Posted on Saturday, September 18, 2004 11:40 AM
A new paper by John McCarthy, introducing a new formalism for Autosegmental Phonology,
on the Rutgers Optimality Archive:
When Optimality Theory (Prince and Smolensky 1993) is applied to autosegmental phonology (Goldsmith 1976), markedness constraints that favor autosegmental spreading are required. The pro-spreading markedness constraints in the current literature have typological problems: they undergenerate because they do not favor partial spreading (McCarthy 2003), or they overgenerate because they can favor unattested alternations, such as 'spreading' by blocking epenthesis (Wilson 2003).
This article proposes a modification of autosegmental representation that is intended to adddress these problems.
As far as I can see, he views autosegmental representations as constituents ('spans'): every segment in the span for a feature [F] takes up the value of F of the 'head' of the domain. McCarthy offers an intriguing thought about what this means for the phonetics-phonology interface: the head of the domain would be the phonetic 'target', so that it is not necessary to assume that all other segments in the span have such a strong realisation. At first sight this seems to make the wrong predictions for vowel harmony (I do not think there is any reason to believe that any of the vowels in a harmonic span is less of a target than any other vowel), but this still is an inspiring thought.
Another interesting aspect is that this seems again part of McCarthy's program (I am not sure whether he has ever made this explicit, but it seems an underlying assumption in all his work of at least the past ten years) to formulate all constraints in terms of the segment. Segments are anchors for everything, and in this sense 'auto'segments do not really exist.