Metrical Grids and Active Edges

Abstract

Theories of word stress assignment differ in the kind of representations they adopt. One family of theories takes stress to be assigned by grouping stress-bearing elements into small units below the level of the word (typically, metrical feet), such that one element in each unit is marked as stronger, hence stressed (e.g., Liberman and Prince 1977; Hayes 1980). Another family of theories, often referred to as grid-only, models stress assignment without appealing to feet or similar bracketed representations above the syllable (Prince 1983; Selkirk 1984; Gordon 2002). While the grid-only approach generates the attested languages with relatively simple representations, it also generates a host of patterns which are very different from those attested in human languages (Hayes 1995; Kager 2012; also see Stanton 2016). This thesis aims to solve a set of overgeneration problems that arise in the grid-only approach. The solution involves three components. The first is a novel class of constraints that are sensitive to word edges but unspecified to the edge they apply to (left or right). The value of this edge, considered the “active” edge, is determined by the ranking between two competing constraints (cf. Richards 2016). The second component involves a specific characterization of alignment constraints and the crucial exclusion of computationally weaker or stronger alternatives. The third component is a set of principled fixed rankings between two classes of constraints. In particular, I propose that constraints sensitive to the active edge systematically outrank constraints that regulate rhythmic alternations (cf. van der Hulst 1997; 2012). The result is a grid-only theory of stress that has a significantly tighter fit to the typology compared to previous theories.