PosterPDF Available

Development of Coarticulation in German Children: Mutual Information



The study aims to investigate the development of coarticulation in 5-year old German children. The main goal was to examine the way different aspects of consonant production vary on a quantitative coarticulation -invariance scale as a function of age. To achieve this goal, we employed Mutual Information (MI), a method that has been used to measure coarticulation degree by quantifying independence between two variables (Iskarous et al., 2013).
Development of Coarticulation in German Children:
Mutual Information
Coarticulation, dened as varying degrees of articulatory overlap between
speech segments, is traditionally measured using Locus Equations (acoustic
and more recently, articulatory).
As an extension of this method,
Iskarous et al. (2013) suggested
Mutual Information (MI) that
measures the amount of information
segment Y contains about segment X,
and thus provides an indication of
how predictable Y is given X.
Test the method for ultrasound - a form of articulatory data quantication
di+erent from EMA
Apply the method to children data as it was only used on adult datasets
Compare the results with those of articulatory LE
Dzhuma Abakarovaa, Khalil Iskarousb, Elina Rubertusa, Mark Tiedec, Jan Riesa, Aude Noiraya
aUniversity of Potsdam, bUniversity of Southern California, cHaskins Laboratories
Acoustic Processing
Semi-automatic labeling
Points of interest for the
art. analysis:
Consonant midpoint (C50)
Consonant o+set (C100)
Vowel midpoint (V50)
Articulatory Processing
Semi-automatic tongue contour
detection at C50, C100, & V50
using SOLLARContours in Matlab
Extraction of horizontal position (x)
of the highest point of the tongue
dorsum (TD)
MI analysis
To see to what extent the position of C can be predicted from the position
of V, we calculate the joint distribution of C050 across V050, and compare
it to the joint distribution that is based on the assumption that V and C are
Figure 1 provides two examples
of MI calculation for /b/ and /d/.
The data were divided into five bins.
(a) shows the joint probability
distribution for /b/, (b) shows
the independent joint distributions.
Lower panels show the same for /d/.
The MI joint distribution for /b/ (a)
appears quite different from (b)
whereas (c) is rather similar to
the independent joint distribution
in (d).
Same pattern for places of articulation in German as found in other
languages : /b/ > /g/ > /d/
Same pattern for places of articulation based on tongue shape as
found for German using point parameters
Same pattern for young children as for adults
The amount of coarticulation is greater for children than for
adults for /b/ and /d/ not for /g/. This pattern holds independent of
bin size for the estimation of joint distribution. This nding does
not exactly follow the one by Rubertus et al (2015). They
analyzed the same data using articulatory LE and have found
greater amount of coarticulation in children for all places of
articulation. Further investigation is needed to understand the
source of this di+erence.
This work was supported by the DFG GZ: NO 1098/2-1
We thank the whole LOLA team for their great work: Helene Killmer-
Rumpf, Lisa Roehle, Liuba Carpova, Michelle Golchert, Stefanos
Tserkezis, & Stella Krüger.
Special thanks go to our adult and child participants (and their parents)
who made it easy and fun for us to collect our data.
Contact Information
Dzhuma Abakarova
Laboratory for Oral Language Acquisition
University of Potsdam
Department Linguistics
[1] Iskarous, K., Mooshammer, C., Hoole, P., Recasens, D., Shadle, C. H., Saltzman, E., &
Whalen, D. H. (2013). The coarticulation/invariance scale: Mutual information as a measure of
coarticulation resistance, motor synergy, an articulatory invariance. JASA, 134.
[2] Elina Rubertus, Dzhuma Abakarova, Mark Tiede, Jan Ries, Aude Noiray (2015) Development
of Coarticulation in German Children: Articulatory Locus Equations
Participants: native German speakers
11 ve-year-olds (4 f, age range: 5,0 – 5,7 y)
4 adults (2 f, age range: 22 – 27 y)
Stimuli: Disyllabic trochaic pseudo words
in carrier phrase: /$aɪnə/ C1VC2@
V is always tense
C1 ≠ C2
C1V (18) x C2 (2) x3 repetitions
108 stimuli
Procedure: Repetition task in SOLLAR
Prerecorded stimuli presented auditorily
Participants’ repetition recorded via
Ultrasound (midsagittal tongue contour for articulatory data)
Microphone (acoustic signal for formant detection &
Videocamera (lip movement and head correction)
/b/ /y:/ /d,g/
/d/+ /u:/ + /b,g/ + / /ǝ
/g/ /a:/ /b,d/
1. Introduction 2. Method
3. Analysis
5. Discussion & Conclusion
4. Results
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
The present study investigates the development of coarticulation in German children between 3 and 7 years of age. To quantify coarticulation degree, we will not only apply the commonly used method of Locus Equations (LE) on the acoustic signal, but also on the articulation recorded with ultrasound, which so far has been rarely done in children (Noiray et al., 2013). This allows us to directly track dynamic movements instead of inferring (co)articulation from the acoustic signal. Coarticulation can be viewed as connecting single speech sounds by varying degrees of articulatory overlap. While some aspects of coarticulation are claimed to be universal, resulting from anatomic properties (e.g., overlap of labial consonants and lingual vowels), others are not that predictable and may be language-specific (e.g., vowel-to-vowel coarticulation). The way children acquire the coarticulatory patterns of their native language has been discussed intensively (i.e., holistic versus segmental theory). The present study extends previous work by investigating coarticulation with a broader set of phonemes, multiple age groups, and in both acoustics and articulation. Five cohorts of monolingual German children (3 to 7 years of age) as well as an adult control group are tested. Stimuli are elicited in a repetition task embedded in a child friendly setting. The prerecorded acoustic stimuli consist of disyllabic pseudo words following the pattern C1V1C2V2, preceded by the carrier word " eine " (/͜ aɪnə/). Within the stressed first syllable (C1V1), C1 is /b/, /d/, /g/, or /z/ and V1 one of the tense, long vowels /i/, /y/, /u/, /a/, /e/, and /o/. The second CV syllable consisting of the same consonant set as C1 plus the neutral vowel /ə/ is added to the syllable of interest such that C2 is never equal to C1, resulting in three different contexts per C1V1. In total, there are 72 different pseudo words. Besides the CV coarticulation within the pseudo word, the carrier phrase enables the investigation of V-to-V anticipatory coarticulation from V1 on the preceding schwa. At Ultrafest VII we will present the first results for CV coarticulation in the cohort of 5 year-olds and adults. During the recordings, children are comfortably seated in an adjustable car seat. They are recorded with a portable ultrasound system (Sonosite Edge, sr: 48Hz) with a small probe fixed on a custom-made probe holder. The probe holder was designed to allow for natural vertical motion of the jaw but prevent motion in the lateral and horizontal translations. It is positioned straight below the participant's chin to record the tongue on the midsagittal plane. Ultrasound video data are collected with synchronized audio speech signal (microphone Sennheiser, sr: 48 KHz) on a computer. In addition to tongue motion, a video camera (Sony, sr: 50Hz) records the participant's face to track the labial articulation as well as head and probe motion enabling us to correct the data from a jaw-based to a head-based coordinate system. As for the analysis, target words in the acoustic speech signal as well as relevant tongue data are extracted using custom-made Praat and Matlab programs. Acoustic LE measures of the CV coarticulation will be based on the F2 transitions between the very onset of V1 and its midpoint, while the articulatory analysis will focus on the highest tongue point's motion between C1 and V1. As the ultrasound allows us to track motion earlier than is visible in the acoustic signal, we will not only use the onset of the vowel but move further into the consonant to find early cues of the vowel's influence on the tongue shape.
Full-text available
Coarticulation and invariance are two topics at the center of theorizing about speech production and speech perception. In this paper, a quantitative scale is proposed that places coarticulation and invariance at the two ends of the scale. This scale is based on physical information flow in the articulatory signal, and uses Information Theory, especially the concept of mutual information, to quantify these central concepts of speech research. Mutual Information measures the amount of physical information shared across phonological units. In the proposed quantitative scale, coarticulation corresponds to greater and invariance to lesser information sharing. The measurement scale is tested by data from three languages: German, Catalan, and English. The relation between the proposed scale and several existing theories of coarticulation is discussed, and implications for existing theories of speech production and perception are presented.
Development of Coarticulation in German Children: Articulatory Locus Equations Participants: native German speakers • 11 five-year-olds (4 f, age range: 5,0 -5,7 y) • 4 adults (2 f, age range: 22 -27 y) Stimuli: Disyllabic trochaic pseudo
  • Elina Rubertus
  • Dzhuma Abakarova
  • Mark Tiede
  • Jan Ries
  • Aude Noiray
Elina Rubertus, Dzhuma Abakarova, Mark Tiede, Jan Ries, Aude Noiray (2015) Development of Coarticulation in German Children: Articulatory Locus Equations Participants: native German speakers • 11 five-year-olds (4 f, age range: 5,0 -5,7 y) • 4 adults (2 f, age range: 22 -27 y) Stimuli: Disyllabic trochaic pseudo words in carrier phrase: //͜ aɪnə/ C 1 VC 2 @