This post is intended as an intellectual provocation. It is the strongest version of a thought I've had knocking around in my head for quite a few years now, but not necessarily a version that I'd be willing to formally defend. Therefore, I urge readers not to lose sight of the fact that this is written in a blog; it is as much an attempt at thinking "out loud" and engaging in conversation as it is an attempt to convince anyone of anything. [Norbert has helped me think through some of these things, but I vehemently absolve him of any responsibility for them, and certainly of any implication that he agrees with me.]
My point of departure for this discussion is the following statement: were the mapping from phonetics to phonology to morphology to syntax to semantics to pragmatics isomorphic – or even 100% reliable – there would be little to no need for linguists. Much of the action, for the practicing linguist, lies precisely in those instances where the mapping breaks down, or is at least imperfect. That doesn't mean, of course, that the assumption that the mapping is isomorphic isn't a valid null hypothesis; it probably is. But an assumption is not the same as a substantive argument.
If you disagree with any of this, I'd be interested to hear it; in what follows, though, I will be taking this as a given.
So here goes...
The last 15-20 years or so have seen a trend in syntactic argumentation, within what we may broadly characterize as the GB/Principles-and-Parameters/minimalism community, of treating facts about the interpretation of an utterance as dispositive in arguments about syntactic theory.
One response that I've received in the past when conveying this impression to colleagues is that all syntactic evidence is inexorably tied to interpretation, because (i) string-acceptability is just the question of whether utterance A is acceptable under at least one interpretation, and so (ii) string-acceptability is not different in kind from asking whether A is acceptable under interpretation X versus under interpretation Y. In fact, this reasoning goes, there really isn't such a thing as string-acceptability per se, since the task of testing string-acceptability amounts to asking a person, "Can you envision at least one context in which at least one of the interpretations of A is appropriate?"
I think this is too simplistic, since as we all know, there is still a contrast between Colorless green ideas sleep furiously and *Furiously sleep ideas green colorless. But even setting that aside for now, I don't think that the fact that an utterance A has at least one interpretation should be treated (by syntacticians) on a par with the fact that it has interpretation X but not interpretation Y. The reason is that the isomorphic mapping from syntax to semantics (or vice versa, for the purposes of this discussion) is a methodological heuristic, not a substantive argument (see above).
Let's illustrate using an example from locality. Evidence about locality can be gleaned in some instances from string-acceptability alone. That (1) is unacceptable does not depend on a particular interpretation – nor does it even depend on a particular theory of what an interpretation is (i.e., what the primitives of meaning are), for that matter.
(1) *What do you know the delivery guy that just brought us?
I therefore consider the unacceptability of (1) dispositive in syntactic argumentation (well, modulo the usual caveats about acceptability vs. grammaticality, I should say). On the other hand, the fact that (2) can only be interpreted as a question about reasons for knowing, not as a question about reasons for bringing, is not the same type of evidence.
(2) Why do you know the delivery guy that just brought us pizza?
To be clear, they are both evidence for the same thing. But they are not evidence of the same kind. And the provocation offered in this post is that they should not be afforded the same status in distinguishing between syntactic theories.
For the sake of argument, suppose we lived in a world where (2) did have both interpretations, but (1) was still bad. I, as a syntactician, would first try to find a syntactic reason for this. Failing that, however, I would be content with leaving that puzzle for semanticists to worry about. (Perhaps, in this counterfactual world, my semanticist friends would conclude that elements like why can participate in the same kind of semantic relationships that regulate the interaction between the logophoric centers of various clauses? I don't know if that makes any sense. Anyway, I won't try too hard to reason about what other people might do to explain something in a hypothetical world.) More importantly, I'd keep the theory of locality exactly as it is in our world. Obviously the other world would be a less pleasing world to live in. The theory of locality would enjoy less support in this hypothetical world than it does in our world. But the support lost in this counterfactual scenario would be circumstantial, not direct; it is the loss of semantic support for a syntactic theory.
There are (at least) two things you might be asking at this juncture. First, is this distinction real? Aren't we all linguists? Aren't we all after the same thing, at the end of the day? I think the answer depends on granularity. At one level, yes, we're all after the same thing: the nature and properties of that part of our mind that facilitates language. But insofar as we believe that the mechanism behind language is not a monolith; that syntax constitutes a part of it that is separate from interpretation; and that the mapping between the two is not guaranteed a priori to be perfect, then no: the syntactician is interested in a different part of the machine than the semanticist is.
Second, you might be asking this: even if these distinctions are real, why are they important? Why should we bother with them? My answer here is that losing sight of these distinctions risks palpable damage to the health of syntactic theory. Above, I noted that in research on syntax, evidence from interpretation should take a back seat to evidence from string-acceptability. But it feels to me like way too many people are content to posit movement-to-spec-of-TargetInterpretationP (or -ScopeP) without the understanding that, as long as the evidence provided is purely from interpretation, this is really just a semantic theory expressed in syntactic terms. (One might even say it is an 'abuse' of syntactic vocabulary, if one's point were to try and provoke.) This will end up being a valid syntactic theory only to the extent that the aforementioned syntax-semantics (or semantics-syntax) mapping turns out to be transparent. But – and this is the crux of my point – we already know that the mapping between the two isn't always transparent. (As an example, think of semantic vs. syntactic reconstruction.) And so such argumentation should be treated with skepticism, and its results should not be treated as "accepted truths" about syntax unless they can be corroborated using syntactic evidence proper, i.e., string-acceptability.