LOT summer school
Ultrasound, phonetics, phonology:
Articulation for Beginners!
James M Scobbie
CASL Research Centre
With special thanks to collaborators
Jane Stuart-Smith & Eleanor Lawson
Joanne Cleland & Zoe Roxburgh
Natasha Zharkova, Laura Black, Steve Cowen
Reenu Punnoose, Koen Sebreghts
Sonja Schaeffler & Ineke Mennen
Conny Heyde
Alan Wrench (aka Articulate Instruments Ltd) for AAA software and UTI hardware
Various funding – thank you to ESRC, EPSRC, QMU
June 2013
Scottish English
Derhoticisation among WC speakers
Rhotic tongue shape
Is it time for some nitty gritty stuff?
Scottish English again
Fronted /u/
Extensions, if time
Northern Irish /u/ and diphthongs
Sociophonetics / Lg var & change
MRI – [o]
MRI – [i]
MRI – [y]
SSE /o/
What about real vowels?
• Example of a UTI vowel space, un-rotated
UTI single speaker for comparison
• Compare
Daniel Jones (1917)
– “frontness” in F2 & “frontness” in mm
– “height” in F1 & “height”
in mm phonetics and
• What are articulatory
its utility
to theand
Nature 100: 96-98.
• Whole tongue shape?
• Constriction degree /
cross-sectional area /
tube diameter?
• High point of the
tongue surface?
What about articulation?
• Vowels
– /u/ in relation to /i/ in terms of “frontness”
– /u/’s similarity to /i/ in tongue shape
• Easy questions are still worth asking!
Some examples of something easy
• A socially-stratified corpus (ECB08) was
collected to examine social variation in postvocalic /r/ articulation WC vs. MC teenagers
• For context, each speaker produced just one
(real word) token of each vowel phoneme
• Labial consonants avoid lingual coarticulation
• 9 monophthongal vowel phonemes
• 3 diphthongs /ai/, /au/, /oi/ were not elicited
• Single word citation forms, no carrier phrase
• One time point was analysed – artic “target”
Analysing minimal data sets
• Video UTI, so only ~30fps
• Averaging acoustics is also hard
– Male and female speakers
– Adolescent speakers of variable vocal tract length
• No opportunity for complex normalisation
What can we get out of this?
• WC n=8, MC n=7… 1 token each
hem hip
How front is Scottish /u/?
• Formants are vocal tract resonances
• A standard approach for 60 years has been to
measure F1 & F2
– Low F1 = “height” & High F2 = “frontness”
– We will come back to these metaphors later
• Nothing is as simple as this metaphor implies,
when you get down to detail
– Higher formants are also important
– Other factors affect these formants
• But they are easy to measure, and plot well…
Formants and frontness
• Hawkins and Midgely, cf Wells, Deterding
SSBE for comparison
• Calculate the F2 distance (Bark) from /i/, the
vowel with the highest F2
– To /o/, /u/ and /e/ for each speaker
– Repeat for a normalised set by treating the /i/-/o/
distance as 100% (corner vowel to corner vowel),
which will make comparison to articulation easier
• Calculate the distance (Bark) from /i/
– To /u/ and /e/ (and /o/) for each speaker
– It was hard to measure F1 for /a/, so no
Acoustic analysis
• /u/ is acoustically “non-back” F2 relative to /i/ (Hz)
• Relative to each
speaker’s /i/ (& /o/),
/u/ is
– Mid F2 (Hz)
– Mid-high F2 (Bark)
– 61% front (from /o/)
• /e/ has high F2
– 94% front
• 2 speakers have
/u/ < 50% front (just)
Acoustic analysis
• /u/ and /e/ are significantly “non-high”
– /u/ not significantly different to /e/ or /o/ in F1
– In 5/14 speakers, /u/ had a higher rel F1 than /e/
– Hz / Bark are almost identical at these frequencies
Acoustic analysis
• /u/ has a raised F2
– 2.6 Bark lower than /i/
– 4.1 Bark higher than /o/
• /u/ has a raised F1
– 0.6 Bark higher than /i/
– Non-distinct from the raised F1 of /e/, 0.4 Bark
/u/ acoustic summary
• We only have mid-sagittal tongue curves
– Not passive articulators (vocal tract tube)
– Not all the tongue surface
– Not all the internal tongue tissue
– Not lips (well, not for this data set)
– One token per speaker (for this data set)
• But unlike EMA
– We are not limited to 3 or 4 anterior points
• And unlike MRI
– UTI is cheap, non-invasive, portable and quick
– We can collect & trace 12 tokens of 5V in half a day
With UTI…
• UTI consistently shows Scottish /u/ is lower and
centralised/fronted compared to other vowels
Vowel space (typical WC)
Vowel space (typical MC)
• What’s “horizontal” about a curving vocal tract?
• What’s the orientation of the probe to the head?
• Images can be rotated by you, looking, for
qualitative understanding, if there is a fixed
aspect ratio on x/y axes
• Images can be rotated for quantitative analysis
of “horizontal” and “vertical” by the analyst
• Occlusal plane is replicable and standard and
provides a reasonable horizontal for the anterior
portion of the vocal tract
High point of tongue
• ECB08 didn’t collect occlusal biteplanes…
• Different shape hard palates don’t help
• Two approaches to estimating “horizontal”
rather than adopting the basic axes of the probe
• common /o i/ tangent
• Assumed occlusal
ECB08 Soc-Lx sample
• /u/ is fronter in articulatory space than acoustics
91%, 2mm
74%, 6mm
90.0%, 2mm
99.6%, <1mm
• /u/ is either more front than central, or fully front
Articulatory analysis
• /u/ is lower in articulatory space than acoustics
• /u/ is not high and may be
• It is lower than /e/ on either
rotation of the space
• /o/ is back… it’s not “lower”
Articulatory analysis
• Analysis of even single tokens with only linear
normalisation on an estimated bite plane (or /i//o/ mean) is at least as valid as acoustic
analysis using normalised F2 (F2=frontness)
– In terms of variation and statistical difference
• Findings
– /u/ is fronted & /o/ is the peripheral corner vowel
– KIT vowel is lower in WC system
– /u/ is much more radically lowered than expected
• Need to improve quantification / averaging /
axes for measurement space
SSE summary and conclusions
• QMU Undergraduate project 2012-13 (Laura
Black) looking at Northern Irish English (NIE)
– /i e u ɔ/ & /au ai oi/
– n=12
– Mix of real & pseudo words, randomised in 3 blocks
– All open syllables with onset /h/, /m/ or /b/
– 3 speakers
– About 1 week of basic analysis from a standingstart, followed by refinement (and checking)
• Northern Irish-accented English .. Infamous for
“confusing” and variable vowels …!
Typical undergrad student project
• What are monophthongs?
– /i/, /u/, /e/, /ɔ/?
– Traditional diphthongs are /ai/, /au/, /oi/
• How front is /u/?
• What singletons are the best phonological
candidates for the offglides in the diphthongs?
• AAA demo
NIE1 monophthongs
NIE1 diphthongs
NIE2 monophthongs
NIE2 diphthongs
NIE3 monophthongs
NIE3 diphthongs
• /u/ ends higher & fronter (almost = /i/) than SSE
• More diphthongal than NIE /i/ or even NIE /e/
• Rounded
N. Irish /u/
But /u/ is a phonetic diphthong…
• /u/ is higher and fronter than ECB08
• Still ok for
• Still not as
high as NIE
Single speaker (SSE) Neutral space
• 28 (of 30) TD children (group 1) incl singletons:
– Minimal vowels set (poop, babe, peep, pope etc)
– DEAP (50w pL subtest and 12w screen)
– Coarticulatory VCV materials (asa, isi etc)
– Some coda /r/ words (hut, hurt, heart)
• Other tasks were part of an experiment to test
the “copyability” of US images on screen in
– Group 1 vs. Group 3, raw vs. ULTRAX enhanced
with presentation by machine and limited interaction
– Group 2a vs. Group 2b raw vs. zero (acoustic only)
with maximised interaction
ULTRAX child corpus
ULTRAX (g1, 2012) TD 11, 23, 24, 31
• Are the ULTRAX kids like the ECB08 corpus
• Yes 
More /u/ (i-o plane)
• Are ULTRAX group 2 like ULTRAX group 1?
• Yes, pretty much, so far (n=15)
More /u/ (i-o plane)
• Rotation of ECB08 20° (assumed biteplane)
– Similar results obtained using /i/-/o/ common
tangent, which is about 45° different in orientation
– Axes provided by the probe are intermediate
– Close “high points” on these curves are tolerant of
axes shift
H -20
Quantifying /u/ frontness

similar documents