From Joe Pater (pater@linguist.umass.edu)
McCarthy, John J., Joe Pater, and Kathryn Pruitt. To appear 2016. Cross-level interactions in Harmonic Serialism. In John McCarthy and Joe Pater, eds. Harmonic Grammar and Harmonic Serialism. London: Equinox Press. http://websites.umass.edu/pater/files/2011/10/McCarthy-Pater-Pruitt-2016-CLI-HS.pdf
Cross-level interactions are phonological processes that make reference to multiple levels of the prosodic hierarchy, such as vowel shortening in the weak position of a foot. Cross-level interactions figure in most arguments for parallelism in Optimality Theory. This chapter demonstrates with several case studies how cross-level interactions can be analyzed in Harmonic Serialism. The key insight is that the relevant constraints may be violated in the course of the derivation, even if they are obeyed in underlying and surface forms. Cross-level interactions require parallelism only if constraints are inviolable, but that is inconsistent with a fundamental premise of Harmonic Serialism and every other version of Optimality Theory. The problems that cross-level interactions pose for serial theories with inviolable constraints are demonstrated through a review of their treatment in pre-OT constraints and repairs theories.
It’s a curious and perhaps alarming fact that in this paper, not a single ranking claim is validly argued.
For each system discussed, the individual arguments vary in the number of constraints cited. Claims of optimality are based on a small selection of competitors, without reference to the set of possible optima. In no case are CON and GEN spelled out.
But the analyst of a system must know what the constraints are (all of them) and know what the candidates are (minimally, the possible optima). Why?
-All constraints in CON for the system must be included, because the definition of optimality is sensitive to the behavior of all of them. Even when a constraint does not participate in a given decision, this fact must be noted: its irrelevance is relevant to the outcome.
– All possibly-optimal candidates must be dealt with, because to be optimal is to be better than all (violation-distinct) competitors.
Consider, by way of contrast, the careful handling of the logic of statistical inference in work that relies on it. Even a sympathetic reader must wonder if we’re going to decide between ‘Serialism’ and ‘Parallelism’ without conducting valid, systematic arguments within either.
So, what’s going on? None of this is controversial; all of it is public knowledge. Is there a kind of implicit methodology or ethos here that deprecates logic in favor of something else, perhaps the way things were done in the historical documents? But since we understand things better now, we must surely aim to do better as well.
Thank you for your comment Alan. I’d like to try to contextualize it within the goal of the paper, and see if I understand what you are asking us to do. The paper aims to show that a type of phonological process previously claimed to provide evidence for parallelism (by two of the authors of the paper, amongst others) can in fact be analyzed in a serial version of OT. A typical example of what we call a cross-level interaction is a foot-conditioned vowel length alternation – iambic lengthening or trochaic shortening. The issue for a serial theory with inviolable constraints is that when we first construct a foot that contains a vowel with the wrong length specification, we violate the relevant stress-weight constraint. Why doesn’t that block foot construction? One approach, taken by Mester and others in pre-OT work, is to have the vowel length “repair” apply in parallel with foot construction. Classic OT adopts both violability and parallelism. We aim to show that violability alone suffices. To show this, it seems it would be enough to provide a simplified example of trochaic shortening or iambic lengthening. If we did this with a fully specified Gen (relatively easy in serial OT) and a full constraint set in every tableau, would that suffice to make our point?
My point is that valid arguments for optimality require certain steps, which I outline.
In your response, you note that you have constructed a thesis about serialism which you wish to refute, and which would be refuted if the arguments you present are sound. This seems like good motivation for trying to establish that your claimed optima are indeed optimal and that the claimed rankings are indeed the ones that render them optimal.
I can think of a couple of reasons why linguists might want to ramp up the level of their OT reasoning, or of their theory-based reasoning in general.
First, there is the simple virtue (and even pleasure) of making justified assertions — ‘news that stays news’ as opposed to the revenant claims and counterclaims of polemic.
Second, and more programmatically, I have the impression that linguistic theories have a lot more in them than is revealed by the hasty encounters of (what we might call) the betterness struggle. To really see what’s there, you have to pay fairly close attention to the theory as it is defined.
Agreed – thanks again.