Desktop version

Home arrow History

Becoming a swan

The {XP}-part of the set-theoretical construct [1], a singleton set of XP, would presumably be born under a self-application of merge (to be entirely precise, self set-merge, to distinguish the operation from self pair-merge proposed by Ishii (2016) to account for syntactic opacity-inducing properties of transfer), a kind of merge which is standardly taken to be illegitimate, but has been admitted under further conditions restricting its admissibility to provide an account of the problem of the beginning of the derivational process (the ‘first merge problem’; see a recent development of the idea in Adger (2013)) and general properties of antisymmetry (together with the distinction between nouns and verbs in Kayne (2011), with an operation directly forming a singleton set; see also Guimaraes (2000) and proposals to replace the self-merge operation with merge with 0 to achieve analogous results as in Fortuny (2008: 18-19), Trotzke and Zwart (2014), De Belder and van Craenenbroeck (2015)). Were the lexicon available for narrow syntax to contain just 0, with its sole element and the operation of self-merge it would be able to generate natural numbers out of nothingness in a Zermelo-like fashion (were von Neumann-like procedure more to its liking instead, it would require external merge in addition)—a connection between the generative engine and the arithmetical capacity noted several times (see Chomsky (2008: 139), Chomsky (2015a: 87-88))—‘spinning extravagant realms of being out of just one single thing’ (Lewis 1991: 13), to use Lewis’s disdainful remark; within the confines of syntax, however, the structure created by self-application of merge may seem problematic—it ‘leads to questions about how to distinguish occurrences of x and what to say about three or more xs merging all at once' Kayne (2011: 332) notes, and the former issue in particular seems pressing in the context of the labeling process (by hypothesis, merge is unavailable for n > 2, hence the latter problem does not arise). Other potential threats to self-merge ceased to be perilous; in particular, once ‘the lingering idea, carried over from earlier work, that each operation has to be motivated by satisfying some demand (Chomsky 2015b: 14), has been abandoned, there is no reason to suspect self-merge on the basis of its ‘satisfying no demand’— more specifically, its involving no feature valuation configuration—anymore.

Furthermore, the assumption that internal merge should operate exclusively on a proper part of a syntactic object is a stipulation: the difference between external and internal merge, preserved—to recall—during the computational process by the phase level memory, consists in the input for the operation being two syntactic objects of which neither is a part of the other or two objects entering into a part- whole relationship:

Merge(X, Y) = {X, Y}. Suppose neither X nor Y is part of the other, as in combining read and books to form the syntactic object {X, Y} corresponding to “read books.” Call that case External Merge EM. Suppose that one is part of the other, say Y is part of X. Then the result of Merge is again {X, Y}, but in this case with two copies of Y, one the original one remaining in X, the other the copy merged with X. Call that Internal Merge IM. Note that both operations come free: it would require stipulation to bar either of them. Furthermore, there are no operations “form copy” or “remerge,” just simple Merge. (Chomsky 2013c: 40)

‘Being part of the other’ should include being an improper part, and the operation of self-merge would thus require no extension of the notion, involving instead elimination of a stipulative restriction of the applicability of the operation. Self-merge is thus nothing more than a subcase of internal merge, not another kind of merge entirely. It is not a subcase of external merge, either: for the operation merge (X, X) as an instance of external merge to take place—putting aside for a moment the definitional property of ‘not being part of the other’, not fulfilled in this case, and assuming that it is required that objects enter the workspace independently to participate in external merge—both X’s would have to be generated separately, irrespectively of their complexity, an expectation as unacceptable as in the case of alleged independent creation of copies: ‘That is multiply wrong. It requires a stipulation that IM is barred. Furthermore, it requires far more computation than copy-formation by Merge, since copies may be arbitrarily complex and would have to be generated separately (...), then matched somehow and distinguished from repetitions’ (Chomsky 2013c: 41). Analogously, there is no reason to stipulate that merge requires distinctness of objects undergoing the operation, thereby excluding the case when X = Y (as in Collins and Stabler (2016: 47-48)), whereas it is merely a special case of the requirement that Y be part of X—improper part included.

Although the fact that no featural requirements need be met for the operation merge to take place frees such cases from the suspicion of illegitimacy in terms of justification of syntactic operations, it does not yet ensure that the output of self-merge is acceptable for the labeling algorithm and subsequently for interpretive purposes. Indeed, it might seem the output of self-merge should exhibit an ambiguity with regard to the determination of the head and the tail of the chain created by this kind of internal merge: since {X, X} = {X}, it would seem that the head and the tail are identified—a syntactic ouroboros is born, as it were. In terms of availability for labeling, X should be then both visible (qua head) and invisible (qua tail) for the labeling algorithm. This conclusion may be too hasty, though, for it depends upon the details of identification of occurrences constituting a chain, its head in particular. Recall that the device used in the minimalist framework to identify syntactic objects as occurrences of an object undergoing internal merge/displacement is the syntactic context in which an object (a copy) appears—a revamping of the solution adopted already in Chomsky (1975b: 109-110) and going back directly to Quine (discussed in Quine (1951: § 56); see also Quine (1987: 218-219)), where linear order was used, a ‘mechanism not available here, as Chomsky (2000a: 145 n. 63) notes— and, in particular, structural sisters of occurrences (see further Chomsky (2000a, 2001), Epstein and Seely (2006) for some discussion of advantages and disadvantages of various ways of implementing the notion of the ‘context of an occurrence’), so that in set- theoretic terms, the only official ones in recent formulations of the minimalist formalism, ‘a “higher” occurrence of a properly contains lower ones’ Chomsky (2000a: 115). Thus, the head of a chain CH of an element a is the occurrence of a which properly contains all other occurrences of a. In the context of self-merge, this definition of the head has it as an immediate consequence that there is no head of the chain in {X, X}: no occurrence of X properly contains the other; there is thus no head of the chain—for the purpose of the labeling algorithm there is only a copy, invisible as copies are.

Seen through such lenses, structures created by self-merge would be indeed troublesome if they were created as syntactic objects supporting the skeleton of a syntactic structure and undergoing labeling as all other objects; without external support they are unable to perform such tasks, since they are invisible for the labeling algorithm (such external aids are explicitly introduced e. g. in Adger (2013); within the system presented there, consecutive self-merge does make a difference, in contrast to the framework of Chomsky (2013c, 2015b), where self-merge going beyond its first application does not change label-wise properties of the object so created); in particular, it cannot be proposed as a solution to the first-merge problem anymore. Yet as a mechanism which prevents parts of complex objects from undergoing labeling—without making them invisible for interpretive purposes, since the C-I component does recognize copies; and without making them internally unlabeled provided that they have undergone the procedure separately—it may be part and parcel of adjunction, contributing to their inertness with regard to labeling and other syntactic processes (not for

  • 2.1 Adjunction: a syntactic ugly duckling
  • 63

interpretive purposes, though, which includes binding-theoretic interactions, if one follows Chomsky (2015a: 88) in assuming that ‘it happens at the conceptual- intentional level’). Sticking to the representation involving the Kuratowski-pair, adjunction so conceived would involve a double application of internal merge to the adjunct-to-be, one of which would be self-merge, as in (2).


This would make the operation more complicated than it has to be—it does not need to involve more than one application of internal merge to XP to achieve the desired effect. Note that this entaglement of internal merge and external merge should be taken to include internal merge as a subroutine rather than as an independent operation occurring after external merge takes place:

  • (3) Operations in phrasal adjunction
  • 1. Take two SO’s XP and YP;
  • 2. self-merge XP;
  • 3. merge the output of 2 and YP.


The operation of adjunction is thus more complicated than a single application of internal or external merge (reminiscent of the proposal of Espanol-Echevarria (2011) to apply ‘downwards’ internal merge to account for the behaviour of adjuncts), but it does not have to be conceptualized as involving a creation of a distinct kind of syntactic object—be it an ordered pair, as on the standard account or a doubly-peaked structure, as per Oseki (2015)—nor does it have to involve replacement of one syntactic object by another (as in Chomsky (2004a: 118), where the operation ‘replaces’ в by (а, в), with further complications arising later due to an operation SIMPL which converts (а, в) to {а, в}); nor does it require that the adjunct be transferred to the interfaces immediately after its merge into the structure, as variously proposed in the literature (see already Raposo (2002) for a proposal of ‘quick spell-out’; Boeckx (2012b: 113) hypothesizes that immediate transfer of adjuncts is responsible for their island behaviour, an idea incorporated into a different web of assumptions also in Boeckx (2015a: 63-69)): making self-merge of the adjunct part of the operation suffices—the phase level memory keeps track of operations occurring during construction of a phase, self-merged objects thus being able to exhibit the desired syntactic properties. Adjunction would thus not be a distinct type of merge; it would not be simpler in any respect than internal or external merge, either (as Boeckx (2008b) proposes, reverting the standard analysis and hypothesizing that it is {а, ft} that ‘fits adjunction like a glove’ (Boeckx 2008b: 100); or as Hinzen (2006: 177) formulates a common attitude towards adjuncts: ‘Adjunction, as an operation more primitive than argument-taking, does not require the apparatus of hierarchical syntax as given through projection, and perhaps has no significant syntax at all,’ speculating tentatively that adjunction may be thus available for non-humans, since it ‘arguably merely consists in the composition of two coordinated rather than subordinated predicates’ (Hinzen 2013: 16)), but should rather be understood as nesting an application of merge (internal, self-merge) inside an application of merge (external, taking XP and YP): once you get merge, you get it in all its glory (precluding thereby scenarios on which operations equivalent to adjunction would be evolutionarily earlier than full-blown merge; see Progovac (2015)).

  • [1] XP}, {XP, YP
< Prev   CONTENTS   Source   Next >

Related topics