Phi features, binding, and A-positions

Preface

This post continues a theme started here and here. Broadly, this series of posts is an attempt to highlight the daylight that exists between syntax and semantics.

I have several motivations for writing these posts. First, writing them, and reading & replying to comments, really helps me sharpen my own thinking on the issues. (Whether I’m convincing anyone but myself is a separate matter, of course.) Additionally, though, it is my impression that when it comes to the syntax-semantics mapping, the working assumption that the mapping in question is transparent – a wholly legitimate research heuristic, of course – is in practice often elevated to the status of ontological principle. This, in turn, licenses potentially problematic inferences about syntax. And it is these cases that I wish to highlight.

I hasten to add that I’m not sure there’s anything different in kind here from what goes on in any other “interface” work. That is, I don’t mean to impugn syntax-semantics work in particular (as opposed to, say, syntax-morphology work or whatever else). It’s just that the particular syntax-semantics inferences I’m talking about are ones that I often bump up against in my own work, and I often get the feeling that they are accorded the status of “established truths” – which places the burden of proof on any proposal that would contradict them. It’s this view that I’d like to challenge here.

Finally, for interesting discussions pertaining to the substance of this post in particular, I’d like to thank Amy Rose Deal – who should not, of course, be held responsible for any of its contents; in fact I’m fairly sure she would disagree!

Okay, let’s get to it...

––––––––––––––––––––

What is an “A-position”? Originally, the ‘A’ was supposed to be a mnemonic for “Argument” – the idea being that an A-position is any position that could, in principle, introduce arguments. A particular set of properties was then shown to correlate with being in, or moving to, an A-position. Most important for our current purposes are the binding-related ones: A-positions were the positions from which one could antecede novel binding dependencies. Hence the well-known kind of asymmetry between (1a) and (1b):

(1)
a. Every smartphone_i seems to its_i owner [to be t indispensable].
b. # Which smartphones_i did their_i owners like t the most?

The difference between (1a) and (1b), phrased in these terms, is that the binder (DET smartphone) in (1a) occupies at least one A-position that c-commands the bound pronoun (its), whereas the highest A-position occupied by DET smartphones in (1b) – the object position of like – still fails to c-command their.

So far so good. But an embarrassment for the concept of “A-position” soon emerged: the canonical subject position in finite clauses – e.g. the A-position responsible for successful binding in (1a) – was quite possibly never an “argument position” in the above sense. With the discovery that even so-called underived subjects were in fact generated within the maximal projection of their predicate (the Predicate-Internal Subject Hypothesis, or in less general terms, the Verb-Phrase-Internal Subject Hypothesis), and only later move to canonical subject position, it was no longer clear that the latter was ever a site where any arguments were first introduced into the derivation.

We therefore now had a putative theoretical category – “A-position” – but we no longer had a way of deriving the set of entities that did and didn’t fall within the purview of that category. Neither SpecIP nor SpecCP ever introduced base-generated arguments; what distinguished them from one another for the purposes of effects like (1a-b) and many others?

In the literature from the late 80’s onwards one finds a variety of attempts to answer that question. One possibility, which I think we can discard quite readily, tries to tie the A-position property to case. The idea would be, roughly, that movement for case purposes yields A-position properties, while (some? all?) other types of movement don’t. There are many potential reasons to reject this view. To mention one that I will not be dwelling on here: intermediate steps en route to a purported “case position” still give rise to A-position properties. This requires a problematically teleological view of the movement in question, one that is a very ill fit for any theory of syntax that invokes notions of cyclicity and locality. But the problems for the case-based view are actually far worse. We know by now that case is not a property of positions, it is a property of certain configurations. (This is not a partisan endorsement of configurational theories of case, though I do endorse those; “c-command by <designated head>” is also a configuration, not a position.) And what is also clear is that the configuration for what we call ‘nominative’ does not require movement to SpecIP (cf. Celtic-type VSO languages, for example). Whether this is because nominative is assigned under c-command by T (Chomsky 2000), because nominative is unmarked case (in the Marantz 1991 sense), or because ‘nominative’ is nothing more than a descriptive label for the absence of valued case features (Kornfilt & Preminger 2015), it is clear that there is no need to move to SpecIP to be ‘nominative’, hence dissolving the case-based explanation for what an A-position is.

Before fully discarding the case-based alternative, though, it is worth pausing to note a potential benefit that it would have delivered, had it been true: by and large, A-movement seems to be a property of nominals, whereas A-bar movement seems available to a wider variety of syntactic categories (perhaps all of them). I say “by and large” because there are fairly clear counterexamples – the psych-verb version of escape for instance is a two-place unaccusative, and yet its Theme argument can surface in canonical subject position even when it is clausal, as in (2). [Also, some analyses of locative inversion involve A-movement of the locative. Others, though, take this to be A-bar movement.]

(2) That there is no hope for abstract case escapes them.

Overall, though, it seems that A-bar movement is less categorically “picky” than A-movement. I bring this up because it leads quite naturally to our next contender: if not case, then what else do nominals possess and other categories lack? Phi features.

[I’ll note in passing that a feature-based approach to the A/A-bar distinction has another clear advantage, involving instances of “mixed” A/A-bar behavior, which can then be construed as involving a mixture of different kinds of features (see Webelhuth 1990, van Urk 2015, and everything in between). Crucially, though, that is a property of any feature-based approach to the A/A-bar distinction, and not just those based on phi features in particular.]

In addition to being something that nominals have and other categories lack, there is another attraction to phi features as the nexus of binding – namely, that in many cases, one sees phi-feature matching between antecedents and their bindees. Now, suppose you believed that binding arises via syntactic phi-feature agreement; it would be natural to bootstrap the notion of “A-position” to the same syntactic mechanism.

There are, I think, two major variants of this view:

Variant One:
XP is in an A-position iff XP is in the specifier position of some head F and XP enters into syntactic phi-feature agreement with F (cf. Kratzer 2009, inter alia)

Variant Two:
XP is in an A-position iff movement of XP to its current position was triggered/preconditioned by syntactic phi-feature agreement (cf. van Urk 2015, inter alia)

For the present purposes, however, these variants can be run together, since they both entail the following corollary, which will be our focus:

Phi-Agreement Corollary (henceforth, PAC):
If XP is in an A-position, XP has participated in syntactic phi-feature agreement.

Thus, for example, if small-clause ECM subjects in Hebrew can bind local reflexives, it follows from PAC that they must have entered into syntactic phi-feature agreement:

(3)
ani tamid hexʃav-ti et Yosi buʃa le-acmo
I always considered-1sg ACC Yosi embarrassment DAT-himself
“I always considered Yosi an embarrassment to himself.”

And here is where the rubber meets the road: the conclusion that objects (and/or small-clause subjects) in Hebrew enter into syntactic phi-feature agreement is at odds with the facts. It’s obviously false as far as overt morphology is concerned; but it is also false, I contend, as it pertains to ‘abstract’, phonologically-null agreement.

The argument comes from the behavior and distribution of PCC effects. In a nutshell: the PCC is a syntactic effect par excellence (in that it cannot be reduced to “morphology” under any contentful definition of how “morphology” differs from “syntax”), and yet it comes and goes with the presence of overt agreement morphology – both crosslinguistically and, more strikingly, even intra-linguistically. Since syntax is not supposed to know whether a particular term will end up overtly exponed or not, it must be that the relevant terms (those associated with syntactic phi-feature agreement) are simply not there when you cannot see them. Thus, there can be no agreement with Yosi (nor with any coreferential null category) in (3), not even abstractly. I have called this the “no-null-agreement generalization.” [If you are interested in the argument in more detail, you can see it for yourself here; but I don’t mean for this post to serve as some kind of promotional pamphlet for that paper; for purposes of this post, you can just as well skip the details.]

So, both Variant One and Variant Two of the phi-agreement approach to A-positions entail PAC, which in turn runs afoul of the no-null-agreement generalization. And so, “A-position” cannot be about syntactic agreement in phi features.

––––––––––––––––––––

What gives? Something, obviously, must.

It could be that there is something wrong with the argument against PAC (e.g. the no-null-agreement generalization is actually wrong); or that there is some “Variant Three” I haven’t noted here; or any other logically-possible avenue for salvaging the premise that A-positions are about syntactic phi-feature agreement.

But the answer, I’d like to suggest, is quite different. It seems to me that the original sin here is in identifying “binding indices” with “morphosyntactic phi features” in the first place. Sure, the two tend to swing together; but the degree to which they coincide has, I think, been vastly overstated. Everybody is familiar with the PP problem, where the binding index seems to reside at the PP level even though there is no evidence of morphosyntactic phi features anywhere above the DP – as is the case with [to him] in (4):

(4) It seems [to him_k/*i] that John_i is on time.

One could attempt to bring these data back into alignment with the collapsing of binding indices and morphosyntactic phi features by assuming, e.g., agreement between the P head (to) and its complement, thus “percolating” the phi features to the PP level. (That would run afoul of the aforementioned no-null-agreement generalization, but let us set that aside for now). Nevertheless, this solution would not work in the general case. As an illustration, consider Basque reflexives. These are formed in a way that seems fairly unremarkable: [PRON.GEN N D], where ‘PRON’ is a pronoun whose phi features track with those of the antecedent, and ’N’ is a designated noun buru (meaning ‘head’, in line with a common crosslinguistic strategy of forming reflexives from possessed body-part nouns). Now, it is (unsurprisingly) the outer DP (i.e., [PRON.GEN N D]) that behaves as an anaphor (e.g. obeying Condition A of the Binding Theory). But given that Basque has overt phi-feature agreement with the absolutive argument, we can directly test the validity of a “percolation”-style solution – i.e., we can test whether the entire DP behaves as a bearer of a binding index by virtue of the morphosyntactic phi features on the genitive-pronoun possessor percolating up to the outer DP layer. And the answer is negative:

(5)
ni-k [ene buru-a] ikusi dut/*naut
I-ERG [my head-DET] seen have.[1sgERG>3sgABS]/*have.[1sgERG>1sgABS]
“I have seen myself.”

That is, the “1sg reflexive anaphor” ene burua behaves as if it is a bearer of 3rd person singular phi features (as any possessed instance of ‘head’ would), showing that the morphosyntactic phi features of the possessor pronoun cannot have “percolated” up to the outer DP level. (See Georges Rebuschi’s chapter in the 2003 Hualde & Ortiz de Urbina volume for a much more detailed discussion.)

The bottom line here is that whatever syntactic mechanism – if any – tracks with the distribution of semantic indices, morphosyntactic phi features ain’t it.

And this is a welcome result, given that we saw before that approaches that tie the notion of “A-position” (and hence, the notion of “potential binder”) to syntactic phi-feature agreement entail PAC, and PAC is not correct.

––––––––––––––––––––

Okay, I’ll stop here. As always with these blogposts, there’s more to be said, and there’s a fairly vast literature that I’ve only scratched the surface of here. Importantly, though, I am currently under the impression that the substantive points made here remain intact even in the harsher light of that literature’s full glory – so if you’re reading this and think that this assertion is wrong, please let me know in the comments (as I’m sure you will).

To state explicitly what I take the bottom line to be: assuming that binding indices and/or “A-positions” are tied to the distribution of morphosyntactic phi features leads to a result (namely, PAC) that requires null agreement wherever there’s an A-position that is not associated with morphologically overt phi-feature agreement. That result stands in contradiction to what emerges from the study of phi-feature agreement, namely the no-null-agreement generalization. It’s possible that the latter is actually wrong; I’m certainly extremely biased. [I will point out, though, that this is not some “the reviewers were mean to me” rant; the paper has been accepted for publication, and the issues I discuss here were not the main thrust of the reviewers’ comments.] But let us examine more closely where the purported tension between the two proposals comes from: it rests on the assumption that binding indices track with morphosyntactic phi features. That is a good null hypothesis; it also appears to be false (see the discussion of Basque, above). So any purported tension between the no-null-agreement generalization and the theory of A-positions depends on elevating the heuristic of “transparent syntax-semantics mapping” to the level of ontological principle. But, as a blanket principle, it does not seem to accord with our current understanding of the facts.