Restrictions and Predictions of Harmonic Serialism: Devil in the Detail? Peter Staroverov [email protected] This will be a discussion session on Harmonic Serialism (HS) – a recent theory by John McCarthy (2006, 2007, 2008, in press). I will present a few problematic issues for the theory, some of them surprisingly undermining McCarthy’s own findings. I would like to discuss how the theory could be enriched to deal with his problem. Ideally, I would like the listeners to contribute towards mutual understanding of how this theory works. The talk will have two parts, at least if we have time for both of them.

Part 1: syncope and too-many-solutions In the first part, I will discuss the results in McCarthy’s recent article (McCarthy 2008) about syncope. I will present the basic features of the model and demostrate that the results McCarthy argues for are not actually guaranteed by the grammar he describes. One of the generalizations that McCarthy addresses is that the syncope constraints can never force stress assignment. In other words, dispreference for vowels in weak positions is never resolved by making the positions strong (this would correspond to a metrical parse of the shape (CV@)(CV$)(CV$)(CV$)(CV$) where brackets indicate feet). To account for this, McCarthy adheres to the view that feet are assigned iteratively one by one in the course of derivation. Coupled with a constraint *V-PlaceWeak that is violated by vowel place features in weak branches of the feet and in extrametrical positions, this assumption predicts that a parse (CV@CV)CVCVCV is a more harmonic way of assigning the first foot. However in section 5.5, McCarthy adopts an additional constraint that only penalizes vowel place in a weak position of the foot: *V-PlaceWeak-In-Foot. A ranking like the one in (1) still reanimates the possibility of stressing every syllable in response to the syncope constraint. (1) *V-PlaceWeak-In-Foot >> Ft-Bin, Max >> *V-PlaceWeak Furthermore, the conception of iterative footing undermines the main point of the article, namely that Harmonic Serialism guarantees that stress should be assigned before syncope can apply. With iterative foot assignment what is guaranteed is more precisely that one foot has to be assigned before syncope can apply. After the first foot is assigned, nothing protects the vowels that are unparsed in the intermediate form like (CV@CV)CVCVCV from randomly deleting. In other words, the vowel that would surface as stressed if furhter prosodic parsing goes on can in fact be deleted after the step when only the first foot has been constructed. After presenting those problems, I would like to discuss the possible solutions available in Harmonic Serialism. Those include but are not limited to fixed rankings and modified markedness constraints. I will describe those possibilities in the light of a broader problem known as too-many-solutions or too-many-repairs problem (cf. Blumenfeld 2006).

Part 2: mutual feeding and opacity The second part of the talk will be devoted to a case where Harmonic Serialism turns out to be potentially too restrictive. In short, two processes can not feed each other in HS. Roughly speaking, for a process A to feed process B, the repair for markedness constraint MA driving A should introduce violations of MB driving B. The requirement of harmonic improvement also imposes a ranking on MA and MB: for A to be harmonically improving, MA should dominate MB. Clearly, a situation where both A feeds B (and hence MA >> MB) and B feeds A (hence MB >> MA) is predicted to be impossible in HS. In fact, this is the reason why HS

predicts processes like “delete all the final segments in a word until you meet a nasal” not to occur (McCarthy 2007: 84-86, McCarthy 2006). This generalization per se is probably on the right track but it yields a somewhat unexpected prediction about opacity. In this domain, a situation where A counterfeeds B while B feeds A is predicted not to occur. I will argue that the latter prediction is probably too strong. In fact, a situation like this is quite expected if in historical terms B ceases to be active at the time A applies to create new violations of MB. I will confirm this expectation by an illustrative case of interaction of apocope and wordfinal debuccalization in Tundra Nenets. In this language, the vowel  gets deleted word finally and before word-final glottal. The consonants t, t, s, s, n, n and  change to  word-finally. The processes apply transparently when the input ends in a C sequence, for instance /timjs/ ‘to rot’ surfaces as timj. However, contrary to the prediction of OT-CC, vowel deletion counterfeeds consonant alternations in the case of vowel-final input. Thus, /xad/ ‘snowstorm’ surfaces as xad, not xa. In short, for the sequence of changes to be harmonically improving in each step, the markedness constraint responsible for debuccalization should dominate the apocope constraint whereas the opposite ranking is required for to go through. I would like to see if any similar cases are attested. The discussion of how we could potentially incorporate these cases in HS, will follow. Like: prec cancelling the harmonic improvement requirement at a very restricted set of situations…

References Blumenfeld, Lev. 2006. Constraints on Phonological Interactions. PhD dissertation. Stanford. McCarthy, John. 2006. Restraint of Analysis. In Eric Bakovic, Junko Ito, and John McCarthy (eds.) Wondering at the Natural Fecundity of Things: Essays in Honor of Alan Prince. Santa Cruz, CA: Linguistics Research Center. Pp. 213-239. Also to appear in Sylvia Blaho, Patrik Bye, and Martin Krämer (eds.) Freedom of Analysis. Berlin: Mouton de Gruyter. McCarthy, John. 2007. Hidden Generalizations: Phonological Opacity in Optimality Theory. London: Equinox McCarthy, John. 2008. The serial interaction of stress and syncope. Natural Language and Linguistic Theory McCarthy, John. In press. The gradual path to cluster simplification. To appear in Phonology