We are two thirds of the way through this years lectures and David continued his Sherman like march through the thicket of Minimalist mechanisms. Here are the slides. The second lecture had three main points; to expose the inadequacies of standard roll-up analyses of certain word order effects (mainly attempts to derive Greenberg's universal 20), to show that what made the standard roll-up approaches successful were exogenous constraints added to an otherwise much too powerful roll-up technology and to argue for a theory which endogenously prevented the unwanted structures from being generated. The latter is a revised (and improved) version of the theory presented in David's book (here).
The second lecture started with David's recasting of the meta-principle that he assumed in lecture 1 and continued to assume here: that movement operations that have no affect on CI are to be discouraged. His entering wedge into the discussion on roll-up was the observation that roll-up is mainly interested in deriving word order effects from hierarchy (in an LCAish context a la Kayne). This violates the meta-principle and so it must go! However, as became clear in the question period of lecture 1, the adequacy of the meta-principle was far from clear (recall Lasnik's objections also pursued by Alexander Williams). So David began lecture 2 by revising his version of the principle. It too ran into some heavy questioning as it seems to detach semantic interpretation operations from the syntactic operations that generated the relevant syntactic structures. This time the semanticists (Valentine Hacquard, Paul Pietroski and Alexander Williams) pushed David very hard on whether so de-tethering CI interoperation from Merge didn't amount to allowing the possibility of what I dubbed "construction semantics." David was appalled at the possibility, but IMO, the semanticists were able to make the case that his revised assumptions seemed to lead in that direction. The question period was taped and I will put a link up to the 3 lectures when the videos are posted.
So today is the final lecture. I expect to be quite challenged for David is gunning for sidewards movement today (a favorite of mine). I'll post the slides tomorrow after I get them. SO far the lectures have been great fun.
So I still want to hold to the principle that non one-to-one hierarchical relations better map onto semantic distinctions. A mother bearing distinct relations to two daughters entails that there's a semantic distinction being made. Equally a syntactic object entering into distinct relations to two containing objects, as happens in Internal Merge, entails there's a semantic distinction being made. With this principle in hand, as a principle that says that the design of he syntactic system is optimised to the semantics in this way, lots of analyses involving copy creation (for head movement and for roll up movement, as I argued) end up being cases of bad design. Completely agree that Howard's `Tabs are likely to be kept on him' A-movement examples are a problem.
ReplyDeleteI think my jet lag was making me stupid in the question period, as I was finding it difficult to see how my proposals were worse than classical proposals (in fact, I tried hard not to say anything new about the semantic side of things in my system as I thought already I was going to have a hard time getting people to buy unary branching structures with no functional heads or head movement, and mirror word order effects being analysed as a switcheroo of the complement lines in Extended Projections).
I think I now understand the issue, but maybe not: if the derived object is complex, and the semantics gets to see the whole thing, then it can do pretty much anything it wants. I thought my semantic rules preclude this, as they operate bottom up, object by object, in a manner that's pretty reminiscent of directly compositional approaches. But the issue that was being further raised is that there are no constraints on what the label of the constructed object can do semantically. So that needs a separate theory. I still think, however, that that places my system in no worse a place than other systems that have functional heads having some semantics: what is the theory that restricts these? Give me some such theory (say that mother daughter relations in an Extended Projection are always functional application, and specifier mother relations are always variable identification, and mothers can never have higher than second order abstraction) and that will be the theory of the syntax-semantics relation, ruling out Construction Semantics (shudder!).
Ok, better go finish my slides for today! How to constrain Merge by making it operate over a teensy-weency memory buffer, thus improving computational complexity, and increasing restrictiveness in making sideways move unstatable.