Naming pictures and
repeating words
Analysis of aphasic production errors and
predictions from computational models
Gary Dell, Nazbanou Nozari,
Audrey Kittredge, and Myrna Schwartz
CogSci 2009
July 30, 2009
Research goal
• Two most common tasks used to study
impaired language production
NAMING
CAT
REPETITION
CAT
“CAT”
• Need a theory of how they are related to
each other
“Parallel Case Series” approach
Computational case series
Real case series
DATA
DATA
Identical statistical analysis
Theories of the
naming-repetition relationship
Naming: the 2-step model of
lexical access
Foygel & Dell (2000)
FOG
f
r
d
Onsets
k
DOG
CAT
m
æ
RAT
o
Vowels
MAT
g
t
Codas
Naming: the 2-step model of
lexical access
Foygel & Dell (2000)
STEP 1
s weights
FOG
f
r
d
Onsets
k
DOG
CAT
m
æ
RAT
o
Vowels
MAT
g
t
Codas
Naming: the 2-step model of
lexical access
Foygel & Dell (2000)
STEP 1
s weights
FOG
f
r
d
Onsets
k
DOG
CAT
m
æ
RAT
o
Vowels
MAT
g
t
Codas
Naming: the 2-step model of
lexical access
Foygel & Dell (2000)
STEP 1
s weights
FOG
DOG
CAT
RAT
MAT
STEP 2
p weights
f
r
d
Onsets
k
m
æ
o
Vowels
g
t
Codas
Naming: the 2-step model of
lexical access
Foygel & Dell (2000)
s weights
FOG
DOG
CAT
RAT
MAT
STEP 2
p weights
f
r
d
Onsets
k
m
æ
o
Vowels
g
t
Codas
Naming: the 2-step model of
lexical access
Foygel & Dell (2000)
s weights
FOG
DOG
CAT
RAT
MAT
STEP 2
p weights
f
r
d
onset
Onsets
k
m
æ
o
vowel
Vowels
g
t
coda
Codas
Naming: cartoon model
Semantics
Words
Phonemes
Repetition:
The lexical route model
e.g. Dell et al. (2007)
NAMING
Semantics
full overlap
Words
Input
phonology
Output
phonology
Repetition:
The non-lexical route model
NAMING
Semantics
little overlap
Words
Input
phonology
Output
phonology
Repetition:
The summation dual route model
e.g. Hillis & Caramazza (1991)
NAMING
Semantics
some overlap
Words
Input
phonology
Output
phonology
Index of step 2 of naming
• Frequency effect
◦ high-frequency words are less error-prone in naming (e.g.
Nickels & Howard, 1994)
• Felt throughout lexical access, but stronger on the
second step (Kittredge et al., 2008)
F
F
• The size of the frequency effect in repetition shows
the degree of overlap with step 2 of naming
Index of step 2 of naming
• Frequency effect on Nonword errors
STEP 2
“DAT”
Parallel Case Series approach
1. Computational case series:
simulate effect of
frequency in both
tasks according to
different theories
Simulations
8 digital “patients”
(lesions to s and p weights)
LEXICAL
ROUTE
NON-LEXICAL
ROUTE
SUMMATION
DUAL
ROUTE
½ trials high-frequency, ½ trials low-frequency
Parallel Case Series approach
2. Real case series:
assess effect of
frequency in both
tasks on group
performance of
aphasics
Aphasic patients
• Non-selective sample of 59 patients
• Left hemisphere CVA
• Naming and repetition with same 175 stimuli
Parallel Case Series approach
3. Apply identical statistical analysis
to simulated and
real patient data
Binomial hierarchical multiple
logistic regression
How large is the frequency effect on Nonword
errors in word repetition, compared to naming?
*
Frequency
effect
*
4
(log odds of not
2
making a
Nonword error)
*
*
SIMULATIONS
repetition
repetition
Lexical
route
Summation
dual route
naming
repetition
Non-lexical
route
Why does the dual route model
behave like the lexical route model?
• Non-lexical route contributes activation
without taking away from the frequency effect
Why does the dual route model
behave like the lexical route model?
SUMMATION DUAL ROUTE REPETITION MODEL
lexical
route
activation of
output phonemes
LOW
FREQUENCY
HIGH
FREQUENCY
Why does the dual route model
behave like the lexical route model?
SUMMATION DUAL ROUTE REPETITION MODEL
activation of
output phonemes
frequency
effect
lexical
route
+
non-lexical
route
LOW
FREQUENCY
HIGH
FREQUENCY
Binomial hierarchical multiple
logistic regression
*
Frequency
effect
*
4
(log odds of not
2
making a
Nonword error)
*
SIMULATIONS
repetition
repetition
Lexical
route
Summation
dual route
naming
repetition
Non-lexical
route
.4
repetition
.2
naming
APHASIC
PATIENTS
Repetition is
fully lexically influenced
• Lexical route is often used repeat words
• Individual patients may augment it with
the non-lexical route
“Parallel Case Series” approach
Computational case series
Real case series
DATA
DATA
Identical statistical analysis
Thank you
•
•
•
•
Gary Dell
Nazbanou Nozari
Myrna Schwartz
Language Production Lab at University
of Illinois
• Funding (NIDCD #DC000191)