Harmonic Serialism (HS) differs from standard parallel OT (P-OT) in two respects: GEN can make only one change at a time; and the output of EVAL loops back into GEN until convergence. Extant arguments in support of HS show that P-OT overgenerates a typology in a way that HS does not or that some generalizations are expressible only at HS’s intermediate derivational steps.

All of these arguments for HS depend on saying precisely what it means for GEN to “make only one change at a time”, but there are inconsistencies in the literature. This talk is an initial attempt to achieve some harmony among differing views of HS’s GEN component. The goal is to understand where parallelism is irreducible — that is, required in both HS and P-OT.