THE EFFECT OF PALATE MORPHOLOGY ON CONSONANT ARTICULATION ... · THE EFFECT OF PALATE MORPHOLOGY ON CONSONANT ARTICULATION IN HEALTHY SPEAKERS . by . KRISTA RUDY . A thesis submitted
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
THE EFFECT OF PALATE MORPHOLOGY ON CONSONANT ARTICULATION IN
HEALTHY SPEAKERS
by
KRISTA RUDY
A thesis submitted in conformity with the requirements for the degree of M.Sc.
The Effect of Palate Morphology on Consonant Articulation in Healthy Speakers
Krista Rudy
Masters of Science, 2011
Department of Speech-Language Pathology
University of Toronto
This study investigated the effect of palate morphology and anthropometric measures of
the head and face on lingual consonant target (positional) variability of twenty one adult speakers
(eleven male, ten female). An electromagnetic tracking system (WAVE, NDI, Canada) was used
to collect tongue movements while each speaker produced a series of VCV syllables containing a
combination of consonants /t, d, s, z, ʃ, tʃ, k, g, j/ and three corner vowel /i, ɑ, u/. Distributions of
x, y, and z coordinates representing maximum tongue elevation during the consonants were used
to represent target variability across contexts. Palate and anthropometric measures were
obtained for each participant. A correlational analysis showed that target variability of the
consonants produced in the front of the mouth (e.g. alveolar and palatal) was explained, to a
degree, by palate morphology. The variability of velar consonants was not explained by the
structural measures.
iii
“Develop a passion for learning. If you do, you will never cease to grow.”
Anthony J. D’angelo
iv
ACKNOWLEDGEMENTS
To my supervisor, Dr. Yana Yunusova: You have been an extremely influential and positive figure in my life over the last two years. Throughout this process, you have challenged me, inspired me to work hard, and provided me with the guidance and encouragement necessary to continue through times of adversity. Your mentorship has afforded me with lessons that will continue to influence my future academic, professional and personal life. For this, I am so thankful.
To the members of my supervisory committee, Dr.’s Pascal van Lieshout and John Daskalogiannakis: I sincerely thank you for your feedback, discussions, and thought-provoking questions, which challenged me to think about this research from different perspectives.
To my dearest friend, Adrienne Chan: You have always been one of my greatest supporters in all aspects of my life. I cannot thank you enough, nor find the words that would adequately convey how important your support has been to the completion of this project.
Finally, I would like to thank my parents, Robert and Deborah Rudy, and my brothers: I owe all of my achievements to you. Thank you for your unconditional love and support. Without your faith in me, and your determination to never give up on me, this would not have been possible, thank you so much.
2002). Anatomical differences between speakers have emerged as a source of articulatory
3
variability as well (Brunner, Fuchs and Perrier, 2005; 2009; Fitch & Giedd, 1999; Mooshammer,
Perrier, Fuchs, Geng and Pape, 2004; Perkell, 1997), in particular, palate morphology. In this
project, the effect of the hard palate on lingual target variability (a form of articulatory
variability) will be examined.
The Hard Palate and Structures of the Head and Face
The hard palate is an important structure in speech production. As the uppermost
boundary of the oral cavity, it is hypothesized that the hard palate passively influences tongue
kinematics (Bhagyalakshmi, Renukarya and Rajangam, 2007; Fuchs, Perrier, Geng and
Mooshammer, 2006; Stone, 1995). According to Stone (1995) the palate serves a variety of
important functions in speech production. These functions include providing a surface to assist
in precise positioning of the tongue during articulation and supplying tactile sensory feedback
about the position of the tongue along the palate within the vocal tract.
The integrity of the hard palate is crucial in the acquisition of spoken language and
healthy development of facial structures. Research indicates that children with Down syndrome
are often born with a very narrow and vaulted hard palate which has been said to contribute to
speech disorder in this population (Bhagyalakshmi et al., 2007; Hamilton, 1993). The hard
palate has been said to play a role in speech development of children with cleft palates. Delayed
surgical repair of the hard palate in individuals born with cleft palate often results in severely
impaired speech and optimal midfacial development (Cosman & Falk, 1980; Holland et al.,
2007; Lohmander-Agerskov, Dotevall, Lith and Soderpalm, 1996); whereas, early surgical repair
of the hard palate during infancy results in a more optimal speech outcome and abnormal
4
midfacial growth (Cosman & Falk, 1980). As a result, morphology of the palate and face may be
interrelated and both may be associated with speech production.
The size of the head and face seem to be related during development (Farkas, 1994;
Farkas, Katic, Hreczko, Deutsch, Munro, 1984). Several regions of the skull appear to influence
the growth of one another while trying to accommodate the growth of other areas both spatially
and functionally, including the face and oral cavity (Polat & Kaya, 2007). For example, the
lower facial height has been negatively associated with pharyngeal distance (Honda, Maeda,
Hashi, Dembrowski and Westbury, 1996). A variety of anthropometric measures of the head
have also been associated with palate length (Kasai, Moro, Kanazawa and Iwasawa, 1995; Polat
& Kaya, 2007), and dental arch width (Solow, 1966). However, research on the relationship
between head size and palate morphology is limited and often includes only a single measure of
the palate.
Documenting the normal relationship between head and face size and palate morphology
in healthy adults is important to better identify anatomical patterns or proportions, which could
be useful for restoration of a face to its previous appearance in areas such a reconstructive
surgery (Wuerpel, 1936). The mouth and face are highly interrelated, such that adjusting
positions of the teeth may affect the contours of the face (Stoner, 1955). In orthodontics, a better
understanding of proportions of the head and palate could influence treatment to correct
malocclusion or dentition. Moreover, the palate is a difficult structure to accurately measure
without having to make an impression or scan, due to its shape and location within the oral
cavity. Identifying the relationship between head and face size and palate measures could enable
a cost-effective and efficient method of inferring palate shape and size based on anthropometric
measures.
5
Palate morphology has been associated with healthy motor speech behaviour. In the view
that palate morphology and head size appear to be interrelated, anthropometric measures of the
head may also be related to speech motor behaviour. Few studies have investigated
anthropometric measures of the head and face and their association to speech motor behaviour.
A single study investigated the association between articulator size and measures of articulatory
velocity and displacement (Kuehn & Moll, 1976), and another examined anthropometric
measures of the head and face and their association to dynamic variability measures of the lower
lip in speech (Riely & Smith, 2003), however, none have investigated the association between
anthropometric measures of the head and face and target variability in lingual consonants.
The Hard Palate as a Source of Articulatory Variability
In one of the first studies to look at lingual articulation and its relationship to palate shape
in healthy speakers, Perkell (1979) investigated tongue adjustments in six speakers producing
vowels /i/ /ɪ/ and /ɛ/. He observed that the participant with the shallowest vault (flat palate)
appeared to have the smallest target variability of tongue position for the three vowels, while the
participant with the deepest vault (domed palate) had the largest target variability in tongue
position across multiple repetitions of the same vowel.
In a more recent study, Mooshammer et al. (2004) evaluated individual differences in
palate shape and token-to-token (repetition) target variability of the same vowel. Three German
speakers produced 14 German vowels (/i:, ɪ, y:, ʏ, e:, ɛ, ǿ:, œ, ɑ:, a, o:, ʊ, ɔ, u:/) in a carrier
phrase 10-11 times. Speech movements were recorded using electropalatography (EPG) and 2D
electromagnetic midsagittal articulography (EMMA). Individual characteristics of tongue-palate
6
contact (EPG) and target variability of the tongue (EMMA) were obtained. Palate shapes were
described for each speaker as either coronally high or flat. The authors found that the speaker
with the flattest palate exhibited the least amount of target variability. Two more studies looked
at the structure of the palate as it relates to articulatory variability (Brunner et al., 2005; 2009).
Results similar to the previous studies were observed (i.e., greater articulatory variability in
domed palates as compared to flat palates), and two different explanations for this phenomenon
were proposed.
The first explanation is the Biomechanical hypothesis, which suggests the degree of
target variability is a direct result of constraints imposed on speech movement by vocal tract
morphology. According to this hypothesis, speakers with flat palates are more likely to have
greater tongue-palate contact during articulation due to greater palate surface area available to
the tongue. This palate shape would provide increased structural support during articulation,
thus, reducing target variability. Earlier, Mooshammer et al. (2004) found that the degree of
tongue-palate contact obtained with EPG was negatively associated with target variability in
vowel production. The speaker in their sample with the flat palate displayed less target
variability and also exhibited greater tongue-palate contact.
To test the Biomechanical hypothesis, Brunner et al. (2005) investigated articulatory
variability, tongue-palate contact, and palate shape in 20 speakers of various linguistic
backgrounds. Articulatory variability and tongue-palate contact were recorded using EPG. Each
speaker produced a number of consonants and vowels (C = /s, ɕ, ј, ʃ/ and V = /i, e, u/) embedded
in carrier phrases 30 times. Articulatory variability was measured as the percent of contact
(POC), which was calculated as the number of electrodes in the EPG that were contacted by the
tongue for each EPG frame over production of the entire acoustically defined sound segment.
7
The resulting normalized mean value was used as a variability coefficient. POC provided
information about the amount of tongue-palate contact, but not about the location of the target.
Tongue-palate contact was measured using a lateral index (Li), which was calculated to illustrate
the percent of lateral linguo-palatal contact. It was hypothesized that greater lateral linguo-
palatal contact would provide more support to the tongue during articulation. Palate shape was
described using a measure of palate curvature (α). Brunner et al. (2005) found significant
negative correlations between articulatory variability (POC) and tongue-palate contact (Li) for
select vowels (/i, e, ɪ, ɛ/) and a consonant glide (/j/), whereby greater tongue-palate contact was
associated with less articulatory variability. However, a strong relationship was not reported
between tongue palate-contact (Li) and palate curvature (α), as would have been anticipated by
the Biomechanical hypothesis. Speakers with flat palates did not exhibit greater tongue-palate
contact during speech. A significant negative relationship between variability (POC) and palate
curvature (α) was found for only a vowel (/i/) and a consonant glide (/j/). However, POC is not a
measure of target variability.
The second explanation proposed to account for these findings is the Speaker-Oriented
Control hypothesis outlined by Brunner et al. (2005). This explanation is based on an
assumption that target variability is influenced by auditory feedback. The Speaker-Oriented
Control hypothesis proposes that palate shape influences the overall cross-sectional area of the
oral cavity, which in turn affects the sensitivity of speech acoustics to changes in articulator
positioning. In this view, speakers with domed palates exhibit greater target variability because
the acoustic output is less sensitive to changes in articulator positioning than it is for speakers
with flat palates. The Speaker-Oriented Control hypothesis was assessed using a 2D
biomechanical tongue model (Brunner et al., 2005), which simulated tongue movements during
8
articulation of three corner vowels (/a, i, u/) in five different palate shape conditions varying in
palate curvature (α). Tongue movements were varied identically for all five palates, and in each
condition, sounds were synthesized and formant values for F1 and F2 were obtained. Brunner et
al. (2005) found that in flat palate conditions, small changes in articulator position had a greater
effect on the cross-sectional area of the oral cavity than domed palates, and thus a greater effect
on the acoustic output. These findings were consistent with the Speaker-Oriented Control
hypothesis; however, it is unclear whether the same findings would be observed in real speech.
In a follow-up study, Brunner et al. (2009) investigated the relationship between palate
shape, articulatory variability, and acoustic variability in speakers using EPG, and found no
relationship between the target variability measure and palate curvature (α). In contrast to their
earlier study (Brunner et al., 2005), Brunner et al. (2009) employed a larger sample size (N=32)
of speakers from various linguistic backgrounds and reduced the target sounds to vowels (/i, e, ɛ,
ɪ/) and a consonant glide (/j/). Two measures of articulatory variability were added, as well as a
spectral analysis of the vowel formant frequencies.
Articulatory variability was measured in three ways: POC, number of contact (NOC), and
center of gravity (COG). In order to calculate NOC, the articulatory target was identified
acoustically. At this target, the row with the greatest number of electrodes contacted in the EPG
was used to calculate a coefficient of variation across repetitions of the sound. NOC provides
information about tongue height variation during articulation. Similarly, to calculate COG, the
articulatory target was acoustically identified, and the point of tongue-palate contact during a
single frame was recorded for each repetition and used to calculate a coefficient of variation.
The COG measure provided information about tongue location variation at the articulatory
target.
9
During vowel production, the acoustic output remained stable for speakers with domed
and flat palates despite differences in articulatory variability. Speakers with domed palates
exhibited greater articulatory variability than speakers with flat palates in the POC measure for
three vowels /ɪ, ɛ, e/ and a consonant glide /j/, but no significant correlations were found in any
other measure of articulatory variability (NOC or COG). COG and NOC would be considered
the most relevant to this study because they are considered measures of target variability. Thus,
the authors found no relationship between palate curvature (α) and their measures of target
variability (COG and NOC) of the tongue during articulation of these sounds.
It is important to note that the Biomechanical hypothesis and the Speaker-Oriented
Control hypothesis predict reduced target variability for speakers with flat palates as compared to
speakers with domed palates. As a result, interpretation of the findings with respect to either of
the hypotheses is challenging.
Limitations of the Previous Research
The existing studies remain inconclusive and contain a number of limitations, therefore,
the relationship between palate morphology and target variability remains unclear.
Sample sizes of the existing studies generally do not exceed six participants speaking the
same language and producing identical stimuli. The studies containing larger samples employ
speakers of various linguistic backgrounds. Varying linguistic backgrounds for this type of study
is problematic because target variability is influenced by phonetic inventories, phonotactic rules,
and phonetic context of different languages (Dromey & Sanders, 2009). Although controlling
for a speaker’s native language may be more crucial to the dynamic measures of speech
10
kinematics, there is not enough known about target variability in speech to understand how a
heterogeneous sample of speakers may affect target variability during consonant production.
The technology employed to measure articulatory variability in the existing studies may
interfere with normal speech kinematics and limit information collected about tongue movement.
Three of the studies used EPG to assess target variability. The 2mm thick EPG covers the entire
palate, which obstructs speech and impedes somatosensory feedback about tongue positioning to
the speaker. Due to the nature of the EPG technology, it only provides information about tongue
positioning at certain points on the palate (Fitzpatrick & Chasaide, 2002). The electrodes
embedded in the EPG are spaced a few millimetres apart resulting in limited spatial resolution.
Also, the EPG does not provide information about the portion of the tongue involved in
articulation, nor does it extend to the soft-palate, preventing collection of information about
tongue dorsum movement (Fitzpatrick & Chasaide, 2002). In comparison to EPG,
articulography provides better spatial resolution with an accuracy that is generally better than
1mm (Hoole & Nguyen, 1997; Yunusova, Green and Mefferd, 2009); however, only one study
(Mooshammer et al., 2004) of those described above also incorporated articulography to collect
information about tongue kinematics.
The speaking tasks employed in previous studies have focused mainly on articulatory
variability in vowels. Vowels involve the least amount of constriction, and in most cases,
consonant articulation requires tongue-palate contact (Fuchs et al., 2006), therefore, the influence
of palate morphology on target variability in consonant production may differ from that of
vowels. In the present study, we focused our investigation on consonants only. A relatively
large sample of speakers, all of whom are homogeneous in their language and dialect, were
recruited. Lingual function in speech was recorded with the WAVE system (NDI, Canada),
11
which operates in a similar fashion to the 3D EMA (AG500) system. One study to date suggests
that the WAVE and 3D EMA systems function with similar error and accuracy scores (Berry,
2010).
Questions and Hypotheses of the Present Study
This study sought to investigate the following main question: are measures of palate
morphology and anthropometric measures of the head associated with target (positional)
variability of the tongue during consonant production? To properly investigate this question, the
following additional questions were addressed: are anthropometric measures associated with the
measures of palate morphology, and do target variability measures differ between consonants
across speakers. It was hypothesized that flat palates would be associated with reduced target
variability in consonants as has previously been observed in vowels (Mooshammer et al., 2004;
Perkell, 1979).
12
METHOD
Participants
Participants were recruited through advertisements on the University of Toronto, St.
George Campus. 21 healthy adult speakers participated in this study (11 male, 10 female). The
balanced division of men and women was necessary because male-female differences in
anthropometric and palate measures were anticipated (Farkas et al., 1984; Farkas et al., 1994;
Ferrario, Sforza, Pizzini, Vogel and Miani, 1993; Ingerslev & Solow, 1974). The average age of
the male group was 32.3 (SD = 8.5, Range = 25-49). The average age of the female group was
28.4 (SD = 6.1, Range = 25-43). All participants were native speakers of Canadian English. All
but five speakers were from Eastern Canada. The remaining five speakers (W03, W05, W07,
W09, W12) were from various parts of Western Canada. Table 1 describes age, sex and dialect
characteristics of the sample population.
All participants passed a pure tone hearing screening and reported no history of speech
disorder, speech therapy, medical conditions or medications affecting speech, and no structural
abnormalities of the vocal tract including ankyloglossia. Participants were selected based on the
absence of any orthodontic apparatus or metal in their mouth at the time of recording. We were
not able to assess the presence of malocclusion; however, none of the participants had grossly
different dentition including significant maxillary spacing irregularities determined during a brief
oral examination.
13
Table 1: Participant ID, sex, age and self-report dialect base. ID Sex Age Dialect
W01 M 26 Ontario W02 M 30 Ontario W03 M 41 Alberta W04 F 26 Ontario W05 F 25 British Columbia W06 F 26 Ontario W07 M 26 British Columbia W08 F 43 Ontario W09 F 25 British Columbia W10 F 36 Ontario W11 M 37 Ontario W12 F 25 Alberta W13 M 49 Ontario W14 M 25 Ontario W15 F 26 Ontario W16 M 42 Ontario W17 M 25 Ontario W18 F 25 Ontario W19 M 26 Ontario W20 M 28 Ontario W21 F 27 Ontario
Speech Sample
Participants were seated in a comfortable chair and asked to repeat symmetrical Vowel –
Consonant –Vowel (VCV) syllables (e.g. ata, asa) embedded in a carrier phrase (“It’s ___
game”) ten times. Each participant was instructed to repeat the sentences at a comfortable
speaking rate and loudness. The syllables contained three corner vowels (/i, ɑ, u/) and nine
lingual consonants (/t, d, s, z, ʃ, tʃ, k, g, j/) associated with different places (alveolar, palatal and
velar) and manners of production (stops, fricatives, affricate and glide). For the purpose of this
study, consonants were classified by place of articulation as either front (alveolar and palatal) or
back (velar). Co-articulation research suggests that surrounding vowels affect the location of
articulation of a consonant (Sussman, Duder and Dalston, 1999). The chosen vowels are
14
considered to be corner vowels, and therefore, were used to provide a better assessment of target
variability for each consonant and speaker.
Data Acquisition and Processing
Data was acquired according to the procedures described in Berry (2010). Articulator
movement was recorded by a 3D electromagnetic tracking system, WAVE (NDI, Canada; see
Figure 1a). WAVE tracked the position of a number of 2mm diameter 5DOF sensors. Sensors
were glued using PeriAcryl Oral Tissue Adhesive, a non-toxic dental surgical glue, to several
structures within and around the oral cavity. Two 5DOF sensors were glued to the mid-sagittal
surface of the tongue blade (TB) and tongue dorsum (TD). A single 6DOF reference sensor was
attached to the bridge of the nose to collect information about head movements during the
recording. Figure 1b illustrates the approximate placement of these sensors. The TB sensor was
placed approximately 1 cm from the tip of the tongue, and the TD sensor was glued
approximately 2cm behind TB. The positions of the sensors were measured using a ruler and
recorded for each speaker. Two 5DOF sensors were glued to the lips, which were positioned at
the midline of the vermillion border of the upper and lower lip. For the purpose of this study,
only data collected with the tongue sensors (TB and TD) will be reported. Sensors were
recorded at the sampling rate of 100Hz. Acoustic signals were acquired simultaneously with
speech movements at 22KHz, using a professional lapel microphone (Countryman B3P4FF05B)
positioned approximately 15 cm away from the mouth.
During acquisition, all data was collected relative to the reference sensor attached to the
head. In an additional recording, the position of the reference sensor was recorded relative to a
15
biteplate containing a 6DOF sensor oriented in respect to the occlusal plane. The biteplate
recording enabled transformation of the original speech data during post-processing in which it
was re-expressed relative to a speaker-specific anatomically-based Cartesian coordinate system.
In this coordinate system, the abscissa is located along the maxillary occlusal plane and the
ordinate is normal to the abscissa and placed where the central maxillary incisors meet the
maxillary occlusal plane. Then, movement data was filtered at 15Hz using a zero-phase digital
filter (8-pole Butterworth).
Figure 1. a) WAVE set up. b) 5DOF sensors on a speaker’s tongue.
Measurements
Recordings were screened by a listener to ensure that only kinematic data of correctly
produced sentences were included for analysis. Target variability measures were obtained by
identifying the point of maximum vertical displacement of the tongue during consonants
embedded in VCV syllables. In the data expressed relative to the occlusal plane, the x-axis
represented vertical movement of the tongue, the y-axis represented mediolateral movement of
the tongue, and the z-axis represented anteroposterior movement of the tongue. Figure 2
illustrates x, y, and z time histories during the segment /ɑtɑ/. In consonants produced primarily
with a front (alveolar or palatal) constriction (/t/, /d/, /s/, /z/, /ʃ/, /tʃ/ /j/), the TB sensor was the
TB
TD
TB
TD
16
point of reference for obtaining the position of the tongue; in production of consonants involving
a back (velar) constriction (/k/, /g/), the TD sensor was used as the point of reference for
obtaining the position of the tongue. x, y, and z coordinates of the tongue position at this point in
time were recorded for each consonant.
Figure 2: A single sentence view of the kinematic and acoustic signal recorded using the WAVE system for the alveolar consonant /d/. The measurement for maximum tongue height during consonant production is identified by the TBx channel. The x, y, z coordinates were obtained for the tongue position at this point in time.
/ It’s ɑ ɑd g εɪme /
17
Target Variability Measures
The following measurements were derived for each consonant:
(1) 3D variability (mm3) represents target variability in 3D and is defined as the volume
contained within the 2SD ellipse fitted around the distribution of the positional
coordinates as illustrated in Figure 3.
(2) Vertical variability (mm) is defined as the x-range of the 2SD ellipse representing target
variability in the vertical dimension.
(3) Mediolateral variability (mm) is defined as the y-range of the 2SD ellipse representing
target variability in the mediolateral dimension.
(4) Anteroposterior variability (mm) is defined as the z-range of the 2SD ellipse
representing target variability in the anteroposterior dimension.
(5) Spread was determined as the distance between the centroids (x, y, z) of the target for /j/,
the most anterior consonant for all speakers, and the centroids of each front consonant
target (/t/, /d/, /s/, /z/, /ʃ/, /tʃ/).
18
Figure 3: 30 repetitions of /t/ produced by a single speaker (W15) and fit with a 2SD ellipse.
Structural Measures: Palate Measures
Palate measures were obtained from individual palate casts that were made by a local
dentist. At the dental office, maxillary impressions were made for each speaker and were poured
with a white stone. Using a pencil, landmarks were identified on each speaker’s palate cast.
These points are shown in Figure 4 and include: (1) Incisive Papilla (IP) which is the point
between the maxillary incisors, (2) Molar Right (MR) which is the point on the gingival margin
at the first permanent molar on the participant’s right side, (3) Molar Left (ML) which is the
point on the gingival margin at the first permanent molar on the participant’s left side, and (4)
Midpoint (M) which is the center point between MR and ML.
13 14 15 16 17 18 19 20 21 22 23 24
10
11
12
13
14
15
16
17
3
4
5
Anteroposterior (mm)
Vertical (mm)
Med
iola
tera
l (m
m)
19
Figure 4: Palate cast obtained from W04 illustrating the landmarks used to calculate palate measures: IP = incisive papilla, MR = molar right, ML = molar left, M = midpoint.
A 3D point digitizer (WAVE, NDI, Canada) was used to collect landmark locations on
each palate cast. 3D point digitization is deemed a more accurate method than manual
measurement (Ferrario, Sforza, Schmitz and Colombo, 1998). Palate measures that were
selected to represent the size and shape of the palate (Brunner et al, 2005; 2009; Ferrario et al.,
1998) included:
(1) Palate Height was calculated as a distance between the point M on the palate and the line
between the ML to MR.
(2) Palate Width was defined as the distance between MR and ML.
(3) Palate Length was defined as the horizontal distance between the IP and M.
(4) Palate Slope was defined as the inclination of the straight line between IP and M.
(5) Palate Curvature (α) was defined as the coronal shape of the hard palate. Palate
curvature was described by an alpha coefficient which was calculated following Brunner
et al. (2005; 2009). Using a point digitizer, a recording was made of a line tracing along
the surface of the palate cast between MR and ML from which a parabolic
IP
ML
M
MR
20
approximation with two coefficients was obtained. The palate shape was then described
by an equation:
y(x) = ax2 + b
The α was calculated using the following equation (Perrier, Boe and Sock, 1992):
Anthropometric measures were determined following procedures outlined in Reily &
Smith (2003). These specific anthropometric measures were employed because they give a
general description of head size, particularly of the lower face, which may indirectly represent
the size of the oral cavity. Landmarks were identified on each speaker’s face and marked with
an eyeliner pencil. A soft measuring tape was used for measurements of tangential linear
distances made along the surface of the face, and a spreading calliper was used to make
measurements for all projective linear distances between these landmarks. Methodology of
anthropometry is explained in more detail elsewhere (Farkas, 1994).
The landmarks are shown in Figure 5 and included (1) Tragions (t) on the right and left
sides of the face, which are defined as the notches on the upper margin of the tragus (Farkas,
1994), (2) Gonions (go), the most lateral point on the mandibular angle on the right and left side
21
of the face (Farkas, 1994), (3) Subnasale (sn) defined as the midpoint where the nasal septum
and upper lip meet (Farkas, 1994), (4) Gnathion (gn) defined as the lowest median landmark on
the lower border of the mandible (Farkas, 1994), (5) Glabella (g) defined as the most prominent
midline point between the eyebrows (Farkas, 1994) and (6) Opisthocranion (op) defined as the
point located in the occipital region of the head, which is the most distant point from the globella
(Farkas, 1994). The anthropometric measures included:
(1) Head Circumference (g-op-g) is defined as the perimeter of the head in the horizontal
plane. This measurement was taken through the eyebrow line identified by the glabella
and through the opisthocranion.
(2) Jaw Size: This oral cavity parameter was determined by measuring:
(a) Mandibular Width (go-go) is defined as the distance between the right and left
gonions and represents the width of the jaw.
(b) Lower Facial Depth (t-gn) is the distance between the tragion and gnathion on the
right and left side.
(c) Mandibular Arc (t-gn-t) is defined as the distance around the chin between the
right and left tragions and represents a perimeter of the mandible.
(d) Lower Facial Height (sn-gn) is defined as the distance between the subnasale and
the gnathion and represents the vertical length of the oral cavity.
22
Figure 5: Landmarks and measurement trajectories for anthropometric measures: t = tragions, go = gonions, sn = subnasale, gn = gnathion (gn), g = glabella, op = opisthocranion.
To estimate the precision of anthropometric measures and intra-measures reliability, all
measures were taken twice within the same session (Marks, Habicht and Mueller, 1989). Table 2
reports the precision and reliability of the anthropometric measures. To report precision, we
calculated the technical error measurement (TEM) for each anthropometric measure. TEM was
calculated according to procedures outlined in Jamison & Ward (1993), using the formula:
Where d is the difference between the first and second measurement. Test-retest reliability of the
anthropometric measures was reported as a reliability coefficient. The reliability coefficient was
calculated as the square of the correlation coefficient (r), where zero represents an unreliable
measure and one represents a reliable measure. These estimates are comparable to published
results (Jamison & Ward, 1993).
t t
gn
sn
go go
g op
23
Table 2: Reliability and precision of anthropometric measures calculated across speakers: Circ = head circumference, Arc = mandibular arc, LFH = lower facial height, LFDR = lower facial depth right, LFDL = lower facial depth left, MW = mandibular width.
Reliability and Precision of Anthropometric Measures Measure Mean TEM r Circ (mm) 582.4 0.38 0.95 Arc (mm) 320.5 0.45 0.91 LFH (mm) 63 0.27 0.75 LFDR (mm) 144.7 0.19 0.93 LFDL (mm) 144.5 0.23 0.89 MW (mm) 115 0.34 0.9
Statistical Analysis
The positional data were examined visually prior to performing calculations or any
analyses. Outliers, defined as data points greater than two standard deviations outside of the
2SD ellipse in two of the 3 dimensions, were removed which composed 4.8% of the total data
set. These outliers may have been caused by random error in the WAVE system. Variability
measures were calculated on the reduced data set.
The distribution of each variable was assessed visually in a histogram plot and tested for
normality using the Shapiro-Wilk test of normality. Both analyses showed that most of the
variables were positively skewed, with the exception of palate length, lower facial height and
mandibular arc, which were normally distributed, and palate curvature and palate slope, which
were bimodally distributed.
Descriptive statistics (medians and interquartile ranges) were calculated across speakers
for each variable. Differences between males and females were assessed using a 2-tailed
Wilcoxon Rank Sum test for all measures. A Generalized Estimating Equations (GEE) approach
was used to test between consonant differences in variability measures with structural measures
24
(palate and anthropometric measures) as covariates. GEE is a semi-parametric technique and is
an extension of the Generalized Linear Models for nested data. The GEE takes into account
interdependencies in clustered data and it is able to model data that is not normally distributed
(Zeger & Liang, 1986). As a result, this model allowed us to maintain the original values of the
data without performing transformations. When the effect of consonant was significant for the
GEE analysis (p < 0.05), a pairwise analysis between consonants was performed. Since, the
majority of the variables were not normally distributed, Spearman correlation coefficients were
computed to assess the association within and between variables.
25
RESULTS
Structural Measures
Anthropometric Measures
Summary statistics for anthropometric measures computed across speakers by sex are
reported in Table 3. A 2-tailed Wilcoxon Rank Sum test revealed that there were significant
differences between males and females for all anthropometric measures with men showing
significantly larger anthropometric measures than women (see Table 3). Box and whisker plots
are presented in Figure 6 to illustrate the distribution of these variables.
Table 3: Median (M) and interquartile ranges (IQR) of averaged anthropometric measures reported by sex: Circ = head circumference, Arc = mandibular arc, LFH = lower facial height, LFDR = lower facial depth right, LFDL = lower facial depth left, MW = mandibular width.
Figure 11: Scatterplot matrix illustrating the associations between all palate measures: PH = palate height, PW = palate width, PL = palate length, PS = palate slope, PC = palate curvature.
Height
28 30 32 34 36 38 40 42 0.0 0.2 0.4 0.6
1012
1416
2830
3234
3638
4042
Width
Length
2530
3540
0.0
0.2
0.4
0.6
Slope
10 12 14 16 25 30 35 40 1.6 1.8 2.0 2.2
1.6
1.8
2.0
2.2
Curv
PH
PW
PL
PS
PC
32
Correlations between Structural Measures
To address the question of the association between structural measures, Spearman
correlation coefficients were computed among the two anthropometric factors and the five palate
measures. Table 7 indicates that the anthropometric factors are not strongly correlated with
palate measures (r = -.03 - .54). Scatterplots of the strongest associations are provided in Figure
12.
Table 7: Spearman correlation coefficients illustrate the association between the two anthropometric factors and the various palate measures: PH = palate height, PW = palate width, PL = palate length, PS = palate slope, PC = palate curvature.
Spearman correlation coefficients indicated that, on average, TB 3D variability is
significantly negatively correlated with palate curvature, vertical variability is significantly
positively correlated with palate length and negatively correlated with palate curvature,
anteroposterior variability is significantly negatively correlated with palate curvature, and spread
is positively correlated with palate width. However, no significant relationship was found
between any of the anthropometric factors and target variability measures or any of the structural
measures and mediolateral variability. Figure 14 and 15 illustrate the significant associations
between the palate and target variability measures.
37
Figure 14: Scatterplots illustrating the association between palate curvature (PC) and target variability measures: 3DV = 3D variability, VV = vertical variability, APV = anteroposterior variability.
Figure 15: Scatterplots illustrating the association of palate length (PL) and palate width (PW) with target variability measures of vertical variability (VV) and spread.
1.6 1.8 2.0 2.2
PC (α)
34
56
7
VV
(mm
)
1.6 1.8 2.0 2.2
PC (α)
3.5
4.0
4.5
5.0
5.5
APV
(mm
)
1.6 1.8 2.0 2.2
PC (α)
1020
3040
50
3DV
(mm
3 )r = -0.51 r = -0.45
r = -0.53
FlatDomed
25 30 35 40
PL (mm)
34
56
7
VV
(mm
)
28 30 32 34 36 38 40 42
PW (mm)
34
56
78
Spre
ad (m
m)
r = 0.46 r = 0.56
38
Speakers with longer palates and domed palates (α < 1.89) exhibited greater vertical
variability. Speakers with domed palates also exhibited greater anteroposterior variability than
those with flat palates (α > 1.89), and speakers with wider palates had greater target spread of the
various consonants.
In addition, speakers with domed palates exhibited greater 3D variability than those with
flat palates. Figure 16 illustrates Median 3D variability of front consonants for speakers with
domed and flat palates. A 2-tailed Wilcoxon Rank Sum test revealed that speakers with domed
palates had significantly greater 3D variability than speakers with flat palates (W(19) = 2.15, p =
0.016).
Figure 16: Median 3D variability for domed and flat palates.
Dome Flat
0
5
10
15
20
25
3DV
(mm
3 )
39
DISCUSSION
The purpose of this study was to investigate the association between individual
differences in palate morphology, head size and lingual target (positional) variability. 21
participants were asked to produce a series of VCV syllables embedded in a carrier phrase ten
times. Consonants /t, d, s, z, ʃ, tʃ, j, g, k/ were produced within the context of three corner vowels
/ɑ, i, u/. The x, y, z coordinates representing maximum tongue height during the consonants
were collected using the WAVE system. Target variability was estimated for every consonant
across context. Palate and anthropometric measures were also obtained for each participant.
Target variability of front consonants was explained, to a degree, by palate morphology, but not
by measures of head size or lower face size.
Structural Measures: Anthropometric and Palate Measures
Head size was described using a metric proposed in Farkas (1994). The measures of
interest included: head circumference, mandibular arc, lower facial height, and lower facial depth
left and right, and mandibular width. These measures were highly inter-correlated (Farkas, 1994;
Farkas et al., 1984; Saunders, Popovich and Thompson, 1980). Similar to the previously
published biometric research (Brown, Barret and Darroch, 1965; Howells, 1951; 1957; Solow,
1966), anthropometric measures were reduced to two factors, one consisting of lower facial
depth left and right, and the other consisting of head circumference, mandibular arc, mandibular
width, and lower facial height; these two composite factors represented head size and lower face
size. A comparison between males and females showed that men, as expected, have larger
40
anthropometric measures on average than women (Farkas, 1984; Farkas et al., 1994; Ferrario et