Whispery voiced nasal stops in rwanda

Full text

EUROSPEECH 2001 - SCANDINAVIA
651
Whispery voiced nasal stops in Rwanda
Didier Demolin
*
and Véronique Delvaux
*+
*
Phonology Laboratory,
+
F.N.R.S.
Université Libre de Bruxelles
ddemoli@ulb.ac.be
Abstract
The paper describes the main phonetic characteristics
(acoustic, aerodynamic and articulatory) of Kinyarwanda
prenasalized stops, focussing on voiceless consonants. We
conclude from instrumental observations that the phonetic
description of these sounds should be redefined. Consonants
previously described as voiceless prenasalized stops in
Rwanda are in fact whispery voiced nasal stops. Finally, the
paper shows that the description of these sounds raises several
important questions about nasal venting and the control of the
velum closure.
1. Introduction
Jouannet [1] describes kinyarwanda, a language spoken in
Rwanda by approximately 7.000.000 speakers, as having
three groups of prenasalized stops in its phonetic inventory,
i.e. (i) a set of voiced and voiceless prenasalized stops
[mb, mv, nd, nz, n, , mh, mf, nh, ns, n, n, h]; (ii)
a set of voiced and voiceless labiovelarized prenasalized
stops [mb, mv, ndw, nzw, nw, gw,
mhn, nhw, nskw, nkw, hw]; and (iii) a set of voiced
and voiceless palatalized prenasalized stops
[mb, nd, , mpy, nh, nsc, hy]. The labiovelarized,
palatalized and voiceless sounds are rather unusual and
present a number of problems that need an accurate
description if one whishes to understand how these sounds are
produced. The voiceless set of sounds
[mh, nh, h, mhn, nhw, hw, nh, hy] shows
that there are voiceless nasals preceding and following the
aspirated part of the consonant. This rather rare phenomenon
must also be demonstrated and explained.
Our instrumental study leads to a modification of
Jouannet’s [1] description of complex nasal consonants in
Rwanda. The main change is that there are no voiceless nasals
except in front of voiceless fricatives and that aspirated
sounds are in fact fully voiced. The paper describes
Jouannet’s voiceless and aspirated nasals as whispery voiced
nasal stops [i.e as
mH, nH, NH, mHw, nHNw, NHw, mHN, nH`, Nhy] and justifies
this modification on an instrumental basis. Material and
method
2. Material and method
Aerodynamic recordings were made using the Physiologia
workstation [2] linked to a data collection system equipped
with different transducers. Oral airflow measurements were
taken with a small flexible silicon mask placed against the
mouth. Nasal airflow was measured at the end of one nostril
via a small tube linked to the data collection system.
Pharyngeal pressure was recorded with a small flexible plastic
tube (ID 2mm) inserted through the nasal cavity into the oro-
pharynx.
Acoustic recordings were made with the same material
via a High Fidelity microphone set on the hardware piece
equipment connecting the transducers to the computer.
Spectrograms and audio waveforms are processed with the
iShell software (www.tribeworks.com).
Seven speakers took part in the experiment (5 women and
2 men, S1 to S7). The material presented to the speakers
consisted of a set of words inserted in a small carrying
sentence “vuga --- icumi” (“say --- ten times”). The words
studied in the experiment are presented in table 1 omitting
tones. The focus of this study is on the following contrasts
and consonants:
[mH/mb, NH/Ng, NHw/ngw and mHN/mN] and [nHNw, nH`].
Several aerodynamic and acoustic measurements of the
data were made : two measures of nasal airflow, the
maximum value observed during the production of these
consonants (M.N.af) and the duration of the increase in
airflow (from the beginning of the increase to the maximum
value) (D.N.af). The maximum value of the oral airflow was
measured after the stop closure release (M.O.af). Pharyngeal
pressure was measured at the maximum value observed
during the production of these consonants (M.Ps) and the total
duration of positive pressure was also measured (D.Ps). The
duration of these sounds was measured on the audio
waveform. The initial point was placed on the audio
waveform after the vowel preceding the nasal consonant and
the final point was placed just before the beginning of the
following vowel. When this point was difficult to establish
because of the labial transitions, a comparison with the
spectrogram was made to decide where to place the
measurement point.
3. Results and measurements
3.1. Aerodynamics
Aerodynamic measurements of whispery voiced nasal stops
[mH, NH, NHw, mHN, nHNw and nH`] (see section 4 for a
discussion of this term) show that there is a very important
amount of nasal airflow when these sounds are produced.
Figure 1 shows aerodynamic measurements, the audio
waveform and a spectrogram for the word [iNHa] ‘cow’.
Aerodynamic plots show a simultaneous increase in nasal
airflow and in pharyngeal pressure, the increase being made
in two phases. The maximum value of nasal airflow and
pharyngeal pressure are reached at the same moment. The oral
airflow plot is slightly negative at the beginning and increases
only slightly when pharyngeal pressure and nasal airflow
decrease.
Table 2 gives the mean value of the different aerodynamic
measurements. The duration of the increase in nasal airflow

652
shows that this increase in airflow takes more time for
whispery voiced nasals stops than for their non-whispery
voiced counterparts (134 ms vs 102 ms), in the oppositions
[mH/mb, NH/Ng, NHw/ngw]. The maximum value of nasal
airflow is always much higher for whispery voiced nasal stops
(146 ml/s) than for the voiced prenasalized stops (40 ml/s).
The maximum value of oral airflow measured after the stop
closure release shows that there is a higher oral airflow after
the non-whispery voiced nasal stops (126 ml/s) than after the
whispery consonant (50 ml/s). Pharyngeal pressure, which
was measured at the maximum value observed during the
production of these consonants, also shows that pressure was
higher during the non-whispery voiced nasal stops (5.2 hPa)
than during the whispery consonant (2.6 hPa). The total
duration of positive pharyngeal pressure measured from the
beginning of the increase in pressure to the return to the
atmospheric pressure value is longer for the whispery
consonants than for the non-whispery counterpart (187.6 ms
compared to 97.4 ms).
3.2. Acoustics
Table 2 gives the duration of these consonants. The main
observation is that there seems to be a timing difference
between the shorter ‘simple’ prenasalized stops and the longer
‘complex’ labio-velarized stops.
Careful observation of the audio waveform shows that
there are two phases during the production of whispery voiced
nasal stops. The first phase corresponds to the initial part of
the nasal and the second phase corresponds to a very
noticeable increase in the voicing amplitude accompanied by
some additional noise.
Spectrograms show the aspiration noise at the end of the
whispery voiced nasal stops (Figure 1). Note however that
most of the turbulence generated by the aspiration is not
present on the spectrogram. This is due to the measurement
system which records the airflow at the end of a nostril and
which therefore does not make the recording of this noise
possible.
One last interesting point worth noticing is the presence of
click bursts in the middle of some nasal consonants
([mHN and nHNw]). These click bursts are clearly visible on
the audio waveforms of the recordings of some speakers.
Figure 2 shows audio waveforms of the word [inHNwaro]
‘weapon’ and compares variations made by three different
speakers. In the first variation there is no click, but in the
second and the third variation there is clear click burst on the
audio waveform. Figure 3 shows one realization of the word
[imHNemHNe] ‘chest hair’ in which two variations of the same
sequence [mHN] can be seen, the second showing a click burst.
The significance of these click bursts will be discussed in
section 4.
4. Discussion
Before discussing the data and the results presented in the
paper, a comparison with some known facts in the world’s
languages is necessary. Ladefoged [3: 13-14] states that
murmured consonants are common in southern Bantu
languages such as Shona, Tsonga and some members of the
Nguni group: ‘In all these languages, during the murmured
sounds the vocal cords seem to me to be slightly closer
together than in the Indian languages, so that there is more
voice and less breath escaping; nevertheless they contrast
clearly with the mode of vibration of the vocal cords which
occurs in regular voiced sounds’. Ladefoged also shows that
in Shona there are voiced murmured nasals and that in the
Nguni languages Ndebele and Zulu the situation is somewhat
similar. There are contrasting voiced and murmured nasals
like those in Shona. ‘These sounds, together with a type of [h]
which is realized as a murmured onset of a vowel (…) form a
phonological class recognizable because they may cause a
noticeable lowering of the tone on the subsequent vowel’.
Traill and Jackson [4] show similar facts in their description
of Tsonga phonation types. Finally, it is important to note that
the data given by Doke [5] present many similarities with the
Rwanda phenomena. Indeed, as in Rwanda, under nasal
influence in Shona : k > H, p > mH, r > nd, Π> nd, t > nH, s >
ts, and √ > mb. More recent discussions of similar
phenomena can be found in Maddieson [6] on Shona and in
Matangwane [7] on Ikalanga.
Laver [8: 236] also notices the Shona data and treats the
phenomenon of murmur as a matter of co-ordination between
a contoïd and a following vocoïd under the heading of voiced
aspiration rather than as an attribute of a single segment.
Laver treats murmur as a phonation type and identifies this
phenomenon as whispery/breathy voiced nasal stops in Shona
and in the languages quoted by Ladefoged, Traill and Jackson
[3,4]. This fact was also noticed by Catford [9: 105-6], who
equates Ladefoged’s murmur with a type of whispery voice.
Our definition of the Rwanda whispery voiced nasal stops
is in accordance with Laver [10: 121], who states that when
whisper combines with another laryngeal setting such as
modal voice (to give the compound phonation type of
whispery voice) there is necessarily a greater amount of inter-
harmonic noise than in the simple modal voice phonation
type, a fact which is observed in Rwanda.
Careful observation of the nasal airflow, oral airflow,
pharyngeal pressure, audio waveform and spectrograms of
Rwanda data presented in figure 1 and in table 2 allow us to
propose an explanation to the way these sounds are produced.
The important increase in nasal airflow observed during the
production of the whispered voiced nasal stops [mH, NH,
NHw, mHN, nHNw and nH`] accounts for the fact that most of
the airflow escapes through the nasal cavities at this time. The
fact that there is no observable oral airflow at the same time
and also the fact that while nasal airflow increases there is
also a small increase in pharyngeal pressure suggest that the
pharyngeal cavity, almost or completely closed at this time,
has its volume slightly reduced. This closure can be achieved
by sealing the space between the velum and the tongue
dorsum, by velum lowering or by tongue dorsum raising.
Several additional articulatory movements, such as an
elevation of the larynx or a contraction of the pharynx, can
lead to a reduction of the volume of the pharyngeal cavity and
account for the observed pharyngeal pressure values. This last
hypothesis obviously needs further investigation to be
demonstrated, but from the aerodynamic data and from
acoustic observations this seems to be a reasonable
explanation.
Variations observed in the whispery voiced nasal stops
show examples where a click burst is observed (Figures 2 and
3). This supports the hypothesis of the temporary separation
between nasal and oral airways. These clicks, which are
clearly audible, but not phonologically relevant, are quite
similar to the examples of Xhosa alveolar clicks shown by

EUROSPEECH 2001 - SCANDINAVIA
653
Ladefoged and Traill [11]. Their presence in phonetic
variations of Rwanda whispery voiced nasal stops are a strong
indication of a temporary separation between the oral and
nasal cavities which also explains the very high nasal airflow
observed.
About the labio-velar nasals, Ladefoged [12] presents a
case in Idoma where in the double articulation [Nm] a bilabial
click accompanies the velar nasal. This click component is
heard when the lip closure is released. Ladefoged adds that
this click component is not very audible and is different from
nasal clicks which occur in Xhosa. In their 1996 book on the
sounds of the world’s languages [13], Ladefoged and
Maddieson re-discuss this case and rephrase Ladefoged’s
1964 interpretation in order ‘to avoid the possibly misleading
implication that these sounds are made in the same ways as
clicks’. The main point for this rephrasing is that in the labio-
velar nasals of Idoma [Nm], there is no salient click burst and
that these sounds are acoustically predominantly resonant in
character. We believe that Rwanda is similar to Idoma and
that Ladefoged’s first interpretation [12] seems more
appropriate. However, a few points must be clarified before
discussing this case. First note that Rwanda descriptions (such
as Jouannet’s [1]) transcribe the labio-velar nasal consonant
as [mhN], indicating that the velar closure comes second. This
point is crucial because the click component observed in the
Rwanda labio-velar nasals [mHN] is produced by the
rarefaction of air in the oral cavity during the double closure.
Since the labial closure is first released with a velar closure
behind, a weak click component can occasionally be heard.
This click is easily observable on the audio waveforms and is
different from the bilabial click [>] of <xoDo). Indeed in the
latter more noise is produced, and it is also less abrupt than
what is observed in Rwanda.
5. Conclusion
An instrumental study of complex nasal consonants in
Rwanda has led to a modification of their description as made
by Jouannet [1]. Our observations show that no voiceless
nasals are observed in this language except in front of
voiceless fricatives and that aspirated sounds are in fact fully
voiced. Therefore, consonants previously described as
voiceless prenasalized stops in Rwanda should now be
described as whispery voiced nasal stops. Finally, the
description of these sounds suggests that there could be an
active control of velum closure during the production of these
whispery voiced nasal stops.
6. References
[1] Jouannet, F., “Phonétique et phonologie des
consonnes du kinyarwanda” In F. Jouannet (ed.). Le
kinyarwanda langue bantu du Rwanda. Etudes linguistiques,
Paris, SELAF, 1983, 55-73.
[2] Teston, B. and Galindo, B., “Design and
development of a workstation for speech production analysis”,
Proceedings of VERBA90 : International conference on
speech technology, Rome, 400-408, 1990.
[3] Ladefoged, P., Preliminaries to Linguistic Phonetics,
University of Chicago Press, Chicago, 1971.
[4] Traill, A. and Jackson, M., “Speaker variation and
phonation type in Tsonga” Journal of Phonetics, 16: 385-400,
1988.
[5] Doke, C.M., A Comparative Study of Shona
Phonetics, The University of Witwatersrand Press,
Johannesburg, 1931.
[6] Maddieson, I., “Shona velarization; complex
segments or complex onsets?” UCLA Working papers in
Phonetics, 72: 16-34, 1990.
[7] Mathangwane, J.T., Ikalanga Phonetics and
Phonology, CSLI Publications, Stanford, 1999.
[8] Laver, J., Principles of Phonetics. Cambridge
University Press, Cambridge, 1994.
[9] Catford., J.C., Fundamental Problems in Phonetics,
Edinburgh University Press, Edinburgh, 1977.
[10] Laver, J., The Phonetic Description of Voice Quality.
Cambridge University Press, Cambridge, 1980.
[11] Ladefoged, P. and Traill, A., “Clicks and their
accompaniments.” Journal of Phonetics, 22: 33-64, 1994.
[12] Ladefoged, P., A phonetic study of West African
languages, Cambridge University Press, Cambridge, 1964.
[13] Ladefoged, P. and Maddieson, I., The Sounds of the
World’s languages, Oxford, Blackwell, 1996.
Table 1. Consonants and words studied in the experiment.
API Rwanda Gloss
[mH] [imHamba]
Food for travelling
[Ng] [iNga‘i]
Mountain gorilla
[NH] [iNHa]
Cow
[mHN] [imHN
mHNmHN
mHNemNe]
Chest hair
[nHNw] [inHNwaro]
Weapon
[NHw] [iNHwano]
Dowry
[Ngw] [iNgwe]
Leopard
[nH`] [inH`ooza]
Eloquent person
[mN] [imHNemN
mNmN
mNe]
Chest hair
[m9f] [im9fiizi]
Bull
[ntS] [intSuti]
Friend
Table 2. Mean value of aerodynamic and acoustic
measurements (N=7): segment duration (Dur., in ms), duration
of the increase in nasal airflow (D.N.af, in ms), maximum
values of nasal airflow, oral airflow and pharyngeal pressure
(respectively, M.N.af, in ml/s, M.O.af, in ml/s and M.Ps, in
hPa) and total duration of positive pressure (D.Ps, in ms).
Dur.
(ms)
D.N.af
(ms)
M.N.af
(ml/s)
M.O.af
(ml/s)
M.Ps
(hPa)
D.Ps
(ms)
[mb]
120 85 30 90 3.39 76
[mH]
108 115 120 50 1.36 145
[Ng]
121 111 50 160 7 103
[NH]
130 155 150 50 4.08 206
[Ngw]
140 111 50 130 5.3 112
[NHw]
164 131 170 50 2.5 210
[nHNw]
156 149 140 30 2.4 216
[mHN]
151 160 100 50 1.3 181
voiced
102 40 126 97.4 5.2
whispery
voiced
134 146 50 187.6 2.6

654
Figure 1. Spectrogram, audio waveform, pharyngeal pressure, nasal and oral airflow for the word [iNHa] ‘cow’.
1) start of the nasal consonant on the audio waveform; 2) stabilization in the nasal airflow and in pharyngeal pressure;
3) maximum of nasal airflow and of pharyngeal pressure; 4) start of an increase in the audio waveform amplitude;
5) maximum oral airflow; 6) start of the next vowel.
Figure 2. Audio waveforms of three variations of the word [inHNwaro] ‘weapon’.
1) Increase in the voicing amplitude. To be compared with : 2) and 3) click bursts.
Figure 3. Audio waveform of the word [imHNemHNe] ‘chest hair’ showing two different variations of the cluster [mHN].