Desktop version

Home arrow History arrow A Minimalist View on the Syntax–Semantics Relationship: Turning the Mind into a Snowflake


Three degrees of defectiveness

Labeling: details of the grand picture

The process of structure building and subsequent labeling following the analysis of Chomsky (2015b) involves the following steps:

(7) a. form R-IA by EM;

b. IM of IA in a (EPP);

c. Merge v*, reaching the phase level;

d. Inheritance;

e. Labelling; a is labelled (ф, ф);

f. R raises to v* forming R with v* affixed, hence invisible, so phasehood is activated on the copy of R, and IA remains in situ, at the edge;

g. Transfer of the copy of IA.

Several properties of the sequence in (7) will turn relevant for the discussion to follow (see sections 2.3.2 and 2.4.3); for the present purpose suffice it to be observed that the procedure gives results expected on the assumption that labeling takes place late in the derivation, testing syntactic structure for interpretability at the C-I interface. Or it seems to do so, until we reflect on the properties of the lowest stratum of the structure and the interpretation that в is supposed to obtain.


Discussing the elimination of selectional features, Chomsky (2004a) comments first on transitive structures, and then on cases like (8):

These problems become more severe if there are no categories, only roots, so that v in (8) determines that the root see is verbal—on many grounds a reasonable assumption, which also yields the otherwise unexplained conclusion that V ^v movement is obligatory. Then given, say, the root arrive, we do not know whether it is verbal (selecting an internal argument IA) or nominal (with no IA required) until the next stage of derivation, at which point it is too late to merge IA (by cyclicity). There are still further problems; for example, how do we know which s-selectional feature must be satisfied first? For a variety of reasons, then, s-selection should be dispensable. (Chomsky 2004a: 112)

Yet this issue cannot be taken to be dissolved this way. Such structures pose a problem for a theory of labeling which relies on the idea that labels are required for the purposes of the C-I component, determining interpretation of syntactic objects on the semantic side of the syntax-semantics divide. A label might not be necessary if labels were assumed to be relevant for syntax-internal purposes only, as when it was supposed to be the case that—since ‘each SO generated enters into further computations. Some information about the SO is relevant to these computations’ (Chomsky 2008: 141)—the information in question should be provided by labels; whence, if an object were otherwise inactivated for the syntactic computation, such information would no longer be necessary (this is the solution Adger (2016) adopts for layers like в in (8))—but it is not so if labels are there for the C-I purpose (as declared in Chomsky (2012a, 2013c, 2015b)): ‘Labeling has to be done, for the same reason that Merge has to be done. Otherwise there is nothing to interpret’ (Chomsky 2015a: 80). Since в is to be interpreted for thematic reasons, and invisibility of the copy of IA is invisibility for the labeling algorithm only, not for interpretive purposes, it stands to reason that it should get a label; therefore, the conjecture that is put forward in Ott (2012), viz.:

Labels are required for thematic interpretation: for a syntactic object Z to be interpreted as an element of the thematic domain (that is, if Z is a 6-marked/selected argument or a vP-internal adjunct), Z must be labeled. (Ott 2012: 60),

should be somewhat modified in the context of the labeling theory of Chomsky (2015b). Both objects merged at the first stage of the derivation of (8) will by the time when в is transferred vacate their original positions; it looks as if it were an instance of the {t, t} structure, for which the labeling algorithm should find no label—copies being invisible for its purposes—and provide therefore no instructions for the C-I component to grapple with its interpretation (it is exactly the invisibility of copies in the case of heads that Goto (2015b) proposes to make use of for labeling of {EA, v*P} structures, although within a different set of assumptions about the role of labels). And yet, it is precisely the canonical configuration of ^-assignment for complements: instead of being banned, it had better be interpretable. The important step is in this case (7g): once all preceding steps have been performed, it is only the complement (the domain) of the derived phase head—phase head by inheritance—that is transferred to the interfaces.


In the case of (9), it is only the copy of IA that undergoes transfer—the copy of R remains in the derivational space as a visible part of the syntactic structure. Transfer, subject to widely differing conceptualizations, may be tentatively assumed to be an operation which provides the C-I component access to the syntactic structure, thereby rendering it inactive for further syntactic operations—otherwise, the interpretation should proceed anew every time the transferred syntactic object undergoes modifications—but not deleting/removing the transferred part literally. This understanding of the operation of transfer is thus a C-I version of the ‘conservative Spell-out’ of Uriagereka (2012), without separation of the structure sent to the interfaces, as the ‘radical Spell-out’ is supposed to do, and seems a minimal way to satisfy the property of the computation that ‘what has been transferred is no longer accessible to later mappings to the interfaces’ (Chomsky 2007: 16):

If H is a phase head with complement Z, then Z is the interior of the phase; the edge is H along with anything merged to {H, Z}. It is the interior that is subject to no further modification. Elements of the edge—H and a sister of {H, Z} (and a sister of the resulting SO, etc.)—can be modified in the next higher phase; for example, they can raise, as in V-to-T or successive-cyclic A'-movement. While Z is immune from further changes, it does not disappear. (Chomsky 2013c: 40)

Being ‘immune from further changes’ does not require being literally absent, as the quote above makes explicit, and as noted in Collins and Stabler (2016) in a discussion of the (PF-)Transfer: ‘Instead of thinking of a workspace as containing a set of syntactic objects, all of which are accessible to Merge, we can think of a workspace as providing access to certain occurrences of syntactic objects. One way to do this is to keep a set of syntactic objects that have been transferred, and then block all access to those transferred elements’ (Collins and Stabler 2016: 73-74). The ‘radical’ kind of approach to the operation of transfer, explored e. g. at length in Narita (2014), leads to problems both at the interfaces (the so-called recombination problem) and within syntax itself (provided we adopt, by and large at least, assumptions made in Chomsky (2013c, 2015b)—it is not an accident that details of the model developed in Narita (2014) differ considerably from the stance taken in Chomsky (2013c, 2015b); literal erasure of the copy of IA in (9) would turn the entire edge of the phase into a complement structure, hence in such system structure building operations would have to operate strictly on the basis of the H-a schema). Note that once phasehood is activated on the copy of R, there are no further operations in its domain that should be performed, whence transfer occurs immediately. All this leaves в still unlabeled, but not transferred—this occurs at the next phase. The requirement that в be labeled is one modification of the labeling conjecture of Ott (2012)—absent there possibly because early formulations of the labeling algorithm do not rely on roots and their properties to the extent and in the way that the theory of Chomsky (2015b) does; suffice it to recall that, whether still considered necessary for further syntactic computation (up to Chomsky (2013b)) or already reconceptualized as required for interpretive purposes, labels in the thematic domain were taken to be assigned under the H-XP schema without positing weakness of the root, hence a ‘V’ label was obtained immediately. The replacement of categorially determined LI’s with roots came together with the general rejection of the idea that labels are necessary for syntactic operations to take place—in particular, external merge does not require, on the most current formulation of the labeling algorithm, that objects undergoing the operations be already labeled, the labeling algorithm operating only once phase level has been reached (as Chomsky (2013c) notes, ‘this is a modification of earlier proposals (of mine in particular) that labeling is a prerequisite for entering into computation. That cannot be, as Samuel Epstein points out, or it would block many cases of EM’ (Chomsky 2013c: 43 n. 30)). Availability of a label for в in (8) is assumed in the framework of Chomsky (2015b) to follow from feature inheritance; in particular, feature inheritance is supposed to lead to strengthening of a head too weak to label a structure on its own: ‘Just as English T can label TP after strengthening by SPEC-T, so R can label RP after object-raising.’ (Chomsky 2015b: 10). The label of в would thus be provided by R itself, as in (10).


That might seem correct insofar as it is actually в that is semantically composed with IA in such structures, its being labeled as ф signals an appropriate configuration for (f, f) labeling. What the analysis hides is the fact that в being labeled f does not help the copy of IA to enter into a relationship with R. Feature inheritance is a process which endowes an XP with interpretively relevant properties, not the head LI of an XP with respect to its sister. Just as А-operators are appropriate devices to model interpretive consequences of internal merge provided that it is remebered that they are reflexes of properties of phrases, not of heads—so that an internally merged EA composes interpretively with a sister predicate ‘TP’, not with its head T—so features inherited via feature inheritance should be so understood; transmitted downwards to ‘weak’ LI’s they are in fact properties of structures such LI’s head, enabling them to receive a label—but they do not straightforwardly ‘strengthen’ them so that they are able to compose correctly with their sisters. In the case of roots as in (10), labeling of в, an intermediate projection in traditional terms, by unvalued f-features provides the C-I component with information about how в behaves upon meeting an XP bearing valued f-features; it does not provide information about how it behaves with respect to an XP which is merged directly with the root. Transfer of the complement of the derived phase head as indicated in (9) defers the problem, but does not solve it—both R and the copy of IA will ultimately meet in interpretation.

Properties exhibited by phasal structures combined with details of feature inheritance and the labeling algorithm suggest that we accept there being degrees of defectiveness of LI’s; more specifically, it invites the idea that although both R and T are weak label-wise, their weakness differs in that R even after feature inheritance is not specified enough as to be interpreted in the C-I component. While ‘weakness’ is obviously reminiscent of the early minimalist distinction between ‘weak’ and ‘strong’ features, both concepts should be kept entirely apart— the weak-strong dimension was primarily concerned with properties of features and their impact on the A-P side of the derivation, strong features standing in need of checking before transfer to the interfaces, weak ones being subject to checking only after spell-out, on the route to the C-I component, so that they need to be checked by LF; label-theoretic weakness, on the other hand, is relevant for the C-I component, given that labeling is understood as a process which checks syntactic objects for featural configurations which provide instructions for the interpretive component to proceed. While labeling in (10) suffices for the ‘VP’ to be interpretable as a whole, it does not suffice for the lowest stratum of the structure to be legible in the C-I component; on the other hand, in the CP- counterpart of this structure, feature inheritance is enough to ensure legibility.


Elucidation of this ‘gradability’ of label-theoretic weakness, proceeding in conformity with general guidelines suggested in section 1.1, takes properties exhibited by LI’s in question in narrow syntax and attempts to find their correlates in the C-I component, seeking to eliminate purely syntax-internal properties and, in accordance with the assumption that externalization is secondary and the relationship with the C-I component primary one, eschewing analyses which would connect such features of LI’s with A-P properties. It does not ground syntactic properties in semantic ones, as it would in fact do if the assumption of the need to satisfy external requirements by narrow syntax were chosen as a null hypothesis.

Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >

Related topics