Introduction

An important goal of phonological theory has been the elucidation of “action at a distance.” This term refers to processes, such as assimilations or dissimilations, in which the trigger segment and affected segment are not string-adjacent; there are segments that intervene, yet seem not to participate in the process. Transparency of this sort raises questions. How and why does it occur? What determines which segments, if any, will be transparent for a given process? The search for answers to such questions has been one of the important forces driving the elaboration of metrical and autosegmental representations.

Consider the case of long-distance feature spreading, or harmony. It is well known that segments within a spreading domain may appear to be nonparticipants, transparent to the harmony process. Various strategies have been proposed to account for such cases of transparency. Within nonlinear phonological frameworks, a property that many approaches have in common is the preservation of locality by relativizing it to what might very generally be called a legitimate target: some notion of “anchor,” “projection,” or “feature-bearing unit.” Locality is obeyed so long as spreading does not skip such a legitimate target. Notable examples of this line of thinking include Goldsmith (1976) and Clements (1980) on the notion “feature-bearing unit,” Halle and Vergnaud (1978) on “projections” of features, Kiparsky (1981) on the notion “harmonic vowel,” and Archangeli and Pulleyblank (1987, 1994) and Anderson and Ewen (1987) on the relativization of adjacency to prosodically or geometrically defined anchors.