This talk presents a theory of markedness hierarchies and their formal expression in Optimality Theory. Informally, the aim is to show that it is not accurate to say "a is always more marked than b" (for any pair of values a and b in a particular markedness hierarchy); instead, it is only valid to claim that "b is never more marked than a." The latter formulation allows for situations of 'markedness conflation': where markedness distinctions are eliminated for particular phenomena (i.e. where a and b are treated as equally marked).

In formal terms, the aim is to show that hierarchy-referring constraints refer to contiguous ranges of scale elements, starting with the most marked element. As an example, the vowel sonority hierarchy | E > i,u > e,o > a | (E=schwa) is related to stressed positions by a series of constraints. Instead of employing a fixed ranking (Prince & Smolensky 1993, Kenstowicz 1996), constraints refer to ranges of the scale: i.e. *Hd/{E}, *Hd/{E,i,u}, *Hd/{E,i,u,e,o}, *Hd/{E,i,u,e,o,a}. These constraints are shown to allow for stress systems like Nganasan's, which ignores the sonority distinction between central vowels [E] and high vowels [i y u] on the one hand, and medial and low peripheral vowels on the other.

The same point is made for faithfulness constraints. For the major place of articulation scale | dorsal > labial > coronal > glottal |, faithfulness constraints are argued to have the form IDENT{dorsal}, IDENT{dorsal, labial}, IDENT{dorsal, labial, coronal} and IDENT{dorsal, labial, coronal, glottal}. These allow for 'faithfulness conflation' - where deviations from the input form incur equal faithfulness violations, allowing markedness constraints to determine the winner. Informally, 'faithfulness conflation' describes a situation where perfect identity to the input form is prevented, and the winning form is chosen entirely for markedness reasons. Evidence will be adduced from consonant coalescence in Pali.