University of Florida Department of Linguistics On the Lack of Fricative Articulations of [B] in the World’s Languages Author: Charles J. Pindziak April 11, 2012
University of FloridaDepartment of Linguistics
On the Lack of Fricative Articulations of[B] in the World’s Languages
Author:Charles J. Pindziak
April 11, 2012
This paper is submitted in partial satisfaction of the honors requirements for the degree Bachelorof Arts in linguistics. It was advised by Dr. Fiona McLaughlin.
Copyright c© 2012 by Charles J. Pindziak.
Contents
1 Introduction 1
2 Methods 42.1 Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2 Recordings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3 Processing and Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.4 Criteria for Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Results 93.1 Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93.2 Selected Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103.3 Ambiguous Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4 Discussion 184.1 Aerodynamics of [B] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184.2 Examples of True Fricative [B] in Ewe Rounded Environments . . . . . . . . . . . . . 204.3 Data from L1 Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5 Modeling [B] 265.1 Straw Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6 Phonological Ramifications 306.1 Issues with the Reclassification of [B] . . . . . . . . . . . . . . . . . . . . . . . . . . . 306.2 Boersma’s (2008) Three-Form Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 326.3 Formalizing [B] as Both an Approximant and a Fricative . . . . . . . . . . . . . . . . 33
7 Conclusion 37
i
List of Figures
1.1 Ewe: [eBe]; “Ewe” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Waveform of [v] in the Kwangali word [evu] “ground” . . . . . . . . . . . . . . . . . 72.2 Spectrogram of the Kwangali word [evu] “ground” . . . . . . . . . . . . . . . . . . . 8
3.1 Bemba: [Baka], [BaBa]; “take care of” – “itch” . . . . . . . . . . . . . . . . . . . . . . 113.2 Kwangali: [veta], [Beta]; “fetch water” – “beat” . . . . . . . . . . . . . . . . . . . . . 123.3 Ewe: [eBe], [eve]; “Ewe” – “two” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123.4 Initial consonant and transition into vowel of Ganda word [Bon:a] “all” . . . . . . . . 133.5 Ganda: [Bon:a]; “all” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.6 Ganda: [kuBa]; “to be” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.7 Logba: [uBa]; “measles” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153.8 Logba: Comparison of waveforms [B] (top) and [v] (bottom) from minimal pair [uBa],
[uva]; “measles” – “side” (amplitude not to scale) . . . . . . . . . . . . . . . . . . . . 16
4.1 Ewe: [Bu]; “to open” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214.2 Waveform of [B] segment from Ewe [Bu]; “to open” . . . . . . . . . . . . . . . . . . . 214.3 Ewe: [eBe]; “Ewe” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224.4 Ewe: [Be(
>tsi)]; “rain” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.1 Model of channel formed at the lips . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.1 Sonority curves with [B] in a complex onset . . . . . . . . . . . . . . . . . . . . . . . 316.2 Mart́ınez-Celdrán’s (2004) classification of approximants . . . . . . . . . . . . . . . . 326.3 Boersma’s (2008) model of the phonology-phonetics interface . . . . . . . . . . . . . 33
ii
List of Tables
2.1 Languages for analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.1 Ewe word list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.1 Straw lengths (in mm) and areas (in mm2) . . . . . . . . . . . . . . . . . . . . . . . 275.2 Channel models and their capability to produce turbulent flow . . . . . . . . . . . . 29
6.1 Phonetic evaluation of [laba] “he washes” . . . . . . . . . . . . . . . . . . . . . . . . 356.2 Evaluation of /kabRa/ “goat” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366.3 Phonetic evaluation of [Bu] “to open” . . . . . . . . . . . . . . . . . . . . . . . . . . 36
iii
Acknowledgements
I am grateful to several people who helped make this project the best that it could be; the errors
that remain are mine alone. First, I thank Fiona McLaughlin, my faculty advisor, for giving me the
independence to pursue my research interests wherever they take me, while at the same time always
giving me support and guidance so that I stay on the right path. I also thank Ratree Wayland,
who taught me everything I know about phonetics and has always kept her office door open to
listen to my ideas and offer friendly feedback on previous drafts of this and other papers. I thank
Jimmy Harnsberger for introducing me to spectrogram analysis, and I am grateful to Sean Witty
for taking a freshman with no experience in linguistics and giving him the confidence to produce
professional-quality research (I hope).
This research also benefited from funding from UF’s University Scholars Program and Depart-
ment of Spanish and Portuguese, which financed my trip to the Universitat de Barcelona to create
recordings of Catalan speakers. I thank Maria-Rosa Lloret for her indispensable help recruiting
speakers and Ana Maria Fernández-Planas for giving me access to the university’s phonetics lab
and equipment. I am also grateful to them and their students for being so welcoming towards me
and making me feel at home throughout my stay in Barcelona.
On a more personal level, I thank my parents for always supporting me and encouraging me
to follow my interests and talents. Lastly, but most importantly, I thank my grandfather, Roland
Chapman, for teaching me at a young age what the square-root button on a calculator does, even
though I had convinced myself that whatever math behind it was too hard to learn. Were it not
for our whimsical discussions from poker probabilities, to perpetual-motion machines, to P?= NP,
I would not have conceived of this project, nor have had the intellectual drive to execute it.
iv
Abstract
Although the voiced bilabial fricative, [B], is purported to exist in several of the world’s languages,
tokens from a sample of 9 languages that are claimed to contain [B] showed no evidence of frication
noise. The lack of noise implies that the bilabial constriction is not narrow enough to create a
turbulent airstream, which is fundamental to the fricative class. Therefore, the bilabial continuant in
these languages should be classified as an approximant. Mathematical modeling and some empirical
evidence of bilabial constrictions with rounded lips indicate that fricative articulations of [B] are
possible in rounded environments, but since [B] is usually realized with the lips spread, true fricative
productions of [B] are much rarer than traditionally claimed.
Chapter 1
Introduction
In the UCLA Phonological Segment Inventory Database (UPSID), the voiced bilabial fricative, [B],
occurs in 55 of the 451 indexed languages, i.e., 12.2 percent of the languages in the database. This is
one of the highest instances among the nonstrident fricatives, surpassed only by [v] (96 languages =
21.29%) and very slightly by [G] (56 = 12.42%). What is puzzling about the frequency of [B] is that
its voiceless counterpart, [F], occurs in only 41 of the languages in UPSID (9.09%). Bilabial is the
only place of articulation of fricatives with a cross-linguistic voiced-to-voiceless ratio greater than
one.1 This trend is a challenge to phonetic markedness since voiced fricatives are more difficult
to articulate than voiceless ones, because air pressure needs to be elevated in two places in the
vocal tract: below the glottis to produce voicing and behind the constriction to produce frication.
Indeed, Ohala (1997) reports a cross-linguistic tolerance of voiced stops over voiced fricatives, with
an estimation of 24 percent of the world’s languages lacking voiced stops in their inventories as
opposed to 38 percent lacking voiced fricatives. Given this markedness and the generalization
that asymmetries in language inventories include the unmarked member over the marked one, it is
puzzling that [B] occurs more often than [F] in language.
Another peculiarity of [B] is evident in an acoustic study of voiced fricatives by Jackson & Shadle
1Maddieson (1984:45), using an earlier version of UPSID containing 317 languages, points out that the voiced-to-voiceless ratio of dental fricatives is also greater than one, with 21 languages containing [D] in their inventories and18 containing [T]. The languages added to UPSID after Maddieson’s publication have tilted this ratio below one.
1
(2000). The researchers used a computer algorithm to separate voiced fricatives into their harmonic
(voicing) and anharmonic (noise/turbulence) components. They discovered that [B], as produced
by their research participants, displayed an anharmonic component with a magnitude very close to
zero, requiring [B] tokens to be excluded from the rest of their analysis. Very little anharmonicity,
thus dominated by a periodic component, would suggest an approximant articulation of [B]. Since
voicing is unmarked in sonorants, an approximant articulation of [B] would be consistent with the
voiced-to-voiceless ratio mentioned earlier.
This raises the question of whether [B] has been misclassified in many of the languages in UPSID
resulting in the puzzling voiced-to-voiceless ratio. Misclassification of approximants as fricatives
is not uncommon. Recasens (1991) and Mart́ınez-Celdrán (2004) have presented spectrographic
evidence that [B] and several other so-called voiced fricatives are articulated as approximants in
Catalan and Spanish. Examining the Basque language, Hualde classified [B] as a fricative in his
phonological investigation (1991), but later classified the sound as an approximant in a phonetically-
based article for the Illustrations of the IPA series (2010). In fact, [B] claimed as a true fricative
rarely, if ever, surfaces in the phonetics literature. While a spectrogram of the Ewe word [eBe] “Ewe”,
appears in Ladefoged & Maddieson 1996:142, the spectrogram does not manifest any frication noise
during the articulation of [B], which is fundamental to the fricative manner.
Recordings taken at the University of Florida of an Ewe speaker’s production of [B] (discussed
in more detail in Chapter 4.2) demonstrate another characteristic of the articulation of [B]. In the
waveform and spectrogram of this speaker’s production of [eBe] (Figure 1.1), after the transition
from the vowel (1), the [B] segment is initially devoiced (2) to facilitate production of turbulent
flow, and then gains voicing (3) shortly before transitioning to the following vowel. At the onset
of voicing, however, the fricative quality is lost. The reduction in flow caused by an adducted
glottis makes adequate pressure build-up for voiced-fricative production phonetically difficult. This
is especially true for [B] because the precision for forming a narrow constriction is more easily
accomplished by the tongue than the lips. Besides the question of whether [B] is often misclassified
in languages, these Ewe recordings and the lack of true fricative [B] in the phonetics literature raise
another question, namely, whether true fricative articulations of [B] (i.e. with simultaneous voicing
2
and frication) can be found in any language.
Figure 1.1: Ewe: [eBe]; “Ewe”
This paper examines the waveforms and spectrograms of tokens of [B] in 9 languages for evidence
of fricative articulations. The languages under examination, the source of the recordings, and the
criteria for fricative classification are all addressed in the Chapter 2. Chapter 3 presents the results
of this examination, showing that fricative articulations of [B] are virtually non-existent in language.
Chapter 4 offers some aerodynamics-based information for the paucity of fricative [B] and displays
a few cases where true fricative [B] tokens surfaced in rounded environments (i.e. adjacent to a
rounded vowel) in a follow-up experiment. Chapter 5 explores modeling production of [B], and
Chapter 6 discusses the effects of non-fricative [B] on phonology.
3
Chapter 2
Methods
2.1 Languages
Most of the recordings for this analysis were obtained from the UCLA Phonetics Lab Archive
(http://archive.phonetics.ucla.edu/). According to the homepage of this archive, the database
contains recordings of over 200 languages with transcriptions from the original field documents. The
interface allows the user to search by phonetic symbol, and the search results return all the languages
in the database that have that phonetic symbol anywhere in the accompanying transcriptions.
A search of [B] was conducted on the database. From the search results, all the language record-
ings of too poor quality for phonetic analysis were discarded. This included, for example, recordings
with high-amplitude background noise and recordings that were sampled at a low rate (< 10kHz).
For all of the languages that had not been discarded yet, the accuracy of the accompanying tran-
scriptions was verified with an outside source, to affirm that the languages indeed are purported to
contain [B] in their inventories. Languages for which the existence of [B] in their inventories could
not be verified were also discarded.
This process eliminated all but the 9 languages that are shown in Table 2.1. The “Reference”
column displays sources that attest that the associated languages indeed have [B] in their inventories.
Note that, as mentioned earlier, Hualde’s 2010 phonetic analysis concludes that Basque’s bilabial
4
continuant is an approximant, contrary to his 1991 phonological analysis (cited in Table 2.1), which
labels this segment as a fricative. Despite Hualde’s subsequent correction, Basque is still included
in the present analysis to offer a second look. Table 2.1 also shows the phonemic status of [B] in
these languages and any significant contrasts with [B] in their inventories.
Table 2.1: Languages for analysis
Language Family Contrasts Reference
Basque isolate (/b/ → [B]) Hualde (1991)Bemba Bantu /B, w/ Sims (1959)
Catalan Romance (/b/ → [B]) Wheeler (2005)Ewe Kwa /B, v, F, f/ Ladefoged & Maddieson (1996)
Ganda Bantu /B, v, f, v:, f:/ Ashton et al. (1954)Kwangali Bantu /B, v/ Ladefoged & Maddieson (1996)
Logba Kwa /B, v, F, f/ Ladefoged & Maddieson (1996)Mazatec (Jalapa) Oto-Manguean /B/ Garellek & Keating (2011)
Venda Bantu /B, v, F, f/ Ziervogel et al. (1972)
2.2 Recordings
All of the recordings of these languages from the UCLA database were created using word lists. The
native speakers were prompted to read one word at a time, and therefore, the recordings contain
careful but not quite hyperarticulated tokens. Since these recordings are void of coarticulatory
pressures that come about from fluid speech, they are most likely to show evidence of fricative
articulations, which require a constricted channel that might not quite be realized all the time in
casual speech.
The details of the recordings are available on the UCLA Archive website. Most were created
on tape and later digitized and stored as WAV files. The UCLA database creators used a sampling
rate of 44kHz to digitize all analog recordings, even though most of the UCLA analog recordings do
not preserve frequencies above 11kHz. Therefore, downsampling these recordings to 22kHz removes
noise introduced during the digitization process without discarding information from the original
signal.
The only language for which the recordings did not come from the UCLA database was Catalan.
5
The Catalan recordings come from a corpus that I recorded at the Universitat de Barcelona in
a sound-attenuated booth in the university’s phonetics lab, with a Shure SM10A head-mounted
microphone connected to a Marantz Professional PMD671 solid-state recorder, sampling at 44kHz.
2.3 Processing and Software
Prior to analysis, all the recordings were downsampled to 22kHz using the command-line program
Sound eXchange (http://sox.sourceforge.net/), since the analog recordings did not contain
information about the signal above 11kHz prior to digitization (as mentioned above). The lack of
information in the signal above 11kHz is not a problem for the present analysis of [B]. All fricatives
show at least part of their frication noise below this 11kHz cutoff. Moreover, the noise spectrum of
[B] is diffuse, spanning many frequencies from low to high (cf. [v]), since it is produced at the lips
with no channel in front to shape the noise.
The analysis of the recordings was done in Praat (http://www.fon.hum.uva.nl/praat/). The
default settings in Praat were changed to produce spectrograms from 0 to 10kHz. This larger
frequency window helps make frication noise more visible.
2.4 Criteria for Classification
Voiced fricatives are comprised of a (quasi)periodic component from the vibration of the vocal
folds as well as an aperiodic component created as air passes through the supraglottal constriction.
Therefore, the presence of both periodic and aperiodic energy is fundamental to the fricative class.
In a waveform, a voiced-fricative articulation is apparent by a sinusoidal component (periodic) with
random deviations from periodicity that look like a jagged overlay (aperiodic), as in Figure 2.1.
In a spectrogram, a voiced fricative is manifested by noise spanning across many frequencies
(aperiodic) modulated by periodic striations, which are residuals of vocal-fold vibration. Aperiodic
noise as well as striations are visible during the articulation of [v] in Figure 2.2, which begins about
halfway into the spectrogram.
6
Figure 2.1: Waveform of [v] in the Kwangali word [evu] “ground”
In the examination of the purported voiced-bilabial-fricative segments in the 9 languages in
Table 2.1, both the waveforms and the spectrograms were consulted. A judgment was made that
classified them as fricatives if and only if the segments manifested both a periodic and an aperiodic
component. The results of these judgments appear in Chapter 3.
It may seem that the harmonics-to-noise ratio (see Qi & Hillman (1997) for an overview, or
Boersma (1993) for the mathematically-minded) may be helpful for this discrimination, but it
ultimately does not aid in making a distinction between fricative and non-fricative. At best, it
computes an accurate value in dB that suggests how much noise is present in the signal, but then
the researcher must impose an arbitrary critical value below which a segment is a fricative and
above which it is a non-fricative. The critical value could be established empirically by measuring
the harmonics-to-noise ratio of many fricative [B] and non-fricative [Bfl] and then demarcating the
midpoint between the two groups as the threshold, but since the present analysis does not assume
that true fricative [B] exists, this empirical basis is not possible. Therefore, harmonics-to-noise ratio
measurements were not consulted for fricative–non-fricative discriminations.
7
Figure 2.2: Spectrogram of the Kwangali word [evu] “ground”
8
Chapter 3
Results
3.1 Details
All tokens of purported [B] had a periodic component from voicing, but overwhelmingly they lacked
an aperiodic, noisy component to justify their classification as fricatives. The results of this dis-
criminatory analysis appear in Table 3.1. The “Environments” column provides a summary of the
contexts in which [B] tokens occurred in the recordings of the language. “/B/-/v/ pairs” indicates
that the researcher had the subject contrast /B/ and /v/ in successive minimal pairs. All tokens
were produced as approximants, i.e. lacking a channel constricted enough to produce aperiodic
noise,1 except the few tokens produced in environments followed by parenthetical fractions in Ta-
ble 3.1. The fractions convey how many of the tokens in that environment were not produced as
approximants, and the manner of articulation of these tokens is listed in the rightmost column,
“Alt. productions”. Of the 126 total tokens, only 3 were produced as possible fricatives, labeled in
the table as “aperiodic continuant” and “weak fricative”. Even so, these 3 cases are ambiguous,
1There is sometimes confusion about the articulation of certain fricatives. Strident (also called sibilant) fricatives(e.g. [s]) are produced with a constriction, creating an aperiodic component, and then the airflow comes in contactwith an obstacle, namely, the teeth. The contact with the obstacle amplifies the frication noise such that stridentfricatives are the noisiest segments. [B] is a non-strident fricative; that is, it is articulated with a constriction,creating an aperiodic component, but the airflow does not make contact with an obstacle. Therefore, non-stridentfricatives contain an aperiodic component, but it does not have as high an amplitude as the strident fricatives. Thisis sometimes confused as non-strident fricative lacking an aperiodic component altogether, but that is not the case;non-strident fricative simply do not have as noisy an aperiodic component as stridents.
9
and are addressed in Chapter 3.3.
Table 3.1: Data
Language Speakers Tokens Environments (non-approx) Alt. productions
Basque male 4 V V (1/2), V u (2/2) stop (3)Bemba female 8 V V, # , u w
Catalan 1 M, 3 F 32 V V, u u, i i (1/4), V w, V R stop (1)Ewe male 8 V V, /B/-/v/ pairs
Ganda male 7 # o (1/1), u a, u i, i i, o u aperiodic cont. (1)Kwangali 2 M, 1 F 23 # , V V, u e, /B/-/v/ pairs
Logba male 3 u a (2/3) weak fricative (2)Mazatec 4 M, 1 F 35 # V
˜Venda male 6 #
3.2 Selected Data
The following waveforms and spectrograms were chosen as representative of the total 126 tokens
under investigation. The Bemba words [Baka] “take care of”, and [BaBa] “itch” in Figure 3.1 show
[B] articulations lacking fricative noise (compare to the noisy release of [k]). Strong formants are
visible in the second [B] articulation in [BaBa]. This is typical of many intervocalic [B] tokens in the
corpus.
Figure 3.2 shows a minimal pair in Kwangali: [veta], “fetch water”, and [Beta] “beat”. Noise
is apparent about halfway through the labiodental fricative, [v], but no noise surfaces in the artic-
ulation of [B]. The duration of [B] is substantially shorter than that of [v], perhaps indicative of a
more glide-like articulation. Short-duration [B] was most common intervocalically in the examined
tokens, but word-initial short-duration [B] was not rare. The formants in [B] show a subtle decrease
in frequency, as predicted by Perturbation Theory for labial constrictions, but it is not as robust
as it would be if it were accompanied by a movement of the back of the tongue towards the velum.
That is, this approximant articulation of [B] is not the same as the labio-velar approximant [w]
because it does not have a velar gesture.
Ewe is perhaps the best known language to contain the voiced bilabial fricative in its inventory,
10
Figure 3.1: Bemba: [Baka], [BaBa]; “take care of” – “itch”
but even Ewe shows approximant-like articulations of [B] instead of fricative.2 In the minimal pair
[eBe] “Ewe”, and [eve] “two”, noise is clearly visible in the labiodental fricative [v] but lacking in
[B] (Figure 3.3).
2It will be shown in Chapter 4.2 that in recordings from a follow-up experiment, a native Ewe speaker produced[B] as a fricative in some rounded environments, but not as a fricative in all the environments where [B] is said tosurface.
11
Figure 3.2: Kwangali: [veta], [Beta]; “fetch water” – “beat”
Figure 3.3: Ewe: [eBe], [eve]; “Ewe” – “two”
12
3.3 Ambiguous Cases
As seen earlier in Table 3.1, all the tokens lacked fricative qualities except a few cases from Ganda
and Logba, which do not quite manifest true fricative qualities but leave some uncertainty. These
tokens are presented in this section.
A close examination of the waveform of the initial consonant of the Ganda word [Bon:a] “all”
(Figure 3.4), reveals some irregularities that could be labeled as aperiodicity. The waveform shows
low amplitude at the onset of the [B] articulation, which then reaches a minimum before the vowel
transition, presumably arising from the greatest degree of constriction during the production. Al-
though this waveform is messy, it does not quite show the particular aperiodicity of fricatives. In
addition, the spectrogram of [Bon:a] (Figure 3.5) does not show any evidence of an initial fricative.
Moreover, other tokens from Ganda show no evidence whatsoever of frication, such as [kuBa] “to
be” (Figure 3.6), where it is difficult to pick out the intervocalic consonant by the waveform or
spectrogram.
Figure 3.4: Initial consonant and transition into vowel of Ganda word [Bon:a] “all”
13
Figure 3.5: Ganda: [Bon:a]; “all”
Figure 3.6: Ganda: [kuBa]; “to be”
14
Another interesting case comes from Logba. The three tokens mentioned in Table 3.1 are all of
the same word: [uBa] “measles”. The informant repeated this word three times in a row, and two
of the tokens showed some weak frication noise. One of those tokens appears in Figure 3.7. The
spectrogram displays some noise, but it is not clear whether it is simply an artifact of the recording.
This word forms a minimal pair in Logba with [uva] “side”, and a comparison of the waveforms
of the two intervocalic fricatives appears in Figure 3.8. The labiodental fricative [v] shows jagged
edges close to the peaks in the period, which are typical of fricatives but lacking in the [B] token.
This comparison with [v] implies that the Logba [B] tokens are not really fricatives, either.
Figure 3.7: Logba: [uBa]; “measles”
15
Figure 3.8: Logba: Comparison of waveforms [B] (top) and [v] (bottom) from minimal pair [uBa],
[uva]; “measles” – “side” (amplitude not to scale)
These few ambiguous cases support a greater point: Even when the benefit of the doubt is given
to questionable fricative segments, upon further scrutiny they fail to exhibit characteristics shared
by other voiced fricatives in the language, or fricatives in general. Given the over 100 tokens from
16
9 different languages across several language families, it seems unlikely that [B] is ever consistently
realized as a fricative.
17
Chapter 4
Discussion
4.1 Aerodynamics of [B]
Fricatives are produced by channeling air through a small constriction. The velocity of the air flow
and the size of the constriction determine whether the resultant flow will be turbulent (fricative) or
laminar (non-fricative). Since a more constricted channel can consequently increase the velocity of
the flow, and since articulations are not simply supraglottal processes but involve control of the lungs
as well, modeling fricatives is complex. Catford (1977:121) concludes that fricative articulations are
typically possible with channels having cross-sectional areas between 3 mm2 and about 20 mm2,
with approximant and resonant articulations above 20 mm2. Similarly, Stevens (2000:33) delineates
the normal range1 for fricative-producing channels as between 5 mm2 and 20 mm2, with vowel-like
productions from 20 mm2 to 30 mm2 to as much as 300 mm2 (= 3 cm2). That is, compared to
approximants and vowels, fricatives are only possible in a very narrow range of constriction sizes.
Forming channels of 20 mm2 or less can be accomplished by the tongue given its musculature, but
the lips are much more restricted. Since [B]2 is normally produced with unrounded lips (Ladefoged
1However, Stevens (2000) cites the MRI research of English fricatives by Narayanan et al. (1995), which showsfricative articulations by some subjects exceeding 20 mm2, though none of the research subjects’ productions exceeded30 mm2.
2One might wonder why this discussion is centered on [B] with no mention of its voiceless counterpart, [F]. Thisis because voiceless fricatives, voiceless approximants, and voiceless vowels are all noisy. Therefore, it is impossible
18
& Maddieson 1996:141), and since the average width of adult lips is 55.60 mm for males and 50.89
mm for females (Ferrario et al. 2000), a constriction distributed across the width of the lips could
have a vertical clearance no greater than 0.36 mm for men and 0.39 mm for women. Achieving this
level of constriction seems unlikely, particularly with the co-articulatory pressures of fluent speech.
It is especially unlikely in languages in which [B] surfaces as the output of a lenition process, since
the realization of a stop seems articulatorily easier.
Although 20 mm2 is just an estimate of the maximal area for fricatives, it is worth noting
that because of the aerodynamics of voicing, voiced-fricative productions with channels substantial
greater than 20 mm2 do not seem possible.3 Voiced fricatives are even more difficult to produce
because they require two locations of high air pressure: below the glottis to stimulate vocal-fold
vibration and behind the constriction to create turbulent flow. The continuous opening and closing
of the vocal folds decreases the mean velocity of the air flow, making turbulence especially diffi-
cult to achieve. Voicing is sometimes temporarily suspended during part of the articulation, and
the environment surrounding the fricative affects the likelihood of the suspension of voicing (Hag-
gard, 1978; Stevens et al., 1992). For example, intervocalic voiced fricatives are more often voiced
throughout their articulation than voiced fricatives adjacent to another voiced fricative. The sus-
pension of voicing facilitates turbulent-flow production, though it is not likely to permit turbulent
flow for much greater constriction sizes.
Even though [B] is usually unrounded at least in the 9 languages under examination, an ar-
ticulation of [B] in a rounded environment would decrease the width of the lips such that the
maximal length of the vertical constriction to produce turbulence would be more reasonable. This
is investigated in the following section.
to distinguish a fricative articulation of [F] and an approximant articulation of [Ffl] solely based on the presence or
absence of noise. Perhaps this distinction could be argued based on the size of the constriction, but it is more difficultto justify than noise presence.
3Indeed, Catford (1977) and Stevens (2000) both came up with the 20-mm2 threshold by looking primarily atvoiceless fricatives.
19
4.2 Examples of True Fricative [B] in Ewe Rounded Envi-
ronments
To test the effects of rounded lips on the articulation of [B], a male native speaker of Ewe was
recorded producing a list of VCV, VCCV, CV, and CCV words, which appear in Table 4.1 tran-
scribed in the IPA. The consonants in these words were stops, fricatives, and approximants, and all
tokens were actual words in Ewe. The speaker read through the list three times. For the present
study, only the tokens containing [B] were analyzed; the others served as distractors. The recording
was made on a Shure SM10A head-mounted microphone in a sound-attenuated room at the Uni-
versity of Florida and was sampled at 44kHz. The tokens from this recording were not included in
the dataset in Table 3.1.
Table 4.1: Ewe word list
Word Meaning Word Meaning
[eãa] you cooked [Bu] to open[eBe] Ewe [eBu] you open[eve] two [exe] you capture[eFa] you polished [Gla] to hide[efa] you was cold [eGla] you hid
[eBlo] mushroom [ewO] you made[eFli] you bought [ehe] you pulls[evlo] you are evil [wu] to kill
[efli] you cut [e>gbO] you came back
[Bã] to move, stir [e>kpO] you saw
[Be>tsi] rain
[B] showed true voiced-fricative characteristics in two out of the three repetitions of [eBu] as
well as all three repetitions of [Bu]. The waveform and spectrogram of one of the of the [Bu] tokens
appears in Figure 4.1. A very noisy component is easily visible in the spectrogram halfway through
the articulation of [B]. A closeup of the waveform of just the [B] portion appears in Figure 4.2. In
the waveform, the onset of frication noise is also easily recognizable starting at the halfway point
of the production.
20
Figure 4.1: Ewe: [Bu]; “to open”
Figure 4.2: Waveform of [B] segment from Ewe [Bu]; “to open”
21
Something curious surfaced (introduced in Chapter 1) in this native speaker’s productions of
[B] in unrounded environments, as well. For example, in the speaker’s production of [eBe] “Ewe”
(Figure 4.3), noise is apparent despite the unrounded articulation. However, the production of
frication noise seems to come at the cost of voicing. In this example, after the transition (1) from
the previous vowel into the fricative segment, unmodulated (i.e. voiceless) frication noise appears
in the spectrogram (2), but once voicing is introduced (3), frication noise vanishes.
Figure 4.3: Ewe: [eBe]; “Ewe”
This interaction between frication noise and voicing makes sense when considering the role of
aerodynamics. Voicing ceases with the abduction of the vocal folds while permitting a larger volume
of air to pass through the glottis. This increase in volumetric flow rate causes frication noise even
though the lips may not be ideally constricted for a bilabial fricative.4 When voicing returns,
however, the adduction of the vocal folds substantially decreases the volumetric flow rate such that
4Similarly, voiceless approximants and voiceless vowels are noisy because of the increase in flow from a spreadglottis, despite the wider constriction.
22
the conditions for frication are no longer met.
This raises the question as to whether this token can really be labeled a voiced bilabial fricative.
Voicing and frication do not occur simultaneously, yet they both appear at some point during the
segment’s production. Haggard (1978) and Stevens et al. (1992) both have documented that periods
of devoicing occasionally occur during the production of voiced fricatives; however, they note that
intervocalic voiced fricatives are the least likely to show periods of devoicing, yet all three tokens
of [eBe] of this speaker had devoicing, with very little overlap between the voiced and fricative
components. The three tokens of [Bu], on the other hand, had simultaneous voicing and frication
throughout.
A few other caveats need to be added to this informant’s productions of [eBe]. The duration
of [B] is much longer than that of a typical intervocalic fricative, approaching close to 200 ms.
Indeed, [B] in Figure 4.3 appears to be almost as long as the surrounding vowels. Moreover, the
token of [eBe] from the UCLA Archive (which appeared in Figure 3.3) is much shorter and shows
no frication, even though both tokens were recorded using similar procedures (list with mono- and
bisyllabic words, mostly VCV). Therefore, this articulation of intervocalic [B] may be idiosyncratic
to this informant.
It is worth noting that not all unrounded articulations of [B] from this speaker showed frication
noise and devoicing. Rather, some articulations were completely periodic, displaying no frication
and maintaining voicing throughout, like most of the tokens analyzed in Table 3.1. Only one of
the three repetitions of [Be>tsi] had a faint amount of voiceless noise during the production of the
initial consonant, whereas the other two tokens were voiced and periodic (as in Figure 4.4, which
shows only the first syllable, [Be]). Hence, even though this informant is more likely to produce
devoiced fricative tokens of [B] than the informant from the UCLA Archive, he still produced some
frictionless exemplars of [B]. Perhaps this number would be greater in flowing speech.
23
Figure 4.4: Ewe: [Be(>tsi)]; “rain”
More native speakers of languages containing [B] are needed to verify the effects of rounding
on [B] productions. Rounding may not be equally as effective in triggering fricative articulations
of [B] in all languages. For example, the rounded vowel preceding [B] in the Ganda token [kuBa]
“to be”, did not facilitate frication noise (Figure 3.6), but the same rounded vowel may have been
responsible for creating weak frication in the Logba tokens of [uBa] “measles” (Figure 3.7).
4.3 Data from L1 Acquisition
While acoustically [B] lacks the characteristics of a voiced fricative, it does not manifest many of the
characteristics of an approximant, either. Approximants are typically high-amplitude segments with
clear formant structure. However, many of the [B] tokens have a low amplitude with weak formants
at best, though there are some tokens that look quite vowel-like, particularly those resulting as
output from a lenition process. Catford (1977) proposes a continuum from fricative to approximant
24
to vowel based on constriction size, so if a voiced consonant is lacking the level of constriction to
produce noise, it would fall into the approximant category, regardless of whether it has the typical
characteristics associated with approximants.
The argument that [B] is phonetically an approximant and not a fricative could be made stronger
if language-acquisition data showed that children acquire [B] around the same time as approximants.
Vowel-like segments are learned by children fairly quickly, but fricatives are difficult to acquire,
especially voiced fricatives. If children produce [B] closer to when they start producing approximants
rather than when fricative productions surface in their speech, then the phonetic classification of
[B] as an approximant is meaningful.
While [B] is in the phonetic inventory of Catalan and Spanish, which have been extensively
studied for L1-acquisition data, similar research is lacking in languages in which /B/ is a phoneme.
Since [B] in Catalan and Spanish surfaces as an allophone of /b/ through a lenition process, it
requires that the child understand a phonological rule, which causes a delay in acquisition that is
not due to the articulatory complexity of the segment alone. Therefore, only acquisition data from
languages with phonemic /B/ can provide insight on the classification of [B].
L1-acquisition data from languages with consonants phonetically similar to [B] in their phonemic
inventory, such as [v] (Ewe, Ganda, Kwangali, Logba, Venda) and [F] (Ewe, Logba, Venda), would
be the most useful for [B] classification. This would be a productive basis for future research.
25
Chapter 5
Modeling [B]
5.1 Straw Models
Modeling [B] in subjects is easier than many other sounds because [B] is produced at the outer
end of the vocal tract (the lips), so a satisfactory model can be created with minimal invasion
and minimal corruption of natural speech conditions. An experiment was conducted with drinking
straws of various sizes to shape the lips of a subject to determine at what cross-sectional area
frication noise appears.
Physical modeling of [B] has certain advantages over pure mathematical models (e.g. with
Reynolds number). For instance, the lungs expel air at a greater flow rate through a less constricted
vocal tract than a more constricted vocal tract (see Prathanee et al. 1994), and this complex trade-
off is difficult to compute. Physical models simplify the involved mathematics to achieve greater
reliability by eliminating uncertain or unknown parameters and thus reducing inaccuracies.
5.2 Methods
Five drinking straws with different diameters were used, which appear in Table 5.1. One male
subject with experience in phonetics participated. To model production of [B], a straw was placed
26
in the subject’s mouth, and he was asked to shape his lips around the straw without rounding or
puckering. The straw was then removed, and the gap between subject’s lips was examined to make
sure it retained the same diameter of the straw. The subject was then asked to place his tongue in
the position for [@] and phonate normally (i.e. the same volume as conversational speech) without
changing the configuration of his lips. This procedure was repeated for each of the five straws. The
speaker’s productions were recorded using a Tascam DR-05 Linear PCM Recorder positioned 30 cm
left of the speaker’s mouth. This ensured that airflow from the mouth did not come in contact with
the microphone. The recordings were analyzed in Praat.
Table 5.1: Straw lengths (in mm) and areas (in mm2)
Straw Diameter Cross-Sectional Area Real Area of Channel
1 3 7.07 13.932 5 19.63 38.713 6 28.27 55.744 7 38.48 75.875 7.5 44.18 87.09
For precision, the straw ideally would be kept in the speaker’s mouth throughout the production;
however, this was not an option for the current experiment. Even when the straw was cut down
to a few centimeters in length, it shaped the noise in the acoustic signal such that it was difficult
to observe the presence or lack of an aperiodic component. When the straw was cut to match the
cross-sectional width of the lips, it could not be held in place by the subject and was ejected shortly
after the onset of phonation.
When drinking, the lips pucker to clamp down on a straw to create an airtight channel, fa-
cilitating suction, but the musculature of unrounded (un-puckered) lips is not able to maintain a
precise, circular gap. Rather, the cross-sectional area of the channel created at the lips with a
straw as a guide is shaped like Figure 5.1. The yellow circle represents the cross-sectional area
of the straw, and the area outside the circle is an approximation of the additional space resulting
from the imperfect closure of the lips around the straw. The angles are estimates but nevertheless
produce a much more accurate model than a simple circle. The area of the model in Figure 5.1 can
be calculated by determining the area of one of the right triangles (Equation 5.1) and multiplying
27
the result by 4 (there are four triangles: two drawn on the right side and the two not drawn on
the left side of the model). After adding the area of the 80◦ of the circle that is not overlapped
by the triangles (Equation 5.2), the area of the whole model is known. The total-area function is
simplified to Equation 5.3. “Real Area of Channel” in Table 5.1 was calculated with this equation,
whereas “Cross-Sectional Area” refers to the cross-section of the straw alone.
Figure 5.1: Model of channel formed at the lips
RR
70◦
70◦
Area of triangle =R2 tan 70◦
2(5.1)
Area leftover in circle =2
9· πR2 (5.2)
A = 2R2(
tan 70◦ +π
9
)(5.3)
5.3 Results
The channel formed by Straw 1, the smallest straw, was the only channel capable of producing
a turbulent airstream at the volume of normal speech (Table 5.2). This is consistent with Cat-
ford’s (1977:121) and Stevens’s (2000:33) calculations that 20 mm2 is the upper bound for the
cross-sectional area of turbulence-producing channels (Chapter 4.1).
28
Table 5.2: Channel models and their capability to produce turbulent flow
Straw Real Area of Channel (mm2) Turbulence?
1 13.93 yes2 38.71 no3 55.74 no4 75.87 no5 87.09 no
In addition to steady, normal-intensity phonation, the subject was asked to produce intensity
(i.e. sound pressure) glides for each of the lip configurations. Reetz & Jongman (1998:68) note that
intensity increases by about the square of the increase in subglottal pressure; increased subglottal
pressure induces a greater volumetric flow rate, such that the conditions for turbulence may be
achieved. After the subject’s lips were configured with the straw, he began to phonate softly and
gradually elevated the intensity of his voice up to the point at which he was yelling behind his
lips. With increased intensity and subglottal pressure, the channel configured with Straw 2 was
able to produce turbulence at 16 dB greater than the intensity level at which turbulence surfaced
in Straw 1. Channels configured with Straws 3–5 did not produce turbulence even at +18 dBStraw 1
(the loudest that the speaker could raise his voice). This reinforces the conclusions in Chapter 4.1
that real-world unrounded bilabial constrictions resulting in turbulent flow cannot be achieved.
29
Chapter 6
Phonological Ramifications
6.1 Issues with the Reclassification of [B]
Even though the phonetic evidence in Chapters 3–5 supports [B] as an approximant rather than
a fricative, the reclassification of [B] is phonologically problematic in languages that require a low
sonority value for [B]. For example, [.Bl] and [.BR] are valid onsets in Catalan and Spanish, but if [B]
were considered a glide-like approximant, it would have a sonority value greater than the following
liquid in the onset. As a fricative, [B] has a low sonority value, yielding a sonority curve that rises
through the following liquid into the vowel (Figure 6.1 (a)), but [B] as an approximant produces
a curve that falls then rises to the vowel (Figure 6.1 (b)). Wheeler (2005) was the first to point
out these effects on sonority and to contest the reclassification of [B] as an approximant in Catalan,
despite Recasens’s (1991) spectrographic evidence that [B] has vowel-like qualities.
Minimal Sonority Distance constraints make rectifying this sonority mismatch even more dif-
ficult. Mart́ınez-Celdrán (2004) advocates for the label spirant to be applied to [Bfl], and defines
spirant as a subclass of approximant, alongside rhotics as central approximants (Figure 6.2). This
granular taxonomy for approximants may justify assigning spirants their own sonority value, which
could be lower than taps and other liquids, such that sonority would rise through [BflR] and [B
fll] on-
sets. However, Catalan and Spanish forbid onset clusters of a sibilant-fricative followed by a liquid,
30
Figure 6.1: Sonority curves with [B] in a complex onset
sonority
segmentB R V
(a) [B] is a fricative
sonority
segmentB R V
(b) [B] is an approximant
e.g. *[.zl] and *[.Zl], whereas these languages allow [.BR] and [.Bl]; therefore, the sonority distance
between [B] and liquids must be greater than that of sibilant fricatives and liquids, and hence, [B]
must be an obstruent.
The observation that [B] must be an obstruent phonologically but is realized phonetically as
an approximant cannot be explained by the simple /underlying form/ → [surface form] model of
phonology. Sonority operates only on the surface form, because this is the level where syllabifi-
cation takes place, but the surface form, which is identical to the phonetic realization in tradi-
tional phonology, shows no characteristics of a fricative articulation that is necessary to account
for syllable structure in Catalan and Spanish. Rather, the behavior of [B] in syllables requires a
phonological model that separates the phonology-driven surface form from the articulatory-minded
phonetic form. This can be accomplished with Boersma’s (2008) model of the phonetics-phonology
interface.
31
Figure 6.2: Mart́ınez-Celdrán’s (2004) classification of approximants
approximants
semi-vowels
[j î 4 w]
consonants
centrals
laterals
[l l L L]rhotics
[ô R]
spirants
[Bfl
V Dfl JflG]fl
6.2 Boersma’s (2008) Three-Form Model
Boersma (2008) proposes a model of phonology where the mapping of discrete segments onto a
continuous phonetic form is nontrivial. The model consists of three levels: the underlying form,
the surface form, and the phonetic form. The underlying form is defined the same way as in classic
phonology; it is the discrete, abstract mental representation of a lexical item. The surface form is
an intermediate form between the underlying form and the phonetic form; it is made up of discrete
units and is where all phonological processing (e.g. assimilation, syllabification) takes place. The
phonetic form1 is a continuous, concrete production of the surface form. The phonetic form can be
represented by its spectrogram, or the set of fully-implemented (i.e. non-abstract) motor commands
that govern the production of the utterance. For simplicity, the phonetic form can be represented
with IPA symbols, keeping in mind that they are placeholders for a fully-specified, continuous
sound. This model is illustrated in Figure 6.3. Note that single brackets are used for the surface
form and double brackets for the phonetic form.
A profound discussion on the validity of this model is beyond the current scope, though it
1Boersma (1998) further divides the phonetic form into the auditory form and the articulatory form, since themapping from perception to production is not trivial, either; however, he shows in his later work (2008) that a singlephonetic form suffices for simpler applications.
32
Figure 6.3: Boersma’s (2008) model of the phonology-phonetics interface
/Underlying Form/
phonological processing
[Surface Form]
mapping to motor commands
[[Phonetic Form]]
/pæt/
aspiration
[phæt^]
has been successful at capturing phonological phenomena that are difficult or impossible to handle
with classic phonology (e.g., see Boersma 2007 for this model’s thorough treatment of h-aspiré in
French). It also captures the observation that syllable boundaries are necessitated by phonology
but have no acoustic correlate; syllable boundaries operate on the surface form and are discarded
during the mapping to the phonetic form. See Boersma (1998, 2008, 2009) for further discussion of
this model.
6.3 Formalizing [B] as Both an Approximant and a Fricative
In Catalan and Spanish, [B] results as the output of a lenition process for the input /b/. Under
Optimality Theory, lenition is motivated my the Lazy constraint, which is violated on a gradient
by the degree of articulatory effort required to articulate the sound (1). Kirchner (1998) defines
articulatory effort in detail, but to summarize, precision and articulator displacement are the two
main factors of effort. For example, sounds that require a precisely constricted channel (e.g. strident
fricatives) take more effort to produce than sounds articulated with a more relaxed vocal-tract
configuration (e.g. approximants). Taking into account the articulatory-effort scores of different
manners of articulation yields the relative hierarchy in (2).
33
(1) Lazy: Minimize articulatory effort. (Kirchner 1998)
(2) Relative articulatory effort: affricates > stridents > stops > non-stridents > approximants
(after Kirchner 1998:ch. 4)
Lazy operates on the phonetic form in Boersma’s (2008) three-form model because its domain
is the mapping of segments onto articulatory gestures. Lazy is at odds with *Map(b, Open), a cue
constraint that disallows the mapping of [b] onto a continuant articulation (3). Cue constraints are
the correlates of faithfulness constraints at the phonetic level; whereas faithfulness constraints are
concerned with preserving the features of an underlying form in the surface form, cue constraints
preserve elements of the surface form in the phonetic form. As explained by Boersma (2008), cue
constraints “express the speaker-listener’s knowledge of the relations between continuous auditory
cues and discrete phonological surface elements.” In the case of *Map(b, Open), the speaker-
listener knows that a good phonetic realization of [b] is one with a full obstruction of the vocal
tract. To account for lenition, Lazy must outrank *Map(b, Open) (4).
(3) *Map(b, Open): Do not map [b] onto an articulatory command without a full closure.
(4) Lazy � *Map(b, Open)
The ranking in (4) is sufficient to handle lenition in Catalan and Spanish. This is demonstrated
in the evaluation of the Spanish word [laba]→ [[laBfla]] “he washes” (Tableau 6.1). The approximant
[[Bfl]] requires the least articulatory effort and is thus optimal. Following the effort hierarchy after
Kirchner (1998) in (2), [[B]] incurs only a slightly more severe violation of Lazy; however, given the
aerodynamic evidence in Chapter 4.1 regarding the production of fricative [[B]], it is clear that this
sound requires a substantially greater amount of effort (i.e. precision and articulator displacement)
outside of a rounded environment. This is why the violation of Lazy by fricative [[B]] has been
marked as the most severe among the candidates listed in Tableau 6.1. It is worth noting that in a
stochastic implementation of OT (see, for example, Boersma & Hayes 2001), the overlap between
Lazy and *Map(b, Open) would be such that *Map(b, Open) would outrank Lazy in a minority
of productions. In these cases, the candidate that does not violate *Map(b, Open) would be
34
realized phonetically, i.e. [[laba]], which is expected and attested (e.g. in Table 3.1, 3 [B] tokens in
Basque and 1 in Catalan were realized as stops).
Tableau 6.1: Phonetic evaluation of [laba] “he washes”
[laba] Lazy *Map(b, Open)
a. laba ∗!
b. laBa ∗!∗∗ ∗
c. + laBfla ∗
d. laza ∗!∗ ∗
The pairing of Boersma’s (2008) model of the phonetics-phonology interface with Kirchner’s
(1998) articulatory-based Lazy makes a powerful statement on lenition, namely, that lenition has
no effect on syllabification. Consider the evaluation of the surface and phonetic forms in parallel for
the Catalan and Spanish word /kabra/ “goat” in Tableau 6.2, in which the constraint responsible
for syllabification, SonSeq (5), has been added. Lazy is concerned only with the [[phonetic form]],
whereas SonSeq sees only the [surface form]. With no motivation to lenite up until the phonetic
realization, the fully-faithful and sonority-satisfactory [ka.bRa] is optimal, which is paired with the
optimal phonetic form [[kaBflRa]] because Lazy � *Map(b, Open). Although the surface form
[ka.BflRa] could be proposed as a candidate by Gen from a rich base (candidate d.), it violates
SonSeq and any relevant Id-Manner-type faithfulness constraints not mentioned in Tableau 6.2.
Indeed, the lenited form is only seen at the phonetic level.
(5) SonSeq: Sonority rises through an onset.
This Lazy-based model can be applied even to languages with /B/ as a phoneme. Since con-
straints are universal, Lazy exists in all languages, but it is lowly ranked in non-leniting languages,
such that it may not be effective at guaranteeing the realization of [[Bfl]]. To overcome this, the
granularization of Lazy, as done by Kirchner (1998:ch. 6), is useful. For a language with /B/ as
a phoneme (realized as [[Bfl]]) that does not have lenition of stops, Lazy(Mid) (6) is ranked low
so that stops are realized with a full closure, while Lazy(High) (7) is ranked high so that /B/ is
35
Tableau 6.2: Evaluation of /kabRa/ “goat”
/kabRa/ SonSeq Lazy *Map(b, Open)
a. [ka.bRa], [[kabRa]] ∗!
b. [ka.bRa], [[kaBRa]] ∗!∗∗ ∗
c. + [ka.bRa], [[kaBflRa]] ∗
d. [ka.BflRa], [[kaB
flRa]] ∗!
articulated as [[Bfl]]. Mid and High could be substituted with real articulatory-effort scores (see
Kirchner 1998:ch. 2). It does not seem unreasonable to expect that languages have some con-
straint that blocks very difficult productions (Lazy(High)). Perhaps this is analogous to English
/nEkst+stAp/ → [[nEkstAp]].
(6) Lazy(Mid): Avoid production of sounds that require a moderate degree of effort.
(7) Lazy(High): Avoid production of sounds that require a very high degree of effort.
In addition, this OT model can explain the minority of productions of true fricative [[B]] in
rounded environments by the Ewe speaker in Chapter 4.2. For example, in the production of [[Bu]],
anticipatory coarticulation causes rounding of [[B]]. This rounding brings the lips close enough
together to facilitate turbulent flow and true fricative production of [[B]] (Tableau 6.3). In fact, it
would require more articulatory effort (i.e. articulator displacement) to produce an approximant
in this case because the lips would either have to be pulled farther apart (candidate b.), or the
coarticulatory pressures brought on by [[u]] would have to be overcome (candidate c.).
Tableau 6.3: Phonetic evaluation of [Bu] “to open”
[Bu] Lazy
a. + Bwu
b. Bflu ∗!
c. Ḃu ∗!
36
Chapter 7
Conclusion
Waveform and spectrographic evidence from a sample of 9 languages has shown that [B] is rarely
realized as a true voiced fricative. The only case where true voiced fricative realizations of [B]
surfaced was in rounded environments (e.g. before the vowel [u]) in a recording of a native Ewe
speaker. The decrease in lip width caused by the rounded environment creates a constriction small
enough to produce frication noise. However, rounding does not appear to have the same effect on
[B] in all languages.
Aerodynamic calculations with Reynolds number and physical models using drinking straws to
simulate the production of [B] at bilabial constrictions of different sizes attest to the impracticality
of true fricative realizations of [B]. While the reclassification of [B] as an approximant may seem
problematic for phonological theory, more detailed phonological models that better capture the
phonetics-phonology interface can handle the mismatched phonetic and phonological characteristics
of [B].
Research on the perception of [B] is needed to provide more insight on its classification. Frication
noise (e.g. onset frequency of noise) is an important perceptual cue for fricative identification and
discrimination. Since [B] lacks this noise component, it would be important to see what other
phonetic qualities of [B] are perceptually relevant, and whether these phonetics qualities coincide
with those used for the perception of approximants.
37
References
Ashton, E., Mulira, E., Ndawula, E. & Tucker, A. (1954). A Luganda grammar . Longmans, GreenLondon.
Boersma, P. (1993). Accurate short-term analysis of the fundamental frequency and the harmonics-to-noise ratio of a sampled sound. In Proceedings of the Institute of Phonetic Sciences. Amster-dam. vol. 17, 97–110.
Boersma, P. (1998). Functional phonology . Ph.D. dissertation, University of Amsterdam.
Boersma, P. (2007). Some listener-oriented accounts of h-aspiré in French. Lingua 117. 1989–2054.
Boersma, P. (2008). Emergent ranking of faithfulness explains markedness and licensing by cue.Rutgers Optimality Archive 954.
Boersma, P. (2009). Cue constraints and their interactions in phonological perception and produc-tion. In Boersma, P. & Silke, H. (eds.), Phonology in perception, Berlin: Mouton De Gruyter.55–110.
Boersma, P. & Hayes, B. (2001). Empirical tests of the gradual learning algorithm. LinguisticInquiry 32. 45–86.
Catford, J. (1977). Fundamental problems in phonetics, vol. 134. Edinburgh University Press.
Ferrario, V., Sforza, C., Schmitz, J., Ciusa, V. & Colombo, A. (2000). Normal growth and devel-opment of the lips: a 3-dimensional study from 6 years to adulthood using a geometric model.Journal of Anatomy 196. 415–423.
Garellek, M. & Keating, P. (2011). The acoustic consequences of phonation and tone interactionsin Jalapa Mazatec. Journal of the International Phonetic Association 41. 185–205.
Haggard, M. (1978). The devoicing of voiced fricatives. Journal of Phonetics 6. 95–102.
Hualde, J. (1991). Basque phonology . Taylor & Francis.
Hualde, J., Lujanbio, O. & Zubiri, J. (2010). Goizueta Basque. Journal of the International PhoneticAssociation 40. 113–127.
Jackson, P. & Shadle, C. (2000). Frication noise modulated by voicing, as revealed by pitch-scaleddecomposition. The Journal of the Acoustical Society of America 108. 1421–1434.
38
Kirchner, R. (1998). An effort-based approach to consonant lenition. Ph.D. dissertation, Universityof California (Los Angeles).
Ladefoged, P. & Maddieson, I. (1996). The sounds of the world’s languages. Wiley-Blackwell.
Maddieson, I. (1984). Patterns of sounds. Cambridge University Press.
Mart́ınez-Celdrán, E. (2004). Problems in the classification of approximants. Journal of the Inter-national Phonetic Association 34. 201–210.
Narayanan, S., Alwan, A. & Haker, K. (1995). An articulatory study of fricative consonants usingmagnetic resonance imaging. The Journal of the Acoustical Society of America 98. 1325–1347.
Ohala, J. (1997). Aerodynamics of phonology. In Proceedings of the Seoul International Conferenceon Linguistics. vol. 92, 97.
Prathanee, B., Watthanathon, J. & Ruangjirachuporn, P. (1994). Phonation time, phonationvolume and air flow rate in normal adults. Journal of the Medical Association of Thailand 77.639–645.
Qi, Y. & Hillman, R. (1997). Temporal and spectral estimations of harmonics-to-noise ratio inhuman voice signals. The Journal of the Acoustical Society of America 102. 537–543.
Recasens, D. (1991). Fonètica descriptiva del català: Assaig de caracterització de la pronúncia delvocalisme i consonantisme del català al segle XX , vol. 21. Institut d’Estudis Catalans.
Reetz, H. & Jongman, A. (2008). Phonetics: transcription, production, acoustics, and perception.Wiley-Blackwell.
Sims, G. (1959). An elementary grammar of Cibemba. Mansa Mission, Christian Missions in ManyLands.
Stevens, K. (2000). Acoustic phonetics, vol. 30. MIT press.
Stevens, K., Blumstein, S., Glicksman, L., Burton, M. & Kurowski, K. (1992). Acoustic andperceptual characteristics of voicing in fricatives and fricative clusters. Journal of the AcousticalSociety of America 91. 2979–3000.
Wheeler, M. (2005). The phonology of Catalan. Oxford University Press, USA.
Ziervogel, D., Wentzel, P. & Makuya, T. (1972). A handbook of the Venda language. University ofSouth Africa (Pretoria).
39
IntroductionMethodsLanguagesRecordingsProcessing and SoftwareCriteria for Classification
ResultsDetailsSelected DataAmbiguous Cases
DiscussionAerodynamics of [B]Examples of True Fricative [B] in Ewe Rounded EnvironmentsData from L1 Acquisition
Modeling [B]Straw ModelsMethodsResults
Phonological RamificationsIssues with the Reclassification of [B]Boersma's boersma2008emergent Three-Form ModelFormalizing [B] as Both an Approximant and a Fricative
Conclusion