Confessions of an adjunct

Adjunction: a syntactic ugly duckling

Labeling adj unction: a challenge

‘Adjunction of YP to XP has had a central place in transformational generative grammar from its origins,’ (Chomsky 1995: 324) states, only to suggest immediately thereafter that, while properties of operations investigated under the umbrella of Move a and its predecessors and successors form the core of processes belonging to narrow syntax, those which belong to the realm of XP-adjunction ‘may not really belong to the system we are discussing here as we keep closely to the first of the two courses just outlined, the one that is concerned with Last Resort movement driven by feature checking within the N ^ X computation’ (Chomsky 1995: 325). The radical move to eliminate adjunction has not been pursued with much effort (save for attempts to assimilate adjuncts and specifiers in the antisymmetry framework of Kayne (1994) and related work, the cartographic enterprise of Cinque (1999) included to an extent), though, given the pervasiveness of adjunction structures in natural language expressions. Difficulties persist, though; the current theoretical setting, after elimination of the X-bar theoretic schema, cannot avail itself of the distinction between segments and categories of May (1985) and Chomsky (1986a); nor, once labels as parts of syntactic representations have been discarded, of differences in explicitly coded labels, as in Chomsky (1995), where, to recall, the label of an adjunction structure is stipulated to be an ordered pair (H(K), H(K)) expressing the segment-like build-up of the category K. The hypothesis that adjunct structures are left unlabeled, as per Hornstein (2009), Pietroski and Hornstein (2009) or Gallego (2010), although doing away with labels, would require further elucidations: given how the labeling algorithm operates, and what filtering role it fulfills in the current theory of the derivational process, such structures can not be marked as ‘immune’ to the labeling failure due to the lack of label, but rather as labeled under an appropriately conceptualized notion of labeling. The solution Chomsky (2004a) opts for is to have a distinct kind of the operation merge, pair-merge, producing as the output of adjunction of a to в an ordered pair (a, ft).

There may arise doubts as to whether the labeling algorithm, working at the level of a phase and inspecting the structure for its conformity with general rules of determination of the interpretive behaviour of its constituents, can properly handle adjunction structures and a fear that it will encounter a problem similar to, but not the same as, that found in so-called {XP, YP} structures: the impossibility of a unique determination of the object to be searched for labels, despite its being rather an ordered pair than a plain set. Given that the ordered pair device does not, contrary to what is standardly implicitly assumed, provide the means to indicate which of the two objects is to serve as the distinguished one for the purpose of label determination, the whole procedure might stop at the moment it starts to look into an adjunction structure. It would be not an insignificant enrichment of the theory of labeling if we assume that there is a built-in definition of an ordered pair which would specify which of the objects is the host and which is the adjunct; yet without such additional constraints, the labeling algorithm might not be able to decide which way to go upon the inspection of an (а, в) structure. Recognizing an ordered pair, it might take it either to be {а, {а, в}} or {в, {а, в}}. Both will do equally well as a set-theoretic representation of the pair (а, в), ensuring that it is an object different from the set {а, в} and that it satisifies basic properties of ordered pairs: (а, в) will be distinct from (в, а) unless а = в, and (а, в) = (у, 8) iff а = у and в = 8. All this, however, does not suffice for the purposes of the labeling algorithm and the determination of the behaviour of both objects constituting such an ordered pair. The asymmetry created by ordered pairs may thus seem to be not enough to identify uniquely one of such objects as the host and the other one as the adjunct, which is the crucial distinction for the interpretation at the interfaces, the C-I interface in particular, and, if constraints on the admissibility of certain syntactic operations are hypothesized to be operative in Narrow Syntax and to rely partly on such structural distinctions, also for the workings of the syntactic component; if the latter class of constraints is to be obeyed at the point of transfer to the interfaces, the halting problem of the labeling algorithm would make it impossible both to label a structure and to check it for correctness with respect to such constraints, with consequences which may be variously taken to encompass derivational failure and/or different kinds of interpretive deviancy, thus making all adjunction structures either strictly underivable or uniformly deviant, an empirically incorrect prediction. It should be stressed that the labeling algorithm does not require that the distinction between the host and the adjunct be determined by structure-building operations in a way which we may intuitively think ‘correct’ from the point of view of the interpretation in the C-I component; all that it requires is that it be established, one way or another. There is no more look-ahead in the case of adjunction structures than there is in all other cases of applications of structure building operations, taking place freely during processes leading to the phase level, without recourse to triggering features or properties required of syntactic objects resulting from such operations on some intended or distinguished interpretation in interpretive components—at least on the most sparse theory of the narrow syntactic component, which eschews imposing constraints on the applicability of syntactic operations. Free availability of structure building operations, beside being the minimal assumption about the requirements on the applicability of merge in its various incarnations, avoids the problems of lookahead and postulation of features which fulfill only the role of ensuring that the derivation proceeds successfully.

< Prev   CONTENTS   Source   Next >