It is often
claimed that representational simplicity accounts for why
particular segments tend to appear in
epenthesis. In Radical Underspecification epenthetic vowels are those
with no feature specification. Glottal stop is often assumed to be Placeless in
order to account for its frequent epenthetic status, and work on coronal
underspecification makes the same assumption about epenthetic coronals.
I
have argued in previous work that the underspecification analysis of consonantal
epenthesis is problematic. It requires that glottal stop have different
representations in different languages, since it clearly has guttural Place in
languages like Arabic. How to
represent the distinction between the coronals and glottal stop is problematic,
and how one is chosen over the other in epenthesis is generally not addressed.
In
Lombardi (1997) I show that these problems can be solved if we assume that both
coronals and glottal stop are fully specified for Place, and that ranked OT
markedness constraints choose the segment with the least marked possible Place
in a given situation. This will
often be glottal stop, which has the
lowest ranked *Place violation. But
conflicting constraints regarding segment sonority, positional
restrictions, etc., may force the optimal epenthetic Place to be the more
marked, but still relatively low marked, Coronal.
Vowel
epenthesis is another situation where languages differ in what segments are
chosen, but the variation and thus the challenge is greater. Many claims have
been made in the past about the relative commonness of various epenthetic
vowels, but they have been based on extremely limited data.
I will claim that the markedness relationships among vowels are the
following and have the following results:
Languages
may vary in whether low or nonlow vowels are less marked.
If low vowels are less
marked the epenthetic vowel will be [a].
If
nonlow vowels are less marked, other constraints choose among them:
Back
vowels are less marked than front vowels
Nonround
vowels are less marked than round vowels
Mid
vowels are marked
This makes
strong predictions about the possible epenthetic vowel given the sound system of
the particular language. I will
show that on the whole these predictions are correct:
If
[H] is present it will be epenthetic: Nonlow, nonmid, back, nonround
If [H] is not present but schwa is, schwa will be epenthetic: Nonlow, back, nonround
If neither
schwa nor [H ] is present, [i] will be epenthetic: Nonlow, nonmid, nonround
A residue of
cases remain outside these
generalizations, involving epenthesis of [e] and of [u].
These are marked vowels - mid and round - under any analysis, so the
point to explain here seems to be how it can be that some languages choose
marked vowels for epenthesis when the vast majority choose unmarked ones.
Since my goal here is to set out the basic markedness relationships among
vowels I leave these cases for future research.
I
show, then, that the ideal analysis of epenthesis does not rely on the relative
representational complexity of different segments.
Under the proposed OT analysis all segments can be fully specified, and
regardless of their complexity, it is the ranking of markedness constraints that
will determine which segments surface in epenthesis.
This type of analysis yields a better account of the cross-linguistic
variation in epenthetic segments, but means that epenthetic segments are of
limited if any use in making arguments about representational complexity.