LEXICAL PROCESSING ANOMALIES IN TASK COMPARISONS

1
LEXICAL PROCESSING ANOMALIES IN TASK COMPARISONS

• Kenneth I. ForsterUniversity of Arizona

2
Any genuine lexical effect should be obtained in
any task that requires lexical access. --Ano
n
Is this really true?
We, and others, have encountered surprising
differences between lexical decision (LD) and
semantic categorization (SC) tasks.
Both tasks clearly involve lexical access.
So, what are the differences?
How are they to be explained?
3
Difference 1Sensitivity to Semantic Effects
SC is more sensitive than LD to semantic effects

• Frenck-Mestre Bueno (1999)
• strong masked priming effects for exemplars
• rifle-pistol whale-dolphin
• (prime duration 28 ms)
• highly unlikely with LD

4
Sensitivity to Semantic Effects (cont.)
Hector (2002) Associative-semantic Priming for
non-exemplars (42 ms prime)
Lexical decision Semantic cat.
(animal) Related 530 570 Unrelated
529 583 -1 13
5
Difference 2Cross-language Translation Priming
In LD, strong L1-L2 priming, no L2-L1 priming

• BUT in SC, priming is symmetric
• Grainger Frenck-Mestre, 1998
• Finkbeiner, Forster, Nicol Nakamura (2002)

6
Difference 3Insensitivity to Orthographic
Effects

• Neighborhood Density (N) Effects

• In lexical decision,
• high-N words are faster (debatable)
• high-N nonwords are slower (non-debatable)

There is no effect for words (Forster Shen,
1996). However this is also being debated.
What happens to nonwords in semantic
categorization?
7
N effects for Nonwords

• Lexical decision

Semantic categorization
(Forster Hector, MC in press)
Neighbors seem to be ignored.
8
Are neighbors really ignored?
Category Animal
GOAN
CADEL
POTHE
loan moan gown goad goat goal
cadet camel
CANDIDATES
SC Times 631 644
607
(Forster Hector, MC in press)
Only the non-animal neighbors are ignored.
9
How is this achieved?
How can you tell which neighbor to evaluate
without testing the semantic properties of each?
This should produce a cost for all neighbors.
10
What does SC have that LD doesnt?
Is it the contextual effect of the category?
This may focus the semantic activation produced
by the prime and the target.
Prime
Target
sense1sense2sense3sense4sense5
sense10sense11sense3sense12sense13
11
What does SC have that LD doesnt?
Is it the contextual effect of the category?
This may focus the semantic activation produced
by the prime and the target.
Prime
Target
sense1sense2sense3sense4sense5
sense10sense11sense3sense12sense13
CONTEXT
12
Semantic Focussing
Context Filter
sense1sense2sense3sense4sense5
word
i.e., this is non-interactive
13
The Focussing Effect
This produces an increase in the proportion of
primed senses. This could explain

• enhanced L2-L1 translation priming
• enhanced semantic priming (for exemplars)

It could not explain

• enhanced semantic priming for non-exemplars
• absence of N effects

14
Difference 4Frequency Effects in SC
Balota Chumbley (1984) No frequency effect for
non-exemplars in SC
NOT SO Monsell, Doyle Haggard (1989)Forster
Shen (1996)
HOWEVER..
15
Category Size Effects
The size of the category affects the frequency
effect for non-exemplars.
LARGE CATEGORIES SMALL CATEGORIES(animal,
living thing) (number, month) Strong
frequency effect No frequency effect
IMPLICATION No decisions for small categories
are reached without lexical access.
16
Category Search

• If a category is very small, and well-learned
• No decisions can be reached by exhaustive
  search of the category
• therefore, no frequency effect
• no masked repetition priming

17
Category Search (cont.)
Categories month, number, body parts, etc.
NON-EXEMPLARS HF REPORT LF TURBAN
Results for Non-exemplars
HF LF Primed 563 583 573 Control 626 610 618
594 596
18
Category Search (cont.)
Could this be a pre-lexical effect? Try again
with a large category.
Category Animal
Results for Non-exemplars
HF LF Primed 558 579 569 Control 579 602 591
569 591
19
Feature Monitoring
Decision maker monitors specific features
S
P
O
Neighbors are irrelevant (unless they activate
the right features)
Nonexemplar decisions are made at semantic level
without waiting for network to settle.
20
But feature monitoring also predicts no frequency
effects for non-exemplars in any category. And,
no priming.
21
Where to next?
Current hypothesis with small categories,
category search is fast enough for the prime to
generate task-relevant output.
Category number turban TURBAN
tentative No output generated
22
Thats all. Thank you.