There is a rather persistent confusion in current linguistic literature that I would like to highlight, between what syntax is (the component of grammar responsible for those computations that are not reducible to sound and to meaning) vs. what syntax looks like (constituents, phrases, complementation, specifiers, c-command, etc.).
One can see this confusion in action in the very existence of proposals where PF and LF “negotiate” with one another not through syntax but through some other channel. I will mention a couple of proposals of this sort shortly, but let’s concentrate first on the architectural matter. Syntax is not defined (or shouldn’t be, at least) as the study of phrases, constituents, etc.; it is defined as the study of everything that does not take place exclusively at PF or exclusively at LF. We might have certain preconceptions about what syntax looks like. And these preconceptions might be backed, at this point, by decades of prior research. But to the extent that someone makes a convincing case that PF and LF need to communicate with one another in a language different than the language of features, constituents, and phrases, what we have is none other than an argument that our preconceptions concerning the elementary vocabulary of syntax need to be revisited.
I must confess, I don’t even understand what the alternative to this is supposed to look like, if I think about it for more than a few moments. The whole point of spellout, to PF and to LF, is that the latter modules have their own proprietary vocabularies, and spellout consists in churning the formerly-syntactic representation, perhaps in several steps, into a representation suited to what the particular extra-syntactic module is about (articulation/perception on one side, meaning on the other). That means that PF and LF are, by the time such cross-modular “negotiations” would purportedly arise, holding onto disparate and potentially irreconcilable representations. To coordinate between PF and LF, presumably one needs a notion of a given object, such that whatever happens to that object at PF will depend on what happens to it at LF, or vice versa. But it’s not clear that the notion of a given object (extant and identifiable) in two incommensurate representations is a logically coherent one.
Note that even if one invents some machinery to overcome the commensurability problem (post-syntactic indexation or what have you), this would address the technical matter in the last paragraph, but not make so much as a dent in the bigger, architectural matter discussed in the penultimate one. Moreover, it is far from trivial to get such a mechanism to actually work, given that spellout to PF can (and often does) pick out a different granularity of syntactic material than spellout to LF does (see here).
So, now is the point in the post where I need to show that I have not constructed a complete straw man. Let me therefore discuss two particular examples of the kind of “extra-syntactic” PF-LF communication I’m talking about. One briefly, the other in a bit more detail.
The first example is Bobaljik & Wurmbrand (2012, LI) – henceforth, BW12 – who propose a (violable) constraint, “ScoT” (for “Scope Transparency”), which favors alignment between X precedes Y at PF, and X scopes over Y at LF. It should already be clear that this requires a given ‘X’ or a given ‘Y’ to be simultaneously identifiable at PF and LF, with all the attendant problems this brings, as discussed above. But crucially, even if all these technical hurdles were somehow cleared (and it is not obvious to me that they can be), we would still be left with the following irreducible fact: BW12 is a proposal for how restrictions on possible PFs map onto restrictions on possible LFs and vice-versa. There is a name for what they are studying, then: syntax. Insofar as their proposal requires a vocabulary that is not in the standard minimalist’s syntactic toolkit, it is an argument for revising that toolkit. It is certainly not an argument that PF and LF have a syntax-bypassing channel of communications; that, given the premises above, would not be a coherent claim.
The second example concerns the status of “roots” in Distributed Morphology. If one considers the Halle-Marantz canon (as summarized in Harley & Noyer’s 1999 State-of-the-Article on DM, available here; see especially the diagram on page 2), there is a line from the postsyntactic portion of the PF branch to the postsyntactic portion of the LF branch. One important reason for this (maybe the only reason? I’m not sure) is the status of roots. In this version of DM, the idea was that roots are entirely opaque to syntax, and thus, for syntactic purposes, there really was only a single ‘root’ object (since, were there to be more than one, syntax would have no way of individuating them from one another anyway). But this creates an immediate problem: what prevents a syntactic derivation in which I utter Sprocket is short, but arbitrarily select, on the LF side, the encyclopedic content corresponding to SMART rather than SHORT for the (single, invariant) adjectival root within the predicative adjective? Clearly, natural language doesn’t work that way, but the basic DM machinery, unadorned, coupled with the assumption that there is only one root object, doesn’t seem to be able to block it. So, one “solution” is to let PF and LF talk to each other extra-syntactically: PF “tells” LF which root entry was selected at PF, so that the corresponding encyclopedic entry can be inserted on the LF branch.
If you’ve been following along, you know what comes next. First, the technical matter. PF “tells” LF, how exactly? What informational primitives does PF deal in that LF also deals in, that can be used to communicate the relevant information? (If you’re thinking to yourself “indices on roots,” stay tuned!) But even if this technical issue could be overcome, the matter remains that this “postsyntactic” PF-to-LF arrow is, well, an arrow between PF and LF. It is therefore (definitionally) a piece of the syntactic apparatus, whether that makes us happy or not (vis-à-vis challenging our preconceptions of what syntactic primitives are / can be). Syntax is not defined by its informational primitives, it is defined by where it is situated in the modularization of grammar; the DMian PF-LF arrow is a piece of syntax; the diagrams that suggest otherwise are an equivocation.
So here’s one consequence of this: there isn’t actually a version of DM where roots are not individuated in the syntax (remember: judge what a proposal actually is, not what it says it is). That’s because even in the DM canon, the PF-LF arrow we’ve been talking about carries some kind of information wherein roots are individuated from one another. And that arrow is, de facto, part of syntax. And so roots are individuated in the syntax. Period.
That’s interesting to note, because, there has been plenty of arguing in the morphosyntax literature of late on whether roots are or aren’t syntactically individuated. (DISCLAIMER: I would absolutely not call these arguments “spilled ink” or anything of the sort; they contain profound, new, fascinating observations and insights, and are extremely valuable to the field. What I am challenging here is a particular corner of the set of theoretical implications that these insights are used to argue for and against.) Thus, Harley (2014a,b) uses verb root suppletion triggered by internal arguments in Hiaki (Uto-Aztecan) to argue that roots do have to be individuated in the syntax, each carrying its own index. In a recent manuscript, Caha, De Clercq & Vanden Wyngaerd (https://ling.auf.net/lingbuzz/004391) argue that there is a way, based on a particular theory of spellout, to account for Harley’s facts without using the kind of indices she suggests.
As should be clear from the above discussion, this back and forth (viewed through the narrow prism of “are roots individuated in the syntax?”) is based on a fallacy. There is no way to prevent Sprocket is short from meaning SPROCKET IS SMART without individuating roots in the syntax, once one comes to terms with what defines ‘syntax’. There is plenty of room to argue, of course, about whether roots need to be individuated in the lexicon (or in List A, as it were). We could therefore take Harley to be arguing that roots come into the syntax already individuated; whereas Caha, De Clercq & Vanden Wyngaerd are arguing that we can maintain a lexicon (or List A) with just one root object, and have roots individuated derivationally – for instance, by having an index added somewhere in the course of the syntactic derivation, e.g. right before the PF-LF split. The point remains, however, that given a principled definition of ‘syntax’ – one based on its place in the modularization of grammar, rather than on the contingent facts of what we’ve discovered is the best model of the relevant computation, so far – there can be no negative answer to the question of whether roots are individuated in the syntax.
So here were two examples of proposals that have posited “extra-syntactic” communication between PF and LF – a notion that I argued, above, should be eliminated – one from BW12, and another from the DM canon (and associated debates). I think these proposals do not wield the term ‘syntax’ coherently, and confuse what syntax is (logically speaking), with what (we think) syntax looks like. And I think it is beneficial to distinguish the two. I think it could help clarify certain aspects of our (overall) theory of grammar, and potentially even help streamline certain debates in the literature.