Minimalist literature Merge applied to X and Y is notated Merge(X, Y), not Merge({X}, {Y}) (Chomsky Reference Chomsky2020a, Reference Chomsky, Franco and Lorusso2020b, Reference Chomsky2021; Collins Reference Collins, Bauke and Blümel2017; Epstein et al. Reference Samuel, Kitahara, Seely, Epstein, Kitahara and Seely2015a, Reference Samuel, Kitahara, Seely, Epstein, Kitahara and Seely2015b, Reference Samuel, Kitahara and Seely2022; Komachi et al. Reference Komachi, Hitakara, Uchibori and Takita2019). Huijbregts (Reference Huijbregts2019: 1) addresses the element-singleton distinction in the following terms:
Standardly, the elements P or Q in Merge(P, Q, WS) are taken to be single syntactic objects (singletons) or terms thereof. Let’s call this the “special theory of merge” (STM). We propose to generalize merge as an operation on WS that is free to select sets as opposed to SOs. The result is a “general theory of merge” (GTM), with special Merge constituting just the limiting case of general Merge where singletons rather than nonunit sets are being selected
The use of “standardly” is somewhat surprising, given the lack of precision with respect to which version of axiomatic set theory is being assumed. In well-founded set theory (i.e., axiomatizations that include the axiom of regularity: every non-empty set contains an element that is disjoint from it), sets that satisfy the equation x = {x} (so-called “Quine atoms”) are excluded, since if x = {x} there is no element in {x} that is disjoint from it. Standardly, ZF includes the axiom of regularity (e.g., Krivine Reference Krivine1971, Kunen Reference Kunen2013). Chomsky et al. (Reference Chomsky, Seely, Berwick, Sandiway Fong, Kitahara, McInnerney and Sugimoto2023: 56) do seem to assume lexical items as urelements:
Consider now the simplest Merge representation adopted by PoP for the VP, namely {V, NP}. The information that it’s a phrase is already (and inherently) encoded by the set brackets {…}. Footnote 5It’s a “phrase” because it’s a set (i.e., it’s not a lexical item).
This passage entails that lexical items are not sets; thus, they are urelements (and set brackets are perhaps part of the grammar’s alphabet). If so, then the axiomatization of set theory assumed in that work cannot be standard ZF, but some variant of it (possibly along Barwise’s lines, a version of Kripke-Platek theory with urelements). In the discussion that follows, I will consider both options in the previously quoted passage from Chomsky (Reference Chomsky2020a) as alternatives: lexical items as urelements (x ≠ {x}) and as Quine atoms (x = {x}).
On a related note, observing a problem that emerges with a strict Bare Phrase Structure implementation of Labelling (where labels are set members created as by-products of Merge, cf. Chomsky Reference Chomsky, Campos and Kempchinsky1994, Epstein Reference Epstein and Epstein2000, Kobele Reference Kobele2023), Rizzi (Reference Rizzi, Domenico, Hamann and Matteini2015: 20) says:
[…] elements drawn from the lexicon bear a [+lex] feature; syntactic objects created by merge normally do not inherit this feature, so heads are objects marked as [+lex] and phrases are not […]. Heads, i.e. elements marked [+lex], but not phrases, are taken into account by the labelling algorithm. Here, I will put aside the problem […]
This feature-based approach to the head-phrase distinction seems to us not only unnecessarily stipulative, but also directly contradictory to how heads are identified under Minimal Search in Chomsky (Reference Chomsky2013, Reference Chomsky, Domenico, Hamann and Matteini2015b, Reference Chomsky2020a, Reference Chomsky, Franco and Lorusso2020b, Reference Chomsky2021) and the aforementioned references. We will thus not consider this option in our discussion.
The formulation of Merge makes no reference to the source of the SO being operated over. Internal and External Merge differ only in terms of whether both SOs involved in the operation are terms of the structure already built or not. Aside from the first step in a derivation (so-called First Merge), in which both SOs are taken from the Lexicon, with EM we introduce a new element to the already existing structure. With IM, on the other hand, we copy and reintroduce a term that belongs to the structure already in the workspace. The strong commitment to unlabelled, unordered sets created by means of an unrestricted operation entails a departure not only from X-bar theory, but also from certain forms of Bare Phrase Structure (e.g., Castillo and Uriagereka Reference Uriagereka and Uriagereka2002; also to an extent Collins and Kayne Reference Collins and Kayne2023). Indeed, Chomsky (Reference Chomsky, Campos and Kempchinsky1994: 63) says
The operation Merge, then, is asymmetric, projecting one of the objects to which it applies.
Current Simplest Merge, in contrast, is presented as fundamentally symmetric (as “Form Set”) and independent from the satisfaction of selectional properties, orthogonal to the representation of predicate-argument relations, which are fundamentally asymmetric (see e.g. Chomsky Reference Chomsky2020a). Pair-Merge makes a return, again to account for what traditionally would have been analyzed using Chomsky-adjunction: non-selected modifiers and adjuncts.
1.2 Defining the EPP under Simplest Merge
Given the abandonment of traditional phrase structure in favour of unordered sets and the fact that grammatical functions play no role in the theory, attempting to define the EPP as a requirement that “specifier” positions be filled or that clauses have “subjects” does not seem appropriate. Consider the core EPP case: Spec-T must be filled. Initially, then, we may want to formulate the EPP under Simplest Merge along the following lines:
EPP (first stab): a syntactic object SO must (Internally/Externally) Merge with a set whose outermost member is T (i.e., a set S such that there is no S’ that properly contains S and T is an element of S, and for any S” such that S” ⊆ S, T does not belong to S”).Footnote 6
A version of this formulation is perhaps that of Epstein, Kitahara, Obata and Seely (Reference Samuel, Kitahara, Obata, Seely, Epstein, Kitahara and Seely2015: 45), where the EPP feature of T requires “[…] that something be overtly merged with the projection headed by T”.Footnote 7Under standard assumptions, however, this “something” must be somehow restricted: for example,
is not a classical EPP-abiding configuration under standard assumptions: we cannot just Merge an overt complementizer, or an adjective, or a preposition, or an adverb. In order to capture the classical EPP configuration, where a DP raises from its VP-internal position to Merge with a segment of T, we may require that the SO that merges with a set whose outermost member is T be a SO endowed with phi-features:
EPP (second stab): a syntactic object SO endowed with phi-features must (Internally/Externally) Merge with a set whose outermost member is T.
Although an advancement over the first formulation, this will not work either: in addition to generating (1), it would also allow for a configuration such as (2):
This is not the kind of output we want. Both C and v, qua phase heads, would be endowed with phi-features, which in Chomsky (Reference Chomsky, Freidin and Carlos Otero2008) are passed on to the non-phase head they take as a complement (i.e., to/with which they Merge) via Feature Inheritance. A reviewer observes that “the irreflexivity of set membership” can be used to filter out (1) and (2), forcing the element that merges with T to be “phrasal”. The accuracy of this observation is unclear, particularly given that whether terminals are Quine atoms or not is a point where Chomsky and others have been inconsistent: in non-well-founded set theory, set membership could be defined as reflexive. But even if we accepted this argument, (2) could be replaced by something like (3), where the equivalent to a traditional VP has raised (especially under free, unconstrained Simplest Merge):
This, again, is not an output we would want, certainly not for purposes of classical EPP satisfaction, our focus in this article. The same argument, so far as we can see, applies to the comment from another reviewer, based on Chomsky (Reference Chomsky, Domenico, Hamann and Matteini2015b: 9):
In languages where T is weak [i.e., non-null-subject languages], and therefore cannot label the projection resulting from Merge with its complement (whatever that is), T must acquire a specifier in order to bring about labeling.
In (3), T merges with a “specifier”, and thus should be labellable (assuming some notion of T “strength”). However, it is not a legitimate syntactic object in English. A proponent of free, unconstrained Merge would not worry about this, but given the explosion of unusable derivations – which are undesirable from a computational viewpoint but also a problem from the perspective of “psycholinguistic responsibility” (Pollard 1997: 11) – and the fact that interpretable objects would be arrived at by pure luck, I prefer a more classical constructivist system (in the sense of Lasnik and Uriagereka Reference Lasnik and Uriagereka2005), as work in Minimalist Grammars does.
Going back to (2), the phi-features in C and v are unvalued and possibly uninterpretable: we can take this to be the source of the problems with (1) and (2). Consider now a particularly relevant case for us: adjectival predicative copular constructions, whose derivation involves – as we will see in detail – a {DP, AP} “small clause”. In these cases, only DP can raise, not AP (Moro and Roberts Reference Moro and Roberts2023: 4, fn. 2 and references therein). This allows us to refine the definition, making explicit reference to interpretability: phi-features are interpretable in DPs, but not in phase heads. Additionally, they enter the derivation valued in the nominal head within DPs, but not on APs (such that an adjective agrees with the nominal it modifies, not the other way around). The revised formulation of the “classical” EPP that results from these considerations is:
EPP (final version): a syntactic object endowed with valued interpretable phi-features must Merge with a set whose outermost member is T.
This definition captures the workings of the EPP in Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001) and much related work: T’s uninterpretable features probe downwards and find a DP in the vP/VP domain, the unvalued phi-features of T get valued under Agree, as a by-product the Case feature of DP is valued Nominative, and Internal Merge of DP satisfies the EPP feature in T.Footnote 8This final revision is also consistent with Chomsky’s (Reference Chomsky, Franco and Lorusso2020b: 164) informal characterization of the notion Spec(ifier): the phase heads C and v (the loci of uninterpretable features) motivate Internal Merge of a DP to the Spec position of their non-phasal complements (T for C, delivering classic EPP effects, and V for v, for raising to object),
Using the conventional term SPEC informally, meaning sister-of XP or of the head X of XP = {X, YP}.
Despite there being no relation of “sisterhood” under set-theoretic Merge, co-membership should do the trick, as in Chomsky et al. (Reference Chomsky, Seely, Berwick, Sandiway Fong, Kitahara, McInnerney and Sugimoto2023: 38):
“Spec-of-R” would be the co-set-member of {R, IA} [R = Root, IA = Internal Argument]
That is, Spec-R would be the syntactic object SO in the configuration {SO, {R, IA}}. This is not entirely unproblematic: assuming unordered sets and symmetric structure building, it is not entirely clear what makes Spec-of an asymmetric relation. Given {{X, YP}, {Z, WP}}, is {X, YP} Spec-of-Z or is {Z, WP} is Spec-of-X? If {X, YP} and {Z, WP} (YP and WP are used only as proxies for sets, as in Seely Reference Seely and Boeckx2006 and Chomsky Reference Chomsky, Domenico, Hamann and Matteini2015b) are built in parallel in a strictly Markovian projection-less system (where there is no record of the derivation, and thus we cannot tell in which order terms have been Merged), there is no way to decide. Indeed, labelling via shared features dissolves the notion of “specifier”, as does the requirement that either SO in a point of symmetry moves (what Moro and Roberts Reference Moro and Roberts2023 call a push-chain), since Spec-of is a representational notion, not a derivational one (Stroik and Putnam Reference Thomas and Putnam2011).
For so-called “generalised EPP” effects, we can, in principle, just replace T in our definition above by any head with some property p (e.g., be a phase head, cf. Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000), and the valuation-interpretability condition with some other property in the relevant phrasal SO: this could give us successive cyclic DP movement through phase edges. Further refinements are of course possible: we may want the SO that is Merged with the set {T, {…}} to be an SO that phi-agrees with T, not just a SO with interpretable phi-features (see Chomsky Reference Chomsky, Franco and Lorusso2020b and Rouveret Reference Rouveret2024 for discussion). Under these assumptions, however, problems arise in a number of cases. Of these, consider for instance expletive there: following Chomsky (Reference Chomsky and Kenstowicz2001: 7), expletive there is phi-defective, having only a lexically valued [3 person] feature but no [Number]. If we require phi-completeness as a prerequisite for labelling via Agree (see Saito Reference Saito2016 for a related view), there is a difficulty: it would be impossible to label {there, {T, …}}, as there is not phi-complete. If this requirement is relaxed, presumably <Person, Person> could do the trick (at the cost of setting there-clauses apart from <phi, phi> clauses, labelling-wise, with unclear effects at the CI interface).
Clausal “subjects” are equally problematic, as it has been proposed that CPs do not have phi-features that T can agree with (see e.g. Picallo Reference Picallo2001 for discussion). However, Davies and Dubinsky (Reference Davies, Dubinsky, Davies and Dubinsky2001), among others, propose that clausal subjects are CPs embedded within null DPs, where the null D bears (valued, interpretable) phi-features. In this case, the remarks about DPs above can be directly applied to clausal subjects. Extensions to the criterial system are also possible for <Q, Q>, <Top, Top>, and <Foc, Foc> labelling, although I will not address them in this article (other than some remarks on <Foc, Foc> in section 2.2). Alternatively, under free Merge, we may drop all requirements other than “something must Merge” with a set with outermost member T and let the interfaces do the work.Footnote 9In that case, however, the relevant output condition(s) that is/are violated in cases such as (1) and (2) must be made explicit.
1.3 Labelling under Simplest Merge
In the Introduction, I mentioned that much recent literature has attempted to eliminate the EPP, expressing EPP effects in terms of solutions to labelling failures: something moves to position P not because of an EPP feature attracting it to P, but due to the need to break an unlabellable configuration somewhere else in the structure. In this article, I will deal with labelling problems in Spanish copular constructions as an empirical point to anchor the theoretical discussion. The labelling account of EPP effects in copular constructions is, roughly, as follows (based on Chomsky Reference Chomsky2013, Reference Chomsky2015; van Gelderen Reference van Gelderen2022; Roy Reference Roy2023). Consider first the formulation of the so-called “Labelling Algorithm” (LA) in Chomsky (Reference Chomsky2013: 43):
Suppose SO = {H, XP}, H a head and XP not a head. Then LA will select H as the label, and the usual procedures of interpretation at the interfaces can proceed. The interesting case is SO = {XP, YP}, neither a head (we return to the only other possibility, {H, H}). Here minimal search is ambiguous, locating the heads X, Y of XP, YP, respectively. There are, then, two ways in which SO can be labeled: (A) modify SO so that there is only one visible head, or (B) X and Y are identical in a relevant respect, providing the same label, which can be taken as the label of the SO.
Chomsky follows Moro’s (Reference Moro2000) analysis of copulas, and takes copular constructions to be structures of the form
Because XP-YP is ambiguous for Minimal Search (but see Krivochen Reference Krivochen2023 for arguments against this position), one of the phrases must move. Given that copies are invisible for the LA, whichever phrase remains in the “small-clause-like” structure will provide the label:
If the copy of XP (indicated by < >) is invisible for labelling, as it is “part of a discontinuous element” (Chomsky Reference Chomsky2013: 44) (also a set-theoretically problematic claim, see Gärtner Reference Gärtner2022), α will receive the label of YP. Having XP raise from {XP, YP} would be motivated by the need to generate a labellable object, and not to satisfy an EPP requirement in some higher head (e.g., T). Essentially the same argument is put forth in Moro and Roberts (Reference Moro and Roberts2023), with more emphasis on the theory of Dynamic Antisymmetry (DA) and thus on the dissolution of structures that are neither linearisable by Kayne’s (Reference Kayne1994) Linear Correspondence Axiom (LCA) nor unlabellable as per Chomsky’s LA. As Moro and Roberts (Reference Chomsky, Seely, Berwick, Sandiway Fong, Kitahara, McInnerney and Sugimoto2023: 5) observe,
for any given POS [Point Of Symmetry] resulting by merging two maximal projections, movement of either XP simultaneously solves both the LCA and the LA related problems
For the purpose of this article, I will focus on the LA argument against EPP, leaving matters of linearisation aside and noting that my arguments against the claim that EPP can be eliminated in favour of LA-based push-chain mechanisms do not automatically extend to the DA analysis.
Of particular relevance to this article is the labelling case Chomsky (Reference Chomsky2013) identifies as (B): the heads X and Y of XP and YP are “identical in a relevant respect”. Suppose that our XP and YP are an argumental DP (a shorthand for a set with outermost D) and a set with outermost T, respectively. Then,
T is similar to roots: T [in English] is too “weak” to serve as a label. With overt subject, the SPEC-TP construction is labeled 
$\langle$ϕ, ϕ 
$\rangle$ by the agreeing features. Therefore, English satisfies EPP. (Chomsky Reference Chomsky, Domenico, Hamann and Matteini2015b: 9)
In other words: because there is phi-agreement between the outermost heads D and T of the aforementioned sets, the output of Merge is labelled 
$\langle$ϕ, ϕ
$\rangle$. Phi-agreeing heads provide a label.
In this article, I call into question the argument that EPP effects in copular sentences can be captured by labelling only (and that EPP can thus be eliminated as redundant). A cautionary remark is at this point necessary: this work is not in and of itself a defence of an EPP feature, nor an account of EPP effects more generally (which – as a reviewer observes – include, for instance, V2 and “residual” V2 phenomena; see e.g. Roberts Reference Roberts and Rizzi2004) and how they may or may not be subsumed to other mechanisms (e.g., V raising; see Alexiadou and Anagnostopoulou Reference Alexiadou and Anagnostopoulou1998 as well as Rouveret Reference Rouveret and Rouveret2018 and references cited there for discussion).Footnote 10It provides an argument against the elimination of EPP in terms of labelling focused on a specific case: Spanish copular constructions. I will evaluate different configurations and consider alternative analyses, and argue that in all cases the “DP raising due to labelling failure” approach makes the wrong predictions. I conclude that, assuming the machinery in sections 1.1 to 1.3, a full reduction of EPP effects to labelling is insufficient. A coexistence of both mechanisms, subject to careful examination of the exact division of labour between them, seems to be a possible and perhaps promising research avenue.
2. Labelling in copular constructions
In this section I analyze two kinds of structure applying the framework outlined in section 1: copular sentences with adjectival predicates and copular sentences with predicative DPs.
2.1 Adjectival copular constructions
Consider the Spanish sentence in (6):
Let us consider some possible alternative derivations for (6) following current Minimalist technology, to test what the LA can do. Assuming to begin with that First Merge involves two syntactic objects, and that these are singletons (i.e., that lexical elements are Quine atoms), we would have the derivation in (7):
Juan (or the pronoun él, as for our purposes they are completely interchangeable) and alto, crucially, must agree on person, number, and gender:
If they agree, then the set {{Juan}, {alto}} should be labelled <phi, phi>. If it is, then there is no labelling motivation for {Juan} to move (note that the same argument holds for cases where the predicative phrase is uncontroversially a set, as in Juan está orgulloso de su hermana ‘Juan is proudMasc.Sg of his sister’; the ser-estar alternation is orthogonal to our present point). Here, the adjective is not an adjunct; in other words, it is not “a tall boy”, the kind of example considered in Chomsky (Reference Chomsky2020a: 49), so Pair-Merge should not apply. Note, however, that even if it did, we would get the ordered set <Juan, alto> (or <Juan, {orgulloso de su hermana}>, abridging the structure of the adjunct), which poses even less of a labelling problem: the adjectival ‘adjunct’ would be, to quote Chomsky, “off in some other dimension” (Reference Chomsky2020a: 50), and thus invisible for labelling. To be sure, the label would be provided by Juan as opposed to alto (the latter, the label that would have been assigned to the SO after Juan-raising), but it is unclear whether this would be a problem at all (certainly, it would not motivate DP raising for labelling reasons), given the turn against constructivist systems in current Minimalist theorizing.
Let us consider an alternative analysis to see if we can still save the labelling motivation for XP raising. Suppose that we considered the phi-set of adjectives to be somehow “defective”, as in Chomsky (Reference Chomsky and Kenstowicz2001: 18), in that they would have Number, Gender, and Case, but no Person. So far as we can tell, there is no restriction to labelling (7a) <Num, Num> or <Gend, Gend>, so there should be no problem there either (perhaps a similar view could be extended to other cases of defectivity, such as expletive there). But suppose, as noted above, that a condition was proposed that only a full set of phi features allows for <phi, phi> labelling. Phi-completeness requirements have a pedigree: Chomsky (Reference Chomsky and Kenstowicz2001: 6), for instance, establishes that “α must have a complete set of ϕ-features (it must be ϕ-complete) to delete uninterpretable features of the paired matching element β”, in the context of a discussion of Case/Agreement systems. Only agreement with non-defective T (i.e., T selected by C) can value Nominative case in a DP goal. For labelling purposes, however, imposing such a completeness requirement would be somewhat strange, in that the label of a complex object would be a feature that does not really exist as such: “phi” is a shorthand for a bundle of features, at the least [person: val] and [number: val], possibly also [gender: val] and [definiteness: val] (see den Dikken Reference den Dikken2011, Preminger Reference Preminger2021). There is, strictly speaking, no ‘phi-feature’. Thus, the label for agreeing {DP, TP} or {DP, AP} should actually be something like <{person, number}, {person, number}>. (This would also be available under a percolation view of feature-based labelling, as in Zeijlstra Reference Zeijlstra, Smith, Mursell and Hartmann2020. Important to this view is the fact that phi features in adjectives may enter the derivation unvalued, but are interpretable, see fn. 15Footnote 15). In any event, suppose that only a full set of phi-features can label, and that adjectives are phi-defective. In this case, <F, F> labelling (F a feature, i.e., an attribute: value pair) would not be available and, assuming that lexical terminals are singletons, one of the XPs (either DP or AP) would need to raise, with DP raising being the only convergent option (cf. *Alto es Juan, with neutral intonation). The labelling approach could have a consistent story in this specific scenario. We will see, however, that copulas with predicative DPs (which should not be phi-defective) present an unavoidable challenge to labelling-based DP raising.Footnote 11
Let us consider briefly how the phi-features of a predicative adjective are valued, such that the adjective agrees morphologically with the nominal it modifies. It is important at this point to note that, following Minimalist practice, a theory with only one operation over features that links valued and unvalued attributes is preferable over a theory with two. In this context, Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001) distinguishes between Agree and Concord, with Concord “involving Merge alone” (Chomsky Reference Chomsky and Kenstowicz2001: 42, fn. 6). In this view, Concord would be a relation that holds between mergemates, whereas Agree links a Probe and a Goal at a structural distance (given co-membership + proper containment, as per the derivational definition of c-command in Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000: 116). The distinction between Agree, which requires at least Minimal Search, plus Agree Link and Agree Copy (see Deal to appear, Smith et al. Reference Smith, Mursell, Hartmann, Smith, Mursell and Hartmann2020: 10), and Concord (which presumably involves only Merge and Copy; see e.g. Carstens Reference Carstens, Smith, Mursell and Hartmann2020: 87ff.) is upheld in some recent literature, and is relevant to the cases under consideration. Smith et al. (Reference Smith, Mursell, Hartmann, Smith, Mursell and Hartmann2020: 4) refer to Concord as “the sharing of phi-features between a head noun and its modifiers”. This kind of sharing should result in labelling, in a configuration like that in (9), where the values of attributes Number and Gender (there being no Person agreement) would be copied onto the head of AP, as argued earlier in this section.
Presumably, Concord is necessarily an operation that applies to co-members of a set, where there is no set that includes only one of those co-members. Thus, a and b can undergo Concord in (10a) but not (10b), since there is a set in (10b) that contains b but not a:
An alternative analysis, presumably more Minimalist in its eliminative spirit, would dispense with Concord altogether, leaving only Agree. Bruening (Reference Bruening2020: 9) adopts such a view, in a sense:
I assume that all agreement, including nominal concord, involves an agreement probe with unvalued features that searches its sister for valued features. When it finds valued features of the type it is looking for, it copies the values. (Emphasis added)
In this context, Concord could apply in (10a), and Agree in (10b). That is just nominalism, though: the operation that applies to both objects, under Bruening’s assumptions, is the same. Bruening’s analysis, which unifies Agree and Concord in terms of the mechanisms that underpin both, is presumably “more minimalist” than having two distinct mechanisms to copy values between attributes in the grammar.
Suppose now that we do not want to identify a singleton set with its member, that we want to keep {alto} distinct from ‘alto’, where the former is a set and the latter is an urelement. Then, we could have the following:
In (12) we assume that the predicate altos is a mergemate of its argument. If the First Merge involves a head and a phrase, the head should label, delivering the structure in (13).
This then Merges with the copula, which is labelled as in (14), and no labelling paradox arises.
The case of copular constructions with a predicative urelement also argues against the DA approach to subject raising: in that approach, symmetry points must be dissolved because they are not LCA-linearisable (Kayne Reference Kayne1994, Moro Reference Moro2000, Moro and Roberts Reference Moro and Roberts2023). In this view, the heads X and Y of {XP, YP} cannot be strictly ordered with respect to each other in terms of asymmetric c-command, and thus in terms of of linear precedence (which is a function of asymmetric c-command). In a sense, DA is the other side of the labelling coin. Supposing that labels are relevant at LF, and thus labelling-driven movement can be looked at as LF-driven movement, then DA-driven movement is PF-driven movement, given that precedence is a total order imposed over terminals. It is important to bear in mind that, unlike the 90s version of the theory, in which Spec-Head relations created by IM were the configurations within which feature checking took place, Agree and IM are currently entirely dissociated: Agree takes place between Probes and Goals at a distance (and even across phasal boundaries). This means that IM is either entirely unconstrained (Chomsky’s position) or entirely interface-driven (in turn, by either LF or PF requirements). This is perhaps one of the reasons why eliminating the EPP, either in specific constructions in specific languages or in general, is such an important aspect of Minimalist practice, and has been so since the early 2000s (see e.g. Boeckx Reference Boeckx2000, Grohmann et al. Reference Grohmann, Drury, Castillo, Billerey and Lillehaugen2000, Bošković Reference Bošković2002, Epstein and Seely Reference Seely and Boeckx2006, among many others). It remains the last stronghold of feature-driven movement.
Going back to our example, and supposing that lexical terminals are not sets (x ≠ {x}), we start by Merging the set {los, chicos} with the non-set ‘altos’. Because only one of the SO involved in the operation is a set, there is no symmetry point, and thus no linearisation or labelling problem. The linearisation of the portion of the structure just built would be as in (15):
However, the fact that it is “deviant”, or, in Epstein et al.’s (Reference Samuel, Kitahara, Seely, Epstein, Kitaharan and Seely2015c: xvi) terms, “gibberish” (“a convergent object that results in some semantically anomalous interpretation”; the term is also used by Chomsky Reference Chomsky, Domenico, Hamann and Matteini2015b: 8, with the same meaning) does not mean that the object {{los, chicos}, altos} is not unambiguously linearisable. As Chomsky (Reference Chomsky, Franco and Lorusso2020b: 166) emphasizes, “deviant expressions” (which do not violate formal conditions) are apparently “required” in the current system, given that Merge (Internal or External) needs no trigger. Operations just “take place”, with some outputs being interpretable and others filtered out at the interfaces. Merge itself is not restricted by specific conditions: restrictions apply to the size of the workspace. For example, it increases by one at every derivational step and cannot decrease (see Fong et al. Reference Fong, Berwick and Ginsburg2019; cf. also Marcolli et al. Reference Marcolli2023, who assume non-planar trees rather than unordered).
2.2 Predicative DPs
A similar argument can be made with DPs instead of APs, but additional considerations about agreement are necessary:Footnote 13
If no functional structure is assumed for the moment, we have two complex SO in a “small clause” symmetric configuration (Chomsky Reference Chomsky2013: 44), shown in (17).
Note that we abstract from the internal structure of the possessive phrase mi amigo since it is inconsequential to our argument. Here too we necessarily have phi-agreement (and, as before, all features must share values), shown in (18).Footnote 14
Evidently, if instead of el hombre or este chico we had Juan, and Juan was an urelement, it could label without any issues, so we will keep to strictly phrasal subjects (bearing in mind that the arguments would extend to Quine atoms). In connection with the previous discussion, note that even if we accepted that adjectives are phi-defective, we cannot say the same for DPs: un ladrón and el hombre should enter the derivation with valued, interpretable phi-features like any other DP (as they undoubtedly would if they appeared, say, as the subject of a transitive sentence). An additional problem appears, though: if we did take both DPs with valued interpretable phi features, we would only get agreeing DPs by chance. Considering only Person, Number, and Gender, we have three possible values for Person, two for Number, and two for Gender. If all combinations were allowed, given free untriggered Simplest Merge, most of the outputs would be “deviant” and would have to be filtered out. Problems do not stop here. Suppose that we wanted to say that predicative DPs enter the derivation with unvalued interpretable phi-features, accepting Pesetsky and Torrego’s (Reference Pesetsky, Torrego, Karimi, Samiian and Wilkins2007) argument that interpretability and valuation are dissociated (also in Zeijlstra Reference Zeijlstra, Smith, Mursell and Hartmann2020).Footnote 15In order to make sure that mi amigo enters the derivation with [_Person], [_ Number] (i.e., unvalued Person and Number, as opposed to [u-Person], [u-Number], which would be uninterpretable but possibly valued), we need to know in advance, before the derivation starts, that mi amigo is going to appear in a context in which it must have unvalued phi features (cf. Juan es mi amigo ‘John is my friend’, or Juan era un ladrón ‘John was a professor’: here, it is unquestionable that mi amigo or un ladrón agrees with Juan, not the other way around, as it is amigo/a/os/as and ladrón/a/es/as that have an inflectional paradigm). That kind of Look Ahead (accessing objects which have not yet been constructed) should not be available in a strongly derivational system, in particular considering Chomsky’s (Reference Chomsky2021) claims about the Markovian character of derivations. Since we are going astray from our EPP discussion, we shall cut this digression short, but the problems noted in this paragraph remain.
Consider now the Italian case analyzed in Moro (Reference Moro2009), of which I provide an intermediate derivational step in (19):
Either DP may raise out of the bracketed small clause, yielding a canonical (20a) or an inverse (20b) copular sentence:
The Spanish facts are exactly parallel: the two DPs in Italian must agree as they do in Spanish. Thus (21a–b) are excluded:
Moro argues that, for purposes of DA and labelling, “there is no necessity to raise either DP to the copula: it is sufficient that either one is raised to any head that merges with {XP, YP}, neutralizing the problem given by the absence of a label” (Moro Reference Moro2009: 18). Indeed, Moro considers VP internal focalization as an example of internal Merge that, without targeting the copula itself, let alone T, dissolves a point of symmetry:
Neither DP must raise further, and after insertion of the copula, Moro says, “pro-insertion can take place successfully, since the postcopular constituent is not unlabelled anymore” (Reference Moro2009: 18).
Let us now consider Moro’s derivations under the assumptions spelled out in section 1. Under Simplest Merge, (22a–b) look like (23a–b), respectively:
How exactly are those structures labelled? If we are able to label {XP, YP} without further raising, there must be a shared feature between the heads X and Y. It is not enough to raise a DP; it must be endowed with an appropriate “prominent” feature shared with the head Foc, otherwise it must keep raising (cf. discussions in Chomsky Reference Chomsky2013 and Epstein et al. Reference Samuel, Kitahara, Seely, Epstein, Kitahara and Seely2015b of successive cyclic wh-movement, which depends on intermediate C heads not bearing a Q feature). However, whereas Q is presumably a lexically valued feature in wh-words, it is harder to justify Foc being assigned in the lexicon (as Moro p.c. also observes), despite both Q and Foc being criterial features. As we just observed, the strictly local character of derivations prohibits the kind of Look Ahead required to foresee that a certain DP will receive a Focus interpretation. But without a Foc feature on either DP, labelling by a shared prominent feature cannot take place, and the prediction is that whichever DP has raised to Spec-Foc should keep moving up – and we have not even emphasized the fact that, as in the Spanish cases, agreement between both DPs is necessary, so nothing blocks <phi, phi> labelling under Chomsky’s assumptions.
A further problem is posed by pro-insertion. The full structure Moro (Reference Moro2009: 18) proposes for the Italian-focused cases is (24):
Note that if we have {{una foto del muro}, {Foc, {{una foto del muro}, {la causa della rivolta}}}} labelled <Foc, Foc>, the introduction of the copula should label VP, as we now have a {H, XP} object. The next derivational step Merges T, and the resulting structure is labelled TP (Italian, as a null-subject language, has strong T). There being no labelling conflict in the structure, {T, {è, {{una foto del muro}, {Foc, {{una foto del muro}, {la causa della rivolta}}}}}} is a well-formed output. If there is pro insertion at all, it cannot be motivated by labelling considerations. Given existing Minimalist tools, an EPP account seems hard to avoid.
Continuing our discussion about predicative DPs in copular sentences of the type este chico es mi amigo, we may consider the logically available possibility that <phi, phi> labelling is only available in cases of Agree, not Concord, if we wanted to keep the two operations separate, as in Chomsky (Reference Chomsky and Kenstowicz2001). If Concord is concomitant to Merge, unlike Agree, then <phi, phi> should not be available in copular constructions, since the morphologically agreeing phrases are mergemates, under the assumptions we have been working with so far. That argument may win the battle, but as far as I can tell it would lose the war: the cost of keeping labelling as a way to obtain DP raising is to admit a new operation, Concord, kept distinct from Agree (distinct enough that labelling can differentiate between their structural descriptions). That seems to me considerably more costly than admitting an EPP feature in what is already a feature-based system.
Alternatively, we may suppose that we want to say that <F, F> labelling (where F ranges over a feature type, such as <phi, phi> or <Q, Q>) is only available for SO that have been related via IM, not EM, to complete the available possibilities. This would require the LA to have access to the derivational history of a structure. Given current assumptions about what may happen at the phase level, this is not entirely unreasonable. If a pair of occurrences can be inspected to see if it respects the theta criterion or not, delivering, for example, the Chomsky (Reference Chomsky2021) raising vs. control distinction, part of that process may also involve marking the head of a chain and making that count for LA. This requires us to allow SOs to remain label-less at least until Transfer. The counter-cyclicity problem does not go away, but otherwise the argument does not seem completely outlandish, other than the fact it must be stipulated that a labelling option that should be straightforward (if you have two sets with common features, label their combination as these common features) must be complicated by having access to information about how a set has been created. In other words, if you have two sets with common features, you must check whether one is the head of a chain. If so, label their combination as these common features. If not, abort labelling. Such information should not be available in strictly Markovian derivations.
An alternative analysis of copular constructions is available, whereby predication is not a direct relation between lexical categories under mergemate conditions. Suppose that, as in Bowers (Reference Bowers1993, Reference Bowers, Baltin and Collins2001), den Dikken (Reference den Dikken2006), Roy (Reference Roy2013), and many others, predication is necessarily mediated by a functional category, call it Pred (following Bowers). Can PredPs be invoked in the labelling-based eliminative argument against EPP? I think not. So far as I can see, PredPs do not help in rescuing the labelling approach to XP raising in copular sentences. Let us see why. Suppose that we Merge the predicative phrase – say, a DP – with a Pred head, as in (25):
Being an {H, XP} object, (25b) can be labelled Pred(P). The same holds, of course, for (16a).
In parallel, we construct the other DP in (26) (as is the case with all complex specifiers, for independently motivated reasons; see e.g. Uriagereka Reference Uriagereka and Uriagereka2002):
Finally, we Merge the output of (26) and the SO defined in (25b), obtaining (27):
At this point, a purported labelling failure would take place, since we have two complex objects as co-members of a set. As a reviewer observes, (27) “[i]s of the form {XP, YP} with no shared features, thus creating a symmetry that ought to be resolved by movement, just as in Chomsky’s LA account of the EPP”. Either phrasal object would have to raise: if {este chico} moves, we obtain este chico es mi amigo (= (16)). However, observe that Pred or not, phi-agreement between este chico and mi amigo must necessarily hold (mutatis mutandis for el hombre era un ladrón). How can we make sure these phrases phi-agree? Given the fact that under a PredP analysis, they are not mergemates, it is unclear how Concord would work (if we kept the mechanisms of Concord separate from those of Agree, though see the quotation from Bruening Reference Bruening2020 in section 2). Without Concord, we could indeed generate {{este, chicoMASC}, {T, {es, {{este, chicoMASC}, {Pred, {mi, amigoMASC}}}}}}, but also, say, *{{este, chicoMASC}, {T, {es, {{este, chicoMASC}, {Pred, {mi, amigaFEM}}}}}} and any other mismatching combination of phi-features. The derivation of copular sentences under these assumptions would entail the generation of multiple unusable structures, which we deem undesirable on theoretical and empirical grounds (see also Ginsburg Reference Ginsburg2016: fn. 14).
Concord cannot apply since the DPs are not mergemates, so let us now turn to Agree. Suppose that the predicative DP enters the derivation with unvalued phi-features. This is in itself a problematic stipulation, since even if we admit that the heads of predicative DPs enter the derivation with unvalued phi-features that get valued via Concord, it means knowing that a D or N head will appear in predicative position even before we started to build structure (again, an evident Look Ahead problem). Unless reverse Agree is allowed (as in Zeijlstra Reference Zeijlstra2012), the unvalued features in the predicative DP cannot probe the valued features in the argumental DP. The argumental DP has no reason to probe, since by hypothesis it enters the derivation with valued phi-features, the features that T will probe for. As a side comment, no labelling-based argument against the EPP that I know of makes use of reverse Agree.Footnote 16
We could, alternatively, say that the unvalued features of the predicative DP percolate to the complex object {Pred, {DP}} and get valued via Concord – but an example such as (28) highlights the problem with such an assumption.
Nothing prevents (28) from being labelled as <phi, phi>, since the values of the phi-features of este chico can be copied to its mergemate, delivering an object with shared features (the same argument holds if instead of a DP we have a complex AP). It is unclear whether Pred bears unvalued phi-features, and I have found nothing that clarifies this issue.
3. Conclusions
A cautionary note is in order. The arguments presented in this article against a reduction of EPP to labelling, all of them based on everyday Spanish data using predicational and specificational copular constructions, do not preclude there being a problem with an XP-YP unit. They only point out that, in these specific constructions, there is no labelling failure that could motivate raising of one of the terms in an XP-YP symmetry point. Thus, the claim that EPP effects in these constructions are subsumed to labelling failures is unsubstantiated. This is not a positive argument in favour of the existence of EPP, nor does it entail that labelling does not have other problems.
Symmetric syntactic objects are problematic in some forms of Minimalism, independently of labelling. For purposes of LCA-based linearisation, XP-YP is still a symmetry point that may trigger independent Spell-Out of either object (Uriagereka Reference Uriagereka and Uriagereka2002, Reference Uriagereka2012; Moro and Roberts Reference Moro and Roberts2023). There is also a type-theoretic problem, as noted in Bowers (Reference Bowers1993) and Roy (Reference Roy2023), in that the relation between the two DPs in a copular sentence cannot be derived compositionally. Those issues, and possible analyses, are beyond the scope of this article. The point I want to make is that, given the fact that there is no labelling issue in the Spanish structures considered here, it is not possible to use labelling as a justification for phrase raising. Thus, the argument that EPP effects in these structures can be reduced to labelling failures carries little if any force. Again, this does not amount to an argument in favour of the existence of an EPP feature (or an EPP “property” of specific features, cf. Goodall Reference Goodall, Davies and Dubinsky2001: 213), which would require independent argumentation.
As a reviewer points out, independent arguments should be provided to show that phrase does indeed raise in the relevant constructions, in that there are reasons to think that the EPP is not active in null-subject languages. We could attempt to justify subject raising to Spec-T by using a periphrastic construction in combination with our copulas:
It seems that no matter what phrase P has auxiliaries as heads (traditionally, INFL/T), the subject raises to a position external to the head of P; following our earlier argument, there is no labelling problem within the VP that motivates raising. However, this does not settle the question of what exactly this “P-external” position is: are we certain Juan appears in Spec-INFL/T? Is there such a position available to begin with?
Indeed, Contreras (Reference Contreras and Rothstein1991), for instance, argues that in Spanish INFL is a lexical category, and as such does not project a specifier (only functional categories project specifiers in his system). Lexical AGR in INFL can also case-mark pro in-situ within the VP, with subjects able to adjoin to IP rather than raise to Spec-IP. The idea is developed in Alexiadou and Anagnostopoulou (Reference Alexiadou and Anagnostopoulou1998) and much subsequent work. However, it is unclear how to translate the claim that AgrS/IP projects no specifier and thus pre-verbal subjects in null subject languages are in an A’ position into Simplest Merge terms: unless labels are set members (as in Epstein Reference Epstein and Epstein2000: 148), capturing the segment-category distinction in the syntax is not straightforward. In other words, under Simplest Merge there seems to be no real way to distinguish between (30a) and (30b):
In both cases, the corresponding set-theoretic object would be (31) below, as IP and I’ are not syntactic objects (Seely Reference Seely and Boeckx2006, Collins Reference Collins, Bauke and Blümel2017, Chomsky et al. Reference Chomsky, Seely, Berwick, Sandiway Fong, Kitahara, McInnerney and Sugimoto2023, among many others):
A relational definition of phrase structure “levels” (such as that advanced in Chomsky Reference Chomsky2015a: 242 and Nunes Reference Nunes1998: 160) refers to projection, a notion no longer available. Many of the early-2000s arguments that attempted to eliminate the EPP as an independent property of the grammar pertained to redundancies: independently motivated mechanisms can do what the EPP does without additional assumptions. For example, Epstein and Seely (Reference Seely and Boeckx2006: 10) identify redundancies between the EPP and the following mechanisms:
• Case valuation
• Predication theory
• Locality
• Derivational morphology
• Null complementizer theory
These specific earlier arguments do not necessarily carry over to contemporary Minimalism, given how some simplicity/economy metrics have changed. For example, Epstein and Seely (Reference Seely and Boeckx2006) work under the assumption, standard at the time, that Merge is simpler than Move (Merge-over-Move, see Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000: 101), whereas Chomsky (Reference Chomsky2020a, Reference Chomsky, Franco and Lorusso2020b, Reference Chomsky2021) argues that Move (IM) is preferable to Merge (EM) insofar as it only works with what is available in the workspace, thus restricting the search space. Similarly, part of Epstein and Seely’s argument based on locality has to do with the idea of Specifier positions being “checking positions” (and the assumption that Case must be checked in a Spec-Head relation, also assumed in Bošković’s (Reference Bošković2002) “Inverse Case Filter” argument), a notion that plays no role under current assumptions. Considering the relation of phi-agreement between T and DP (which should be followed by IM of DP to Spec-T), Epstein et al. (Reference Samuel, Kitahara, Seely, Epstein, Kitahara and Seely2015a: 20) say:
What forces IM to raise the goal to the specifier of the phi-probing head – the residue of Extended Projection Principle (EPP) – is still an open question.
They assume with Chomsky (Reference Chomsky, Domenico, Hamann and Matteini2015b, Reference Chomsky, Franco and Lorusso2020b) that IM (like EM) needs no motivation, featural or otherwise; that anything accessible in WS can be internally Merged at the root, freely, without regard for creating an interpretable object or repairing local violations. Merge applies blindly, Merge applies blindly – as Ovid would say, sponte sua, sine lege.
Other arguments require additional assumptions that are controversial in their own right. Gallego (Reference Gallego2017), for instance, assumes that C and T are one and the same category, and that the invisibility of T for labelling purposes in English is not due to T being “weak”, but due to it being a “copy” of C. Space restrictions prevent us from dealing with this proposal in depth, but note that it is necessary to assume T independently from C, for CP-less structures. If T is a copy of C, then no C would mean no T, and no EPP effects. However, if we consider raising-to-object structures (Postal Reference Postal1974, Chomsky Reference Chomsky2021: 24), under an ECM analysis, the raised DP must appear in Spec-TP in the downstairs clause, preceding infinitival to:
If the clausal complements of raising-to-object verbs are structurally defective in not having a C layer (following Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky2021), the fact that Beatrix does not stay within the vP domain but raises (at least) to Spec-to remains unexplained. But suppose that raising-to-object is analyzed as involving DP raising to the matrix VP, which would make it very hard to diagnose the collapse of C and T into a single category. If this is the case, can Gallego’s hypothesis be maintained? I believe not, due to independent problems. Assuming that C and T are “bundled” in languages like Spanish makes it difficult to provide adequate analyses for sequences of auxiliary verbs (so-called auxiliary chains) and the positions of subjects in cases such as (33) (see Krivochen and García Fernández Reference Krivochen and García Fernández2019 for detailed discussion and analysis):
In my opinion, Gallego’s proposal exemplifies the scenario in which the elimination of EPP on labelling grounds causes more problems than it appears to solve.
Some final considerations. The “direction” of redundancy elimination is problematic: if principles or mechanisms P and Q (say, EPP and Case, or Labelling) are redundant, in that they cover (at least partially) the same empirical ground, how do we know which one to eliminate? In the domain of our inquiry, the usual answer has been “eliminate EPP”, but that is certainly not the only logical option. For example, consider EPP vs. Case. As mentioned above, Bošković (Reference Bošković2002) adjudicates in favour of Case theory with his Inverse Case Filter, which relies on Spec-Head relations available in checking theory (rather than Agree or Move-F) and requires additional assumptions about the relation between structural and inherent Case – for instance, that accusative case can be either structural (for a DP in Spec-AgrOP) or inherent (in Compl-V) (see Bošković Reference Bošković2002: 212). McFadden (Reference McFadden2002), in contrast, proposes that the redundancies between EPP and Case should be resolved by eliminating syntactic Case rather than the EPP, making it exclusively post-syntactic, along the lines of Marantz (Reference Marantz2000) (see also Chomsky et al. Reference Chomsky, Gallego and Ott2019). Considering the arguments put forth here against a pure labelling approach to XP raising in copular constructions, and the unclarity of the applicability of the “redundancy” argument schema to the same cases, the reports of the EPP’s death seem to have been greatly exaggerated.
Open access
