Please put comments and questions on the Week 4 readings on vowel harmony in Harmonic Serialism (McCarthy) and Harmonic Grammar (Potts et al.) and any aspect of class discussion here. If your comment has to do with the feature geometry discussion, which was scheduled for last week but we are only getting to this week, it’s probably best to put that in last week’s thread.
Having read McCarthy 2011, I can quite easily understand the appeal of Harmonic Serialism with regard to autosegmental spreading, but I was left somewhat confused by some of his arguments in favor of SHARE[F] over AGREE. Many of his arguments against AGREE only work if AGREE is applied through a parallel OT analysis.
An example of this would be the argument that he makes based on segmental deletion. As he states, the segmental deletion of /r/ would harmonically bound the attested candidate that stops the spreading of the nasal feature at the liquid. However, this would not be the case in HS. If we consider that only one change can be made at a time, the deletion of /r/ would serve no purpose as GEN cannot predict the next change after liquid deletion. Based, on this definition of HS, the candidates at this step in the derivation should be: mãw̃ã.a vs mãw̃ã.ra. Both would produce the same violations, if there is no IDENT constraint restricting segmental deletion, and there should be no motivation for liquid deletion. So my question is, is he restricting AGREE constraints to parallel OT?
Just as SHARE[F] typological predictions seem to only work in HS, i think the same might be said of AGREE-R.
In his discussion of the “sour-grapes” effect of Agree McCarthy argues with respect to (7) that the intended winner (d) is harmonically bounded by the candidate with total spreading (f). However, since (d) violates Agree-R([nas]) and (f) doesn’t this argument doesn’t follow through fully.
What are the current predictions of the field about the division of labour, if any, between the parallel OT and HS components of the grammar? Do we predict that there will be some languages where all phonological processes would be derivable from from constraints applying to candidates determined by only one of these two types of GEN?
Logically, it is possible that the GEN can make N changes at a time, and that the value of this N may be determined independently. Since a significant number of current analyses may be compatible with the various values of N, what rules out such possibilities?
Given that deletion is understood to be ‘a series of reductive neutralization processes’, should epenthesis, being its reverse, be considered a series of differentiation processes? Has this been explored within the HS paradigm?
On page 7 there are some examples of how Share(nas) can be violated. One of them is [ma] with the nasalized vowel where both segments are linked to two different autosegments. Why would this happen?
On how HS’s Gen does operations: inserting a feature and a single association line to pre-existing structure is one change. Also inserting a single association line linking two elements of pre-existing structure is one change. How can this work? If feature addition is “free” surely this makes bad predictions?
Can we go over exactly how Share(f) assigns violations? It says assign a violation mark for every pair of adjacent elements are are not linked to the same token of [f]. How does m|a|w|a|r|a get 5 violations when the nasal segment is only adjacent to one other segment?
After reading Potts et al. (2010) I was curious about the margin of separation constant a. Is it merely a mathematical convenience to define minimality, or does it have some theoretical notion behind it? The paper indicates that the actual value of a doesn’t matter for the calculation of constraint weights for observed data, but can have consequences for generalization to new, unseen data. What kinds of theoretical implications does this have for learning? It’s not uncommon for learners to come across new data forms. Would it be possible to test for the effects of a?
Sorry for posting this here — there doesn’t seem to be a page for Week 7 yet.
I am confused with the formal definition of the harmony function: I don’t see why, for a real language, the sum is guaranteed to be finite. Are not we supposed to sum over the whole set of constraints in existence?
I think I’m not quite clear about finite-state-implementable vs. non-finite-state-implementable constraints either (as mentioned on p. 84 of Potts et al.)
And do I understand correctly that ganging effects for Lango arise for this specific set of constraints? Or is there an argument showing that this is a problem to be encountered with no matter what the choice of constraints is?