Incomplete Neutralization and the (A)symmetry of Paradigm

Incomplete Neutralization and the (A)symmetry of Paradigm Uniformity
Abby Kaplan, University of Utah, [email protected]
It is well known that neutralization may be incomplete, such that speakers produce a small
difference between apparently neutralized segments for at least some cues. In German, for example,
devoiced final obstruents in words like /vEöd/ ∼ [vEöt] ‘become.imp’ have shorter closure durations,
are preceded by longer vowels, have shorter releases, and have more voicing than non-alternating
obstruents in words like /vEöt/ ∼ [vEöt] ‘be worth.imp’ (Port and O’Dell 1985; the precise cues
involved vary by study). Similar effects have been found for a variety of contrasts in many languages.
A prominent hypothesis regarding the source of these contrasts points to the fact that the final [t] of [vEöt] ‘become.imp’ corresponds to [d] in morphologically related forms such as [vEöd@n]
‘become.inf’. Perhaps the [t] is slightly voiced because it is under pressure to resemble the corresponding [d]. This view is attractive because there is an observable tendency more generally
for stems in the same paradigm to be realized in a uniform way (e.g., Kenstowicz 1996). If this
hypothesis is correct, then we predict that such intra-paradigmatic influence might go in the other
direction as well: the [d] of [vEöd@n] ought to be slightly devoiced under the influence of the [t] of
[vEöt]. This prediction has been little discussed in the literature (Ernestus and Baayen 2007 is, to
my knowledge, the only example) and has not been tested experimentally.
The experiments reported here fill this gap in the literature: they test whether a non-neutralized
form (such as [vEöd@n]) can be influenced by a related neutralized form (such as [vEöt]). I explore
this question in two different neutralization patterns: final devoicing in Afrikaans (Experiment 1),
and unstressed vowel reduction in Russian (Experiment 2).
Participants in Experiment 1 were eight native speakers of Afrikaans. Stimuli consisted of the
singular and plural forms of 28 Afrikaans nouns; half ended in an alveolar stop ([t] or [d]) and half
in an alveolar stop followed by schwa. (Sample stimuli are shown in Table 1.) The stop-final stimuli
form the plural by adding [@]; the schwa-final stimuli, by adding [s]. Afrikaans has final devoicing
much like German; thus, stem-final voiced stops are realized as voiceless in the singular, while
non-final voiced stops do not alternate. If non-neutralized forms are influenced by paradigmatically
related neutralized forms, then the alternating [d] of hoede ([hut] ∼ [hud@]) should be slightly less
voiced than the non-alternating [d] of roedes ([rud@] ∼ [rud@s]). But, as shown by the following
results, alternating and non-alternating stops did not differ in the relevant cues (preceding vowel
duration, closure duration, etc.).
As shown in Figure 1, vowels were longer before voiced /d/ than before voiceless /t/, even in
final position (sproet vs. hoed, p = 0.000896 in a linear model of singular consonant-final stimuli
with fixed effects of Voicing and Log Frequency and random effects of Subject, Stem, and Vowel);
thus, neutralization of the vowel length cue is incomplete. But there is no evidence that the vowel
of hoede was shorter than the vowel of roedes (p = 0.368 in a linear model of plural stimuli with
/d/ with fixed effects of Stem-Final? and Log Frequency and the same random effects as above).
For consonant closure duration, shown in Figure 2, neutralization as a result of final devoicing
was complete: the /d/ of hoed was no shorter than the /t/ of sproet (p = 0.815 in a linear model with
the same effects described above). There was also no evidence of a difference between alternating
hoede and non-alternating roedes (p = 0.205 in a model with the same effects described above).
Data collection for Experiment 2 is ongoing; three native speakers of Russian have participated
so far. Stimuli consist of the nominative and instrumental singular forms of 46 monosyllabic Russian
nouns, each with one of the four vowels /i/, /e/, /a/, or /o/. (Sample stimuli are shown in Table 2.)
Half of the stimuli keep stress on the stem when the instrumental suffix /-om/ is added; the other
Final
Non-Final
Voiceless
sproet ‘freckle’
roete ‘route’
Voiced
hoed ‘hat’
roede ‘rod’
Stress Shift
No Shift
hoed(e)
/o/
kot ‘cat’
grot ‘grotto’
Table 2: Sample Russian stimuli
Table 1: Sample Afrikaans stimuli
sproet(e)
/a/
vrač ‘doctor’
krax ‘crash’
sproet(e)
0.30
hoed(e)
0.20
0.25
0.15
0.15
0.10
0.10
0.05
0.05
roete(s)
0.30
roede(s)
0.25
CDur
VDur
0.20
0.20
roete(s)
roede(s)
0.15
0.20
0.15
0.10
0.10
0.05
0.05
sing
plur
sing
plur
sing
plur
sing
plur
Figure 1: Vowel duration by condition in Exper- Figure 2: Consonant closure duration by condiiment 1
tion in Experiment 1
half shift stress to the suffix. Neutralization results from vowel reduction in unstressed syllables:
/i, e/ → [i]; /a, o/ → [a].
Preliminary results for Experiment 2 provide evidence that unstressed vowel reduction in Russian is in fact incomplete. Reduced /a/ and /o/ differ on both F1 and F2, while reduced /i/ and
/e/ differ on F1 only (all differences significant at p < 0.05, from models of F1/F2 in instrumental
forms of nouns with a stress shift with random effects of Subject and Lemma). By contrast, alternating and non-alternating [o] (e.g., kot [kot] ∼ [ka."tom] vs. grot [grot] ∼ ["gro.tam]) do not differ
from each other; nor do alternating and non-alternating [e] (p < 0.05 for all models of F1/F2 in
nominative forms of nouns with [o]/[e], with the same random effects described above).
Both Experiment 1 and Experiment 2 fail to yield any evidence that a non-neutralized form can
be influenced by a neutralized form in the same morphological paradigm. This is unlikely to be the
result of insufficient power, since both experiments were able to detect ordinary incomplete neutralization. The evidence suggests that sub-phonemic paradigm uniformity effects are asymmetrical:
a non-neutralized form can influence a neutralized form (as in ordinary incomplete neutralization),
but effects in the opposite direction are not attested.
References
Mirjam Ernestus and R. Harald Baayen. Intraparadigmatic effects on the perception of voice. In Jeroen van de
Weijer and Erik Jan van der Torre, editors, Voicing in Dutch: (De)voicing – Phonology, Phonetics, and Psycholinguistics, number 286 in Current Issues in Linguistic Theory, pages 151–172. John Benjamins Publishing Company,
Amsterdam, 2007.
Michael Kenstowicz. Base identity and uniform exponence: Alternatives to cyclicity. In Jacques Durand and Bernard
Laks, editors, Current Trends in Phonology: Models and Methods, volume 1, pages 365–395. European Studies
Research Institute, Salford, 1996.
Robert Port and Michael O’Dell. Neutralization of syllable-final voicing in German. Journal of Phonetics, 13:455–471,
1985.