Wednesday, October 2, 2013

What's Next? An Invitation.

Friday saw the launching of the new Maryland Language Science Center (here) at UMD. As part of the festivities, Colin Phillips asked various people to try and divine the next big questions. I was asked to prognosticate about syntax and accepted only because I was assured that the performance was capped at five minutes, thereby precluding any real opportunity for self-embarrassment. As this fear abated, I came to enjoy thinking the question through, at least a little bit, and as such I have decided to opine even more publically.  What I really hope is that my coming out of the intellectual closet will provoke others to do so as well. So here’s the plan: I would like to invite you to join me in thinking about “what’s next in X” where you fill in X with your own subdomain of linguistics broadly construed (e.g. phonology, Morphology, semantics, acquisition etc.). If you want to do so in the comments section, great. But, if you want a little more space to expatiate (and room for some notes), send me a 2-3 page Word document which I will happily post on your behalf on the blog (if I like it) and let you handle comments in any way you wish. 

This is the sort of thing that colleagues often do over a couple of beers at conferences and workshops, right after dissecting indispensible gossip and advancing the careers of deserving protégées. My proposal is that we go public and let everyone in on our hunches. What do you think are the questions we should be going after in the immediate future and why?  What are the problems that you see as ripe for the picking now and why (especially why now?) How do you see the intellectual landscape in linguistics in general or your sub-domain of interest over the next 5 years? 10 years?  These are the sorts of questions we (or at least I and my colleagues at UMD) have standardly asked those seeking tenure track employment here. It only seems fair that if we are asking these questions of others that we be ready to speculate ourselves.  So without further ado, here’s a few of my kickoff answers.

Let’s start with some basics. Syntax starts from the observation that grammatical structure has two sources: (i) the combinatoric system (viz. the rules) and (ii) the elements combined (viz. linguistic atoms).  IMO, Generative Grammar over the last 50 years has discovered a bunch of invariant “laws” about the combinatoric system that underlies human linguistic facility. The success of the program has been in finding and articulating the fine structure of these “formal” universals (e.g. bounding, binding, control, minimality, etc.).  Given this, here are some projects that I think should be on the immediate research agenda.

1.     Unification

Minimalism (MP), as I see it, has aimed to unify the disparate modules of UG. The project has been roughly analogous to what Chomsky (in ‘On Wh Movement’) accomplished wrt Ross’s theory of islands. This time, the “islands” are the modules of GB and the aim is to see how the laws of grammar proprietary to each module are actually manifestations of a very small number of simple principles. This project has been partially implemented. So for example, (beware: the following is tendentious in that it reflects my views, not necessarily the field’s consensus) Merge based accounts have unified the following modules: case, theta, control, binding (in part)[1], and phrase structure. However, large parts of GB remain outside this Merge-unified framework. The two obvious ones are Island/bounding theory and the ECP.  Of these, phases can be ornamented to incorporate bounding theory (i.e. there is a pretty easy translation between phase theory and the old subjacency account of islands). It has problems form a minimalist perspective, but it can at least be made to fit in essentially the same way that it fit in GB.

It is far less clear how to fit the GB ECP into a minimalist mold, especially the “argument/adjunct” asymmetries that were the focus of discussion for well over a decade.[2] It is still quite unclear, at least to me, how to code the differences between those chains that require a very strict licensing condition (roughly that provided by successive cyclic adjunct chains) and those that allow more liberal licensing (roughly those that allow traces in comp to be deleted without disrupting the licensing).  It is even unclear what the ECP is a condition on. In GB, it was on traces, null elements considered to be grammatically problematic and so requiring licensing of some sort. But with the Copy Theory eliminating traces, so within MP it is conceptually unclear why something like the ECP is needed. How (or whether) to unify the bounding and ECP modules with the others is, hence, a good problem for MPists interested in grammatical unification.

2.     Exceptions to discovered invariances

On a more empirical front, it is worth considering how much variation our GB principles tolerate. Why? Because where invariances break down, learning is required and learning needs data that is plausibly often absent. Let me explain by giving two examples.

It has been claimed that islands are not uniform cross-linguistically.  In the early days, Italian and English were distinguished wrt to their inventory of bounding nodes. Pretty quickly, however, we discovered that many of the sentences that allowed extraction out of apparent islands in Italian, were pretty good in English as well (Grimshaw noted this pretty early on).  Recently, Dave Kush has done the same for CNPC/Relative Clause (RC) violations in Swedish versus English. He has demonstrated pretty conclusively, IMO, that the acceptability of some RC violations in Swedish coincide with upgraded acceptability on analogous RC structures in English. In other words, whatever is going on in Swedish is not limited to Swedish but appears in English as well. This is a good result. Why? If it’s true, then it is what we would expect to be true given pretty standard PoS considerations: data concerning variability in extraction from RCs should be pretty sparse so we should expect variation here. If Kush is right, we don’t find it. Good.[3] 

Question: what of other claimed cases of variation, e.g. fixed subject constraints (aka, that-t effects)? Are they real or only apparent? This is worth nailing down and now is a good time to do this.[4] Why?

Well, I believe that settling these issues may require using slightly more refined methods of data collection that has been our syntactic habit. Kush, for example, did standard rating studies and used them to find that despite differences in absolute ratings between Swedish and English speakers, the relative improvements in the same contexts were comparable.  This is not fancy stats, but it is, in this case, very useful. Jon Sprouse used similar methods to provide further evidence for the grammatical “reality” of islands.  At any rate, these methods are easy enough to apply and deciding how much variation there actually is among the principles described in the GB modules is important for PoS reasons.

3.     Substantive Universals

The GB achievements noted above concern formal universals. From where I sit, we have had relatively little to say about substantive universals.  Recently, due to the work of Cinque and colleagues, it has been proposed that there is a small(ish) inventory of potential functional heads available for grammatical use and from which grammars can select. More interesting still is the proposal that the order of these heads is invariant. Thus, the relative hierarchical position of these functional heads is constant cross linguistically, e.g.. T is always above v and below C. Of course, were it only C, T, v then this might not be that interesting. However, Cinque has considerably fattened up this basic inventory and has provided reasons for thinking that the invariance extends to the position of adverbs, modals, aspects and more.  In effect, this is a slightly more sophisticated version of the old universal base hypothesis. And if true, it is very interesting. So, two questions: is it true? People are working on this already. And, theoretically more interesting, why is it true, if it is? Thus, for example, why exactly must the hierarchy be C over T over v?  There are theories that treat the semantics as effectively conjunctive. Thus, a sentence is effectively one long conjunction of properties. If this is so, why need the conjuncts embed with Cish information higher than Tish higher than vish? Or, why must theta domains be inside case domains, inside C-info domains?

It is, perhaps, worth further observing that this program is in tension with MP. How? Well tt effectively enriches FL/UG with a bunch of linguistically very specific information.  That’s part of what makes it so interesting.

4.     Lexical Atoms and Constructions

Scratch a thoroughly modern generativist and s/he will rail against constructions. What GB showed is that these should be treated epiphenomenally, as the interaction of simpler interacting operations and conditions. However, constructions are still alive and well in lexical semantics. So for example, we still tend to treat subcategorization as a reflex of semantic selection, the former syntactically projecting “information” coded in the latter, e.g. give projects three DPish arguments, believe one DP external argument and a propositional internal argument.  This effectively reflects the view that Lexical Items are “small” sentences.  This also reflects an effectively “Fregean” conception of lexical meaning which divides the world semantically into n-ary predicates/concepts and arguments/objects that saturate them. 

Until recently, this has been the only game in town. However, lately neo-Davidsonians have offered another “picture”: lexical items, even predicates, are semantically very simple. All they do is denote properties. Thus, syntactic structure is not a projection of lexical information, but a construction from basically unstructured 1-place predicates.  What makes an object an object interpretatively on this view is not that it saturates the internal y variable, but that it has been merged with a predicate of events and has thereby been type lifted in an event modifier. 

The availability of these two different pictures of what combination amounts to semantically has raised the question of what a lexical item really is. Or more specifically: given that the properties of a linguistic expression are the joint contribution of the contents of the items plus the grammatical combinatorics, how much ought we attribute to each? IMO, a very interesting abstract question that is ripe for theoretical investigation. There is already empirical work that bears on these issues by Higginbotham, Kratzer, Pietroski, Schein, (A) Williams a.o. However, the general implications of these discussions have not been foregrounded as much as they deserve to be.

There is another reason to pursue this question. Syntacticians have generally assumed that LIs are very simple and that the action comes from the combinatorics.  However, it does not take that much tinkering with the contents of LIs to get them to allow in through the lexical back door what the grammar prohibits.  Alex C and Thomas G have walked us through this for some feature passing technology. However, Paul Pietroski (over lunch) noted that thinking about the long distance relations that lambdas can code, coupled with a rich enough lexical content for the meaning of a terminal can allow one to generate perfectly fine representations (using just simple merge/lambda conversion) in which, say, The doctor rode a horse from Texas could mean that the doctor was from Texas (not a possible reading of this sentence).[5] For our syntactic explanations to work, then, we need to make sure that they don’t sneak in through the lexical back door and this means gaining a better understanding on what our lexical primitives look like.

These are some of the questions and problems I’d like to see tackled. I could give others, but, hopefully, this will grease the wheels. Again, please feel free to add your 2 cents.  It’s a discussion that is both worth having and fun to have, or so I hope. Looking forward to your input.

[1] Here, I believe, there is still a big question of what to do about Principle B effects. I’ve written on this elsewhere but the proposals don’t even meet my low standards satisfactorily.
[2] I put “argument/adjunct” in scare quotes for it is pretty clear that this is not exactly the right cut, though it serves current purposes.
[3] Kush notes that there are residual issues to be solved, but recall this discussion is taking place over beer.
[4] I mention this one as some UMD grad students are working on this now (Dustin Chacon and Mike Fetters). It seems that the variation may not be quite as reported.
[5] Say the lexical content of ride is roughly ly lx lz [Agent (x,e) & ride (e) & patient (y,e) & from (x,z)]. We don’t allow these right now. Why not?  What blocks these?


  1. Here is a monster that's prowling outside the gates: how are the uses of socially conditioned variants of grammatically relevant things such as pronoun forms learned, in spite of the dim prospects for finite parameterization of the conditioning factors? For example, the endlessly subtle variations in the use of polite vs familiar pronoun forms in European languages, and different kinds of complications found in other places. such as systems of kinship depending on the three-way relationships between speaker, addressee and person referred to ('kunderbi';, and various other sources).

  2. It would be terrific to have a discussion on the extended version of the Berwick/Hauser/Tattersall paper that had been mentioned a little while ago. Evaluating the minimalist solution of the language evolution problem should certainly be of interest to any BIOlinguist

  3. I have this intuition that a lot of the current stuckness of syntax is due in part to the fact that we're beholden to particular frameworks which unintentionally sneak in a lot of assumptions. We use these as tools for thinking about the phenomena, but they're really not generic tools at all, and I think that prevents us from really addressing the issues. There's a lot of stuff at the bottom of these frameworks -- the primitive, low level stuff -- which are assumed, but which also force us to think about problems in a very particular way, and I suspect that if we could pull back to a more general theory, we'd make some more progress. So I'd like to think that what's next is a de-framework-ification, but who knows.