On Explaining the Link between Coda Consonants and Onset Clusters
Stuart Davis and Karen Baertsch
Indiana University and Southern Illinois University, Carbondale

	Japanese strictly obeys the coda condition (i.e. codas are
place-linked to a following onset);  Japanese lacks onset clusters.  Is
that an accident?  In the development of Pali from Sanskrit (Zec 1995),
Pali codas became more restricted so that a coda can only be a consonant
that is place-linked to the following onset (like Japanese).  Moreover, in
Pali, historical onset clusters have been reduced to singletons.  Is that
an accident?  In the development of Campadanian Sardinian from Latin
(Bolognesi 1998), coda laterals have become rhotics (e.g. [arba]  > ALBUS
'white'); also laterals in onset clusters have become rhotics (e.g. [prus]
> PLUS 'more).  Previous researchers viewed these changes as independent;
but are they related?  This paper formally accounts for a view that
diachronic changes making codas more restrictive have the consequence of
making onset clusters more restrictive as well.
	There is a similarity of patterning connecting the second segment
of an onset cluster with a singleton coda consonant; clearest is the
preference for each of these positions to be realized with a high sonority
consonant.  There are a variety of other links between them such as the
phonotactic link found in English disallowing identical consonants
flanking a vowel in CCVC sequences but not in CVC or CVCC sequences; thus
English disallows sequences like *plil but allows 'pop' and 'lilt'.   Such
observations linking the second member of an onset with a coda have defied
previous explanation.  In accounting for this similarity of patterning, we
propose to augment Prince & Smolensky's (1993) Margin Hierarchy so as to
distinguish between structural positions that prefer low sonority and
those preferring high sonority.  Prince & Smolensky's Margin Hierarchy,
which gives preference to segments of low sonority, applies to singleton
onsets; this is our M1 hierarchy.  Our proposed M2 hierarchy applies both
to the second member of an onset and a singleton coda.  The M2 hierarchy
differs from the M1 hierarchy in giving preference to consonants of high
sonority.  Splitting the Margin Hierarchy into the M1 and M2 hierarchies
allows us to explain the phonotactic restriction noted above as involving
an OCP type constraint on the M2 position.  A prediction of the split
margin approach is that a (diachronic) change making codas in a language
more restrictive should have the effect of limiting onset clusters in that
language as well.  As we formally show, this prediction is corroborated by
the Sardinian and Pali examples described above.