Desktop version

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

Adjunctions and labels

Invisibility for the purpose of labeling, together with lack of featural interactions with the host, make adjuncts seem to be ‘on a different plane’ as far as syntactic relationships are concerned; it also exempts thereby such structures from the filtering effect that labeling has on syntactic objects. To be sure, the adjunct as such needs to be labeled, but the adjunction structure as a whole does not have its interpretive viability checked by the labeling algorithm. There two consequences of such an approach to adjuncts relevant for the present discussion. First, creation of an adjunction structure suffices for a structure {{XP}, YP} to go through the labeling gate and enter the kingdom of C-I processes. This is consistent with the current understanding of labeling as searching for appropriate configurations and not modifying a structure with label annotations, but it goes farther than in cases falling under the labeling algorithm as studied for structures created by single applications of merge (set-merge, in other words), in which the relationship sought by the labeling mechanism involves featural makeup (possibly even exclusively featural content, as hypothesized in Chomsky (2013c: 45)): all that remains in the case of {{XP}, YP} structures is to check that the structure—whose derivational history is kept in the phase level memory—satisfies demands of minimal search (ultimately ‘kind of a representational construal principle’ (Chomsky 2015a: 89)) and thereby conforms to the requirement that the relationship so established be in accordance with the principle of minimal computation, the central property hypothesized for the language faculty (Chomsky 2016a). Intuitively, the labeling algorithm ensures in this case that the structure may possibly undergo procedures of interpretation along the lines outlined above, with the host being interpreted with respect to the adjunct, the latter internalized by the interpretive process as constraining parameters of evaluation. It does not, however, ensure that such procedure is in fact applicable. Merge being free, narrow syntax generates various kinds of structures of which only some may pass the labeling test; yet the latter is not a procedure guaranteeing that a structure will not be interpretively deviant. A second consequence of the view on adjunction discussed above is that it opens the way for more varieties of interpretive deviancy than other applications of merge do, the latter being constrained by feature-based relationships. This is consistent, though, with the general stance towards various kinds of linguistic deviancy as it gradually took its shape once the notions of grammaticality and acceptability became clearly separated in Chomsky (1965) and it came to be more frequently and explicitly noted that background concepts taken over from the study of formal languag- es—in particular, the notion of language as a set of well-formed expressions, with the grammar being ‘a device that generates all of the grammatical sequences of L and none of the ungrammatical ones’ (Chomsky 1957: 13), and the concomitant notion of weak generation—could be used merely as tools to study formal properties of certain classes of grammars (in the sense made explicit in the preceding quote), without having an official place in the proper generative theory of language, as it was repeatedly stressed in the GB period and at the dawn of the minimalist era in Chomsky (1980, 1986b, 1993):

One might be tempted to interpret the class of expressions of the language L for which there is a convergent derivation as “the well-formed (grammatical) expressions of L.”

But this seems pointless. The class so defined has no significance. The concepts “well- formed” and “grammatical” remain without characterization or known empirical justification; they played virtually no role in early work on generative grammar except in informal exposition, or since. (Chomsky 1993: 44 n. 7)

That did not mean even as much as a shift in perspective, far less a more fundamental change in approach, for the explanatory inadequacy of weak generation and related concepts had been known and discussed from the beginning of the generative period—one of the points of disagreement between Chomsky and Quine, the latter famously denying relevance of the distinction between weak and strong generative capacity and accepting extensionally equivalent systems of rules ‘fitting’ the observable linguistic behaviour as the limit of inquiry in this area (Quine 1972), the former insisting that ‘Quine’s sole point amounts to the observation that there will always be distinct theories that are compatible with all the evidence at hand. That is true if we restrict ourselves, as he sometimes unreasonably suggests, to weak generation of sentences’ (Chomsky 1976: 314)—and concepts and techniques associated with weak generative capacity used to serve purely technical purposes (as in Chomsky (1956, 1975b) and in the extensive discussion in Chomsky (1955-56: § *57), which did not find its way into the published version), but more unequivocal statements made no room for confusion as to the place of such notions and their being inappropriate for the study of linguistic competence (as Chomsky (2016b: 130 n. 4) speculates, ‘A source of misunderstanding may be that in early work, “language” is sometimes defined in introductory expository passages in terms of weak generation, though the usage was quickly qualified’; see further remarks in Chomsky (2015a: 83-87) and Fukui (2015)). Indeed, the very notion of ‘deviancy’ may be used only as an informal qualification of various expressions which are generated by narrow syntax:

Merge can apply freely, yielding expressions interpreted at the interface in many different kinds of ways. They are sometimes called “deviant,” but that is only an informal notion. (...) expressions that are “deviant” are not only often quite normal but even the best way to express some thought (...) That includes even expressions that crash, often used as literary devices and in informal discourse, with a precise and felicitous interpretation at the interfaces. The only empirical requirement is that SM and C-I assign the interpretations that the expression actually has, including many varieties of “deviance” (Chomsky 2008: 144)

Although the mainstream conception of grammar rejects its being ‘crash-proof’ (as postulated by Frampton and Gutmann (2002), Putnam and Stroik (2010); Stroik (1999, 2009); Stroik and Putnam (2013); see Ott (2010), Boeckx (2010a), Chomsky (2004a: 111)), the labeling algorithm does have a filtering effect on the output generated by narrow syntax, in accordance with the conception of labels as required for interpretive purposes, and not for syntax-internal reasons: structures for which the labeling algorithm incurs a halting problem, as in pure {XP, YP} objects without a featural relationship between X and Y, do not get a label due to the indeterminacy of the source of a label, failing to meet requirements of minimal search, and, given that labeling is part and parcel of the computation at the phase level immediately preceding delivery to the interfaces, it stands to reason that they never reach interpretive components. Adjunction structures are acceptable for the labeling algorithm because the adjunct is made invisible for its search mechanism, and the host is thereby uniquely specified as the part of the structure responsible for its interpretive behaviour in the C-I component; they may therefore be transferred to the interfaces. This does not necessarily mean, though, that the adjunct is susceptible of an interpretation properly integrating it into interpretation of the host along the lines sketched earlier—as in cases due to type-theoretic mismatch—and they may therefore remain ‘on a separate plane’ forever, in the C-I component as well, thereby leading to one kind of interpretive deviancy which is allowed for by the labeling algorithm.

It is to be noted that from the perspective of the labeling-relevant operations, properties of {XP, YP}-structures which are labeled (are in a configuration, to be exact) as (F, F) and {{XP}, YP} adjunction structures, are to some degree analogous: in both cases the presence of both consitutents is required for interpretive purposes—in the former case, for the label to be determined as (F, F), in the latter case for the lack of label to be accepted by the labeling procedure. Thus, adjuncts and agreeing specifiers, to use the traditional terminology, actually are expected to be frozen for the same underlying reason: were they displaced, the configuration necessary for the labeling algorithm would be destroyed. The freezing effect due to the labeling process discussed by Chomsky (2015b) thus holds for adjuncts with an important qualification: whereas in the case of ‘specifiers’ the ban on displacement affects objects which have not yet undergone labeling (while internal merge of objects which have entered the relationship of feature-based labeling does not lead to labeling issues provided that (i) they do not enter the same type of relationship again—so that A-bar movement from an A-position is possible—and (ii) internal merge occurs while the label is kept in the phase level memory), for adjuncts, which are accepted by the labeling algorithm because of their internal structure, and not for feature valuation reasons, there is no real point of availability of internal merge and extending their chain; it is thus a consequence of the account of their properties outlined above that there are no actual cases of internal merge of adjoined objects beyond their self-merge. It is also worth noting in this context that the current perspective on the mechanism of labeling, its place in the derivational procedure, consequences for syntactic derivation and its import for interpretive purposes, invite a specific view on aspects of the island-related phenomena. While the framework pre-dating Chomsky (2013c) incorporated labeling as a condition on applications of syntactic operations, so that labeling was forced to take place before a syntactic object could enter into further syntactic relationships (a view mantained e. g. in Rizzi (2015a,b, 2016), Boskovic (2016, To appear)), the current view embraces free merge in its entirety, making labeling a prerequisite for interpretive operations (a kind of via media is proposed in Goto (2015a,b,c, 2016), who takes labeling to be required for the procedure of search, while mantaining the freedom of merge). Adjoined XPs are on this view perfectly visible in narrow syntax—except as sources of labels; their consitutents may therefore be supposed to have the freedom of undergoing internal merge, leaving the adjunction structure (unless the option is otherwise excluded by the PIC, of course) and, if they appear in a structure in a way obeying labeling requirements, their displacement seems possible—although island conditions would tell otherwise. It seems that the view on labeling taken in Chomsky (2015b) invites understanding of several isl and effects rather in terms of violations of interpretive principles than in terms of syntax-internal prohibitions—a possibility of reconceptualization of a large part of island constraints as interface-related argued for in much research (see an overview of various proposals in Boeckx (2012b)), and consistent both with the hypothesis of free merge and with a representational rather than derivational account of the nature of dependencies (whereas attributing to island constraints a derivational nature would force understanding them as constraining applicability of syntactic operations). In the case of adjuncts, the source of trouble at the interpretive, C-I side, may be tentatively put down to the difficulty with the interpretation of a chain involving occurrences; foreshadowing the account of chains we turn to in chapter 3, occurences of an object constituting a discontinuous syntactic object are interpretively linked by counterpart relation(s). In the case of extraction from adjuncts, they appear partly on the main structural skeleton and are partly absorbed together with the interpretation of the object they belong to into a restriction on the assignment function: this immediately causes a trouble with the local counterpart relation linking both parts of the chain, and also a trouble with the counterpart relation which connects occurrences on the mainland and on the island along the lines outlined in section 3.6. It may be expected that in cases in which such interpretive difficulty is considerably diminished, extraction from adjuncts will be easier or even entirely impeccable (provided that internal merge does not proceed from a domain made opaque for independent reasons)—as seems to be the case, witness effects investigated in Truswell (2011), which testify to the relevance of the kind and degree of interpretive integration of the adjunct and the host.

< Prev   CONTENTS   Source   Next >

Related topics