|Title:||Some New Generalizations Concerning Geminate Inalterability|
|Comment:||LSA handout (10 pages), requires IPAKiel font|
|Abstract:||Some New Generalizations Concerning Geminate Inalterability
It is a familiar observation that phonological processes frequently fail
to apply to geminates. For example, in Tigrinya, post-vocalic spirant-
ization of velar stops yields [/a-xalIb] \'dogs\' (cf. [k@lbi] \'dog\'), but
[f@kk@r@] \'boasts\', not [f@xk@r@] nor [f@xx@r@]. A number of proposals
(e.g. Hayes 1986, Schein and Steriade 1986) accounted for this geminate
inalterability in terms of a distinction between singly- and multiply-
linked autosegments. Moreover, these approaches correctly account for
heteromorphemic geminates in Tigrinya, which pattern with singletons
(e.g. [mIrax-ka] \'calf-2sg.m.\'). Subsequent research, however, has ob-
served that geminate inalterability is inviolable only in the domain of
lenition processes (Churma 1988, cf. Inkelas and Cho 1993), a generali-
zation uncaptured by the autosegmental approaches. (I characterize
lenition as temporal or spatial reduction of an articulatory gesture.)
Building upon Churma 1988, as well as the lenition surveys of Lavoie
1996 and Kirchner (in progress), I have identified the following addi-
a. No process converts an underlying non-continuant (geminate or other-
wise) to a geminate continuant.
b. No process converts an underlying voiceless consonant (geminate or
otherwise) to a voiced geminate obstruent.
c. \"Partial geminates\" (homorganic nasal/stop or lateral/stop clusters)
behave like full geminates with respect to spirantization; however,
they readily undergo voicing.
d. No lenition process (other than degemination) exclusively targets
An account of these generalizations is couched within a general Optimal-
ity Theoretic treatment of lenition, in which a scalar effort minimiza-
tion constraint, LAZY, interacts with a set of lenition-blocking con-
straints, including faithfulness constraints. The geminate inalterabil-
ity generalizations follow from this effort-based approach to lenition,
coupled with certain assumptions concerning the effort involved in
geminates and their lenited counterparts. Specifically,
a. A geminate fricative is more effortful than a geminate stop (the
reverse of the situation in singletons), due to the isometric ten-
sion involved in maintaining a partial constriction for a prolonged
temporal interval. Therefore it is never optimal to spirantize a
geminate. This assumption is supported by a computational mass-
spring model of articulatory movement.
b. A voiced geminate obstruent is marked relative to a voiceless
geminate, due to the aerodynamic conditions required to sustain
voicing (see Ohala 1983, in The Production of Speech (MacNeilage,
ed.)). Therefore it is never optimal to voice a geminate obstruent.
The behavior of heteromorphemic geminates in Tigrinya follows from in-
teraction with paradigmatic (output-output) faithfulness constraints
(e.g. Benua 1995, Papers in OT, UMass), without resort to a phonetically
unmotivated representational distinction between \"true\" and \"fake\"
geminates: the candidate [mIraxka] defeats *[mIrakka], due to the for-
mer\'s greater faithfulness to the base form, [mIrax].