Two arguments for a positive vowel harmony imperative

In this paper I provide arguments from two case studies in support of Kimper’s (2011) new Trigger Competition approach to vowel harmony, which is based on autosegmental linking and the positively formulated Spread constraint. The first argument is based on Hungarian vacillation, in which sequences of vowels that individually do not participate in backness harmony can conspire to create free variation or to participate fully. Trigger Competition’s violable preference for local harmony over longdistance harmony, as well as its formalization of differing degrees of vowel transparency, make this simple to account for. The second is based on an account of Seto backness harmony, a system which has not yet been addressed in the constraint-based phonology literature, and which I claim presents a distinctly difficult case for conventional theories due to its paired neutral vowels. 1 An introduction to Trigger Competition Trigger Competition is a novel framework that accounts for vowel harmony by proposing that any segment in a word can spread its feature values onto any other segment, but that locality, vowel quality, and other factors contribute to deciding which segment, if any, can trigger harmony on each potential target. This paper presents two pieces of evidence for a version of this system with two implementational changes. The competition referred to by the name Trigger Competition is implemented using the positive Spread constraint, which assigns rewards (rather than violations) to candidates in which segments share their feature values. Factors which impact how desirable a particular instance of spreading is, like the proximity between the segments involved, are encoded as multipliers which affect the reward score assigned by the constraint: rather than stipulating additional constraints and mechanisms on top of harmony, Trigger Competition enriches the harmony imperative directly. Kimper proposes a multiplier which scales the reward according to the quality of the vowel that triggers spreading. This multiplier is meant to capture how well cued the harmonic feature is for a given segment, with the understanding that the grammar uses harmony to suppress some of the ambiguity associated with these weakly cued segments by allowing them to jointly cue the same feature value. This innovation allows the grammar a much richer vocabulary of possible behaviors when dealing with the notorious problem of neutral vowels—those which are not systematically subject to harmonic alternation. In particular, this makes it possible for the grammar to treat vowel transparency as a scalar phenomenon 1 Samuel R. Bowman Two arguments for a positive vowel harmony imperative rather than a binary one, allowing it to account for cases where generally transparent vowels become opaque and propagate harmony in certain environments. The constraint is implemented within Serial Harmonic Grammar (SHG, Pater et al., 2008, Pater, 2010, Mullin, 2010), a variant of the weighted constraint system Harmonic Grammar which adds stepwise evaluation from Harmonic Serialism. A grammar in SHG makes incremental changes to a linguistic form until no more productive changes can be made. At each iteration, the grammar generates every possible single change to the input. The grammar then chooses to make the change which has the highest harmony score: the weighted sum of the change’s violations of the grammar’s various constraints. This repeats until there are no changes which improve upon the harmony score of the existing form. At this point, that form is chosen as the output, and optimization ends. What constitutes a single change for the purposes of a derivation step is an open question, and an essentially empirical one. I follow Kimper’s approach, never fully enumerated, which seems to reflect a tentative consensus that changing contrastive feature values possibly by way of linking is a valid single step, as is the epenthesis of unmarked and minimally specified segments, and that the epenthesis of a fully specified, autosegmentally linked segment is not a valid step. 1.1 The constraint and representation Serial Harmonic Grammar, under Kimper’s formulation, inherits Harmonic Serialism’s tolerance of positively formulated constraints, which reward candidates for demonstrating certain structures, rather than penalizing them. The constraint enforcing harmony is one of these: (1) Spread(±F): For a feature F, assign +1 for each segment linked to F as a dependent. (Kimper, 2011) In the representation used here, one of the segments linked to a feature value node F is the head, and additional segments linked to it are dependents. A set of segments is considered to be linked if they all share the same specification for F on the tier for feature F, and thereby all display the same value of F. A candidate can establish a link between two segments with different values of F, but this forces one of them to change in its value for F (represented as in 2), and introduces a faithfulness violation. The grammar does not require that segments that have identical values for F be linked to one another. It is possible for each to have its own specification, or for the two to share one, as in (3):