Abstractness in phonology: the case of virtual geminates
Philippe Ségéral, Tobias Scheer · 2001
segments in the sense of the early generative debate thus fall into two different species if autosegmental representations are assumed: segments that are abstract because 1) their melodic identity or a part of it does never occur on the surface, or 2) their association to another autosegment has not the usual phonetic consequences. By the time the discussion on abstractness took place, autosegmental representations were not part of phonological theory, or were just emerging. Hence, the formal distinction between different abstract segments could not be made, and indeed is absent from the literature. We would like to suggest that the majority, if not all cases where analyses recurring to abstract segments are compelling involve an association between two autosegments which is not phonetically executed. Philippe Sgral and Tobias Scheer 330 In epistemological terms, this analysis is cumulative in the sense of Encrev (1997) and Durand and Laks (1996). When new theories arise and science moves on, open questions that are related to and were debated in the old framework are often left behind. This would not be harmful if people expected the new theory to provide tools that allow to view the unresolved questions in a new light after a while. New theories are cumulative if they do not forget about old problems that still beg the question. They are non-cumulative if they ignore old insights and questions that have been raised prior to their birth and with no relation to them. Unfortunately, generative phonology in the past 30 years bears quite a bit of a non-cumulative character. Rulebased theories such as SPE were mainly concerned with procedural issues, i.e. rule-ordering. They had their typical set of data, which was different from the ...