Blurb:

Wiese (2005) and Bobaljik (2012) have suggested that stem allomorphy obeys an *ABA constraint (for ablaut and comparative/superlative formation, respectively); this issue has received a lot of attention in the recent morphological literature. *ABA can be derived in standard morphological theories by assuming morphological exponence to be governed by specificity, but only if the features involved are privative, and if the different syntactic contexts (e.g., infinitive, past participle, finite past tense; or positive, comparative, superlative) are characterized by subset-superset relations. I show that there is reason to doubt both these assumptions. Given this state of affairs, it turns out that an approach to *ABA in terms of harmonic serialism accounts for the absence of this pattern without further ado, even if binary features and non-subset-superset-related syntactic contexts are assumed: To derive an ABA pattern, a stem exponent (A) that has initially been taken out of the morphological array (emerging as optimal at an early optimization step), and that has subsequently been removed in favour of a stem exponent B (because B turns out as optimal at a later step) woul dhave to come back again, which it cannot do because it is irrevocably gone by this time. As a consequence, an unfaithful ABB pattern will be optimal in the end in this case (with Neutralization plus, possibly, input optimization occurring since the ABB pattern can also be derived in a faithful way with a different initial input).