Here’s a nice little thing to ponder. Suppose syntactic computation is bound by some cyclic domains – we can call them “phases” (though I am on record with the opinion that phase theory is largely a facsimile of Subjacency, and many of its more recent developments, a facsimile of Barriers). And suppose that we encounter a configuration like (1) in which H is able to access XP but unable to access YP, where XP and YP both seem, as far as we can tell, featurally suited to H’s needs:
(1) H ≫ XP ≫ YP (where ‘≫’ represents asymmetric c-command)
In principle, the explanation could be phases: there is a cyclic domain that includes YP but not XP (and not H), and that’s what prevents H from talking to YP.
But quite often, that is not the route that is taken. Instead, we appeal to minimality. (I’m going to call this ‘minimality’, and not ‘Relativized Minimality’, even though what I mean is feature-relativized minimality; I do this to make clear that the theory being appealed to is not the theory in Rizzi 1990; but the phenomenon in question is the same one, and often goes by the name ‘relativized probing’ nowadays. See Bejar 2003, Nevins 2007, Preminger 2014, and others.) On this kind of account, there is nothing about YP’s position as such that prevents H from accessing it; it’s only the presence of XP that obstructs a relation between the two.
Based on this description alone, you might think that the two possible accounts (phases and minimality) would be easy enough to distinguish, empirically. Just remove XP, and see if YP becomes accessible or not! In practice, though, things are not always that easy. Suppose you were looking at finite T’s inability to agree with the direct object in lieu of the subject, in some language. Is it because the subject intervenes(=minimality), or because (some category in) the transitive verb phrase is a phase? The answer of “just remove the subject and see” doesn’t quite work here since, according to some of the literature, the phasehood of the verb phrase covaries with the presence of the subject. So if removing the subject (e.g. by switching to an anticausative) facilitates agreement with the object, is it because the subject no longer intervenes, or because the verb phrase is no longer a phase?
One can imagine a more general version of this question. How many of the results accounted for by minimality can be restated in terms of phasehood, or vice versa? At the limit, one might envision a complete reduction of one of these to the other: a world in which there is no minimality per se, only phases; or a world in which there are no phases, only minimality.
The proposal of Abels 2003 (later rejected in Abels 2012), where phase heads contain an instance of every feature available in the language, might be a way of deriving phasehood entirely from minimality – since for any probe located outside the phase and seeking feature f, there will be an instance of f on the phase(-head), which is closer to the probe than anything inside the phase. I won’t pursue this any further here, though. The question I’d like to turn to is the converse one, namely, whether we can get rid of minimality entirely and recoup its effects via phasehood alone.
Before turning to my eventual answer (“no”), there’s one thing we need to acknowledge: if we sufficiently reduce the size of phases, then, at the limit, the two explanations become identical. Here’s what I mean: if literally every maximal projection is a phase (Müller 2004), then phasehood is minimality. If XP asymmetrically c-commands YP, then at minimum, there is some head W that c-commands YP but not XP; and since WP is by hypothesis a phase, then both phase theory and minimality predict YP to be inaccessible. So the question posed above is only interesting to extent that there are maximal projections that are not phases. This assumption will figure in what I have to say below, so I wanted to flag it here.
I think a useful empirical domain with which to probe this question (no pun intended) is two-place unaccusatives, a.k.a. applicative unaccusatives. As Albizu (1997) and Rezac (2008) have shown, two-place unaccusatives in Basque come in two varieties, one in which the DAT argument asymmetrically c-commands the ABS argument, and one in which these hierarchical relations are reversed. Furthermore, probes located outside the verb phrase whose search criterion is just “find me a nominal” can access the ABS argument in ABS≫DAT verbs, but not in DAT≫ABS ones. Finally, evidence from clitic doubling suggests that even in those configurations where the ABS argument is inaccessible to such probes (i.e., DAT≫ABS configurations), the DAT argument is still accessible to some verb-phrase-external operations. In other words, it is not the case that for DAT≫ABS verbs, both internal arguments are properly contained in a phase that excludes higher verbal projections.
a. H ≫ ABS-DP ≫ DAT-DP (ABS-DP is accessible to H)
b. H ≫ DAT-DP ≫ ABS-DP (ABS-DP is inaccessible to H)
Given that the DAT argument in (2b) is still accessible to external operations (e.g. clitic doubling), the only way I can see for phase theory to account for the inaccessibility of the ABS argument is to assume that there is a phase boundary in between the DAT and ABS arguments in (2b). But what on earth could this boundary possibly be? It needs to be absent in (2a) (since, there too, the DAT argument is accessible to clitic doubling), and it needs to be lower than vP (since the DAT argument is probably introduced in Spec,ApplP, with ApplP as the complement of v; remember, also, that these are all unaccusatives we’re talking about here, both in (2a) and in (2b)). While it is technically possible to assume that, e.g., ApplP is a phase, this is now veering perilously close to a trivial theory of phasehood, where every projection is a phase, and the subsuming of minimality under phase theory is definitional. If ApplP is a phase, what isn’t a phase? And why? How do we know, and, more importantly, how does the child know?
Here’s another way of saying all this. If phasehood is nontrivial (i.e., there are at least some maximal projections that are not phases), then ApplP looks like a heck of a good candidate for a category that has no business being phasal. And if that is true, then there is no way to subsume (2a–b) under phase theory, meaning that, more generally, minimality cannot be subsumed under phase theory.
Personally, I think this is rather good news, for the following reason: from my vantage point, the theoretical “health” of minimality is much better than that of phase theory. That is to say, I think we understand the former way better than the latter. (Which is not to say that we understand minimality perfectly, of course; only that it’s less of a mess than phase theory.) So the fact that phase theory cannot replace minimality is good news, since we can continue to reason about effects like (1–2) in terms of something that is relatively well understood.
(Thanks to Patrick Niedzielski for a question that helped me think through this.)