101
vowels, unstressed vowels and perhaps low toned vowels, are most susceptible to
devoicing in final position. The gradual nature of the decline in subglottal pressure
throughout the utterance is also compatible with the fact that devoicing in many
languages (e.g. Acoma, Big Valley Shoshoni, Turkana, Nyangumata) affects not
only the final vowel but also may extend farther back from the end of the domain in
gradient fashion.
Interestingly, the decline in subglottal pressure is in direct competition with
another common cross-linguistic property of final position: final lengthening (cf.
Wightman et al. 1992). A priori one might expect, by analogy with the blocking of
devoicing by phonemic long vowels and accented vowels in final position in many
languages, that the additional phonetic length of final vowels would block devoicing.
However, the gestures associated with final lengthening are different from those
associated with other types of lengthening. Final lengthening does not involve an
increase in gestural magnitude, unlike lengthening associated with accent (Beckman
et al. 1992) or presumably, phonemic length. It is thus not surprising that, whereas
phonemic long vowels and accented vowels inhibit final devoicing, final lengthening
typically does not. In fact, final voiceless vowels, like their non-final counterparts,
are usually described as being quite short, shorter than even non-final voiced vowels.
Thus, the subglottal pressure decline not only inhibits final lengthening, it also
appears to induce final shortening.
5.
ARTICULATORY
/
AERODYNAMIC VS
.
PERCEPTUAL FACTORS
. In summary, a
combination of articulatory overlap and the decline in subglottal pressure in final
position account for many of the devoicing patterns found cross-linguistically.
Interestingly, the articulatory and aerodynamic factors which induce devoicing are in
conflict with the perceptual factors militating against devoicing. Devoicing of
vowels is articulatorily and aerodynamically natural under certain conditions as
shown in the last section; however, voiceless vowels are perceptually less salient than
voiced vowels as argued in section 2. The conflict between articulatory/aerodynamic
factors and perceptual considerations is evident when we compare Hupa, in which
long but not short vowels devoice, with the many languages (e.g. Cheyenne) in
which short but not long vowels undergo devoicing. This conflict can be modeled in
the grammar by assuming different ranking of the relevant constraints in the two
language types. In Hupa, *N
ON
-
MODAL
S
HORT
V
OWELS
is ranked above the
relevant constraint forcing devoicing, whereas, in Cheyenne, *N
ON
-
MODAL
S
HORT
V
OWELS
is ranked lower than the relevant constraint driving vowel devoicing
between voiceless consonants. By shifting the rankings slightly we get other
patterns. For example, the Acoma pattern, in which both short and long vowels
devoice, is derived by ranking both *N
ON
-
MODAL
S
HORT
V
OWELS
and *N
ON
-
MODAL
L
ONG
V
OWELS
below the constraint driving devoicing. If, on the other
hand, we rank both *N
ON
-
MODAL
S
HORT
V
OWELS
and *N
ON
-
MODAL
L
ONG
V
OWELS
above constraints requiring devoicing, we get a language without devoicing
of any vowels. The rankings which generate the attested patterns are shown in (1).
For expository purposes, the set of constraints which force devoicing are collapsed
as a single constraint devoice. Although space limitations preclude doing so in this
paper, these constraints can easily be divided into narrower (or broader, as in Hupa)
constraints capturing the further asymmetries discussed in this paper.
(1) Ranking
Chia sẻ với bạn bè của bạn: