Wednesday, October 31, 2018

FL and the envelope of variability

I read a nice little paper that I would like to bring to your attention. The article(by Alison Henry) is ostensibly about Q-float in varieties of Irish English and it elaborates a point made in earlier work by Jim McCloskey (2000). Jim’s paper used the distribution of quantifiers floated off of WH elements to provide evidence for successive cyclicity (the Qs could be stranded in what are effectively phase edges (aka “escape hatches”)). You are all likely more familiar with this work than I am so I won’t review it here. Henry’s interesting paper makes three additional interesting points:

1.     The paper observes that the Q stranding provides evidence for vP as a phase edge as it seems possible to strand material in this position in several dialects.
2.     The parallel between the stranding facts in A’ movement configurations with that of Q float in A-movement configurations suggests that these should be treated uniformly andthis implies that Q-float in A-movement configurations piggy backs on movement as Sportiche originally suggested rather than these Qs being based generated like adverbials in vP edge positions for semantic reasons.
3.     It suggests an interesting way of parsing these Q stranding effects to shed light on what wrt the phenomenon reflects universal features of FL/UG and what is more G specific.

Like I said, this is a nice little paper and a quick and illuminating read. Before ending, let me say a word or two about the points above.

First, if correct, it provides as the paper notes, interesting evidence for the claim that vP is a phase edge. There is also counter evidence for this claim coming from Keine and Bhatt’s work on non-local agreement in Hindi (thx to Omer for bringing this to my attention). However the Henry data seems compelling, at least to me, and clearly points to something like a landing site under CP for A’-movement. At the very least this now sets up a nice research question for some ambitious syntax grad student: how to reconcile the Irish English data with the Hindi data. Good luck.

The paper’s second point also seems to me quite solid. If the stranding data under A-movement is a proper subset of that under A’-movement then it is hard to see what could motivate treating them as generated by entirely different mechanisms. In fact, theoretically, this would seem to me to be a disaster, invoking the worst kind of constructionism. Linguists have a habit of theoretically reifying surface data in their generative procedures. This leads to multiplication of G operations that have similar effects, which requires enriching the structure of FL/UG. This is a habit to be resisted, IMO. In fact, as a working hypothesis, I believe that we should standardly assume that FL can only do things in one way. There are never two roads to Rome! Henry’s paper shows how productive this kind of assumption can be empirically.

Third, and this IMO is the paper’s most interesting feature, it proposes a very reasonable view of one rich source of variation. Henry’s paper notes that we can find Qs stranded in anyphase edge (and base position) when we take the unionof all the dialects. No singledialect appears to allow Qs to surface in every position. Thus, it might seem as if each dialect has a different Q float mechanism. And, of course, in one sense this is correct. Each G must have somedifference or there would be no dialectal differences. However, as Henry’s paper argues, we can see this another way. The data points to the conclusion that FL/UG actually permits stranding in anyposition but specific LADs acquire Gs with further restrictions. In other words, FL/UG provides an envelope of possibilities that particular Gs further restrict. How? Via learning from the input. The paper makes the plausible point that PLD could fix the specific landing spots allowing the dialect specific G to use the FL/UG provided options as templates for where Qs could appear. This seems to me like a very reasonable idea and allows us to use the full range of variation as a window into the properties of FL/UG.

Two points: First, I have no idea how robust the Q float data are in the PLD and whether there is enough there to fix the various dialects.[1]However, Henry’s speculation can be tested. We are talking about data that should be easy to spot in a CHILDES data base for Irish English (if there is one).  One nice feature of this data: it will all fall into the domain of 0+learning (discussed here) and so be the right rain size to be acquirable via PLD. 

Second, the idea that Henry’s paper illustrates with Q float is one that others (e.g. Idsardi and Lidz and yours truly) have suggested for other syntactic phenomena.[2]We know that I-Merge generates copies in many places and which copy is pronounced should have an impact on surface order given standard linearization procedures. We can put these things together in Henryish fashion and note that what FL/UG provides via Merge is an envelope of possibilities that PLD then winnows out to provide some basic word order templates. On this view, FL/UG provides representations for the class of possible dependencies and PLD provides evidence for selecting among these possibilities wrt linearization. If this is correct, then specificlinearizations in specific Gs are not going to reflect much on the structure of FL/UG though the full range of typological options attested might well do so. At any rate, Henry provides a nice case study of the logic that Idsardi and Lidz were proposing more generally.

Enough said. Like I said, Henry’s paper is interesting and very well written and reasonably compact. Wish I had written it. 

[1]In fact, I have a sneaking suspicion that the range of variation might be more idiolectal than dialectal, but I really do not know enough to ground this suspicion.
[2]Eric Raimy and Lidz have suggested something like this for phonological phenomena as well. They argue that phonological structures are graphs, not strings, and so linearization is as much an issue in phonology as it is in syntax. If you haven’t read this stuff, you should take a look. It’s quite cool.


  1. I don't understand what any of this has to do with phases. Phases tell you where you have to stop; this data speaks to where you can stop. We already know that not every instance of the latter is an instance of the former. For example, there is good evidence that successive-cyclic subject-to-subject raising can stop at every intermediate SpecTP (in cases like John seems to appear to himself to have been awarded the prize). Nobody, I hope, takes this as evidence that non-finite TP is a phase.

    So the way to reconcile Keine's observation (from his 2017 paper in the Kyle Johnson festschrift) that there is no verb-phrase-level phase boundary with the observations about all-stranding is to recognize that the latter tell us nothing (or at least very little) about phasehood.

    The same, incidentally, apply to the arguments from reconstruction for the phasehood of this, that, or the other (as in, for example, Fox and Legate's arguments about the phasehood of transitive and non-transitive vPs, respectively). Those just show us where phrases can pass, not where they have to pass, and the theory of phases is a theory of the latter, not of the former.

    1. Not sure that I fully agree. First, the Spec T example you provide does not in itself show that there was movement via the intermediate spec T. As you know the relevant examples (due to Fox) are far more complex: John seems to the men to appear to himself to be.... Here it looks like you need something below 'the men' but above 'himself.' These examples, however, are not pristine and it is not clear to me that subbing a pronoun for the reflexive changes the status much. But I could be wrong. At any rate, that is the relevant example.

      As regards the argument for phases from this, I am more inclined to think that it shows something precisely in those cases where the 'all' is stranded in an intermediate (non-local) vP edge. Why would that ever happen but for some restriction on cyclicality? I have nothing against thinking that there is optionality, but this is very weird. Last point, as you know, to the degree that we can attribute this to phases it would be a net theoretical plus. I know you think that there is data against this, but maybe there is another way to have one's cake and eat it.

      Here is a suggestion: Can we say that the Keine&Bhatt data indicate that v is not a phase head in Hindi? We have had theories where what is ab funding node is language relative, son why not phase heads. The issue then would be if there is anything analogous to 'all' stranding at v edges in Hindi. My recollection is that there might be (don't WHs sit at what look to be v edges?). I also dimly recall that one needs a copy at every phase edge when there is long distance movement (or maybe I am wrong here). At any rate, if one found that one needs an unbroken string of these then the phase-hood of the v would be better established.

      SO, yes, you are right that we can assume v is not a phase at all, but then we are stuck with having to explain why things stop there. IMO, one EPP problem is more than enough. But, if you are ok with these, then yes, this is not dispositive.

    2. Yeah, so, we agree for the most part on the logical lay of the land, we just disagree (as usual ��) on which way we think we should therefore proceed. I can't answer your question of how robust the evidence is that you need vP phases for other aspects of Hindi syntax. But I am deeply skeptical of a parametrized theory of phases, and let me say why. You often talk about the "0+" PLD: the data available to the child from monoclausal utterances plus maybe the edge of embeddings, but not more than that. I think figuring out whether vP is a phase or not from 0+ PLD, especially given the "weak PIC" (a.k.a. "PIC2"), is going to be quite a challenge.

      Lastly, a nit to pick about John seems to the men to appear to himself to be...: This example indeed requires a position for John above himself and below the men, i.e., in the infinitival SpecTP of the clause anchored by appear. In my judgment, replacing himself with him here is quite bad. BUT, even if it is good, it doesn't change my point: that SpecTP is a position where you can stop (as the well-formedness of himself attests), and is not a phase edge on anybody's theory.

  2. Omer: "I don't understand what any of this has to do with phases" --- well, movement by one fell swoop would seem to be the zero hypothesis, given that the most direct evidence on the surface is just for the filler and the original gap, and phase theory unlike that zero hypothesis predicts an intermediate trace at the vP edge. So I would say it has something to do with phases.

    In the other direction from one fell swoop, a SLASH alternative potentially predicts intermediate reconstruction sites at every point along the path between the filler and the gap. Abels and Bentzen 2009 argue that intermediate traces are *not* available at every point along the path, where they can be detected at the vP edge, scoring an additional point for phase theory against alternatives.

  3. Concerning Keine's observation about Hindi long-distance agreement, couldn't the higher v be copying phi features from the next v down? Finite C doesn't agree, explaining why you can't agree this way across a finite CP boundary.

    1. Agreement via v is what I suggested for Icelandic long-distance agreement in my 2004 paper 'On the edge'

  4. Peter, Bhatt (2005) had arguments that this is not how LDA in Hindi works.

    1. It also wouldn’t work for Basque (see my 2009 LI and 2011 NLLT papers).

    2. Okay, I checked Bhatt (2005) and you’re right, he has an argument against LDA being v-to-v; but he also says that LDA only holds across restructuring clauses, which quite plausibly have a defective v --- in Wurmbrand’s analysis they can even lack v altogether. So the agreement data hardly suggests that Hindi vP is not a phase in general, only possibly that restructuring vP isn’t.

    3. Yeah, but Keine also has arguments from the licensing of wh in situ, which is not sensitive to restructuring.

    4. Yes, we’ve had a discussion about wh-in-situ before; it is a problem in more than just Hindi.