Targeting underspecified segments: A formal analysis of feature-changing and feature-filling rules

In this paper, we discuss how the difference between feature-changing and feature-filling processes has not been adequately addressed in rule-based, derivational phonology. We explore two different theories of these processes. One theory analyzes feature-changing in two steps-set subtraction and then set unification. Another analyzes feature-changing in one step-a function that alters the polarity of features. We argue that empirical evidence favours the two-step over the one-step process. In particular, the two-step process provides an account for the lack of rules that target specified segments independently of underspecified segments (so-called by-passing rules). It also explains why there are no attested examples of purely phonological rules that switch the polarity of certain features (e.g., changing /t/ and /d/ to [d] and [t] respectively).