Kendall, Fridland, & Farringon 2013: More on the production and perception of regional vowel differences in the U.S.
1. More
on
the
produc-on
and
percep-on
of
regional
vowel
differences
in
the
U.S.
Tyler
Kendalla
Valerie
Fridlandb
Charlie
Farringtona
a
Dept.
of
Linguis;cs,
University
of
Oregon
b
Dept.
of
English,
University
of
Nevada,
Reno
ExApp
2013
|
Copenhagen
|
21
March
2013
1
2. @
ExAPP
2010
• We
presented
some
results
of
an
ongoing
vowel
percep;on/vowel
produc;on
study
addressing
the
ques;on:
– How
does
variability
in
speech
produc;on
relate
to
variability
in
speech
percep;on,
in
the
context
of
current
US
vowel
shiVs?
• Based
on
data
from
three
regions
of
the
US
– South
(Memphis,
TN,
and
to
a
lesser
extent
Blacksburg,
VA)
– Inland
North
(Oswego,
NY)
– West
(Reno,
NV)
• Which
are
characterized
by
different
vowel
systems
in
produc;on
2
3. Three
major
regional
US
vowel
shiVs
Southern
Vowel
ShiV
(SVS)
Northern
Ci;es
ShiV
(NCS)
bat?
Elsewhere
ShiV
a.k.a.
Canadian
Vowel
ShiV
a.k.a.
California
Vowel
ShiV
a.k.a.
Columbus
Vowel
ShiV
NCS:
Eckert
1988,
2000,
Evans
2001,
Gordon
1997,
Labov
1991,
1994,
2001,
Labov
et
al
2006,
Thomas
1997b,
2001;
SVS:
Feagin
1986,
Fridland
2000,
2001,
2003a,
2003b,
2004,
Fridland
and
Bartled
2006,
Labov
1991,
1994,
2001,
Labov
et
al
2006,
Thomas
1989,
1997a,
2001;
Elsewhere:
Clarke
et
al
1995,
Luthin
1987,
Labov
et
al
2006,
Thomas
2001
3
Figures
from
Gordon
“Do
you
speak
American?”
hdp://www.pbs.org/speak/ahead/change/changin/
4. About
our
study
• Web-‐based
percep;on
survey
– Developed
by
Bartek
Plichta
(hdp://bartus.org/)
~
e
• Vowel
con;nua
synthesized
from
a
single
talker’s
natural
vowels
as
endpoints
• Five
vowel
con;nua,
two
contexts
each
/e/
~
/ɛ/
/i/
~
/ɪ/
/æ/
~
/ɑ/
/ɪ/
~
/u/
/ʌ/
~
/o/
ɛ
~
• Iden;fica;on
task
– Listeners
heard
4
repe;;ons
of
each
of
7
steps
in
random
order
• A
subset
of
the
percep;on
– Listeners
had
to
iden;fy
the
word
they
par;cipants
also
read
a
heard
from
two
choices
(Hillenbrand
et
al
passage
and
a
word
list
con-‐
1995,
Strange
1995,
Thomas
2002)
– E.g.
BAIT
or
BET,
DATE
or
DEBT
taining
vowels
and
phone;c
contexts
of
interest
4
5. Our
previous
findings
• Focused
on
the
mid-‐front
vowels
and
the
/e/
~
/ɛ/
con;nuum
• Our
results
indicated
that
a
percep;on/produc;on
link
exists
so
that:
BAIT
BET
1. Regional
shi4s
involve
not
only
differing
produc>on
but
also
percep-on
DATE
DEBT
Fridland
&
Kendall.
2012.
The
effect
of
regional
vowel
differences
on
vowel
percep;on
and
produc;on:
Evidence
5
from
U.S.
vowel
shiVs.
Lingua
122/7:
779-‐793.
6. Previous
findings
• Focused
on
the
mid-‐front
vowels
and
the
/e/
~
/ɛ/
con;nuum
• Our
results
indicated
that
a
percep;on/produc;on
link
exists
so
that:
2. Speakers
showing
more
+SVS
+NCS
evidence
of
par>cipa>on
produc>vely
in
the
SVS
and
NCS
also
show
shi4ed
percep>on
compared
to
those
in
their
regions
with
less
produc>on
shi4
Fridland
&
Kendall.
2012.
The
effect
of
regional
vowel
differences
on
vowel
percep;on
and
produc;on:
Evidence
6
from
U.S.
vowel
shiVs.
Lingua
122/7:
779-‐793.
7. Expanding
our
inquiry
• Since
ExAPP
2010
(Lingua
2012)
our
project
has
expanded:
– We’ve
examined
new
aspects
of
our
collected
data
allowing
us
to
ask
here:
• To
what
extent
do
other
parts
of
the
vowel
space
paKern
like
the
mid-‐front
vowels?
– We’ve
gathered
data
from
subjects
in
new
field
sites
allowing
us
to
ask:
• How
robust,
or
variable,
are
the
paKerns
within-‐region?
7
8. Total
subjects
included:
Percep-on
N
=
298
8
Produc-on
N
=
48
(-‐1)
ANAE
Map
11.15:
Labov,
Ash,
&
Boberg
2006:
148
9. Produc;on
data,
briefly:
West
&
North
Legend
/i/
&
/ɪ/:
green
/e/
&
/ɛ/:
blue
/æ/:
red
/ɑ/
&
/ɔ/:
orange
All
vowels
normal-‐
ized
using
Lobanov
method
(Kendall
and
Thomas
2012)
West
shows
evidence
North
shows
evidence
of
elsewhere
shiV
of
NCS
9
10. Produc;on
data,
briefly:
South
(3
sites)
TN
(original
data
from
NC
shows
some
SVS
VA
shows
some
SVS
Lingua
2012)
shows
par;cipa;on,
but,
e.g.,
par;cipa;on,
but,
e.g.,
greatest
par;cipa;on
in
low-‐back
merger
less
proximate
mid-‐
and
SVS
high-‐
front
vowels
than
TN
and
NC
South
shows
evidence
of
SVS
These
paKerns
are
in
line
with
other
findings
of
the
retreat
of
the
SVS
in
many
parts
of
the
But
variability
across
South
(Fridland
1999,
Baranowski
2008,
10
the
three
field
sites
Prichard
2010,
Dodsworth
&
Kohn
2012,
…)
11. Current
inquiry
1. How
robust
are
our
previous
findings
(on
/e/
~
/ɛ/)
when
considered
in
terms
of
sub-‐regions
and
our
new
data?
2. How
do
the
findings
obtained
for
/e/
~
/ɛ/
relate
to
other
parts
of
the
vowel
space?
– Here:
/i/
~
/ɪ/
&
/æ/
~
/ɑ/
11
12. 1.
/e/
~
/ɛ/
regional
paderns
• Our
earlier
results
(Lingua
Percep;on
of
BAIT
~
BET
2012)
– 217
subjects
• Southerners
hear
significantly
less
/ɛ/
than
North
&
West
Percep;on
of
DATE
~
DEBT
BAIT
~
BET
Model
Results
Log-‐odds
Std.
p
(Kendall
&
Fridland
2012)
Est.
Err.
(Intercept)
-‐9.615
0.647
<
0.000001
Con;nuum
Step
2.123
0.128
<
0.000001
North
vs.
South
2.983
0.891
<
0.001
West
vs.
South
3.583
0.828
<
0.0001
Ext.
Spkrs
vs.
Headphones
-‐0.766
0.477
=
0.11
Int.
Spkrs
vs.
Headphones
-‐1.354
0.481
<
0.01
Step
x
North
vs.
South
-‐0.416
0.179
<
0.05
Step
x
West
vs.
South
-‐0.540
0.159
<
0.001
Not
showing
results
for
DATE
~
DEBT
12
13. 1.
/e/
~
/ɛ/
regional
paderns
• With
the
new
data:
Percep;on
of
BAIT
~
BET
• Southerners
hear
significantly
less
/ɛ/
than
North
&
West
– I.e.:
Quite
similar
results
Percep;on
of
DATE
~
DEBT
BAIT
~
BET
Model
Results
Log-‐odds
Std.
p
Est.
Err.
(Intercept)
-‐9.504
0.608
<
0.000001
Con;nuum
Step
2.019
0.099
<
0.000001
North
vs.
South
1.073
0.588
=
0.068
West
vs.
South
2.403
0.707
<
0.001
Ext.
Spkrs
vs.
Headphones
-‐1.033
0.364
<
0.01
Int.
Spkrs
vs.
Headphones
-‐0.910
0.311
<
0.01
Step
x
North
vs.
South
-‐0.251
0.121
<
0.05
Step
x
West
vs.
South
-‐0.489
0.136
<
0.001
Not
showing
results
for
DATE
~
DEBT
13
14. 1.
/e/
~
/ɛ/
sub-‐regional
paderns
• Broken
down
by
sub-‐ Percep;on
of
BAIT
~
BET
region
(states):
• There
are
within-‐region
differences,
but
these
ul;mately
appear
in
line
with
the
larger
regional
Percep;on
of
DATE
~
DEBT
paderns
– E.g.,
the
three
Southern
sites
are
significantly
different
from
one
other
but
s;ll
padern,
together,
differently
than
the
other
regional
sites
14
15. 1.
/e/
~
/ɛ/
direct
link
• What
about
the
curvilinear
rela;on-‐
ship
between
/e/-‐/ɛ/
Euclidean
distance
and
vowel
percep;on?
• As
reported
in
Lingua
2012
+SVS
+NCS
15
16. 1.
/e/
~
/ɛ/
direct
link
• What
about
the
curvilinear
rela;on-‐
ship
between
/e/-‐/ɛ/
Euclidean
distance
and
vowel
percep;on?
• In
new
data:
Generally
similar
results,
but
somewhat
mi;gated
– Logis;c
mixed-‐effect
model
on
subset
data
for
BAIT
~
BET
indicates
that
South
is
sig.
different
from
North
but
not
West
and
that
/e/-‐/ɛ/
distance
as
a
polynomial
is
sig.
(though
polynomial
term
is
marginal)
– The
Virginians
in
par>cular
are
much
more
West-‐like
in
their
mid
vowel
produc>ons,
and
somewhat
flaKen
out
the
paKern…
16
17. 2.
/i/
~
/ɪ/
regional
paderns
• Not
as
differen;ated
as
the
/e/
~
/ɛ/
Percep;on
of
BEAD
~
BID
percep;ons
– Both
in
terms
of
regional
differences
and
the
range
of
the
psychometric
func;ons
• But
Southerners
do
hear
significantly
more
/i/
than
the
other
regions
• These
/i/
~
/ɪ/
and
/e/
~
/ɛ/
percep>on
findings
are
in
line
with
SVS’
more
Percep;on
of
DEED
~
DID
centralized
front
tense
vowels
DEED
~
DID
Model
Results
Log-‐odds
Std.
p
Est.
Err.
(Intercept)
-‐4.605
0.303
<
0.000001
Con;nuum
Step
0.792
0.048
<
0.000001
North
vs.
South
0.592
0.348
=
0.089
West
vs.
South
0.891
0.437
<
0.05
Step
x
North
vs.
South
-‐0.133
0.061
<
0.05
Step
x
West
vs.
South
-‐0.137
0.067
<
0.05
Not
showing
results
for
BEAD
~
BID
17
18. 2.
/i/
~
/ɪ/
direct
link
• For
the
subset
produc;on
subjects,
we
take
as
a
relevant
produc;on
measure
/i/-‐/ɪ/
Euclidean
distance
and
consider
the
percep;on
data…
• Although
regional
paderns
do
exist
in
produc;on
and
percep;on,
no
direct
produc;on-‐percep;on
rela;onship
• …
Mean
percep;on
of
BEAD
~
BID,
18
ordered
by
subjects’
/i/-‐/ɪ/
distance
19. 2.
/æ/
~
/ɑ/
regional
paderns
• Also
not
as
differen;ated
as
Percep;on
of
SAD
~
SOD
the
/e/
~
/ɛ/
percep;ons
– Again,
both
in
terms
of
regional
differences
and
the
range
of
the
psychometric
func;ons
• But
Northerners
do
hear
significantly
more
/ɑ/
than
the
other
regions
Percep;on
of
PAD
~
POD
– In
line
with
NCS
fronted
/ɑ/
SAD
~
SOD
Model
Results
Log-‐odds
Std.
p
Est.
Err.
(Intercept)
-‐4.878
0.324
<
0.000001
Con;nuum
Step
1.024
0.060
<
0.000001
South
vs.
North
-‐1.239
0.445
<
0.01
West
vs.
North
-‐1.364
0.478
<
0.01
Step
x
South
vs.
North
0.201
0.085
<
0.05
Step
x
West
vs.
North
-‐0.024
0.087
=
0.782
Not
showing
results
for
PAD
~
POD,
also
19
not
showing
a
significant
effect
of
speaker/headphone
factor
20. 2.
/æ/
~
/ɑ/
direct
link
• For
the
subset
produc;on
subjects,
we
take
as
a
relevant
produc;on
measure
/ɑ/-‐/ɔ/
Pillai
score,
a
measure
of
merger
status
(Hay
et
al.
2006,
Hall-‐Lew
2010)
and
consider
the
percep;on
data
• Similar
results
as
found
for
/e/
~
/ɛ/!
– We
find
significant
effects
for
both
region
and
for
merger
status
– North
hears
more
/æ/
• (Yes,
opposite
from
full
dataset
results!?)
– But
also
curvilinear
direct
rela;onship
between
produc;on
and
percep;on
• Subjects
in
middle
of
the
Pillai
range
most
likely
to
hear
/ɑ/,
those
with
lowest
Pillai
most
likely
to
hear
/æ/
Mean
percep;on
of
SAD
~
SOD,
20
ordered
by
subjects’
/ɑ/-‐/ɔ/
Pillai
21. 2.
More
on
/æ/
~
/ɑ/
direct
link
• But
/ɑ/-‐/ɔ/
Pillai
is
actually
a
“weird”
predictor
for
performance
on
this
con;nuum
• And,
e.g.,
/æ/-‐/ɑ/
Euclidean
distance
seems
like
a
reasonable
metric
for
the
low
vowel
percep;on
data
– And
actually
is
the
parallel
to
our
/e/-‐/ɛ/
work
• Indica;ons
of
significance
here
too!
– With
an
interac;on
between
/æ/-‐/ɑ/
distance
and
/ɑ/-‐/ɔ/
Pillai
– But
/ɑ/-‐/ɔ/
Pillai
has
stronger
effect
• And
the
model
on
previous
slide
outperforms
this
model
Mean
percep;on
of
SAD
~
SOD,
21
ordered
by
subjects’
/æ/-‐/ɑ/
distance
22. In
closing
• A
lot
more
to
do!
– We
are
con;nuing
to
gather
new
produc;on
and
percep;on
data
in
these
and
addi;onal
field
sites
• Our
produc;on
results,
in
par;cular,
for
VA
and
NC
will
likely
change
as
we
flesh
out
the
number
of
analyzed
speakers
– And
collec;ng
new
percep;on
data
in
a
social
condi;on
(ala
Niedzielski
1999,
Hay
et
al.
2006)
• But:
– A
larger
dataset,
with
more
regionally
variable
subjects,
con;nues
to
show
the
same
overarching
paderns
for
/e/
~
/ɛ/
– Most
importantly,
perhaps,
we
have
also
found
more
evidence
for
a
curvilinear
rela;onship
between
vowel
produc;on
and
vowel
categoriza;on
in
percep;on
• /æ/
~
/ɑ/
shows
the
same
kind
of
padern
as
/e/
~
/ɛ/
-‐
individuals
who
are
in
the
middle
of
the
produc;on
spectrum
appear
to
behave
differently
than
those
on
the
extremes
–
even
though
/æ/
&
/ɑ/
are
engaged
in
different
kinds
of
shiVs
22
23. Thank
you
Research
funded
by
NSF
grants
#
Selected
References
BCS-‐0518264
&
BCS-‐1123460
(PI
• Baranowski,
Maciej.
2008.
The
Southern
ShiV
in
a
marginally
Southern
dialect.
Pennsylvania
Working
Papers
in
Linguis>cs
14.2:
35-‐43.
Fridland),
and
BCS-‐1122950
(PI
• Dodsworth,
Robin
and
Mary
Kohn.
2012.
Urban
rejec;on
of
the
vernacular:
Kendall)
The
SVS
undone.
Language
Varia>on
and
Change
24:
221-‐245
• Fridland,
Valerie.
1999.
The
Southern
ShiV
in
Memphis,
Tennessee.
Language
Varia>on
and
Change
11:
267-‐285.
• Fridland,
Valerie.
2001.
The
social
dimension
of
the
Southern
Vowel
ShiV:
We
are
grateful
to
Craig
Fickle
at
the
Gender,
age
and
class.
Journal
of
Sociolinguis>cs
5,
233-‐253.
University
of
Oregon
and
Sohei
• Fridland,
Valerie
and
Tyler
Kendall.
2012.
The
effect
of
regional
vowel
differences
on
vowel
percep;on
and
produc;on:
Evidence
from
U.S.
vowel
Okamoto
at
the
University
of
shiVs.
Lingua
122/7:
779-‐793.
Nevada,
Reno
for
support
with
• Kendall,
Tyler
and
Valerie
Fridland.
2012.
Varia;on
in
the
produc;on
and
percep;on
of
mid
front
vowels
in
the
US
Southern
Vowel
ShiV.
Journal
of
various
aspects
of
this
research.
Phone>cs
40:
289-‐306.
• Kendall,
Tyler
and
Erik
Thomas.
2012.
Vowels:
Vowel
manipula>on,
We
also
thank
Haley
Lee,
Kristen
normaliza>on,
and
plofng
in
R.
R
package,
version
1.2.
[
URL:
hdp://cran.r-‐
project.org/web/packages/vowels/
]
Mankosa,
and
Ken
Konopka
for
• Labov,
William,
Sharon
Ash
and
Charles
Boberg.
2006.
The
Atlas
of
North
help
conduc;ng
fieldwork
for
American
English:
Phone>cs,
Phonology
and
Sound
Change.
Berlin:
De
Gruyter.
this
project.
• Gordon,
Madhew
J.
2005.
The
Midwest
and
West.
In
Handbook
of
Varie>es
of
English:
The
Americas
and
Caribbean,
Vol
I:
Phonology,
ed.
E.
Schneider,
338–
350.
Berlin:
Mouton
de
Gruyter.
• Prichard,
Hillary.
2010.
Linguis;c
Varia;on
and
Change
in
Atlanta,
Georgia
Pennsylvania
Working
Papers
in
Linguis>cs
16,
141-‐149.
• Thomas,
Erik.
2001.
An
Acous>c
Analysis
of
Vowel
Varia>on
in
New
World
English.
Publica;on
of
the
American
Dialect
Society
85.
Durham,
NC:
Duke
University.