non-linear phonology

In PHONOLOGY, any model which avoids a linear REPRESENTATION of the phonological structure of a word. ‘Linear’, in this context, refers to a representation of structure as a series of SEGMENTS occurring in a strict horizontal sequence, each segment being analyzed (vertically) as an unordered column of FEATURES – [s], for example, being [−syllabic], [−voice], [−nasal], [+coronal], etc. A recognized weakness of these two-dimensional matrices is their inability to handle features which extend over DOMAINS greater than an individual segment (e.g. certain properties of TONES, vowel HARMONY); another is their inability to represent structure relationships (functional or ‘natural’ classes) within columns of features. The result has been the development of non-linear models, such as Firth’s PROSODIC PHONOLOGY, and (since the 1970s) such models as METRICAL PHONOLOGY, AUTOSEGMENTAL PHONOLOGY and DEPENDENCY PHONOLOGY. In these approaches, features which extend over domains greater than a single segment are taken out of feature matrices and represented on separate levels (TIERS) of their own. There is now a large class of non-linear models in contemporary phonology.