<%BANNER%>

The Effects of Feature Type on Semantic Priming of Picture Naming in Normal Speakers

Permanent Link: http://ufdc.ufl.edu/UFE0042185/00001

Material Information

Title: The Effects of Feature Type on Semantic Priming of Picture Naming in Normal Speakers
Physical Description: 1 online resource (84 p.)
Language: english
Creator: Del Toro, Christina
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: features, memory, priming, semantic
Communication Sciences and Disorders -- Dissertations, Academic -- UF
Genre: Communication Sciences and Disorders thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The aim of this study was to investigate the roles of shared and distinctive features on conceptual activation. A picture naming paradigm was employed to measure speech reaction time during feature-to-concept activation of animals, tools, and vehicles. Fifty-nine young adults and 47 older adults completed the priming task with an interstimulus-interval of 200msec and 600msec, in two different sessions. Results indicate that regardless of semantic category, distinctive feature primes resulted in the fastest reaction times compared to shared features, a combination of distinctive and shared features, and neutral primes. In general, manipulation of ISI did not produce changes in SRT. Overall, the results showed that as stated in the CSA, distinctive features have a privileged role in concept activation. However, the proposed differential roles of shared and distinctive features in living and nonliving things, was not confirmed. Additionally, the results indicate that in general, multiple feature primes do not require more time to activate a concept.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Christina Del Toro.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Kendall, Diane L.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042185:00001

Permanent Link: http://ufdc.ufl.edu/UFE0042185/00001

Material Information

Title: The Effects of Feature Type on Semantic Priming of Picture Naming in Normal Speakers
Physical Description: 1 online resource (84 p.)
Language: english
Creator: Del Toro, Christina
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: features, memory, priming, semantic
Communication Sciences and Disorders -- Dissertations, Academic -- UF
Genre: Communication Sciences and Disorders thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The aim of this study was to investigate the roles of shared and distinctive features on conceptual activation. A picture naming paradigm was employed to measure speech reaction time during feature-to-concept activation of animals, tools, and vehicles. Fifty-nine young adults and 47 older adults completed the priming task with an interstimulus-interval of 200msec and 600msec, in two different sessions. Results indicate that regardless of semantic category, distinctive feature primes resulted in the fastest reaction times compared to shared features, a combination of distinctive and shared features, and neutral primes. In general, manipulation of ISI did not produce changes in SRT. Overall, the results showed that as stated in the CSA, distinctive features have a privileged role in concept activation. However, the proposed differential roles of shared and distinctive features in living and nonliving things, was not confirmed. Additionally, the results indicate that in general, multiple feature primes do not require more time to activate a concept.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Christina Del Toro.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Kendall, Diane L.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042185:00001


This item has the following downloads:


Full Text





THE EFFECTS OF FEATURE TYPE ON SEMANTIC PRIMING OF PICTURE NAMING
IN NORMAL SPEAKERS




















By

CHRISTINA MARIA DEL TORO


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2010

































2010 Christina Maria del Toro





















To my family who has given me unending love, support, guidance and joy, this work is
dedicated to you









ACKNOWLEDGMENTS

There are many people who have helped and supported me throughout the

completion of my degree and dissertation. I thank my parents, Jorge and Margarita del

Toro and my sister Jennifer McGugan, for their constant encouragement and support

and most importantly understanding in what has been a unique journey for us all. I am

forever grateful to Michelle Troche for being an academic partner, study buddy,

roommate and friend over many years and degrees at UF. I cannot imagine my UF

years without her. I am thankful for Lauren Bislick who I met at the beginning of the

PhD journey and who has become a wonderful friend and support especially in the final

year. I thank my roommates who have lived this journey with me, Joshua Troche,

Heather Beck, and Renee Dickinson. They were there through it all and reminded me

to live beyond school.

I must thank the UW lab members who have provided endless moral support,

technical support, intellectual support, and of course, doughnuts: JoAnn Silkes,

Rebecca Hunting Pompon, Laine Anderson, Wesley Allen, Megan Oelke, and Mike

Mackinnon. There are three people without whom I could not have completed data

collection: Erica Gonzalez, Brittney Hayes, and. Matt Lacourse. Also, Lorraine

Gonzalez who not only recruited and scheduled nearly everyone she knows, but

graciously allowed us to use her house for weekend data collection. I am forever

grateful to all of them for many hours spent in participant recruitment, scheduling,

testing, and traveling.

My committee members deserve many thanks for years of guidance and support.

I thank Bruce Crosson, for introducing me to new literatures and providing support along

the way; also for invaluable guidance on methodology. I am thankful to Lisa Edmonds









for allowing me to work in her lab for my first semester and gain invaluable experience

in treatment research; also, for her mentorship in my first ever teaching experience; and

guidance in stimuli development for my dissertation. I thank Leslie Gonzalez-Rothi for

providing a supportive environment at the Brain Rehabilitation Research Center where I

learned how to become a researcher starting as an undergraduate; also, for always

pushing me beyond my comfort zone and encouraging me the whole way.

To my chair and friend, Diane Kendall I offer more thanks than I can express. For

years of research experience, from doing reliability on her treatment study to designing

stimuli alongside her for the SAPA test. I have not only learned skills for research,

teaching, and mentoring but through her example and her words I have learned skills for

balancing an academic life with a fulfilling personal life. I look forward to many more

years of academic collaboration and friendship.

Lastly, there is a large group of people without whom I would not be able to write

this dissertation: the participants. I especially thank those that gave their weekend time

to help me. I am so thankful for their time and energy and willingness to complete all

the tasks I asked of them.









TABLE OF CONTENTS

page

A C K N O W LE D G M E N T S ................................................................................. .. .... 4

LIST O F TA B LES ...................................................................................... 8

LIST OF FIGURES.................................. ......... 10

LIST O F A BBR EV IAT IO N S ........................................................................... 11

ABSTRACT .................................... ................................... ........... 12

CHAPTER

1 INT R O D U C T IO N ....................................................... .......... .......... 13

S e m a ntic M e m o ry ............................................................................ 13
The Conceptual Structure Account ............................. ....................... .............. 16
Evidence for the conceptual structure account ......................................... 19
Limitations of the conceptual structure account................ ................... 21
C onfirm ing the C SA ............................................................. .... ........ 23
S em antic P rim ing ..................................................... ............. 24
Prim ing in D distributed Netw orks ................................................. ............... 25
M u ltip le P rim e s ........................................................................ 2 6
Statem ent of the Problem .................................... ........................ ............... 28
Research Q questions and Hypotheses................................................. ............... 28

2 METHODS........................................................ 33

Data Collection ................................... .................. ................ 33
D e sig n ...................................... ..................... ...... ........... ....... 3 3
Dependent variables ........................................ ........................ ........ 33
Independent variables........................................ .................. 33
Trial structure .................................... ........................... .......... 33
T im in g ................................................................................................ 3 4
Task ............................................................................................. 34
Settings and Equipm ent .............................................................. ............. 35
Stimuli.................. ............................. ........... 35
S e m a ntic ca te g o rie s ................................................................... .......... 3 5
P rim e ty p e ............................................................................... ...... ........ 3 6
Participants................... .. ................................ 37
Inclusion criteria ...................................................................... 38
Screening............................................. 38
Data Analysis............................ ........................... 39
Transcription and Scoring................................................. .................... 39
D a ta T rim m in g .................................................................................. ...... ...... 3 9


6









Reliability ................... ............ ....................... 39
S tatistica l A na lysis ........................................................................ ......... 4 0

3 R ESU LTS ......... ..... ......... ....................................................................... 44

R liability ............... ........................................................................ ...... 44
Research Question 1 ........... ............................. 44
Younger Adults ............. ............................... 44
Older Adults............................................ ............... 45
Research Question 2 ........................................... 45
Research Question 3 ............. .............................. 46
Research Question 4 ............. .............................. 47
W ord Frequency Effects ................................. ......... ............................. 47

4 D ISC U SS IO N ............. ......... .................................................................... 63

Priming Effects from Related Features and Neutral Primes.................................. 63
The Effect of Feature Type ................................. ......... ............................. 64
The Effect of Number of Features................... ....... .................... 68
The Effect of Timing on Multiple Feature Primes .......................................... 69
The Effect of Word Frequency............................................ 71
Lim stations ................................. ...... ... .. ........ ................... 71
Implications for Anomia Treatment .......... ...... ......... ....................... 72
F u tu re D ire c tio n s .................................................................................................... 7 3
S um m ary .............. ......... ................................................................. ...... 74

APPENDIX

A PRIME-TARGET STIMULI........................... ............... 76

B TARGET WORD FREQUENCY ...................... .... ..................... 79

LIST OF REFERENCES .............. ............................... 80

BIOGRAPHICAL SKETCH ................................. ............................. 84









LIST OF TABLES


Table page

1-1 The distribution and correlations of shared and distinctive features in living
and nonliving things, according to the CSA ............................... ... ................ 31

1-2 Summary of Research Questions (RQ) and Predictions ............... ............... .. 32

2-1 Participant demographic information including age, years of education,
gender and average (AVE) test scores with standard deviations (SD) for the
Mini-Mental Status Exam (MMSE; Folstein, Folstein & McHugh, 1975), Test
of Word Reading Efficiency (TOWRE; Torgesen, Wagner, & Rashotte, 1999),
Pyramids and Palm Trees Test (P&P; Howard & Patterson, 1992), American
National Adult Reading Test (ANART; Nelson, 1982), and estimated IQ
based on ANART Scores (Grober & Sliwinski, 1991). .................... .. .......... .. 41

3-1 Raw mean SRT for younger adults for comparison of neutral, shared
features, distinctive features, and combination of shared and distinctive
features prime conditions. Asterisks indicate significant pairwise
comparisons which are described in parentheses ............................................ 49

3-2 Raw mean SRT for older adults from comparison of neutral, shared features,
distinctive features, and combination of shared and distinctive features prime
conditions. Asterisks indicate significant pairwise comparisons which are
described in parentheses. ...................................................... ............... 50

3-3 Raw mean SRT from comparison of shared versus shared-shared prime
conditions. There were no significant effects of prime condition......................... 51

3-4 Raw mean SRT from comparison of distinctive and distinctive-distinctive
prime conditions. There were no significant effects of prime condition .............. 52

3-5 Raw mean SRT from comparison of 200msec and 600msec ISI for the prime
conditions of neutral, shared features, distinctive features, and combination
of shared and distinctive features. There were no significant effects of ISI....... 53

3-6 Raw mean SRT from comparison of 200msec and 600msec ISI for the
shared feature prime conditions. Asterisks indicate significant pairwise
comparisons which are described in parentheses ................ .............. 54

3-7 Raw mean SRT from comparison of 200msec and 600msec ISI for the
distinctive feature prime conditions. There were no significant effects of ISI..... 55

A-1 Animal Stim uli.......................................... .......... 76

A-2 Tool Stimuli........................................... .......... 77









A -3 V e h ic le S tim u li ............... ......... .... ......... .. ..... .............................................. 7 8

B-1 Taret Word Frequency................................... ............... 79









LIST OF FIGURES


Figure page

2-1 Example of priming task trial structure displaying the shared-distinctive
condition .................. ................. ......... ....... ..........42

2-2 Priming task trial structure for one complete trial displaying timings of each
screen. All trials within one session are administered with an ISI of 200 or
600msec for the blank screens between primes and between the second
prime and target picture. The target picture remains on the screen until the
voice key is triggered by the participant's verbal response.............................. 43

3-1 Log-transformed mean SRT of young adult responses from comparison of
neutral, shared, distinctive and combined prime conditions. Brackets indicate
significant pairwise comparisons. Asterisks indicate the prime condition
which produced significantly faster SRTs .................................. ................... 56

3-2 Log-transformed mean SRT of older adult responses from comparison of
neutral, shared, distinctive and combined prime conditions. Brackets indicate
significant pairwise comparisons. Asterisks indicate the prime condition
which produced significantly faster SRTs....................................... .... ........... 1

3-3 Log-transformed mean SRT of older adult responses from comparison of
shared and shared-shared prime conditions. There were no significant
effects of prime condition. ..... ................................ ... ..... .......... 58

3-4 Log-transformed mean SRT of older adult responses from comparison of
distinctive and distinctive-distinctive prime conditions. There were no
significant effects of prime condition ....... ...................... .............. 59

3-5 Log-transformed mean SRT of older adult responses to all prime conditions
(collapsed across shared, distinctive, combined, and neutral primes) at 200
ISI versus 600 ISI. There was no significant difference in SRT between ISI
conditions. .............. ... ......... ... ........................ ... ......... 60

3-6 Log-transformed mean SRT of older adult responses to shared primes
(collapsed across shared and shared-shared) primes at 200 ISI versus 600
ISI. Brackets indicate significant pairwise comparisons. Asterisks indicate
the prime condition which produced significantly faster SRTs............ ........... 61

3-7 Log-transformed mean SRT of older adult responses to distinctive primes
(collapsed across distinctive and distinctive-distinctive) primes at 200 ISI
versus 600 ISI. There was no significant difference in SRT between ISI
conditions. .............. ... ......... ... ........................ ... ......... 62









LIST OF ABBREVIATIONS

ANART American National Adult Reading Test

CSA Conceptual Structure Account

ISI Inter-stimulus Interval

MMSE Mini Mental Status Examination

msec Milliseconds

P&P Pyramids & Palm Trees

SEC Seconds

SRT Speech Reaction Time

TOWRE Test of Word Reading Efficiency









Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

THE EFFECT OF FEATURE TYPE ON SEMANTIC PRIMING OF PICTURE NAMING
IN NORMAL SPEAKERS

By

Christina M. del Toro

August 2010

Chair: Diane Kendall
Major: Communication Sciences & Disorders

The aim of this study was to investigate the roles of shared and distinctive

features on conceptual activation. A picture naming paradigm was employed to

measure speech reaction time during feature-to-concept activation of animals, tools,

and vehicles. Fifty-nine young adults and 47 older adults completed the priming task

with an interstimulus-interval of 200msec and 600msec, in two different sessions.

Results indicate that regardless of semantic category, distinctive feature primes resulted

in the fastest reaction times compared to shared features, a combination of distinctive

and shared features, and neutral primes. In general, manipulation of ISI did not

produce changes in SRT. Overall, the results showed that as stated in the CSA,

distinctive features have a privileged role in concept activation. However, the proposed

differential roles of shared and distinctive features in living and nonliving things, was not

confirmed. Additionally, the results indicate that in general, multiple feature primes do

not require more time to activate a concept.









CHAPTER 1
INTRODUCTION

Semantic Memory

Semantic memory is our cumulative knowledge of things, people, places, and

events and is considered one of the most critical aspects of human cognition

(Hutchison, 2003). Many theories and models have been put forth to elucidate the

structure of conceptual representations and how the representations are processed.

One group of models, referred to here as semantic feature theories, is based on the

assumption that concepts, or conceptual information, are instantiated in a distributed

neural system comprised of smaller units. The smaller units represent semantic

features or properties and are the lowest level of representation in semantic memory.

Concepts emerge from overlapping patterns of activation across related feature units

(Plaut, 1996). For example, the activation of the conceptual representation of tree,

requires simultaneous activation of related feature units such as branch, leaves, trunk,

bark, etc. Thus, a concept is not represented as a discrete unit, but instead emerges

from activated features. This leads to distributed representations of concepts over

unique patterns of activation (Plaut, 1996).

The relationship of conceptual feature units and their co-activation have been

described by many researchers using varying terminology and specifications. However,

despite such differences, it is generally agreed upon that the connections between

conceptual representations are based on similarity and differences of features. Such

connectivity has been described in terms of semantic neighborhoods (Mirman and

Magnuson, 2008) and typicality effects (Rosh, 1975; Plaut, 1996).









Mirman and Magnuson (2008) propose that the structure of semantic memory is

based on how close or distant conceptual representations are from one another in terms

of semantic relatedness. If concepts are closely related, i.e. share overlapping features,

such as cow and bull, they are considered to be 'near' neighbors. On the other hand,

conceptual representations that share fewer overlapping features, such as cow and

tiger are considered 'distant' neighbors. In this semantic neighborhood model, patterns

of activation correspond to the distance between features based on similarity. Mirman

and Magnuson have shown, in computational and behavioral experiments, that

semantic processing is slowed by dense, near neighbors and speeded by far neighbors.

These results are attributed to inhibitory effects from near neighbors and facilitative

effects from distant neighbors. For example, when asked to name a picture of a cow,

prior presentation of a far neighbor tiger, will be facilitative. Presentation of tiger will

activate its features which will include a few shared features with cow; however, when

cow is activated, the semantic system can inhibit the activation of tiger because there is

little overlap of activation patterns for tiger and cow. Inhibiting the activation of tiger will

not turn off a critical amount of features for cow. Conversely, prior presentation of a bull

(a near neighbor) will be inhibitory when cow is presented. The presentation of bull will

activate numerous features that are shared with cow but also features that are unique to

bull. When cow is presented, the semantic system must activate its features which will

overlap with the already activated features of bull. Additionally, the system must inhibit

the features of bull which now do not match the activation pattern of cow. The greater

number of shared features between bull and cow causes difficulties in inhibiting only bull

and in turn, activation of cow is slowed.









Conceptual representations in semantic memory have also been purported to be

connected through semantic typicality (Rosch, 1975). This view suggests semantic

categories are built on a hierarchy of typicality, or how close each exemplar is to the

prototype of that category. For example, robin is a typical member of the bird category,

while ostrich is considered an atypical exemplar. The notion of typicality has been

tested computationally (Plaut, 1996). The results of damaging and retraining the model

revealed that generalization to untrained items was greater when the trained items were

atypical members of a category. For example, training ostrich, resulted in

generalization to untrained items such as bluejay, and robin; but training bluejay did

not generalize to the untrained penguin. This suggests that activation of atypical

exemplars activates features which are unique to that conceptual representation, as

well as features which are common to most members of that category. On the other

hand, activation of typical exemplars activates the features common to most members

and not the unique features of the atypical members. Thus, if typical items are trained,

the distinguishing features of atypical members will not be activated and access to them

will not be improved.

Based on the theory that a concept is comprised of a distributed representation of

features, and processed via simultaneous activation of those features, techniques for

remediating word retrieval deficits have been developed. Two such treatments are

semantic feature analysis (SFA; Boyle & Coehlo, 1995) and the complexity account of

treatment efficacy (CATE; Thompson, Shapiro, Kiran, & Sobeck, 2003). In SFA

treatment patients are asked specific question about the features of a pictured object

and then asked to name the object; for example, while viewing a picture of an apple, the









patient is asked where do you find it? What do you do with it? The goal is to improve

access to the concept (object to be named) by activation of the features. The GATE

treatment is based on Plaut's (1996) computational model which investigated Rosch's

(1976) typicality effect. As in SFA, the objective of GATE is to improve activation of a

conceptual representation by activating the related features. The typicality effect is

employed through the use of atypical (instead of typical) features of the concept to

improve generalization to untrained items.

Although semantic treatments have had positive treatment effects, generalization

to untrained items and maintenance effects long after therapy concludes can be

improved (Nickels, 2002). Therefore, it is necessary to refine semantic feature

treatments. One approach is to gain a better understanding about the relationship of

the feature units on the activation of conceptual representations, so that this relationship

can be exploited in word retrieval therapies. One theory that provides further specificity

regarding feature types and their connections in the semantic system is called the

Conceptual Structure Account (CSA; Tyler & Moss, 2001; Taylor et al., 2007).

The Conceptual Structure Account

The CSA, like other semantic feature theories (Mirman & Magnuson, 2008; Plaut,

1996), is based on the assumption that semantic memory is a distributed connectionist

network comprised of units that represent semantic features and the processing of a

concept is the result of overlapping patterns of activation across semantic feature units.

CSA extends and adds further specificity to the above mentioned accounts of semantic

memory through two essential points. First, the degree to which a feature is shared by

different concepts varies and second, the frequency of co-occurrence of features varies.

These two points will be described below and are summarized in Table 1-1.









In regards to the degree to which a feature is shared, the CSA identifies two types

of features, shared and distinctive. The idea is that shared and distinctive features

differentially activate conceptual representations. Shared features are defined as those

features that are common to related concepts, while distinctive features are unique to

each concept. That is, most living things share features of eyes, ears, breathes, legs;

thus, these features are only indicative of category membership. Conversely, fewer

living things have stripes, trunk, mane, or an udder making these distinctive features.

Distinctive features belong to fewer concepts and provide more information about a

specific concept.

An example given by Taylor and colleagues is tiger. To activate the concept of

tiger, the shared features which define animals and, specifically cats, such as four legs,

teeth, and tail, will be activated but, until stripes is activated, the concept of tigerwill not

be complete and therefore, not activated above other types of cats. Thus, shared

features reflect category membership and are not helpful in identification, and distinctive

features provide more information about a particular concept and are critical to

identification.

The CSA goes on to stipulate that activation of only shared features will lead to

activation of several concepts sharing those features (e.g., activation of four legs, teeth,

and tail would activate animals in addition to tiger). Activation of a distinctive feature

with shared features will highlight the target concept above its neighbors (e.g. activation

of stripes and not mane will highlight tiger over lion). Thus, the distinctive feature is

needed to fully activate the target concept (Taylor et al., 2007).









The second postulate of the CSA regarding semantic structure is the frequency of

co-occurrence of features. The CSA proposes a Hebbian structure in which units that

"fire together, wire together"; in other words, connection strengths are built through

simultaneous activation (Munakata & Pfaffly, 2004). Thus, features that co-occur

frequently have stronger connections than features which co-occur infrequently. For

example, most things which breathe have eyes but not all things that breathe have

stripes. In a Hebbian system, consequently, breathe and eyes will have a stronger

connection than breathe and stripes and therefore, the activation of breathe is more

likely to lead to the activation of eyes than stripes.

The frequency of feature co-occurrence, or correlation as referred to by Taylor et

al., (2007), differs for living and nonliving things. In living things, shared features are

highly correlated to each other (e.g. things which breathe typically have eyes and ears)

while distinctive features, on the other hand, are weakly correlated to other features

(e.g. most things with eyes and ears don't have pouches, udders, or eight legs). In the

example of tiger given above, the features describing nearly all living things, breathes,

legs, hears, eyes, ears are strongly correlated shared features while stripes is a

distinctive feature that is weakly correlated with the other features of tiger. Nonliving

things on the other hand, have distinctive features which are highly correlated (things

that have a blade also cut) and fewer shared features with lower correlations (most tools

have a handle but could be made of wood, plastic, or metal; be used for hitting, cutting,

or turning). Thus, knife has the distinctive features of blade and cuts which are strongly

correlated to each other as opposed to the shared features of handle and metal which









describe several tools but are weakly correlated with each other and the distinctive

features of blade and cuts.

Evidence for the conceptual structure account

The claims of CSA have been supported by several studies (McRae, de Sa, and

Seidenberg, 1997; McRae, Cree, Westmacott, and de Sa, 1999; Randall, Moss, Rodd,

Greer, and Tyler, 2004) using neurologically healthy individuals. Study tasks have

included feature generation, priming in a feature verification task, and priming with

targets and primes that have correlated features.

The goal of feature generation studies is to understand more about the distribution

of feature type in living and nonliving things. Participants are asked to list all the

features they can think of for a given concept. From these lists, shared and distinctive

values are calculated based on the number of occurrences of each feature for all

concepts in the list. Distinctiveness, or the degree to which a feature is distinctive to

one concept, is calculated in one of two ways: either by taking the inverse of the number

of concepts for which a feature was produced or by identifying a cut-off a priori (e.g.

distinctive features occur in only 1 or 2 concepts). Results of a feature generation study

for living and nonliving things by Taylor and colleagues (2007) supported the proposed

division of high and low correlations among shared and distinctive features in living and

nonliving things. That is, living things had a greater number of highly correlated shared

features compared to nonliving things; and, nonliving things had more highly correlated

distinctive features than living things (Taylor et al., 2007).

McRae et al. (1997) have conducted two priming studies to investigate the effect

of feature correlation. Priming studies are based on the principle that presentation of a

word (prime) will facilitate recognition of a subsequent word (target) (Meyer,









Schvaneveldt, and Ruddy, 1974). McRae et al. used an online written word

comprehension task in which participants viewed a prime concept followed by target

features which were either weakly or strongly correlated to other features of the

concept. Participants were required to answer questions about the semantic

relatedness of the prime to the target, such as, is it animate? In this word

comprehension task, reaction time to living things was faster compared to nonliving

things when the correlations between prime and target were higher; thus, supporting

the claims of the CSA that correlation of shared features is higher in living things than in

nonliving things (McRae et al.).

In the second study by McRae et al. (2007) investigating the effect of feature

correlation, participants viewed prime concepts and target features, such as deer-is

hunted, and were asked if the features belonged to the concept. Reaction times

decreased as the correlation of prime and target increased; conversely, reaction times

increased as the correlation of prime and target decreased. The authors concluded that

the simultaneous activation of correlated features leads to a faster initial rise in

activation, allowing strongly correlated features to settle into a stable activation pattern

faster than weakly correlated features (McRae et al., 1999).

Randall et al. (2004) used the results from McRae et al. (2007) to investigate if, in

living things, distinctive features would indeed be activated more slowly than shared

features due to the weaker correlation of distinctive features (when compared to a

stronger correlation of distinctive features in nonliving things). This prediction was

supported by their results from a timed feature verification task in which participants had

to respond to the target within a set time limit (Randall et al). However, in an untimed









version of the task there was no such difference between distinctive features of living

and nonliving things indicating that correlation strength of features affects only the initial

rise time of activation and not the final level of activation when the network is in a stable

state.

Limitations of the conceptual structure account

While the CSA provides a framework upon which testable hypotheses about the

structure of semantic memory can be tested, two critical limitations remain: 1) the

distribution of features in semantic categories and 2) the specificity of the semantic

concepts to which the theory can be applied.

In regards to the distribution of features in semantic categories, the CSA proposes

a distinction between living and nonliving things. As described earlier, there are more

shared than distinctive features in living things and more distinctive than shared

features in nonliving things. However, there are categories of living and nonliving things

which violate this distinction. This is a fact acknowledged by the authors of the CSA.

Vehicles, for example, although a nonliving category, are processed more like living

things because they have many shared features (wheels, engines, tires, doors, fuel, etc)

and fewer distinctive features (wings, sail, etc.) which are necessary to distinguish a

plane from a car and a car from a truck (Taylor, Moss, Tyler, 2007).

This difference in the structure of features for vehicles is also supported by

anatomical studies investigating the processing of different domains of concepts. In an

fMRI study of the contribution of category and visual attributes to semantic knowledge in

the fusiform gyrus, Wierenga et al., (2009) compared activation from naming pictures of

animals, tools, and vehicles. To compare the contribution of visual attributes, the

investigators varied the amount of visual information available by providing increasing









details to pictured images. At lower levels of spatial frequency only global form, which

can be likened to shared features, was seen compared to local detail or distinctive

information given as the spatial frequency information increased. Results indicated that

animals required the least visual detail (distinctive features) while tools required the

most and vehicles fell between the two (Wierenga, et al., 2009). These results align with

the claims of the CSA that animals have many shared features so there is information

available for identification from just shared features whereas tools have fewer shared

features and need more local details or distinctive features for identification. Likewise,

as predicted by the CSA, vehicles did not behave entirely like living or nonliving objects

but instead required both global form and local details for identification (Wierenga et al.,

2009).

Taylor and colleagues (2007) also point out that the features of fruits and

vegetables are not distributed in the same manner as animals, though all three

categories are considered living things. Fruits and vegetables have fewer distinctive

features which have even lower correlations than animals. Thus, the distribution of

shared and distinctive features and their weight for identifying objects cannot be solely

attributed to a distinction of living and nonliving categories; within these categories,

further distributional differences can be found. Taylor and colleagues (2007) posit that it

is the internal structure of the properties of concepts which determines the relationship

between shared and distinctive features, not just the category or domain to which the

concepts belong. Therefore, investigation of the distribution of shared and distinctive

features and their interaction in categories more specific than living or nonliving things

may reflect the truer structure of the system.









The second limitation of the CSA, is the specificity of objects which must be used

in order to account for the differential effects of shared and distinctive features. That is,

the CSA provides experimental evidence that, with access to distinctive features alone,

an animal will be identified more slowly due to fewer connections between features

related to the concept. However, this point may not be entirely accurate. The specificity

of the concept may influence the relationship of shared and distinctive features. For

example, meow can be considered a distinctive feature of cats, which according to the

CSA should activate the concept slower than a shared feature; but, meow would likely

be able to activate the concept of cat on its own without other information. This issue

has not been addressed by the authors of CSA. Their examples and stimuli from their

studies appear to be specific objects or members of a category (e.g. tiger) and not

general concepts or subcategories (e.g. cat). For example, Taylor and colleagues

(2007) discussed the difference between a knife and a tiger, using stripes as the

distinctive feature of tiger which would not apply to all cats. When presented alone,

stripes would not have the same effect in eliciting tiger as meow would in eliciting cat.

More information would be needed to activate tiger but meow could activate cat on its

own. Until the authors address this issue investigations of the CSA are limited to

employing specific items as stimuli and more critically the results of such studies can

only be applied to specific exemplars of a category.

Confirming the CSA

The principles of the CSA may be applied to anomia treatments, such as SFA, by

providing further insight into the nature of feature-to-concept activation. Such

information could potentially improve treatment outcomes by specifying the type of

stimuli that could be used in an experimental therapy program. However, the claims of









the CSA cannot be applied to these treatments without empirical evidence that the

principles of the CSA can be used to affect lexical access in the manner used in

aphasia therapy. Typical tasks employed in anomia treatments involve, object

description, picture naming, or naming objects by description which engage feature-to-

concept activations. In other words, when shown a picture of a hammer, individuals

with aphasia are typically asked to "tell me what you do with this?". However, CSA

experiments, thus far, have only been tested using concept-to-concept activation (e.g.

using eagel to prime hawk). In order for principles of the CSA to provide insight into

rehabilitation of anomia, the principles must first be addressed in feature-to-concept

activation. To date, no study has used features as the prime and concepts as the

target. Since a priming paradigm will be used in this study, relevant semantic priming

literature will be reviewed here.

Semantic Priming

Meyer, Schvaneveldt, and Ruddy (1974) first showed that presentation of a word

(prime) will facilitate recognition of a subsequent word (target). Since then, priming

effects have been shown to be stronger when the prime word and target word are

related or share semantic information compared to a neutral or unrelated prime-target

pair (Balota, & Paul 1996; Bueno & Frencke-Mestre, 2008; de Groot, 1984; McNamara,

2005; Moss, Ostrin, Tyler, & Marslen-Wilson, 1995; Neely, 1991; Seidenberg, Waters,

Sanders, & Langer, 1984). Most commonly, primes and targets are concepts related or

interacting in some manner. For example, lake may prime ocean because both are

bodies of water. Such concept-concept pairs reliably produce priming effects given a

semantic relationship and a semantic-based task (Hutchison, 2003).









Priming in Distributed Networks

Priming effects occur in distributed networks for two possible reasons. McNamara

(2005) refers to the first as learning models. These models predict priming effects as a

result of gradual learning by the network which leads to increased probability of

producing the same response each time a specific input is recognized. Priming effects

under this model occur over long lags of time as the system learns (McNamara, 2005).

The second explanation of priming effects in distributed networks comes from the

proximity models (Cree, McRae, & McNorgan, 1999; McRae, de Sa, & Seidenberg,

1997). These models suggest priming occurs because related words (primes and

targets) are closer and more strongly connected than unrelated words. In priming, a

target is processed after the processing of the prime; that is, a pattern is activated for

the prime which in turn activates a pattern for the target. Thus, processing of the target

is faster following processing of the prime when there are connections between the two

patterns than when there are no connections between the patterns (McNamara, 2005).

The proximity model of priming in distributed networks particularly describes the

process of priming between words which share features, because activating features for

a prime will also activate features for the target. Consequently, when the target is

processed, some of the features are already activated. Furthermore, the proximity

model predicts that exposure to a feature, or set of features, will activate those features

and connected features, thereby increasing the speed of reaching a stable pattern of

activation for a subsequent concept (McNamara, 2005). Processing of core, seeds, red

will activate these units and the units each is connected to (stem, tree, worm) so when

the picture of apple is seen activation for the name is already in process ,and thus, will

more quickly achieve a stable pattern. The effect of multiple primes, as illustrated in the









above example is an empirical question which has been studied (Balota & Paul, 1996;

Milberg, Blumstein, Giovanello, and Misiurski, 2003). However, these studies,

described below, have not used semantic features as primes. Such a paradigm would

provide further evidence that activation of features is the mechanism for concept

activation. Furthermore, determining the type of feature and number of each feature

type which most strongly primes concepts will provide a framework for devising

rehabilitation approaches for semantic impairments.

Multiple Primes

The effect of multiple primes has been studied based on spreading activation

theories which posit that activation of one unit (from one prime) will spread automatically

to related and connected parts of the network (even areas not directly connected to the

original node). Thus, activation of lion will spread to stripes through the connection of

tiger. There are three possible influences from multiple primes: additive, underadditive,

and overadditive (Balota & Paul, 1996). An additive influence occurs when the

facilitation of multiple primes is equal to the sum of the facilitation of each prime

presented individually. An underadditive influence is seen when the facilitation of

multiple primes is less than the total facilitation of each prime individually. Lastly,

overadditivity is when the facilitation from multiple primes is greater than the total

facilitation from each prime individually (Balota & Paul, 1996). Balota and Paul (1996)

investigated this effect using two sequential primes in a lexical decision task and in a

speeded word naming task. Additionally, they manipulated duration of the first primes

and degraded the target stimulus in separate conditions. All conditions resulted in

additive priming. Thus, in their study, providing several primes in succession was no

different than presenting each prime individually. However, the series of experiments









conducted by Balota & Paul (1996) used concept primes and concept targets.

Consequently, the effect of multiple features on priming a concept is not known and the

effect of time on multiple feature primes is unknown.

Milberg, Blumstein, Giovanello, and Misiurski (2003) have conducted a multiple

prime study using triplets in which the third word was the target of a lexical decision

task. The interstimulus interval (ISI), or time between the offset of a prime and the

onset of the next prime, was varied between 200msec and 600 msec. ISI was varied to

determine at what time point an overadditive effect could be achieved. There were four

types of triplets: definitionally related triplets (e.g. meal, morning, breakfast), triplets with

a categorical prime and nonword (e.g. meal, foncern, breakfast), triplets with a nonword

and featural prime (e.g. jarm, morning, breakfast), and triplets with two nonwords (e.g.

jarm, foncern, breakfast). The authors concluded that the priming effect from the

definitionally related triplets seen in the 200ms ISI condition was additive; that is, the

same as adding the priming effects from each individual prime (categorical and nonword

prime triplet + nonword and featural prime triplet). The priming effect from the

definitionally related triplet in the 600ms ISI condition, however, was overadditive, or

greater than the effect from each individual prime. Thus, the longer ISI provided time

for the convergence of meaning of the multiple primes and, subsequently, the

enhancement of the priming effect and semantic facilitation (Milberg et al., 2003).

Milberg and colleagues' (2003) work suggests that, given enough time, exposure

to multiple related words can aid reaction to a target. Whether the priming at 200 msec

ISI and 600 msec ISI is an automatic or controlled process is not discussed by Milberg

et al. and was not the aim of the study. Moreover, while understanding if the potent









mechanism is automatic or controlled processing is important, it was not the primary

focus of the current study. Instead, the aim was to determine if there is a benefit to

presenting multiple features prior to naming a picture and timing may be an important

factor in order to allow for consolidation of the meanings of each prime.

Statement of the Problem

The CSA appears to be a solid theory of semantic memory which suggests a

unique structure based on shared and distinctive features. Shared features of living

nouns are more abundant and more strongly correlated with other features, while

distinctive features are more weakly correlated and fewer in number. Conversely, non-

living nouns have more distinctive features which have strong correlations and fewer

shared features with weaker correlations. The distribution of shared and distinctive

features in living and nonliving concepts could be used to further specify the stimuli

used in anomia treatments; however, it is unknown if shared and distinctive features can

be used to activate concepts because previous studies have used concept-to-concept

activation and not feature-to-concept activation. Thus, the aim of the current study was

to test the roles of shared and distinctive features, as proposed by the CSA, in feature-

to-concept activation. Additionally, there is evidence that more time is necessary to

allow convergence of meanings from multiple primes which in turn leads to increased

priming and semantic facilitation. To determine if time is a factor in activating different

types of features, the current experiment was conducted at two inter-stimulus intervals.

Research Questions and Hypotheses

Table 1-2 summarizes the research questions and respective predictions. Specific

Aim #1: According to the CSA, shared features will boost performance in naming living

nouns due to more numerous connections to strongly correlated shared features.









Furthermore, distinctive features result in slower speech production time due to weaker

correlations with other features. Conversely, for non-living nouns, distinctive features

boost performance over shared features due to the stronger correlations and greater

number of distinctive features; and shared features produce slower speech production

time as a result of weak correlations to other features. The following null hypothesis

was investigated: There is no significant difference in naming living and non-living

nouns as measured by speech reaction time (SRT) when primed with shared or

distinctive features or a combination thereof.

The research question is: Is there a significant difference in naming animals, tools,

or vehicles as measured by SRT when primed with shared or distinctive features or a

combination there of compared to neutral primes? The prediction is: Based on the

CSA, it was predicted in living things (animals and vehicles), that shared features would

result in faster SRTs compared to distinctive features. For non-living nouns (tools),

distinctive features were predicted to produce faster SRTs over shared features. For all

domains of nouns, a combination of shared and distinctive features was predicted to

result in the fastest SRTs.

Specific Aim #2: The CSA proposes different correlation strengths between

shared and distinctive features for living and nonliving things. The strength of these

correlations is largely based on the number of features for each concept, indicating that

the more features, the stronger the correlations between them; and the stronger the

correlations. Thus, activation of features or multiple features with strong correlations

results in faster activation of the concept. The null hypothesis was: There is no









difference in SRT for naming living and non-living nouns between conditions varying

number shared and distinctive features.

This hypothesis was addressed with two research questions: Does SRT for

naming animals, tools, or vehicles change linearly as number of shared features

changes from 1-2? Does SRT for naming animals, tools, or vehicles change linearly as

number of distinctive features changes from 1-2? The prediction is: SRT following

shared features will be faster for living nouns compared to non-living nouns; SRT

following distinctive features will be faster for non-living nouns compared to living nouns.

Specific Aim #3: Using multiple features to prime conceptual activation requires

the consolidation of the information from each prime. This consolidation may occur over

longer time intervals than is required for one prime. The null hypothesis was: There is

no difference in priming effects over time.

The research is: Is there a difference in priming effects over time when comparing

an ISI of 200msec to an ISI of 600msec? The prediction is: Patterns of priming effects

will be different over time. Specifically, an ISI of 200msec would have an additive

influence and an ISI of 600msec would have an overadditive influence.









Table 1-1 The distribution and correlations of shared and distinctive features in living
and nonliving things, according to the CSA.
Living things Non-living things
Shared features High in number Fewer in number


High correlation
(e.g. things which breathe
typically have eyes and ears)

Fewer in number

Weak correlation (e.g. most
things with eyes and ears do
not have pouches, udders, or
eiaht leas)


Weak correlation
(e.g. most tools have a
handle but could be used for
hitting, cutting or turning)
High in number

High correlation
(e.g. things with blade also
cut)


Distinctive features










Table 1-2. Summary of Research Questions (RQ) and Predictions
Specific Aims Feature Type Predictions for Predictions for Nonliving Things
Living Things


#1
RQ:
Is there a
significant
difference in
naming animals,
tools, or vehicles
as measured by
SRT when
primed with
shared or
distinctive
features or a
combination
there of
compared to
neutral primes?
#2
RQ 1: Does SRT
for naming
animals, tools, or
vehicles change
linearly as
number of shared
features changes
from 1-2?
RQ2: Does SRT
for naming
animals, tools, or
vehicles change
linearly as
number of
distinctive
features changes
from 1-2?
#3
RQ:
Is there a
difference in
priming effects
overtime when
comparing an ISI
of 200msec to an
ISI of 600msec?


Neutral,
Shared,
Distinctive, and
Combination
(shared &
distinctive)


One shared
feature

Two shared
features




One distinctive
feature

Two distinctive
features


1. Combination of
shared and distinctive
features will lead to
significantly faster
SRT than shared or
distinctive features
alone

2. Shared features
alone will lead to
significantly faster
SRT than distinctive
features alone


Two shared features
will lead to
significantly faster
SRT





There will be no
significant difference
between one and two
distinctive features in
SRT


1.Combination of shared and distinctive
features will lead to significantly faster SRT
compared to shared or distinctive features
alone

2. Distinctive features alone will lead to
significantly faster SRT than shared
features alone


There will be no significant difference
between one and two shared features in
SRT






Two distinctive features will lead to
significantly faster SRT


Prediction
Patterns of priming effects will be different overtime; namely,
an ISI of 200msec would have an additive influence
and an ISI of 600msec would have an overadditive influence.









CHAPTER 2
METHODS

Data Collection

Design

A linear mixed effects model was employed to measure the effects of prime type

(distinctive, shared, combination) and semantic category (animals, tools, vehicles) in

two conditions (200 and 600 ISI) on speech reaction time during a picture naming task.

The details of the experimental paradigm and data collection sessions are presented

below.

Dependent variables

The dependent variable was SRTs for picture naming. SRTs were measured from

the onset of the picture to the initiation of speech. Practice trials were administered to

familiarize participants with the task and to minimize false starts. False starts and

incorrect naming responses were excluded from analysis.

Independent variables

There were three independent variables. The first was combination of prime types

which consisted of nine prime conditions (described below). The second independent

variable was the semantic categories of target pictures (animals, tools, vehicles). The

third independent variable was ISI (200msec vs. 600msec).

Trial structure

Trials consisted of two orthographic word primes and one picture target. All

picture targets were concepts. Word primes where either semantic features or a neutral

prime. The semantic feature primes were either a shared or distinctive feature (defined

below) related to the target. The neutral prime was the word blank. The combination









and order of feature type and neutral primes were randomized across all trials while

limiting each target to a single appearance per participant. Thus, in one trial a feature

type was presented from 0-2 times. Figure 2-1 shows an example trial structure. The

prime pair conditions were:

Shared-shared
Shared-neutral
Neutral-shared
Shared-distinctive
Distinctive-shared
Distinctive-distinctive
Distinctive-neutral
Neutral-distinctive
Neutral-neutral

Timing

Each participant completed the experiment twice in two separate sessions in the

same week, separated by no less than two days. The purpose of two test sessions was

to compare the priming effects when ISI was manipulated. One administration

presented the items with 200msec ISI and the other administration presented the items

with 600msec ISI. The order of ISI over the two sessions was randomized. ISI was

measured from the offset of a stimulus to the onset of the next stimulus. Figure 2-2

displays the trial structure with time intervals.

Each trial began with a fixation point of a (+) for 500 milliseconds. Two prime

words appeared sequentially, each with duration of 200msec and an ISI of either 200

msec or 600 msec. Then a target picture appeared until the participant named the item

aloud. The inter-trial interval was 3 seconds. The next trial began with a fixation point.

Task

Participants were given the following instructions: "You will see words and

pictures. Just watch the words. When a picture appears, name it as fast as you can."









Participants completed 10 practice items followed by further instructions from the

examiner if needed.

Settings and Equipment

All testing was conducted in a quiet room with participants seated at a comfortable

distance from the computer screen. E-prime software (E-Prime version 2.0, Psychology

Software Tools, Pittsburgh, PA) was used for stimulus presentation and collection of

SRTs. Verbal responses were recorded by a C420 PP MicroMic head-mounted

microphone connected to a Tube MP preamplifier which activated the voice key of a

Serial Response Box interfaced with E-prime software. Reaction times were stored in

by the E-prime software. Verbal responses were also recorded with a Marantz PMD671

digital recorder.

Stimuli

Semantic categories

Target items for this study were animals, tools, and vehicles. Targets can be

found in appendix A. Additionally, a set of distracter items comprised two-thirds of the

total items. These items were from several semantic categories with the exclusion of

the three target categories (animals, tools, vehicles), in order to prevent participants

from anticipating an animal, tool, or vehicle. Prime types for distracter items were the

same as for target items. Distracter items were not analyzed.

Target pictures. Targets were presented as black and white line drawings.

Pictures were collected from: the CRL International Picture-Naming Project (Bates et

al., 2003), edupics.com, and Google Images. Target pictures were presented in the

vertical and horizontal center of the screen.









Prime type

Neutral primes. The word blank was used as the neutral prime to compare to

the effect of semantic feature primes. de Groot, Thomassen, and Hudson (1982) found

using blank as a neutral prime did not result in a significant difference in reaction times

to targets following blank versus meaningful primes.

Semantic feature primes. Semantic features were selected from the corpus of

541 living and nonliving things created by McRae, Cree, Seidenberg, & McNorgan

(2005). This corpus was normed on over 700 neurologically-healthy participants. The

features include nouns and verbs. McRae et al. categorized the types of features

provided by the participants as: functional, visual-motor, visual form and surface, visual

color, sound, taste, smell, tactile, encyclopedic, taxonomic. The data include concept

production frequencies, which are the number of concepts in which each feature occurs.

Semantic features can be found in the appendix listed by target and coded as shared

(S) or distinctive (D).

The definition of shared and distinctive features is based on a modified version of

the McRae et al. (2005) concept production frequency. Concept production frequency

(CPF) is a measure from the McRae et al. database which is defined as the number of

times the feature was produced in the entire corpus of items. Any feature with a CPF

greater than two is considered shared and two or less is considered distinctive.

However, the McRae et al. database calculates the CPF over all items in the database

not items within semantic category. Because the hypotheses of the CSA regarding

shared and distinctive features is based on categories, the CPF was re-calculated for

this study to represent the number of times a feature was produced within semantic









category (animals, tools, vehicles). Shared and distinctive feature primes can be seen

in appendix A.

As seen in appendix A, in order to construct the distinctive-distinctive and shared-

shared prime conditions, each concept has two shared and two distinctive features.

In constructing the shared-distinctive condition, only one pair of four possible pairings

was chosen for each target. The most shared and most distinctive features were

chosen. If the value for shared and distinctive was equal, then the decision was based

on consensus from three outside raters asked to choose the most distinctive/shared

feature of the two.

Stimuli were chosen from the McRae et al. (2005) database based on number of

features and syllable length. Specifically, each target (pictured concepts to be named)

had to have at least two distinctive and two shared features. Targets were controlled for

syllable length, specifically one to four syllables, to minimize variance in speech motor

programming.

Participants

Two groups of neurologically healthy individuals served as participants. Forty-

seven individuals between the ages 50-80 comprised the older group. Fifty-nine

University of Washington undergraduate students between the ages of 18-30 comprised

the younger group. Neurologically healthy participants were chosen because feature-to-

concept priming has not previously been employed in priming studies and the presence

of priming effects should be established in a healthy brain before investigating effects in

a pathologic population.

The older group is the population of interest in this study because this is the typical

age range of stroke survivors and people with aphasia. The younger group was chosen









as a methodologic control group because the majority of previous research in semantic

priming has been conducted on young adults; thus, it is an empirical question if priming

effects from the current paradigm will be found in older adults. Demographic

information can be found in Table 2-1.

Inclusion criteria

Participants were right-handed, monolingual, English speaking adults without a

history of neurologic conditions or disease and/or developmental cognitive disorders as

measured by participant report. Such disorders included, but were not limited to,

dyslexia or alexia, phonologic impairments, memory impairments, language

impairments, and vision impairments (excluding corrected vision).

Screening

Prior to data collection, each participant was administered the following tests in

order to determine eligibility for participation:

Mini-Mental Status Exam (MMSE; Folstein, Folstein & McHugh, 1975) was administered
to determine intact memory based on a minimum score of 27/30. No participants
failed the MMSE.

Visual acuity screening was administered using a Snellen Eye Chart to ensure
participants were able to perceive the stimuli based on minimum criteria of reading
the stimuli for 30 feet from a distance of 20 feet. No participants failed the vision
screening.

The following tests were administered to describe the participants' reading and

semantic abilities:

The Test of Word Reading Efficiency (TOWRE; Torgesen, Wagner, & Rashotte, 1999)
was administered to describe the participants' abilities to read single words. This
test was designed to measure the ability to sound out words and to read words as
a whole unit (sight words).

The Pyramids and Palm Trees Test (P&P; Howard & Patterson, 1992) was used to
describe participants' ability to access semantic information from picture stimuli.









Lastly, the American National Adult Reading Test (ANART; Nelson, 1982) was

used to estimate intelligence quotient (IQ) in order to establish similar intelligence

between the younger and older adults. IQ was calculated using the number of errors on

the ANART and years of education as described by Grober & Sliwinski (1991). Average

test scores can be found in Table 2-1.

Data Analysis

Transcription and Scoring

Digital audio-recordings of participant responses were transcribed and scored as

correct or incorrect. Responses were scored correct when there was a single utterance

of the name of the target picture. Responses were scored incorrect if they included

verbal preparation output such as "um, huh" or self-corrections (ex. screw- no drill).

Data Trimming

Only correct responses were analyzed. This resulted in an exclusion of 8.8% of

the responses (5304 total response, 468 excluded responses). Reaction times less

than 200msec were excluded based on previous studies which have used this cutoff to

minimize the inclusion of false triggering of the voice key (e.g. breathing, lip smacking,

etc.). This resulted in an exclusion of 2.8% of the correct responses (4765 total

responses, 133 responses).

Reliability

Five independent raters performed inter-rater reliability on 25% of each

participant's responses. Raters were trained to score items as correct or incorrect

based on the original criteria explained above. Training took place in a one-hour

session with the primary investigator. Raters demonstrated understanding of the criteria

by completing reliability on a single participant's responses. The reliability was









reviewed by the primary investigator and when determined to be accurately completed

the training was complete.

Statistical Analysis

A linear mixed effects model was employed to measure the effects of prime type

on SRT in each semantic category. The nine prime conditions described above were

collapsed into six prime conditions by coding conditions which were only different by

order of feature primes as the same conditions. This resulted in the following six prime

conditions:

1. Shared-shared
2. Shared-neutral\ Neutral-shared
3. Shared-distinctive\Distinctive-shared
4. Distinctive-distinctive
5. Distinctive-neutral\Neutral-distinctive
6. Neutral-neutral


Distinct analyses were conducted for each group, and within categories (animals,

tools, vehicles). Log-transformed data were used to reduce the effect of outliers. In all

models, participants were included as a random factor and target frequency was

included as a fixed factor. Frequency of targets was included as a fixed factor because

frequency was not controlled for a priori, but the literature provides strong evidence that

frequency interacts with priming. Frequency was entered into the models as a

categorical variable, using criteria commonly seen in the literature (low frequency was

less than ten per million, medium frequency between ten and twenty per million, high

frequency over twenty per million. Frequency ratings were taken from Brysbaert and

New (2009) and are displayed in appendix B. All analyses were conducted with

Bonferroni correction for multiple comparisons.









Table 2-1 Participant demographic information including age, years of education,
gender and average (AVE) test scores with standard deviations (SD) for the
Mini-Mental Status Exam (MMSE; Folstein, Folstein & McHugh, 1975), Test
of Word Reading Efficiency (TOWRE; Torgesen, Wagner, & Rashotte, 1999),
Pyramids and Palm Trees Test (P&P; Howard & Patterson, 1992), American
National Adult Reading Test (ANART; Nelson, 1982), and estimated IQ based
on ANART Scores (Grober & Sliwinski, 1991).
Age Education Gender MMSE TOWER P&P ANART Estimated
Female (raw (standard (raw (raw IQ based
(F) score out score with score out score out on
Male of 30) a range of of 50) of 50) ANART
(M) 35-165) score
Young AVE 22 15 25 F 30 108 49 37 114.93
Adults (SD) (2) (2) 34 M (1) (11.6) (1.79) (6.5) (8.0)

Older AVE 60 16 28 F 29 97 51 40 119.23
Adults (SD) (7) (2) 18 M (1) (18) (1) (8) (7.8)











+ Shared [blank screen] Distinctive [blank screen]
feature feature


Target
Picture


Figure 2-1. Example of priming task trial structure displaying the shared-distinctive
condition














500MSEC +

200MSEC PRIME

200MSEC/600MS

200MSE( PRIME

200MSEC/600MSE E

TARGET



3SEC

+
500MSE [NEW TRIAL]


Figure 2-2. Priming task trial structure for one complete trial displaying timings of each
screen. All trials within one session are administered with an ISI of 200 or
600msec for the blank screens between primes and between the second
prime and target picture. The target picture remains on the screen until the
voice key is triggered by the participant's verbal response.















43









CHAPTER 3
RESULTS

Reliability

Inter-rater reliability was performed on 25% of each participant's responses for

each session. An Intra-class correlation was calculated. Inter-rater reliability was 96%.

Research Question 1

Is there a significant difference in naming animals, tools, or vehicles as measured

by SRT when primed with shared or distinctive features or a combination there of

compared to neutral primes? A linear mixed effects model was conducted for each

category to determine the main effect of prime condition on SRT of picture naming.

Each model included frequency and prime condition as fixed factors and participant as a

random factor. Four prime conditions were entered into the model, shared (shared-

blank, blank-shared, shared-shared), distinctive (distinctive-blank, blank-distinctive,

distinctive-distinctive), combination (distinctive-shared, shared-distinctive), and neutral

(blank-blank). Targets were coded for high, medium, or low frequency. Frequency

effects are presented separately at the end of the results.

Younger Adults

Table 3-1 and Figure 3-1 display mean SRT and standard errors for each prime

condition with each target semantic category (animals, tools, vehicles). The analysis of

SRT of naming animals resulted in a main effect of prime condition F(3, 874.86) = 6.26,

p=.000. Pairwise comparisons indicate distinctive primes resulted in significantly

(p<.05) faster responses than the neutral primes and faster than the shared primes

(p=.001). The main effect of prime condition on naming tools was not significant F(3,









861.85) = 2.34, p>.05. The main effect of prime condition on naming vehicles was not

significant F(3, 745.81) <1.

Older Adults

Table 3-2 and Figure 3-2 show mean SRT and standard error for each prime

condition within the target semantic category (animals, tools, vehicles). The analysis of

SRT of naming animals resulted in a main effect of prime condition that was significant,

F(3, 674.78) = 5.16, p<.05, Pairwise comparisons revealed that reaction times to

distinctive features were significantly faster (p=.001) than reaction times to shared

features. The main effect of prime condition on naming tools was significant F(3,

675.82) = 4.78, p<.05. Pairwise comparisons indicate reaction times following distinctive

features were significantly faster (p<.05) than responses to neutral primes, shared

feature primes, and combination primes (distinctive-shared or shared-distinctive). The

main effect of prime condition on naming vehicles was significant F(3, 598.88) = 5.47,

p<.01. Pairwise comparisons revealed that reaction times following distinctive features

were significantly (p=.001) faster than reaction times following shared features.

Results for research question one indicate similar priming patterns for the younger

and older adults. Specifically, distinctive features have greater priming effects than

shared thus, validating the use of this picture naming priming task with older adults.

Given that older adults are the population of interest, only results from the older adult

group will be presented and discussed for the remaining research questions.

Research Question 2

Does SRT for naming animals, tools, or vehicles change linearly as number of

shared features changes from 1-2? A linear mixed effects model was conducted for

each category to determine the main effect of prime condition on SRT of picture









naming. Each model included frequency and prime condition as fixed factors and

participant as a random factor. Two prime conditions were entered into the model,

shared (shared-blank, blank-shared) and shared-shared to compare the effects of the

number of shared features.

Table 3-3 and Figure 3-3 display the mean SRT and standard error for the two

prime conditions within each semantic category. The analysis of SRT of naming

animals did not produce a significant main effect of prime condition, F(1, 211.38) <1.

Likewise, the main effect of prime condition was not significant for SRT of naming tools,

F(1, 220.24) <1 or vehicles, F(1, 183.80) <1.

Research Question 3

Does SRT for naming animals, tools, or vehicles change linearly as number of

distinctive features changes from 1-2? A linear mixed effects model was conducted for

each category to determine the main effect of prime condition on SRT of picture

naming. Each model included frequency and prime condition as fixed factors and

participant as a random factor. Two prime conditions were entered into the model,

distinctive (shared-blank, blank-shared) and distinctive-distinctive to compare the effects

of the number of distinctive features.

Table 3-4 and Figure 3-4 display the mean and standard error of SRT for the two

prime conditions within each target semantic category. The main effect of prime

condition was not significant for SRT of animal picture naming, F(1, 235.59) <1.

Similarily, the main effect of prime condition was not significant for tools, F(1, 226.62) <

1, or vehicles, F(1, 205.32) <1.









Research Question 4

Is there a difference in priming effects between 200 and 600msec ISI? The

comparisons of priming conditions in research questions 1-3 were conducted with ISI as

a fixed factor and participant as a fandom factor to determine the main effect of ISI on

SRT of picture naming. Tables 3-5 through 3-7 and figures 3-5 through 3-7 display

means SRTs with standard errors for comparisons of ISI.

Comparison of shared, distinctive, combination, and neutral primes did not

produce significant effects of ISI for animals, F(1, 663.65) = 1.704, p=.192, tools F(1,

663.60)<1, or vehicles, F(1, 586.59) = 1.74, p=.188. Comparisons of shared and

shared-shared conditions produced a significant main effect of ISI for animals, F(1,

197.55) = 4.89, p<.05. Pairwise comparisons revealed that reactions times in the 600

ISI condition were significantly (p<.05) faster than reaction times in the 200 ISI

condition. Responses to tools did not produce a significant main effect of ISI, F(1,

208.60) = 1.44, p=.23, neither did response to vehicles, F(1, 179.82)<1. Likewise,

comparisons of distinctive and distinctive-distinctive prime conditions did not result in

significant main effects of ISI for animals, F(1, 228.9) = 1.43, p=.23; tools, F(1,

222.70)<1; or vehicles, F(1, 210.33) = 2.95, p=.09.

Word Frequency Effects

When comparing SRTs from prime conditions of shared, distinctive, combined and

neutral primes, only responses to animal targets revealed significant effects of

frequency, F(2, 665.90) = 6.43, p<.05. Pairwise comparisons showed that reaction

times to high frequency animals were faster than reaction times to medium frequency

animals. Frequency did not have a significant effect on SRT for picture naming of tools

F(1, 666.76) <1, or vehicles F(2, 589.10) = 2.09, p=.12.









Comparison of SRTs of picture naming when primes were one shared feature and

two shared features revealed no significant effects of frequency. The results from

neither animals, F(2, 194.88) = 2.87, p=.059, tools, F(1, 205.54) = 1.05, p=.307, nor

vehicles, F(2, 179.22) <1 was significant. Analysis of frequency effects on SRT's

following primes of one distinctive and two distinctive features, resulted in significant

effects from only animal targets, F(2, 230.88) = 4.18, p<.05. Pairwise comparisons

indicate reaction times to high frequency animals were significantly (p<.05) faster than

reaction times to medium frequency animals. Frequency effects from neither tools, F(1,

220.04) <1 nor vehicles, F(2, 204.27) = 1.24, p=.293 was significant.









Table 3-1. Raw mean SRT for younger adults for comparison of neutral, shared
features, distinctive features, and combination of shared and distinctive
features prime conditions. Asterisks indicate significant pairwise comparisons
which are described in parentheses.
Category Prime Condition Raw Mean SRT Standard Error
Animals Neutral 677.83 235.70

Shared 695.18 541.28

Distinctive* 637.81 308.88

(Faster than neutral primes, p<.05

Faster than shared primes, p=.001)

Distinctive & Shared 665.34 331.52

Tools Neutral 775.27 424.63

Shared 783.72 483.57

Distinctive 767.27 466.62

Distinctive & Shared 768.28 354.28

Vehicles Neutral 722.76 245.66

Shared 687.80 259.64

Distinctive 679.26 365.92

Distinctive & Shared 715.57 304.34









Table 3-2. Raw mean SRT for older adults from comparison of neutral, shared features,
distinctive features, and combination of shared and distinctive features prime
conditions. Asterisks indicate significant pairwise comparisons which are
described in parentheses.
Category Prime Condition Raw Mean SRT Standard Error
Animals Neutral 661.38 207.21

Shared 695.52 272.11

Distinctive* 634.34 303.19

(Faster than shared features, p=.001)

Distinctive & Shared 647.25 243.64

Tools Neutral 769.70 303.69

Shared 706.65 276.82

Distinctive* 665.89 254.50

(Faster than neutral primes, p<.05

Faster than shared features, p<.05

Faster than combined features, p<.05)

Distinctive & Shared 771.42 360.65

Vehicles Neutral 701.04 236.43

Shared 788.50 450.97

Distinctive* 656.21 307.39

(Faster than shared primes, p=.001)

Distinctive & Shared 656.18 227.95









Table 3-3. Raw mean SRT from comparison of shared versus shared-shared prime
conditions. There were no significant effects of prime condition.
Category Prime Condition Raw Mean SRT Standard Error


Animals Shared

Shared & Shared


Tools


Shared


Shared & Shared

Vehicles Shared

Shared & Shared


696.30

694.25

713.52

690.18

774.75

810.19


270.81

275.83

304.50

196.16

412.43

507.90









Table 3-4. Raw mean SRT from comparison of distinctive and distinctive-distinctive
prime conditions. There were no significant effects of prime condition.
Category Prime Condition Raw Mean SRT Standard Error

Animals Distinctive 625.89 289.44

Distinctive & Distinctive 651.35 330.27

Tools Distinctive 669.67 248.42

Distinctive & Distinctive 658.38 267.60

Vehicles Distinctive 671.48 325.08

Distinctive & Distinctive 630.53 275.01









Table 3-5. Raw mean SRT from comparison of 200msec and 600msec ISI for the prime
conditions of neutral, shared features, distinctive features, and combination of
shared and distinctive features. There were no significant effects of ISI.
Category ISI Raw Mean SRT Standard Error

Animals 200 679.62 294.65

600 638.87 246.72

Tools 200 713.23 281.64

600 715.41 311.25

Vehicles 200 711.60 330.48

600 695.20 360.63









Table 3-6. Raw mean SRT from comparison of 200msec and 600msec ISI for the
shared feature prime conditions. Asterisks indicate significant pairwise comparisons
which are described in parentheses.
Category ISI Raw Mean SRT Standard Error


Animals 200


740.33

653.85


600*


303.57

232.88


(Faster than 2001SI, p<.05)


Tools


200

600


702.39

711.18

799.11

777.35


Vehicles 200


600


249.74

304.03

464.01

438.95









Table 3-7. Raw mean SRT from comparison of 200msec and 600msec ISI for the
distinctive feature prime conditions. There were no significant effects of ISI.
Category ISI Raw Mean SRT Standard Error

Animals 200 657.65 325.99

600 609.32 275.77

Tools 200 654.86 210.44

600 673.55 281.56

Vehicles 200 667.80 240.45

600 642.39 372.62









3.2
3.15
3.1
3.05
3
2.95
2.9
n or -i--


.OJ0
2.8
2.75
2.7
2.65
2.6
2.55
2.5


Animals

Animals


V c 06
Vehicles


5 506
Tools


Figure 3-1. Log-transformed mean SRT of young adult responses from comparison of
neutral, shared, distinctive and combined prime conditions. Brackets indicate
significant pairwise comparisons. Asterisks indicate the prime condition
which produced significantly faster SRTs.


-- T TI-


i~












3.2
3.15
3.1
3.05

2.95
2.9
2.85
2.8
2.75 -
2.7 -
2.65 -
2.655
2.55
2.5


.> ._1)


L5 506
Animals


U)
5 506
Tools


SCO
a
to,
i506


Vehicles


Figure 3-2. Log-transformed mean SRT of older adult responses from comparison of
neutral, shared, distinctive and combined prime conditions. Brackets indicate significant
pairwise comparisons. Asterisks indicate the prime condition which produced
significantly faster SRTs.










3.2
3.15
3.1
3.05
3
2.95
2.9
2.85
2.8
2.75
2.7
2.65
2.6
2.55
2.5


Shared Shared &
Shared


Tools


Shared Shared &
Shared


Vehicles


Figure 3-3. Log-transformed mean SRT of older adult responses from comparison of
shared and shared-shared prime conditions. There were no significant effects of prime
condition.


Shared Shared &
Shared
Animals


I I .TT =












I _____ __


t I I


3.2
3.15
3.1
3.05
3
2.95
2.9
2.85
2.8
2.75
2.7
2.65
2.6
2.55
2.5


Distinctive


Animals


Distinctive Distinctive


Distinctive


Distinctive Distinctive
&
Distinctive


Tools


Vehicles


Figure 3-4. Log-transformed mean SRT of older adult responses from comparison of
distinctive and distinctive-distinctive prime conditions. There were no
significant effects of prime condition.


Distinctive Distinctive


-
-
-
-
-
-
-











3.2
3.15
3.1
3.05
3
2.95
2.9
2.85
2.8
2.75
2.7
2.65
2.6
2.55
2.5


Animals


1 2


200 ISI


600 ISI


Tools


200 ISI 600 ISI


Vehicles


Figure 3-5. Log-transformed mean SRT of older adult responses to all prime conditions
(collapsed across shared, distinctive, combined, and neutral primes) at 200
ISI versus 600 ISI. There was no significant difference in SRT between ISI
conditions.


200 ISI 600 IS


T T














3.2
3.15
3.1
3.05
3
2.95
2.9
2.85
2.8
2.75
2.7
2.65
2.6
2.55
2.5


TI T


TI I


200 ISI


Animals


600 ISI


200 ISI 600 ISI


Tools


Vehicles


Figure 3-6. Log-transformed mean SRT of older adult responses to shared primes
(collapsed across shared and shared-shared) primes at 200 ISI versus 600 ISI.
Brackets indicate significant pairwise comparisons. Asterisks indicate the prime
condition which produced significantly faster SRTs.












































61


I


-


-
-
-
-
-
-
-
-
-
-
-


I










3.2
3.2
3.1
3.1
3.0
3.0
2.9
2.9
2.8 -
2.8
2.7 -
2.7
2.6 -
2.6
2.5


200 ISI 600 ISI


Tools


Vehicles


Figure 3-7. Log-transformed mean SRT of older adult responses to distinctive primes
(collapsed across distinctive and distinctive-distinctive) primes at 200 ISI
versus 600 ISI. There was no significant difference in SRT between ISI
conditions.


200 ISI 600 IS


T T 1=


Animals


_L









CHAPTER 4
DISCUSSION

The purpose of this study was to investigate the roles of shared and distinctive

features, as proposed by the CSA, in conceptual activation. The principles of the CSA

could be used to further specify the stimuli used in anomia treatments however, it is

unknown if shared and distinctive features can be used to activate concepts because

previous studies have used concept-to-concept activation and not feature-to-concept

activation, the route used in anomia treatments. Additionally, there is evidence that

more time is necessary to allow convergence of meanings from multiple primes which in

turn leads to increased priming and semantic facilitation. Thus, the current experiment

was conducted at two inter-stimulus intervals.

Overall, the results showed that indeed, as stated in the CSA, distinctive features

have a privileged role in concept activation. However, the proposed differential roles of

shared and distinctive features in living and nonliving things, was not confirmed.

Additionally, the results indicate that more time was not necessary to consolidate the

information from two primes. The results are discussed below with respect to

theoretical and clinical implications.

Priming Effects from Related Features and Neutral Primes

SRT was compared in priming conditions of related features (shared, distinctive, or

combined) and neutral primes (blank) to establish the validity of feature-to-concept

priming. Only the tools category revealed a significant difference in SRTs when primes

were related compared to SRT's when the prime was the word blank. This is most likely

due to the difference in frequency between tools and; animals and vehicles. As seen in

appendix B, animals and vehicles had greater average frequencies (21.39 and 86.16,









respectively) than tools (5.96). Consequently, participants' greater exposure to animals

and vehicles left little need for feature primes resulting in non-significant differences in

SRT between feature primes and neutral primes. On the other hand, participants'

reduced exposure to tools allowed the feature primes to boost concept activation. Most

importantly, there were significant differences in the priming effects of the feature types

for all semantic categories.

The Effect of Feature Type

The first analysis investigated the hypothesized difference between the roles of

shared and distinctive features, in the representation of living and nonliving things, by

measuring the difference in SRT for picture naming following primes of shared or

distinctive features or a combination thereof. Based on the CSA (Tyler & Moss, 2001;

Taylor et al., 2007), it was predicted that in living nouns (animals and vehicles), priming

with shared features would result in shorter SRTs than priming with distinctive features,

due to greater spread of activation from strongly correlated shared features and weaker

correlation of distinctive features to other features. In the activation of non-living nouns

(tools), distinctive features were predicted to boost performance over shared features

due to the stronger correlations and greater number of distinctive features than shared

features. For all domains of nouns, a combination of shared and distinctive features

was predicted to be the optimal condition for performance because the semantic

category is activated as well as a feature unique to that concept which will lead to

identification of the concept from its category neighbors. Table 1-1 displays these

predictions.

Overall, the predictions of the current study were not confirmed. Analysis of

reaction times of animal, tools, and vehicle picture naming revealed that distinctive









feature primes consistently had a greater effect compared to shared feature or

combination of shared and distinctive feature primes. Although, the CSA proposes that

distinctive features play a critical role in concept activation, the level of information

about a concept carried by distinctive features is proposed to be different in living and

nonliving things. The CSA claims that because of weaker correlations between

distinctive features and other features, distinctive features are less informative of living

things than shared features with stronger correlations to other features; conversely,

distinctive features of nonliving things are more informative and have stronger

correlations to other features than shared features of nonliving things. Consequently, it

was not surprising that in activation of tools, distinctive features produced the fastest

SRTs. Although, the same result in living things was surprising it is not unique to this

study. Cree, McNorgan and McRae (2006) reported identical results. They used two

feature verification tasks, one with concept primes and feature targets and importantly

one with feature primes and concept targets (as in the current study). Distinctive

features consistently led to faster feature verification regardless of the conceptual

classification as living or nonliving. Additionally, Cree et al. conducted a computer

simulation similar to a feature verification task. The simulation showed that distinctive

features activated the target concepts more strongly than did shared features. The

findings of Cree et al. and the current study may be attributed to the high level of

distinctiveness used in both studies.

The distinctive features used in this study, and those used by Cree et al. (2006)

were not only distinctive but were unique to a single concept, thus exposure to this

information alone (without a shared feature) could activate one concept only. For









example, the distinctive features of quacks, cocoon, or udders activate only the

concepts of duck, butterfly, or cow, respectively. Likewise, the distinctive features of

vehicles used in this study were also unique to each concept (e.g. hovers-helicopter,

mast-sailboat, caboose-train) resulting in only one possible concept. Thus, it appears

that when a feature is unique to a concept and not ambiguous, it will activate the

concept without need of activating the category explicitly, through shared features.

Furthermore, some of the shared features used in this study had very high concept

production frequencies. As described in the methods, concept production frequency is

the count of how many times the feature was listed in the norms by McRae et al. (2005).

Shared features with high concept production frequencies may have provided minimal

information on category membership. For example large was seen for animals as well

as vehicles, consequently, large supplied minimal information regarding the target. In

comparison, metal, a shared feature for tools, at the least indicated the target was a

nonliving thing, thereby activating fewer semantic neighbors than large which could

refer to a living or nonliving thing.

This raises an important issue which should be addressed in future studies

investigating the CSA. It appears that the proposed distinction, between the roles of

shared and distinctive features in living and nonliving things, is relative to the level of

distinctiveness of the feature. A highly shared feature does little to activate the

semantic category or neighborhood, of a living or nonliving thing. On the other hand, a

highly distinctive feature, of a living or nonliving thing, can activate the concept directly;

or, has stronger connections than proposed, to activate related features and lead to









activation of the concept. Further specifying the effects of different levels of shared and

distinctive features is critical to the applicability of the CSA.

The most surprising finding was the lack of a significant difference in reaction

times following, combined shared and distinctive feature primes, compared with

distinctive feature primes alone, or shared feature primes. It was hypothesized that a

combination of shared and distinctive features would result in the fastest SRT because

the participant would have information on category membership and information unique

to the target concept prior to seeing the target. The results however, suggest that

shared features did little to activate the concept, whether presented alone or in

combination with a distinctive feature.

One explanation is that, shared features activated large sets of semantic

neighbors which could not be inhibited by the activation of the distinctive feature. In

recent work, Mirman and Magnuson (2009) explain this effect at the featural level using

the mechanisms of excitation and inhibition, as presented in the semantic neighborhood

model. In the process of activating lion, shared features which also belong to tiger are

activated (e.g. predator, teeth, wilderness, and circus), thus, tiger is partially activated.

The activation of distinctive features of lion (mane) leads to inhibition of tigerwhile

adding to the excitation of lion. In regards to the features of the current study, the

shared feature prime of four legs would activate most animals to some degree. If the

distinctive feature of humps has weak correlations to other features of camel, as

suggested by the CSA, it may not provide enough additional activation to signal the

system to inhibit all other four-legged animals while further exciting camel. Thus, the









distinctive feature of hump, seen without a shared feature, was just as effective in

activating camel.

The Effect of Number of Features

The second analysis investigated the effects of multiple shared features on SRT

of picture naming to determine if number of features had an effect on concept activation.

The findings for animals, tools, and vehicles were identical; there was no difference in

priming effects from one or two shared features, or from one or two distinctive features.

For living things (animals and vehicles), it was predicted that multiple shared

features would increase priming because these features have strong connections with

other features; as such, direct activation of several features should result in activation of

several other features. However, it appears that this activation did not reach a

measurable level. The prediction for nonliving things (tools) was that multiple shared

features would not affect priming because these features have weak connections to

other features. Results from analyses of research question one showed that shared

features with or without a distinctive feature did not significantly prime picture naming for

any category. Therefore, it is not surprising that increasing the number of shared

features shows no difference. If four-legged activated animals the addition of fur does

little to focus the activation within the semantic neighborhood of animals given so many

four-legged animals have fur. The CSA model states that shared features reflect

category membership only; these results suggest that multiple references to the

category do not increase activation of that category.

In regards to the effects of multiple distinctive features, it was predicted that for

activation of tools, increasing the number of distinctive features would increase priming

because these features have strong connections to other features; thus, activating









several features would result in activation of several other features prior to viewing the

target. On the other hand, for animals and vehicles it was predicted that increasing the

number of distinctive features would not increase priming because these features have

weak connections to other features. In the present paradigm, there was no significant

difference in SRT when the primes were one or two distinctive features for animals,

tools, or vehicles. Given the findings in research question one analyses that activation

from distinctive features was greater than from shared features or a combination of

shared and distinctive, it appears that multiple distinctive features are not necessary to

activate a concept in any category. If the distinctive feature is unique to the concept,

and therefore, will only activate one concept, additional distinctive features are not

necessary to boost activation. The null effects of multiple features have interesting

implications for anomia treatments which will be address in clinical implications below.

The Effect of Timing on Multiple Feature Primes

The comparison of priming effects at 200 and 600 ISI was motivated by the work

of Milberg and colleagues (2003) who found overadditive priming effects at 600 ISI and

addivitve effecs at 200 ISI. Additive effects indicate the priming effect from two

sequential primes is the same as adding the priming effect of each prime when

presented individually; thus, there would be no significant difference in SRTs from

conditions with one or two primes. Overadditive effects suggest the priming from two

sequential primes is greater than adding the priming effects of each prime presented

individually; therefore, there would be a significant difference in SRT from conditions

with one or two primes. Additionally, multiple primes can result in underadditive effects

which occur when the facilitation of multiple primes is less than the total facilitation of

each prime individually. The current findings suggest multiple feature primes produce









underadditive effects, i.e. there is no benefit to presenting multiple feature primes.

There is one exception to this finding. There were overadditive effects for naming

animals when primed with only shared features. The implications of these results are

discussed below.

In all prime conditions, except shared conditions, there was no difference in SRT

when ISI was increased from 200msec to 600msec. This result is particularly

interesting for the conditions containing a distinctive feature because in earlier analyses

where ISI was collapsed, distinctive features produced the strongest priming effects.

Consequently, it would seem plausible that this effect would increase over time.

However, it appears that distinctive features do not need more time to activate the

concept. As discussed earlier, this may be attributed to the high level of distinctiveness

of the distinctive feature primes. If the distinctive primes of periscope and underwater

can only activate submarine, then more time is not needed; i.e., there are not several

concepts which could emerge from the activation patterns of periscope and underwater

which would require time to reach a stable state. Consequently, the priming effects from

such highly distinctive multiple primes do not change over time.

The one significant finding in analysis of ISI was in naming animals when primed

with shared features; SRTs from the 600 ISI condition were faster than in the 200 ISI

condition. Given that earlier analyses suggested the effects of shared features of any

semantic category are not measurable in this task, this finding is surprising. The lack of

priming effect from shared features in earlier analyses was partially attributed to the

highly shared features used in this study (e.g. large or metal which both belong to many

concepts). Again, the high shared value of the features may be implicated in the null









effects of ISI. Highly shared features (e.g. legs or large) may require more time to

consolidate with other highly shared features (e.g. furor feathers) because of the large

semantic neighborhood each feature activates. When each feature is activated, related

features are partially activated. Highly shared features will have many related features;

consequently, a large semantic neighborhood is partially activated. Much of this

neighborhood must be inhibited in the process of activating the target. Thus, with more

time (600msec versus 200msec) the system is able to inhibit the features unrelated to

the target and increase activation of the overlapping units from the shared primes.

The Effect of Word Frequency

In general, there was minimal influence of frequency on reaction times. Only

SRTs to animal targets were affected by frequency, in the expected direction of faster

reaction times following high frequency items. This effect was seen in analyses

comparing the prime conditions of shared, distinctive, combined, and neutral; as well as

analysis comparing shared and shared-shared prime conditions. The lack of frequency

effect for vehicles and tools compared to animals is likely due to differences in category

size. There are many more animals in semantic memory than vehicles and tools. Word

frequency may play a larger role in semantic activation when there is larger semantic

neighborhood and therefore, more competitors. However, there were no significant

interactions between frequency and prime condition; therefore the significant priming

effects are valid regardless of frequency differences.

Limitations

The comparison of only one and two features, may have limited the effects of

combining feature types. A difference in priming may be seen with a greater number of

features. Specifically, the effects of multiple shared features may be measurable when









comparing one shared feature to three or five shared features. Likewise the

comparisons of one feature type to combined features are limited by having used only

combinations of one shared and one distinctive. Comparing several shared and

distinctive features, and even varying the number of each paired together (e.g. 1

shared, distinctive; 2 shared, 3 distinctive), could provide further insights into the level of

information carried by each feature type.

With regards to the effects of multiple primes over time, the findings are limited to

the use of 200msec and 600msec. As mentioned earlier, the aim of neither the Milberg

et. al (2003) study nor the present work was the delineation of automatic and controlled

processes. Certainly, the mechanisms of feature-to-concept activation, and the roles of

shared and distinctive features would be further clarified by comparing the results from

priming paradigms administered under controlled and automatic processes. However,

the goal of the current study was to establish the validity of feature-to-concept activation

and test the role of shared and distinctive features in such a paradigm.

Additionally, the current study did not employ a timed response. Randall et al.

(2004) found slower activation from distinctive features of living things compared to

distinctive features of nonliving things only when there was a response deadline. In an

untimed version of the task there was no difference. Consequently, the differential roles

of shared and distinctive features in living and nonliving things may be time dependent.

The interaction of time and feature-to-concept activation is important for the clinical

application of the CSA; however, more empirical evidence of an interaction is required.

Implications for Anomia Treatment

Two issues addressed in the current study have implications for anomia treatment:

the use of multiple features for activation of a concept and the effects of shared and









distinctive features on concept activation. In regards to the former, the current study

would suggest that asking a patient to name several features of a target concept or

providing the patient with multiple features of a target is not the optimal condition for

concept activation. In fact, the work of Mirman and Magnuson (2008) would further

suggest that in the case of shared features, especially, multiple primes will lead to

semantic competition. Consequently, semantic feature treatments may be improved in

terms of treatment outcomes, if the number of semantic questions per target item were

reduced.

The second issue regarding clinical treatments which the current findings can

address is the feature type used in treatment. Clearly the current findings suggest that

distinctive features have the greatest effects on feature-to-concept activation. The

findings by Mirman and Magnuson (2008) regarding inhibitory effects of near neighbors

and facilitative effects of distant neighbors, as well as the CATE approach (Thompson

et. al, 2003), similarly suggest that distinctive features are more beneficial for feature-to-

concept activation. Certainly, semantic feature treatments do include activation of

distinctive features; however, a more explicit focus on these features may improve

treatment outcomes.

Future Directions

The Conceptual Structure Account (Tyler & Moss, 2001; Taylor et al., 2007) is

currently modeled in healthy adults and individuals demonstrating category- specific

semantic deficits resulting from herpes-simplex encephalitis or other degenerative

disease. Extending this theory to aphasia may add specificity to the claims of the CSA

regarding the structure of the semantic system. Specifically, a comparison of

individuals with aphasia and individuals with semantic dementia on a paradigm similar









to the current experiment may be useful in specifying effects of feature correlation on

activation. Persons with a degrading semantic system may not be able to use feature

correlations as well as someone with lexical-semantic access impairment. In addition,

several question related to methodology remain. Answering these questions will further

specify the claims of the CSA:

Are the results of this study different when there is a response deadline?

How do the roles of shared and distinctive features compare when used in automatic
versus controlled processing?

Does order of prime type (shared-distinctive, distinctive-shared) affect SRT in a picture
naming paradigm?

Do other psycholinguistic variables (number of phonemes, phonotactic probability,
familiarity etc.) of the target items affect the difference in SRT between prime
conditions?

Does manipulation of the concept production frequencies (i.e. manipulating the degree
to which features are distinctive or shared) change the effect of the prime
conditions on SRT?

Summary

In a series of analyses of speech reaction times following multiple feature primes

during a picture naming task, the most consistent finding was greater priming after

distinctive feature primes as compared to shared primes, distinctive and shared primes,

or neutral primes. One of the most central hypotheses of the Conceptual Structure

Account (Tyler & Moss, 2001; Taylor et al., 2007) is the unique and critical role of

distinctive features in semantic space and conceptual activation. The current study

supports this claim and validates it in verbal production, a language behavior not

previously used in investigations of the CSA. The lack of support for the postulates of

the CSA regarding differences in the distribution and correlations of shared and

distinctive features in living and nonliving things, suggests further research is required to









elucidate these relationships. The unique contribution of the current work is the strong

evidence that features prime concepts; suggesting, features can be used to activate the

semantic network and indeed different feature types activate the network to varying

degrees.










APPENDIX A
PRIME-TARGET STIMULI

Table A-1. Animal Stimuli
Concept Feature Feature Type CPF
butterfly pollinates D 1
cocoon D 1
flies S 46
small S 121
camel spits D 1
two humps D 1
legs S 44
large S 106
chicken clucks D 1
pecks D 1
wings S 44
edible S 78
cow udder D 1
produces milk D 2
four legs S 49
large S 106
duck quacks D 1
waddles D 2
feathers S 38
edible S 78
elephant trunk D 1
tusks D 2
four legs S 49
large S 106
ostrich large eggs D 1
buries head D 1
wings S 44
large S 106
pig squeals D 1
curly tail D 1
four legs S 49
edible S 78
tiger roars D 2
large teeth D 2
ferocious D 2
fur S 22
large S 106
D indicates distinctive features, S indicates shared features. CPF is the concept production
frequency.










Table A-2. Tool Stimuli
Concept Feature Feature Type CPF
airplane crashes D 1
engine D 2
large S 106
metal S 133
bus many seats D 1
fare D 1
transportation S 33
large S 106
car four doors D 1
steering wheel D 1
wheels S 23
transportation S 33
motorcycle two people D 2
helmet D 2
fast S 33
loud S 34
canoe paddles D 1
tips over D 1
wood S 79
long S 81
helicopter hovers D 1
propellers D 2
loud S 34
metal S 133
sailboat mast D 1
wind D 2
floats S 7
transportation S 33
skateboard board D 1
tricks D 1
wood S 79
long S 81
submarine periscope D 1
underwater D 1
large S 106
metal S 133
train caboose D 1
conductor D 1
fast S 33
transportation S 33
unicycle balance D 1
one wheel D 2
seat S 7
transportation S 33
D indicates distinctive features, S indicates shared features. CPF is the concept production
frequency.










Table A-3. Vehicle Stimuli
Concept Feature Feature Type CPF
axe metal blade D 2
chopping D 2
heavy S 27
handle S 42
drill bits D 1
makes holes D 1
electrical S 22
loud S 34
hammer hits nails D 1
metal head D 2
heavy S 27
handle S 42
hoe tilling soil D 1
metal bland D 2
handle S 42
long S 81
rake gardening D 1
prongs D 2
long S 81
metal S 133
scissors blades D 1
cuts D 1
sharp S 18
metal S 133
screwdriver tightens D 1
turns D 2
handle S 42
long S 81
shovel digging D 1
scooping D 1
handle S 42
metal S 133
wrench tightens bolts D 1
turns bolts D 2
heavy S 27
metal S 133
D indicates distinctive features, S indicates shared features. CPF is the concept production
frequency.









APPENDIX B
TARGET WORD FREQUENCY


Table B-1. Taret Word Frequency
Category Target Brysbaert and New (2009) Frequency Frequency
(per million words) Category
Animals


Average
Tools


Average
Vehicles


Average


Butterfly
Camel
Chicken
Cow
Duck
Elephant
Ostrich
Pig
Tiger


Axe
Drill
Hammer
Hoe
Rake
Scissors
Screwdriver
Shovel
Wrench

Bus
Canoe
Car
Helicopter
Motorcycle
Sailboat
Submarine
Train


5.51
5.02
61.73
25.51
24.76
11.37
0.94
39.14
18.53
21.39

4.88
13.75
12.47
0.92
2.98
6.69
0.06
6.84
3.96
5.96
74.18
3.57
483.06
15.80
8.92
1.61
7.10
95.06
86.16


Low
Low
High
High
High
Medium
Low
High
Medium


Low
Medium
Medium
Low
Low
Low
Low
Low
Low

High
Low
High
Medium
Low
Low
Low
High









LIST OF REFERENCES


Balota, D. A., & Paul, S.T. (1996). Summation of activation: Evidence from multiple
primes that converge and diverge within semantic memory. Journal of
Experimental Psychology Learning, Memory, Cognition, 22(4), 827-845.

Bates, E., D'Amico, S., Jacobsen, T., Sz6kely, A., Andonova, E., Devescovi, A., Herron,
D., Lu, C-C., Pechmann, T., Pl6h, C., Wicha, N., Federmeier, K., Gerdjikova, I.,
Gutierrez, G., Hung, D., Hsu, J., lyer, G., Kohnert, K., Mehotcheva, T., Orozco-
Figueroa, A., Tzeng, A., & Tzeng, 0. (2003). Timed picture naming in seven
languages. Psychonomic Bulletin & Review, 10(2), 344-380.

Boyle, M., & Coehlo, C.A. (1995). Application of semantic features analysis as a
treatment for aphasic dysnomia. American Journal of Speech-Language
Pathology, 4(4), 94-98.

Brysbaert, M., & New, B. (2009). Moving beyond Kucera and Francis: A critical
evaluation of current word frequency norms and the introduction of a new and
improved word frequency measure for American English. Behavior Research
Methods, 41(4), 977-990.

Bueno, S., & Frenck-Mestre, C. (2008). The activation of semantic memory: Effects of
prime exposure, prime-target relationship, and task demands. Memory &
Cognition, 36(4), 882-898.

Cree, G. McNorgan, C., & McRae, K. (2006). Distinctive features hold a privileged
status in the computation of word meaning: Implications for theories of semantic
memory. Journal of Experimental Psychology, 32(4), 643-658.

Cree, G., McRae, K., & McNorgan, C. (1999). An attractor model of lexical conceptual
processing: Simulating semantic priming. Cognitive Science: A Multidisciplinary
Journal, 23(3), 371-414.

de Groot, A. M. (1984). Primed lexical decision: Combined effects of the proportion of
related prime-target pairs and the stimulus-onset asynchrony of prime and target.
The Quarterly Journal of Experimental Psychology A: Human Experimental
Psychology, 36(2): 253-280.

de Groot, A.M., Thomassen, A.J., & Hudson, P.T. (1982). Associative facilitation of word
recognition as measured from a neutral prime. Memory & Cognition, 10: 358-370.

Folstein M.F., Folstein S.E., & McHugh P.R. (1975). Mini-mental state: A practical
method for grading the cognitive state of patients for the clinician. Journal of
psychiatric research, 12(3): 189-98.

Grober, E., & Sliwinski, M. (1991). Development and validation of a model for
estimating premorbid verbal intelligence in the elderly. Journal of Clinical and
Experimental Neuropsychology, 13, 933-949.









Hampton, J. A. (1979). Polymorphous concepts in semantic memory. Journal of Verbal
Learning & Verbal Behavior. 18(4), 441-461.

Howard D, & Patterson K. (1992). Pyramids and palm trees: A test of semantic access
from pictures and words. Bury St Edmunds, Suffolk: Thames Valley Test
Company.

Hutchison, K. A. (2003). Is semantic priming due to association strength or feature
overlap? A microanalytic review. Psychonomic Bulletin & Review. 10(4), 785-813.

Kiran, S., & Thompson, C.K. (2003). The role of semantic complexity in treatment of
naming deficits: Training semantic categories in fluent aphasia by controlling
exemplar typicality. Journal of Speech, Language, and Hearing Research, 46,
773-787.

Larochelle, S., & Pineau, H.I. (1994). Determinants of response times in the semantic
verification task. Journal of Memory and Language. 33(6), 796-823.

McNamara, T. (2005). Semantic priming: Perspectives from memory and word
recognition. Psychology Press.

McRae, K., Cree, G.S., Seidenberg, M.S., and McNorgan, C. (2005). Semantic feature
production norms for a large set of living and nonliving things. Behavior Research
Methods, Instruments, & Computers, 37, 547-559.

McRae, K., Cree, G.S., Westmacott, R., & de Sa, V.R. (1999). Further evidence for
feature correlations in semantic memory. Canadian Journal of Experimental
Psychology, 53, 360-373.

McRae, K., de Sa, V.R., and Seidenberg, M.S. (1997). On the nature and scope of
featural representations of word meaning. Journal of Experimental Psychology:
Genera, 126, 99-130.

Mervis, C., Catlin, J., & Rosch, E. (1976). Relationships among goodness-of-example,
category norms, and word frequency. Bulletin of the Psychonomic Society, 7(3),
283-284.

Meyer, D. E., Schvaneveldt, R.W., Ruddy, M.G. (1974). Functions of graphemic and
phonemic codes in visual word-recognition. Memory & Cognition. 2(2), 309-321.

Milberg, W., Blumstein, S., Sullivan Giovanello, K., & Misiurski, C. (2003). Summation
priming in aphasia: Evidence for alterations in semantic integration and activation.
Brain and Cognition, 51(1), 31-47.

Mirman, D., & J. S. Magnuson (2009). Dynamics of activation of semantically similar
concepts during spoken word recognition. Memory & Cognition. 37(7), 1026-1039.









Mirman, D., & J.S. Magnuson (2008). Attractor dynamics and semantic neighborhood
density: Processing is slowed by near neighbors and speeded by distant
neighbors. Journal of Experimental Psychology: Learning, Memory, and Cognition,
34(1), 65-79.

Moss, H., Ostrin, R., Tyler, L., & Marslen-Wilson, W. (1995). Accessing different types
of lexical semantic information: Evidence from priming. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 21(4), 863-883.

Munakata, Y., & Pfaffly, J. (2004). Hebbian learning and development. Developmental
Science, 7(2), 141-148.

Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective
review of current findings and theories. In D. Besner & G. Humphreys (Eds.),
Basic processes in reading: Visual word recognition, (pp. 264-336). Hillsdale, NJ:
Erlbaum.

Nelson, H.E. (1982). National Adult Reading Test (NART). Windsor, Berkshire, England:
The NFER-NELSON Publishing Company.

Nickels, L. (2002). Therapy for naming disorders: Revisiting, revising, and reviewing,
Aphasiology, 16(10), 935-979.

Plaut, D. C. (1996). Relearning after damage in connectionist networks: Toward a
theory of rehabilitation." Brain and Language. 52(1), 25-82.

Randall B., Moss H.E., Rodd J.M., Greer M., & Tyler L.K.. (2004) Distinctiveness and
correlation in conceptual structure: Behavioral and computational studies. Journal
of Experimental Psychology Learning Memory Cognition, 30, 393-406.

Rosch, E. (1975). Cognitive representations of semantic categories. Journal of
Experimental Psychology: General. 104(3), 192-233.

Seidenberg, M., Waters, G., Sanders, M., & Langer, P. (1984). Pre- and postlexical loci
of contextual effects on word recognition. Memory & Cognition, 12(4), 315-328.

Taylor, K.I., Moss, H.E., Tyler, L.K. (2007). The conceptual structure account: A
cognitive model of semantic memory and its neural instantiation. In J. Hart, Jr., &
M.A. Kraut (Eds.), Neural Basis of Semantic Memory, (pp 265-301). Cambridge:
Cambridge University Press.

Thompson, C.K., Shapiro, L.P., Kiran, S., & Sobeck, J. 2003. The role of syntactic
complexity in treatment of sentence deficits in agrammatic aphasia: The
complexity account of treatment efficacy (CATE). Journal of Speech, Language,
and Hearing Research, 46, 591-607.

Torgesen, J.K., Wagner, R. K., & Rashotte, C.A. (1999). Test of Word Reading
Efficiency. Austin, TX: PRO-ED Publishing, Inc.









Tyler, L.K., & Moss, H.E. (2001). Towards a distributed account of conceptual
knowledge. Trends in Cognitive Sciences, 52, 244-252.

Wierenga, C., Perlstein, W., Benjamin, M., Leonard, C., Gonzalez Rothi, L., & Conway,
T. (2009). Neural substrates of object identification: Functional magnetic
resonance imaging evidence that category and visual attribute contribute to
semantic knowledge. Journal of the International Neuropsychological Society,
15(2), 169-181.









BIOGRAPHICAL SKETCH

Christina del Toro received her bachelor's and master's degrees in communication

sciences and disorders from the University of Florida in 2004 and 2006, respectively.

She received clinical training as a speech-language pathologist at the Malcom Randall

Veterans Affairs Medical Center in Gainesville, FL. She has been involved in research

at the VA Brain Rehabilitation Research Center since her undergraduate years. She

has taught undergraduate and graduate level courses at the University of Florida and

the University of Washington. She plans to continue research and teaching in speech-

language pathology as an academician.





PAGE 1

1 THE EFFECTS OF FEATURE TYPE ON SEMANTIC PRIMING OF PICTURE NAMING IN NORMAL SPEAKERS By CHRISTINA MARIA DEL TORO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010

PAGE 2

2 2010 Christina Maria del Toro

PAGE 3

3 To my family who has given me unending love, support, guidance and joy, this work is dedicated to you

PAGE 4

4 ACKNOWLEDGMENTS There are many people who have helped and supported me throughout the completion of my degree and dissertation. I thank my parents, Jorge and Margarita del Toro and my sister Jennifer McGugan, for their constant encouragement and support and most importantly und erstanding in what has been a unique journey for us all. I am forever grateful to Michelle Troche for being an academic partner, study buddy, roommate and friend over many years and degre es at UF. I cannot imagine my UF years without her. I am thankful f or Lauren Bislick who I met at the beginning of the PhD journey and who has become a wonderful friend and support especially in the final year. I thank my roommates who have lived this journey with me, Joshua Troche, Heather Beck, and Renee Dickinson. The y were there through it all and reminded me to live beyond school. I must thank the UW lab members who have provided endless moral support, technical support, intellectual support, and of course, doughnuts: JoAnn Silkes, Rebecca Hunting Pompon, Laine And erson, Wesley Allen, Megan Oelke, and Mike Mackinnon. There are three people without whom I could not have completed data collection: Erica Gonzalez, Brittney Hayes, and. Matt Lacourse. Also, Lorraine Gonzalez who not only recruited and scheduled nearly everyone she knows, but graciously allowed us to use her house for weekend data collection. I am forever grateful to all of them for many hours spent in participant recruitment, scheduling, testing, and traveling. My committee members deserve many thank s for years of guidance and support. I thank Bruce Crosson, for introducing me to new literatures and providing support along the way; also for invaluable guidance on methodology. I am thankful to Lisa Edmonds

PAGE 5

5 for allowing me to work in her lab for my fi rst semester and gain invaluable experience in treatment research; also, for her mentorship in my first ever teaching experience; and guidance in stimuli development for my dissertation. I thank Leslie Gonzalez Rothi for providing a supportive environment at the Brain Rehabilitation Research Center where I learned how to become a researcher starting as an undergraduate; also, for always pushing me beyond my comfort zone and encouraging me the whole way. To my chair and friend, Diane Kendall I offer more thanks than I can express. For years of research experience, from doing reliability on her treatment study to designing stimuli alongside her for the SAPA test. I have not only learned skills for research, teaching, and mentoring but through her example and her words I have learned skills for balancing an academic life with a fulfilling personal life. I look forward to many more years of academic collaboration and friendship. Lastly, there is a large group of people without whom I would not be able to wr ite this dissertation: the participants. I especially thank those that gave their weekend time to help me. I am so thankful for their time and energy and willingness to complete all the tasks I asked of them.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ ........ 10 LIST OF A BBREVIATIONS ................................ ................................ ........................... 11 ABSTRACT ................................ ................................ ................................ ................... 12 C H A P T E R 1 INTRODUCTION ................................ ................................ ................................ .... 13 Semantic Memory ................................ ................................ ................................ ... 13 The Conceptual Structure Account ................................ ................................ ......... 16 Evidence for the conceptual structure account ................................ .......... 19 Limitations of the conceptual structure account ................................ ......... 21 Confirming the CSA ................................ ................................ ................... 23 Semantic Priming ................................ ................................ ................................ .... 24 Priming in Distributed Networks ................................ ................................ ....... 25 Multiple Primes ................................ ................................ ................................ 26 Statement of the Problem ................................ ................................ ....................... 28 Research Questions and Hypotheses ................................ ................................ ..... 28 2 METHODS ................................ ................................ ................................ .............. 33 Data Collection ................................ ................................ ................................ ....... 33 Design ................................ ................................ ................................ .............. 33 Dependent variables ................................ ................................ .................. 33 Independent variables ................................ ................................ ................ 33 Trial structure ................................ ................................ ............................. 33 Timing ................................ ................................ ................................ ........ 34 Task ................................ ................................ ................................ ........... 34 Settings and Equipment ................................ ................................ ................... 35 Stimuli ................................ ................................ ................................ ............... 35 Semantic categories ................................ ................................ .................. 35 Prime type ................................ ................................ ................................ .. 36 Participants ................................ ................................ ................................ ....... 37 Inclusion criteria ................................ ................................ ......................... 38 Screening ................................ ................................ ................................ ... 38 Data Analysis ................................ ................................ ................................ .......... 39 Transcription and Scoring ................................ ................................ ................. 39 Data Trimming ................................ ................................ ................................ .. 39

PAGE 7

7 Reliability ................................ ................................ ................................ .......... 39 Statistical Analysis ................................ ................................ ............................ 40 3 RESULTS ................................ ................................ ................................ ............... 44 Reliability ................................ ................................ ................................ ................ 44 Research Question 1 ................................ ................................ .............................. 44 Younger Adults ................................ ................................ ................................ 44 Older Adults ................................ ................................ ................................ ...... 45 Research Question 2 ................................ ................................ .............................. 45 Research Question 3 ................................ ................................ .............................. 46 Research Question 4 ................................ ................................ .............................. 47 Word Frequency Effects ................................ ................................ ......................... 47 4 DISCUSSION ................................ ................................ ................................ ......... 63 Priming Effects from Related Features and Neutral Primes ................................ .... 63 The Effect of Feature Type ................................ ................................ ..................... 64 The Effect of Number of Fea tures ................................ ................................ ........... 68 The Effect of Timing on Multiple Feature Primes ................................ .................... 69 The Effect of Word Frequency ................................ ................................ ................ 71 Limitations ................................ ................................ ................................ ............... 71 Implications for Anomia Treatment ................................ ................................ ......... 72 Future Directions ................................ ................................ ................................ .... 73 Summary ................................ ................................ ................................ ................ 74 APPENDIX A PRIME TARGET STIMULI ................................ ................................ ...................... 76 B TARGET WORD FREQUENCY ................................ ................................ ............. 79 LIST OF REFERENCES ................................ ................................ ............................... 80 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 84

PAGE 8

8 LIST OF TABLES Table page 1 1 The distribution and correlations of shared and distinctive features in living and nonliving things, according to the CSA. ................................ ....................... 31 1 2 Summary of Research Questions (RQ) a nd Predictions ................................ .... 32 2 1 Participant demographic information including age, years of education, gender and average (AVE) test scores with standard deviations (SD) for the Mini Mental Status Exam (MMSE; Folstein, Folstein & McHugh, 1975), Test of Word Reading Efficiency (TOWRE; Torgesen, Wagner, & Rashotte, 1999), Pyramids and Palm Trees Test (P&P; Howard & Patterson, 1992), American N ational Adult Reading Test (ANART; Nelson, 1982), and estimated IQ based on ANART Scores (Grober & Sliwinski, 1991). ................................ ........ 41 3 1 Raw me an SRT for younger adults for comparison of neutral, shared features, distinctive features, and combination of shared and distinctive features prime conditions. Asterisks indicate significant pairwise comparisons which are described in parentheses. ................................ ............. 49 3 2 Raw mean SRT for older adults from comparison of neutral, shared features, distinctive features, and combination of shared and d istinctive features prime conditions. Asterisks indicate significant pairwise comparisons which are described in parentheses. ................................ ................................ .................. 50 3 3 Raw mean SRT from comparison of shared versus shared shared prime conditions. There were no significant effects of prime condition. ........................ 51 3 4 Raw mean SRT from comparison of distinctive and distinctive distinctive prime conditions. There were no significant effects of prime condition. .............. 52 3 5 Raw mean SRT from comparison of 200msec and 600msec ISI for the prime conditions of neutral, shared features, distinctive features, and combination of shared and distinctive features. There were no significant effects o f ISI. ...... 53 3 6 Raw mean SRT from comparison of 200msec and 600msec ISI for the shared feature prime conditions. Asterisks indicate significant pairwise comparisons which are described in parentheses. ................................ ............. 54 3 7 Raw mean SRT from comparison of 200msec and 600msec ISI for the distinctive feature prime conditions. There were no significant effects of ISI. .... 55 A 1 Animal Stimuli ................................ ................................ ................................ ..... 76 A 2 Tool Stimuli ................................ ................................ ................................ ......... 77

PAGE 9

9 A 3 Vehicle Stimuli ................................ ................................ ................................ .... 78 B 1 Taret Word Frequency ................................ ................................ ........................ 79

PAGE 10

10 LIST OF FIGURES Figure page 2 1 Example of priming task trial structure displaying the shared distinctive condition ................................ ................................ ................................ ............. 42 2 2 Priming task trial structure for one complete trial displaying timin gs of each screen. All trials within one session are administered with an ISI of 200 or 600msec for the blank screens between primes and between the second prime and target picture. The target picture remains on the screen until the voice key is triggered ................................ 43 3 1 Log transformed mean SRT of young adult responses from comparison of neutral, shared distinctive and combined prime conditions. Brackets indicate significant pairwise comparisions. Asterisks indicate the prime condition which produced significantly faster SRTs. ................................ .......................... 56 3 2 Log transformed mean SRT of older adult responses from comparison of neutral, shared, distinctive and combined prime conditions. Brackets indicate significant pairwise comparisions. Asterisks indic ate the prime condition which produced significantly faster SRTs. ................................ ............................ 1 3 3 Log transformed mean SRT of older adult responses from comparison of shared and shared shared prime conditions. There were no significant effects of prime condition. ................................ ................................ ................... 58 3 4 Log transfor med mean SRT of older adult responses from comparison of distinctive and distinctive distinctive prime conditions. There were no significant effects of prime condition. ................................ ................................ .. 59 3 5 Log transformed mean SRT of older adult responses to all prime conditions (collapsed across shared, distinctive, combined, and neutral primes) at 200 ISI versus 600 ISI. There was no significant difference in SRT between ISI conditions. ................................ ................................ ................................ .......... 60 3 6 Log transformed mean SRT of older adult responses to shared primes (collapsed across shared an d shared shared) primes at 200 ISI versus 600 ISI. Brackets indicate significant pairwise comparisions. Asterisks indicate the prime condition which produced significantly faster SRTs. ........................... 61 3 7 Log transformed mean SRT of older adult responses to distinctive primes (collapsed across distinctive and distinctive distinctive) primes at 200 ISI versus 600 ISI. There was no significant difference in SRT between ISI conditions. ................................ ................................ ................................ .......... 62

PAGE 11

11 LIST OF ABBREVIATION S ANART American National Adult Reading Test CSA Conceptual Structure Account ISI Inter stim ulus Interval MMSE Mini Mental Status Examination msec Milliseconds P&P Pyramids & Palm Trees SEC Seconds SRT Speech Reaction Time TOWRE Test of Word Reading Efficiency

PAGE 12

12 Abstract of Dissertation Presented to the Graduate School of the University of Florid a in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THE EFFECT OF FEATURE TYPE ON SEMANTIC PRIMING OF PICTURE NAMING IN NORMAL SPEAKERS By Christina M. del Toro August 2010 Chair: Diane Kendall Major: Communication Sciences & Disorders The aim of this study was to investigate the roles of shared and distinctive features on conceptual activation. A picture naming paradigm was employed to measure speech reaction time during feature to concept activation of animals, tools, and vehicles. F ifty nine young adults and 47 older adults completed the priming task with an interstimulus interval of 200msec and 600msec, in two different sessions. Results indicate that regardless of semantic category, distinctive feature primes resulted in the fastes t reaction times compared to shared features, a combination of distinctive and shared features, and neutral primes. In general, manipulation of ISI did not produce changes in SRT. Overall, the results showed that as stated in the CSA, distinctive feature s have a privileged role in concept activation. However, the proposed differential roles of shared and distinctive features in living and nonliving things, was not confirmed. Additionally, the results indicate that in general, multiple feature primes do not require more time to activate a concept.

PAGE 13

13 CHAPTER 1 INTRODUCTION Semantic Memory Semantic memory is our cumulative knowledge of things, people, places, and events and is considered one of the most critical aspects of human cognition (Hutchison, 2003) Many theories and models have been put forth to elucidate the structure of conceptual representations and how the representations are processed. One group of models, referred to here as semantic feature theories, is based on the assumption that concepts or conceptual information, are instantiated in a distributed neural system comprised of smaller units. The smaller units represent semantic features or properties and are the lowest level of representation in semantic memory. Concepts emerge from over lapping patterns of activation across related feature units (Plaut, 1996). For example, the activation of the conceptual representation of tree, requires simultaneous activation of related feature units such as branch, leaves, trunk, bark, etc. Thus, a c oncept is not represented as a discrete unit, but instead emerges from activated features. This leads to distributed representations of concepts over unique patterns of activation (Plaut, 1996). The relationship of conceptual feature units and their co activation have been described by many researchers using varying terminology and specifications. However, despite such differences, it is generally agreed upon that the connections between conceptual representations are based on similarity and differences of features. Such connectivity has been described in terms of semantic neighborhoods (Mirman and Magnuson, 2008) and typicality effects (Rosh, 1975; Plaut, 1996).

PAGE 14

14 Mirman and Magnuson (2008) propose that the structure of semantic memory is based on how cl ose or distant conceptual representations are from one another in terms of semantic relatedness. If concepts are closely related, i.e. share overlapping features, such as cow and bull, near concept ual representations that share fewer overlapping features, such as cow and tiger of activation correspond to the distance between features based on similarity. Mirman and M agnuson have shown, in computational and behavioral experiments, that semantic processing is slowed by dense, near neighbors and speeded by far neighbors. These results are attributed to inhibitory effects from near neighbors and f acilitative effects from distant neighbors. For example, when asked to name a picture of a cow prior presentation of a far neighbor tiger, will be facilitative. Presentation of tiger will activate its features which will include a few shared features with cow ; however, whe n cow is activated, the semantic system can inhibit the activation of tiger because there is little overlap of activation patterns for tiger and cow Inhibiting the activation of tiger will not turn off a critical amount of features for cow. Conversely prior presentation of a bull (a near neighbor) will be inhibitory when cow is presented. The presentation of bull will activate numerous features that are shared with cow but also features that are unique to bull. When cow is presented, the semantic syste m must activate its features which will overlap with the already activated features of bull Additionally, the system must inhibit the features of bull which now do not match the activation pattern of cow. The greater number of shared features between bul l and cow causes difficulties in inhibiting only bull and in turn, activation of cow is slowed.

PAGE 15

15 Conceptual representations in semantic memory have also been purported to be connected through semantic typicality (Rosch, 1975). This view suggests semantic categories are built on a hierarchy of typicality, or how close each exemplar is to the prototype of that category For example, robin is a typical member of the bird category, while ostrich is considered an atypical exemplar. The notion of typicality h as been tested computationally ( Plaut 1996). The results of damaging and retraining the model revealed that generalization to untrained items was greater when the trained items were atypical members of a category. For example, training ostrich resulted in generalization to untrained items such as blue jay, and robin ; but training blue jay did not generalize to the untrained penguin This suggests that activation of atypical exemplars activates features which are unique to that conceptual representation as well as features which are common to most members of that category. On the other hand, activation of typical exemplars activates the features common to most members and not the unique features of the atypical members. Thus, if typical items are trai ned, the distinguishing features of atypical members will not be activated and access to them will not be improved. Based on the theory that a concept is comprised of a distributed representation of features, and processed via simultaneous activation of those features, techniques for remediating word retrieval deficits have been developed. Two such treatments are seman tic feature analysis (SFA; Boyle & Coehlo, 1995) and the complexity account of treatment efficacy (CATE; Thompson, Shapiro, Kiran, & Sobeck, 2003 ). In SFA treatment patients are asked specific question about the features of a pictured object and then aske d to name the object ; for example, while viewing a picture of an apple, the

PAGE 16

16 patient is asked where do you find it? What do you do with it? The goal is to improve access to the concept (object to be named) by activation of the features. The CATE treatment (1976) typicality effect. As in SFA, the objective of CATE is to improve activation of a conceptual representation by activating the related features. The typicality effect is empl oyed through the use of atypical (instead of typical) features of the concept to improve generalization to untrained items. Although semantic treatments have had positive treatment effects generalization to untrained items and maintenance effects long a fter therapy concludes can be improved (Nickels, 2002) Therefore, it is necessary to refine semantic feature treatments One approach is to gain a better understanding about the relationship of the feature units on the activation of conceptual represent ations, so that this relationship can be exploited in word retrieval therapies. One theory that provides further specificity regarding feature types and their connections in the semantic system is called the Conceptual Structure Account (CSA; Tyler & Moss 2001; Taylor et al., 2007). The Conceptual S tructure A ccount The CSA, like other semantic feature theories (Mirman & Magnuson, 2008; Plaut, 1996), is based on the assumption that semantic memory is a distributed connectioni st network comprised of units that represent semantic features and the p rocessing of a concept is the result of overlapping patterns of activation across semantic feature units. CSA extends and adds further specificity to the above mentioned accounts of semantic memory through two ess ential points. First, the degree to which a feature is shared by different concepts varies and second, the frequency of co occurrence of features varies. These two points will be described below and are summarized in Table 1 1.

PAGE 17

17 In regards to the degr ee to which a feature is shared, the CSA identifies two types of features, shared and distinctive. The idea is that shared and distinctive features differentially activate conceptual representations. Shared f eatures are defined as those features that are common to related concepts while distinctive features are unique to each concept. That is, most living things share features of eyes, ears, breathes, legs; thus, these features are only indicative of category membership. Conversely, fewer living things have stripes, trunk, mane, or an udder making these distinctive features Distinctive features belong to fewer concepts and provide more information about a specific concept. An example given by Taylor and colleagues is tiger To activate the concept of tiger, the shared features which define animals and, specifically cats, such as four legs, teeth, and tail will be activated but until stripes is activated the concept of tiger will not be complete and therefore not activ ated above other types of cats. Thus, shared features reflect category membership and are not helpful in identification, and distinctive features provide more information about a particular concept and are critical to identification. The CSA goes on to stipulate that a ctivation of only shared features will lead to activation of several concepts sharing those features (e.g., activation of four legs, teeth, and tail would activate animals in addition to tiger). Activation of a dis tinctive feature with s hared features will highlight the target concept above its neighbors (e.g. activation of stripes and not mane will highlight tiger over lion ) Thus, the distinctive feature is needed to fully activate the target concept (Taylor et al 2007).

PAGE 18

18 The secon d postulate of the CSA regarding semantic structure is the frequency of co occurrence of features. The CSA proposes a Hebbian structure in which units that simultaneous activation (Munakata & Pfaffly, 2004). Thus, features that co occur frequently have stronger connections than features which co occur infrequently. For example, most things which breathe have eyes but not all things that breathe have stripes In a Hebb ian system, consequently, breathe and eyes will have a stronger connection than breathe and stripes and therefore, the activation of breathe is more likely to lead to the activation of eyes than stripes. The frequency of feature co occurrence, or correlation as referred to by Taylor et al., (2007), differs for living and nonliving things. In living things shared features are highly correlated to each other (e.g. things which breathe typically have eyes and ears) while distinctive features on the other hand, are weakly correlated to other features (e.g most things with eyes and ears pouches udders, or eight legs) In the example of tiger given above, the features describing nearly all living things, breathes, legs, hears, eyes, ea rs are strongly correlated shared features while stripes is a distinctive feature that is w eakly correlated with the other features of tiger Nonliving things on the other hand, have distinctive features which are highly correlated (things that have a bl ade also cut ) and fewer shared features with lower correlations (most tools have a handle but could be made of wood plastic, or metal; be used for hitting, cutting, or turning ) Thus, knife has the distinctive features of blade and cuts which are strongly correlated to each other as opposed to the shared features of handle and metal which

PAGE 19

19 describe several tools but are weakly correlated with each other and the distinctive features of blade and cuts. Evidence for the conceptual structure account The claims of CSA have been supported by several studies ( McRae, de Sa, and Seidenberg, 1 997 ; McRae, Cree, Westmacott, and de Sa, 1999; Randall, Moss, Rodd, Greer, and Tyler, 2004 ) using neurologically healthy individuals. Study tasks have includ e d feature generation, priming in a feature verification task, and priming with targets and primes that have correlated features The goal of feature generation studies is to understand more about the distribution of feature type in living and nonliving things. Pa rticipants are asked to list all the features they can think of for a given concept. From these lists, shared and distinctive values are calculated based on the number of occurrences of each feature for all concepts in the list. Distinctivenes s or the degree to which a feature is distinctive to one concept, is calculated in one of two ways: either by taking the inverse of the number of concepts for which a feature was produced or by identifying a cut off a priori (e.g. distinctive features occ ur in only 1 or 2 concepts ) Results of a feature generation study for living and nonliving things by Taylor and colleagues (2007) supported the proposed division of high and low correlations among shared and distinctive features in living and nonliving t hings. That is, living things had a greater number of highly correlated shared features compared to nonliving things ; and nonliving things had more highly correlated distinctive features than living things (Taylor et al., 2007). McRae et al (1997 ) have conducted two priming studies to investigate the effect of feature correlation Priming studies are based on the principle that pres entation of a word (prime) will facilitate recognition of a subsequent word (target) ( Meyer,

PAGE 20

20 Schvaneveldt, and Ruddy, 1974) McRae et al. used an online written word comprehension task in which participants viewed a prime concept followed by target features which were either weakly or strongly correlated to other features of the concept. Participants were required to an swer questions about the semantic relatedness of the prime to the target, such as, is it animate? In this word comprehension task, reaction time to living things was faster compared to nonliving things when the correlations between prime and target were h igher; thus, supporting the claims of the CSA that correlation of shared features is higher in living things than in nonliving things (McRae et al ) In the second study by McRae et al. (2007) investigating the effect of feature correlation, participant s viewed prime concepts and target features such as deer is hunted and were asked if the features belonged to the concept. Reaction times decreased as the correlation of prime and target increased; conversely, reaction times increased as the correlation of prime and target decreased The authors concluded that the simultaneous activation of correlated features leads to a faster initial rise in activation allowing strongly correlated features to settle into a stable activation pattern faster than weakly correlated fe atures (McRae et al., 1999) Randall et al. (2004) used the results from McRae et al. (2007) to investigate if, in living things distinctive features would indeed be activated more slowly than shared features du e to the weaker correlation o f distinctive features (when compared to a stronger correlation of distinctive features in nonl iving things) This prediction was supported by their results from a timed feature verification task in which participants had to respond to the target within a set time limit ( Randall et al ). However, in an untimed

PAGE 21

21 version of the task there was no such difference between distinctive features of living and nonliving things indicating that correlation strength of features affects only the initial rise time of act ivation and not the final level of activation when the network is in a stable state. Limitations of t he conceptual structure account While the CSA provides a framework upon which testable hypotheses about the structure of semantic memory can be tested, two critical limitations remain : 1) the distribution of features in semantic categories and 2) the specificity of the semantic concepts to wh ich the theory can be applied In regards to the distribution of features in semantic categories, the CSA propose s a distinction between living and nonliving things. As described earlier, there are more shared than distinctive features in living things and more distinctive than shared features in nonliving things. However, there are categories of living and nonlivi ng things which violate this distinction. This is a fact acknowledged by the authors of the CSA. Vehicles, for example, altho ugh a nonliving category, are processed more like living things because they have many shared features ( wheels, engines, tires, d oors, fuel etc) and fewer distinctive features ( wings, sail etc.) which are necessary to distinguish a plane from a car and a car from a truck (Taylor, Moss, Tyler, 2007). This difference in the structure of features for vehicles is also supported by a natomical studies investigating the processing of different domains of concepts. In an fMRI study of the contribution of category and visual attribute s to semantic knowledge in the fusiform gyrus, Wierenga et al., (2009) compared activation from naming pi ctures of animals, tools, and vehicles. To compare the contribution of visual attributes, the investigators varied the amount of visual information available by providing increasing

PAGE 22

22 details to pictured images. At lower levels of spatial frequency only gl obal form, which can be likened to shared features, was seen compared to local detail or distinctive information given as the spatial frequency information increased. Results indicated that animals required the least visual detail (distinctive features) w hile tools required the most and vehicles fell between the two (Wierenga, et al., 2009 ). These results align with the claims of the CSA that animals have many shared features so there is information available for identification from just shared features wh ereas tools have fewer shared features and need more local details or distinctive features for identification. Likewise, as predicted by the CSA, vehicles did not behave entirely like living or nonliving objects but instead require d both global form and l ocal details for ident ification (Wierenga et al., 2009 ). Taylor and colleagues (2007) also point out that the features of fruits and vegetables are not distributed in the same manner as animals though all three categories are considered living things. Fruits and vegetables have fe wer distinctive features which have even lower correlations than animals. T hus, the distribution of shared and distinctive features and their weight for identifying objects cannot be solely attributed to a distinction of living and nonliving categories ; within these categories further distributional differences can be found. Taylor and colleagues (2007) posit that it is the internal structure of the properties of concepts which determines the relationship between shared and di stinctive features not just the category or domain to which the concepts belong. Therefore, investigation of the distribution of shared and distinctive features and their interaction in categories more sp ecific than living or nonliving things may reflect the truer structure of the system

PAGE 23

23 The second limitation of the CSA is the specificity of objects which must be u sed in order to account for the differential effects of shared and distinctive fea tures. That is, the CSA provides experimental evidence t hat with access to distinctive features alone, an animal will be identified more slowly due to fewer connections between features related to the concept. However, this point may not be entirely accurate. The specificity of the concept may influence the relationship of shared and distinctive features. For example, meow can be considered a distinctive feature of cats which according to the CSA should activate the concept slower than a shared feature; but meow would likely be able to activate the concept of cat on i ts own without other information This issue has not been addresse d by the authors of CSA. Their examples and stimuli from their studies appear to be specific objects or members of a category (e.g. tiger) and not general concepts or subcategories (e.g. cat) For example, Taylor and colleagues (2007) discuss ed the difference between a kn ife and a tiger, using stripes as the distinctive feature of tiger which would not apply to all cats When presen ted alone stripes would not have the same effect in eliciting tiger as meow would in eliciting cat More information would be needed to activate tiger but meow could activate cat on its own. Until the authors address this issue investigations of the CSA are limited to employing specific items as stimuli and more critically the results of such studies can only be applied to specific exemplars of a category Confirming the CSA The princip les of the CSA may be applied to anomia treatments, such as SFA, by providing further insight into the nature of feature to concept activation. Such information could potentially improve treatment outcomes by specifying the type of stimuli that could be used in an experimental therapy program. However, the claims of

PAGE 24

24 the CSA cannot be applied to these treatments without empirical evidence that the principles of the CSA can be used to affect lexical access in the manner used in aphasia therapy. Typical tasks employed in anomia treatments involve, object description, pictur e naming, or naming objects by description which engage feature to concept activations. In other words, when shown a picture of a hammer, individuals However, CSA experiments, thus far, have only been tested using concept to concept activation (e.g. using eagel to prime hawk ). In order for principles of the CSA to provide insight into rehabilitation of anomia, the principles must first be addressed in feature to concept activation. To date, n o study has used features as the prime and concepts as the target. Since a priming paradigm will be used in this study, relevant semantic priming literature will be reviewed here. Semantic Priming Meyer, Schvaneveldt, and Ruddy ( 1974) first showe d that presentation of a word (prime) will facilitate recognition of a subsequent word (target). Since then, priming effects have been shown to be stronger when the prime word and target word are related or share semantic information compared to a neutral or unrelated prime target pair (Balota, & Paul 1996; Bueno & Frencke Mestre, 2008; de Groot, 1984; McNamara 2005 ; Moss, Ostrin, Tyler, & Marslen Wilson, 1995; Neely, 1991; Seidenberg, Waters, Sanders, & Langer, 1984). Most commonly, primes and targets are concepts related or interacting in some manner. For example, lake may prime ocean because both are bodies of water. Such concept concept pairs reliably produce priming effects given a semantic relatio nship and a semantic based task (Hutchison, 2003).

PAGE 25

25 Priming in Distributed Networks Priming effects occur in distributed networks for two possible reasons. McNamara (2005) refers to the first as learning models. These models predict priming effects as a result of gradual learning by the network which leads to increased probability of producing the same response each time a specific input is recognized. Priming effect s under this model occur over long lags of time as the system learns (McNamara, 2005). The second explanation of priming ef fects in distributed networks comes from the proximity models (Cree, McRae, & McNorgan, 1999; McRae, de Sa, & Seidenberg, 1997). These models suggest priming occurs because related words (prime s and targets) are closer and more strongly connected than unrelated words. In priming, a target is processed after the processing of the prime; that is, a pattern is activated for the prime which in turn activates a pattern for the target. Thus, processing of the target is faste r fo llowing processing of the prime when there are connec tions between the two patterns than when there are no connections between the patterns (McNamara, 2005). The proximity model of priming in distributed networks parti cularly describes the process of priming between words which share features because activating features for a prime will also a ctivate features for the target. Consequently, when the target is processed some of the features are already activated. Furthermore, the proximity model pred ict s that exposure to a feature or set of features will activate those features and connected features, thereby increasing the speed of reaching a stable pattern of activation for a subsequent concept (McNamara, 2005) Processing of core, seeds, red wil l activate these units and the units each is connected to ( stem, tree, worm) so when the picture of apple is seen activation for the name is already in process and thus will more quickly achieve a stable pattern The effect of multiple primes, as illustr ated in the

PAGE 26

26 above example is an empirical question which has been studied (Balota & Paul, 1996; Milberg, Blumstein, G iovanello, and Misiurski, 2003). However, these studies, described below, have not used semantic features as primes. Such a paradigm wou ld provide further evidence that activation of features is the mechanism for concept activation Furthermore, determining the type of feature and number of each feature type which most strongly primes concepts will provide a framework for devising rehabil itation approaches for semantic impairments. Multiple Primes The effect of multiple primes has been studied based on spreading activation theories which posit that activation of one unit (from one prime) will spread automatically to related and connected p arts of the network ( even areas not directly connected to the original node). T hus, activation of lion will spread to stripes through the connection of tiger There are three possible influences from multiple primes: additive, underadditive, and overaddi tive ( Balota & Paul, 1996) An additive influence occurs when the facilitation of multiple primes is equal to the sum of the facilitation of each prime presented individually. An underadditive influence is seen when the facilitation of multiple primes is less than the total facilitation of each prime individually. Lastly, overadditivity is when the facilitation from multiple primes is greater than the total facilitation from each prime individually (Balota & Paul, 1996). Balota and Paul (1996) investiga ted this effect using two sequential primes in a lexical decision task and in a speeded word naming task. Additionally, they manipulated duration of the first primes and degraded the target stimulus in separate conditions. All conditions resulted in addi tive priming Thus, in their study, providing several primes in succession was no different than presenting each prime individually. However, the series of experiments

PAGE 27

27 conducted by Balota & Paul (1996) used concept primes and concept targets. Consequent ly, the effect of multiple features on priming a concept is not known and the effect of time on multiple feature primes is unknown Milberg, Blumstein, Giovanello, and Misiurski (2003) have conducted a multiple prime study using triplets in which the third word was the target of a lexical decision task The interstimulus interval (ISI), or time between the offset of a prime and the onset of the next prime, was varied between 200 msec and 600 msec. ISI was varied to determine at what time point an overadditive effect could be achieved. Th ere were four types of triplets: definitionally related triplet s (e.g meal, morning, breakfast), triplets with a c ategorical prime and nonword (e.g meal, foncern, breakfast), triplets with a nonword and featural prime (e.g jarm, morning, breakfast) an d triplets with two nonwords (e.g jarm, foncern, breakfast). The authors concluded that the priming effect from the definitionally related triplets seen in the 2 00ms ISI condition was additive; that is the same as adding the priming effects from each individual prime (categorical and nonword prime triplet + nonword and fe atural prime triplet). T he priming effect from the definitionally related triplet in the 600ms ISI condition however, was overadditive or gre ater than the effect from each individual prime. Thus, the longer ISI provided time for the convergence of meaning of the multiple primes and subsequently the enhancement of the priming effect and se mantic facilitation (Milberg et al 2003). Milberg to multiple related words can aid reaction to a target. Whether the priming at 200 msec ISI and 600 msec ISI is an automatic or controlled process is not discussed by Milberg et al. an d was not the aim of the study. Moreover, while understanding if the potent

PAGE 28

28 mechanism is automatic or controlled processing is important, it was not the primary focus of the current study Instead, the aim was to determine if there is a benefit to presen ting multiple features prior to naming a picture and timing may be an important factor in order to allow for consolidation of the meanings of each prime. Statement of the Problem The CSA appears to be a solid theory of semantic memory which suggests a unique structure based on shared and distinctive features. Shared features of l iving nouns are more abundant and more strongly correlated with other features, while distinctive features are more weakly correlated and fewer in number. Conversely, non livi ng nouns have more distinctive features which have strong correlations and fewer shared features with weaker correlations. The distribution of shared and distinctive features in living and nonliving concepts could be used to further specify the stimuli us ed in anomia treatments; however, it is unknown if shared and distinctive features can be used to activate concepts because previous studies have used concept to concept activation and not feature to concept activation. Thus, the aim of the current study was to test the roles of shared and distinctive features, as proposed by the CSA, in feature to concept activation. Additionally, there is evidence that more time is necessary to allow convergence of meanings from multiple primes which in turn leads to in creased priming and semantic facilitation. To determine if time is a factor in activating different types of features the current experiment was conducted at two inter stimulus intervals. Research Questions and Hypotheses Table 1 2 summarizes the resear ch questions and respective predictions. Specific Aim #1: According to the CSA shared features will boost performance in naming living nouns due to more numerous connections to strongly correlated shared features.

PAGE 29

29 Furthermore distinctive features resu lt in slower speech production time due to weaker correlations with other features. Conversely, f or non living nouns, distinctive features boost performance over shared features due to the stronger correlations and greater number of distinctive f eatures; and shared features produce slower speech production time as a result of weak correlations to other features T he fol lowing null hypothesis was investigated: There is no significant difference in naming living and non living nouns as measured by speech re action time (SRT) when primed with shared or distinctive features or a combination thereof. The research question is : Is there a significant difference in naming animals, tools, or vehicles as measured by SRT when primed with shared or distinctive featur es or a combination there of compared to neutral primes? The prediction is: Based on the CSA it was predict ed in living things (animals and vehicles), that shared features would result in faster SRTs compared to distinctive features. For non living nouns (tools) distinctive features were predicted to produce faster SRTs over shared features For all domains of nouns, a combination of shared and distinctive features was predicted to result in the fastest SRTs. Specific Aim #2: The CSA proposes di fferent correlation strengths between shared and distinctive features for living and nonliving things. The strength of these correlations is largely based on the number of features for each concept, indicating that the more features, the stronger the corr elations between them; and the stronger the correlations. Thus, activation of features or multiple features with strong correlations results in faster activation of the concept. The null hypothesis wa s: There is no

PAGE 30

30 difference in SRT for naming living and non living nouns between conditions varying number s h ared and distinctive features. This hypothesis was addressed with t wo research questions: Does SRT for naming animals, tools, or vehicles change linearly as number of shared features changes from 1 2? Does SRT for naming animals, tools, or vehicles change linearly as number of distinctive features changes from 1 2? The prediction is: SRT following shared features will be faster for living nouns compared to non living nouns; SRT following distinctive fe atures will be faster for non living nouns compared to living nouns Specific Aim #3: Using multiple features to prime conceptual activation requires the consolidation of the information from each prime. This consolidation may occur over longer time inte rvals than is required for one prime. The null hypothesis was : There is no difference in priming effects over time. The research is: Is there a difference in priming effects over time when comparing an ISI of 200msec to an ISI of 600msec? The predictio n is: P atterns of priming effects will be different over time. Specifically an ISI of 200msec would have an additive influence and an ISI of 600msec would have an overadditive i nfluence.

PAGE 31

31 Table 1 1 The distribution and correlations of shared and distinc tive features in living and nonliving things, according to the CSA. Living things Non living things Shared features High in number High correlation ( e.g. things which breathe typically have eyes and ears ) Fewer in number Weak correlation ( e.g. most tools have a handle but could be used for hitting, cutting or turning ) Distinctive features Fewer in number Weak correlation ( e.g. most things with eyes and ears do not have pouches, udders, or eight legs) High in number High correlation ( e.g. things wi th blade also cut )

PAGE 32

32 Table 1 2 Summary of Research Questions (RQ) and Predictions Specific Aims Feature Type Predictions for Living Things Predictions for Nonliving Things # 1 RQ: Is there a significant difference in naming animals, tools, or vehicles as measured by SRT when primed with shared or distinctive features or a combination there of compared to neutral primes? Neutral, Shared, Distinctive, and Combination (shared & distinctive) 1. Combination of shared and distinctive features will lead to sig nificantly faster SRT than shared or distinctive features alone 2. Shared features alone will lead to significantly faster SRT than distinctive features alone 1.Combination of shared and distinctive features will lead to significantly faster SRT compared to shared or distinctive features alone 2. Distinctive features alone will lead to significantly faster SRT than shared features alone #2 RQ 1: Does SRT for naming animals, tools, or vehicles change linearly as number of shared features changes from 1 2? One shared feature Two shared features Two shared features will lead to significantly faster SRT There will be no significant difference between one and two shared features in SRT RQ2: Does SRT for naming animals, tools, or vehicles change linearly as number of distinctive features changes from 1 2? One distinctive feature Two distinctive features There will be no significant difference between one and two distinctive features in SRT Two distinctive features will lead to significantly faster SRT # 3 RQ: Is there a difference in priming effects over time when comparing an ISI of 200msec to an ISI of 600msec? Prediction Patterns of priming effects will be different over time; namely, an ISI of 200msec would have an addit ive influence and an ISI of 600msec would have an overadditive influence.

PAGE 33

33 CHAPTER 2 METHODS Data Collection Design A linear mixed effects model was employed to measure the effects of prime type (distinctive, shared, combination) and semantic category (animals, tools, vehicles) in two conditions (200 and 600 ISI) on speech reaction time during a picture naming task. The details of the experimental paradigm and data collection sessions are presented below. Dependent variables The dependent variable was SRTs for picture naming SRTs were measured from the onset of the picture to the initiation of speech. Practice trials were administered to familiarize participants with the task and to minimize false starts. False starts and incorrect naming responses were excluded from analysis. Independent variables There were three independent variables. The first was combina tion of pr ime types which consisted of nine prime conditions (described below). The second independent variable was the semantic categories of target pictures (animals, tools, vehicles) The third independent variable was ISI (200msec vs. 600msec). Trial structur e Trials consisted of two orthographic word primes and one pict ure target. All picture targets were concepts. Word primes where either semantic features or a neutral pr ime. The semantic feature primes were either a shared or distinctive feature (defined below) related to the target. The neutral prime was the word blank. The c ombination

PAGE 34

34 and order of feature type and neutral primes were randomized across all trials while limiting each target to a single appearance per participant. Thus, i n one trial a f eature type was p resented from 0 2 times. Figure 2 1 shows an example trial structure. The prime pair conditions were: Shared shared Shared neutral Neutral shared Shared distinctive Distinctive shared Distinctive distinctive Distinctive neutral Neutral distinctive Neutral neutral Timing Each participant completed the experiment twice in two separate sessions in the same week separated by no less than two days. The purpose of two test sessions was to compare the priming effects when ISI was manipulate d. One administration presented the items with 2 00msec ISI and the other administration presented the items with 600msec ISI. The o rder of ISI over the two sessions was randomized. ISI was measured from the offset of a stimulus to the onset of the next stimulus. Figure 2 2 displays the trial structure with time intervals. Each trial began with a fixation point of a (+) for 500 milliseconds. Two prime words appear ed sequentially each with duration of 200msec and an ISI of either 200 msec or 600 msec. Then a target picture appeared until the participant named the item aloud. The inter trial interval was 3 seconds. The next trial began with a fixation point. Task Participants were given the following instructions :

PAGE 35

35 Participants completed 10 practice items followed by further instructions from the examiner if needed. Setti ngs and Equipment All testing was conducted in a quiet room with participants seated at a comfortable distance from the computer screen. E prime software (E Prime version 2.0, Psychology Software Tools, Pittsburgh, PA) was used for stimulus presentation a nd collection of SRTs. Verbal responses were recorded by a C420 PP MicroMic head mounted microphone connected to a Tube MP preamplifier which activate d the voice key of a Serial Response Box interfaced with E prime software Reaction times were stored in by the E prime software. V erbal responses were also recorded with a Marantz PMD671 digital recorder. Stimuli Semantic categories Target items for this study were animals, tools, and vehicles. Targets can be found in appendix A. Additionally, a set of distracter items comprised two thirds of the total items. These items were from several semantic categories with the exclusion of the three target categories (animals, tools, vehicles), in order to prevent participants from anticipating an animal, tool or vehicle. Prime types for distracter items were the same as for target items. Distracter items were not analyzed. Target pictures Targets were presented as black and white line drawings. Pictures were collected from: the CRL Inter national Picture Naming Project (Bates et al., 2003), edupics.com, and Google Images. Target pictures were presented in the vertical and horizontal center of the screen.

PAGE 36

36 Prime type Neutral primes. The word blank was used as the neutral prime to compare to the effect of semantic feature primes de Groot, Thomassen, and Hudson (1982) found using blank as a neutral prime did not result in a significant difference in reaction times to targets following blank versus meaningful primes. Semantic feature primes S emantic features were selected from the corpus of 541 living and nonliving things created by McRae, Cree, Seidenberg, & McNorgan (2005). This corpus was normed on over 700 neurologically healthy participants. The features include nouns and verbs. Mc Rae et al categorized the types of features provided by the participants as: functional, visual motor, visual form and surface, visual color, sound, taste, smell, tactile, encyclopedic, taxonomic. The data include concept production frequencies which ar e the number of concepts in which each feature occurs. Semantic features can be found in the appendix listed by target and coded as shared (S ) or distinctive (D). Th e definition of shared and distinctive features is bas ed on a modified version of the McRa e et al. (2005) concept production frequency. Concept production frequency (CPF) is a measure from the McRae e t al. database which is defined as the number of times the feature was produced in the entire corpus of items. Any feature with a CPF greater tha n two is considered shared and two or less is considered distinctive. However, the McRae et al. database calculates the CPF over all items in the database not items within semantic category. Because the hypotheses of the CSA regarding shared and distinct ive features is based on categories, the CPF was re calculated for this study to represent the number of times a feature was produced within semantic

PAGE 37

37 category (animals, tools, vehicles). Shared and distinctive feature primes can be seen in appendix A. As seen in appendix A, in order to construct the distinctive distinctive and shared shared prime conditions, each concept has two shared and two distinctive features. In constructing the shared distinctive condition only one pair of four possible pairi ngs was chosen for each target. The most shared and most distinctive features were chosen. If the value for shared and distinctive was equal then the decisi on was based on consensus from three outside raters asked to choose the most distinctive/shared f eature of the two. Stimuli were chosen from the McRae et al. (2005) database based on number of features and syllable length. Specifically, e ach target (pictured concepts to be named) had to have at least two distinctive and two shared features. Target s were controlled for s yllable length, specifically one to four s yllables, to minimize variance in speech motor programming. Participants Two groups of neurologically healthy individuals served as participants. Forty seven individuals between the age s 50 8 0 comprised the older group. Fifty nine University of Washington undergraduate students between the age s of 18 30 comprised the younger group. Neurologically healthy participants were chosen because feature to concept priming has not previously been emplo yed in priming studies and the presence of priming effects should be established in a healthy brain before investigating effects in a pathologic population. The older group is the population of interest in this study because this is the typical age range of stroke survivors and people with aphasia The younger group was chosen

PAGE 38

38 as a methodologic control group because the majority of previous research in semantic priming has been conducted on young adults; thus it is an empirical question if priming effects from the current paradigm will be found in older adults Demographic information can be found in Table 2 1. Inclusion criteria Participants were right handed, monolingual English speaking adults without a h istory of neurologic conditions or disease and/or developmental cognitive disorders as measured by participant report. Such disorders included, but were not limited to dyslexia or alexia, phonologic impairments, memory impairments, language impairments, and vision impairments (excluding corrected vision) Screening Prior to data collection, each participant was administered the following tests in order to determine eligibility for participation: Mini Mental Status Exam ( MMSE; Folstein, Folstein & McHugh, 1975) was administer ed to determine intact memory based on a minimum score of 27/30. No participants failed the MMSE. Visual acuity screening was administered using a Snellen Eye Chart to ensure participants were able to perceive the stimuli base d on minimum criteria of reading the stimuli for 30 feet from a distance of 20 feet No participants fail ed the vision screening. semantic abilities: The Test of Word Reading Efficiency ( TOWRE; Torgesen, Wagner, & Rashotte, 1999 ) was administered to This test was designed to measure the ability to sound out words and to read words as a whole unit (sight words). The Pyr amids and Palm Trees Test ( P&P; Howard & Patterson, 1992) was used to describe

PAGE 39

39 Lastly, the American National Adult Reading Test ( ANART; Nelson, 1982) was used to estimate intellig ence quotient (IQ) in order to establish similar intelligence between the younger and older adults. IQ was calculated using the number of errors on the ANART and years of education as described by Grober & Sliwinski (1991). Average test scores can be found in Table 2 1. Data Analysis Transcription and Scoring Digital audio recordings of participant responses were transcribed and scored as correct or incorrect. Responses were scored correct when there was a single utterance of the name of the target pict ure Responses were scored incorrect if they included verbal preparation output um, huh or self corrections (ex. screw no drill). Data Trimming Only correct responses were analyzed. This resulted in an exclusion of 8.8% of the responses (5304 total response, 468 excluded responses). Reaction times less than 200msec were excluded based on previous studies which have used this cutoff to minimize the inclusion of false triggering of the voice key (e.g. breathing, lip smacking, etc.). This result ed in an exclusion of 2.8% of the correct responses (4765 total responses, 133 responses). Reliability Five independent raters performed inter rater reliability on 25% of each orrect based on the original criteria explained above. Training took place in a one hour session with the primary investigator. Raters demonstrated understanding of the criteria ity was

PAGE 40

40 reviewed by the primary investigator and when determined to be accurately completed the training was complete. Statistical Analysis A linear mixed effects model was employed to measure the effects of prime type on SRT in each semantic category. The nine prime conditions described above were collapsed into six prime conditions by coding conditions which were only different by order of feature primes as the same conditions. T his resulted in the following six prime conditions: Shared shared 1. Shared neutral \ Neutral shared 2. Shared distinctive \ Distinctive shared 3. Distinctive distinctive 4. Distinctive neutral \ Neutral distinctive 5. Neutral neutral 6. Distinct a nalyses were conducted for ea ch group, and within categor ies (animals, tools, vehicles) Log t ransformed data were used to reduce the effect of outliers. In all models participants were included as a random factor and target frequency was included as a fixed factor. Frequency of t argets was included as a fixed factor because frequency was not controlled for a priori but the literature provides strong evidence that frequency interacts with priming. Frequency was entered into the models as a categorical variable using criteria com monly see n in the literature ( low frequency was less than ten per million, medium frequency between ten and twenty per million, high frequency over twenty per million. Frequency ratings were taken from Brysbaert and New (2009) and are displayed in appendi x B. All analyses were conducted with Bonferroni correctio n for multiple comparisons.

PAGE 41

41 Table 2 1 Participant demographic information including age, years of education, gender and average (AVE) test scores with standard deviations (SD) for the Mini Menta l Status Exam ( MMSE; Folstein, Folstein & McHugh, 1975) Test of Word Reading Efficiency ( TOWRE; Torgesen, Wagner, & Rashotte, 1999 ) Pyramids and Palm Trees Test ( P&P; Howard & Patterson, 1992) American National Adult Reading Test ( ANART; Nelson, 1982), and estimated IQ based on ANART Scores (Grober & Sliwinski, 1991). Age Education Gender Female (F) Male (M) MMSE (raw score out of 30) TOWER (standard score with a range of 35 1 65) P&P (raw score out of 50) ANART (raw score out of 50) Estimated IQ based on ANART score Young Adults AVE (SD) 22 (2) 15 (2) 25 F 34 M 30 (1) 108 (11.6) 49 (1.79) 37 (6.5) 114.93 (8.0) Older Adults AVE (SD) 60 (7) 16 (2) 28 F 18 M 29 (1) 97 (18) 51 (1) 40 (8) 119.23 (7.8)

PAGE 42

42 Figure 2 1. Example of priming task trial structure displaying the shared distinctive condition [blank screen] + Shared feature Target Picture Distinctive feature [blank screen]

PAGE 43

43 Figure 2 2. Priming task trial structure for one complete trial displaying timings of each screen. All trials within one session are administered with an ISI of 200 or 600msec for the blank screens between primes and between the second prime and target picture. The target picture remains on the screen until the esponse. 500MSEC 200MSEC 200MSEC/600MS EC 200MSEC 200MSEC/600MSE C 3SEC 500MSEC TARGET PICTURE + PRIME PRIME + [NEW TRIAL]

PAGE 44

44 CHAPTER 3 RESULTS Reliability Inter each session. An Intra class correlation was calculated. Inter rater reliability was 96%. Research Question 1 Is there a significa nt difference in naming animals, tools, or vehicles as measured by SRT when primed with shared or distinctive features or a combination there of compared to neutral primes? A linear mixed effects model was conducted for each category to determine the main effect of prime condition on SRT of picture naming. Each model included frequency and prime condition as fixed factors and participant as a random factor. Four prime conditions were entered into the model, shared (shared blank, blank shared, shared share d), distinctive (distinctive blank, blank distinctive, distinctive distinctive), combination (distinctive shared, shared distinctive), and neutral (blank blank). Targets were coded for high, medium, or low frequency. Frequency effects are presented separ ately at the end of the results. Younger Adults Table 3 1 and Figure 3 1 display mean SRT and standard errors for each prime condition with each target semantic category (animals, tools, vehicles). The analysis of SRT of naming animals resulted in a main effect of prime condition F ( 3, 874.86 ) = 6.26, p =.000. Pairwise comparisons indicate distinctive primes resulted in significantly ( p <.05) faster responses than the neutral primes and faster than the shared primes ( p =.001). T he main effect of prime condition on naming tools was not significant F ( 3,

PAGE 45

45 861.85 ) = 2.34, p >.05. The main effect of prime condition on naming vehicles was not significant F ( 3, 745.81 ) <1. Older Adults Table 3 2 and Figure 3 2 show mean SRT and standard error for each prime c ondition within the target semantic category (animals, tools, vehicles). The analysis of SRT of naming animals resulted in a main effect of prime condition that was significant, F (3, 674.78) = 5.16, p <.05, Pairwise comparisons revealed that reaction times to distinctive features were significantly faster ( p =.001) than reaction times to shared features. The main effect of prime condition on naming tools was significant F (3, 675.82) = 4.78, p <.05. Pairwise comparisons indicate reaction times following distin ctive features were significantly faster ( p <.05) than responses to neutral primes, shared feature primes, and combination primes (distinctive sh ared or shared distinctive). The main effect of prime condition on naming vehicles was significant F (3, 598.88) = 5.47, p <.01. Pairwise comparisons revealed that reaction times following distinctive features were significantly ( p =.001) faster than reaction times following shared features. Results for research question one indicate similar priming patterns for th e younger and older adults. Specifically, distinctive features have greater priming effects than shared thus, validating the use of this picture naming priming task with older adults. Given that older adults are the population of interest, only results f rom the older adult group will be presented and discussed for the remaining research questions. Research Question 2 Does SRT for naming animals, tools, or vehicles change linearly as number of shared features changes from 1 2? A linear mixed effects mod el was conducted for each category to determine the main effect of prime condition on SRT of picture

PAGE 46

46 naming. Each model included frequency and prime condition as fixed factors and participant as a random factor. Two prime conditions were entered into the model, shared (shared blank, blank shared) and shared shared to compare the effects of the number of shared features. Table 3 3 and Figure 3 3 display the mean SRT and standard error for the two prime conditions within each semantic category. The analy sis of SRT of naming animals did not produce a significant main effect of prime condition, F (1, 211.38) <1. Likewise, the main effect of prime condition was not significant for SRT of naming tools, F (1, 220.24) <1 or vehicles, F (1, 183.80) <1. Research Question 3 Does SRT for naming animals, tools, or vehicles change linearly as number of distinctive features changes from 1 2? A linear mixed effects model was conducted for each category to determine the main effect of prime condition on SRT of picture naming. Each model included frequency and prime condition as fixed factors and participant as a random factor. Two prime conditions were entered into the model, distinctive (shared blank, blank shared) and distinctive distinctive to compare the ef fects of the number of distinctive features. Table 3 4 and Figure 3 4 d isp lay the mean and standard error of SRT for the two prime conditions within each target semantic category. The main effect of prime condition was not significant for SRT of animal picture naming, F (1, 235.59) <1. Similarily, the main effect of prime condition was not significant for tools, F (1, 226.62) < 1, or vehicles, F (1, 205.32) <1.

PAGE 47

47 Research Question 4 Is there a difference in priming effects between 200 and 600msec ISI? The comparisons of priming conditions in research questions 1 3 were conducted with ISI as a fixed factor and participant as a fandom factor to determine the main effect of ISI on SRT of picture naming. Tables 3 5 through 3 7 and figures 3 5 through 3 7 d isplay means SRTs with standard errors for comparisons of ISI. Comparison of s hare d, d istinctive, c ombination, and n eutral primes did not produce significant effects of ISI for animals, F (1, 663.65) = 1.704, p =.192, tools F (1, 663.60)<1, or vehicles, F (1 586.59) = 1.74, p =.188. Comparisons of shared and shared shared conditions produced a significant main effect of ISI for animals, F ( 1, 197.55 ) = 4.89, p <.05. Pairwise comparisons revealed that reactions times in the 600 ISI condition were significantly ( p <.05) faster than reaction times in the 200 ISI condition. Responses to tools did not produce a significant main effect of ISI, F (1, 208.60) = 1.44, p =.23, neither did response to vehicles, F ( 1, 179.82 ) < 1. Likewise, comparisons of distinctive and distinctive distinctive prime conditions did not result in significant main effects of ISI for animals, F ( 1, 228.9 ) = 1.43, p =.23; tools, F (1, 222.70)<1; or vehicles, F ( 1, 210.33 ) = 2.95, p =.09. Word Frequency Effects When comparing SRTs from prime conditions of shared, distinctive, combined and neutral primes, only responses to animal targets revealed significant effects of frequency, F ( 2, 665.90 ) = 6.43, p <.05. Pairwise comparisons showed that reaction times t o high frequency animals were faster than reaction times to medium frequency animals. Frequency did not have a significant effect on SRT for picture naming of tools F ( 1, 666.76 ) <1, or vehicles F ( 2, 589.10 ) = 2.09, p =.12.

PAGE 48

48 Comparison of SRTs of picture n aming when primes were one shared feature and two shared features revealed no significant effects of frequency. The results from neither animals F (2, 194.88) = 2.87, p=.059, tools F (1, 205.54) = 1.05, p =.307 nor vehicles, F ( 2, 179.22 ) <1 was significa following primes of one distinctive and two distinctive features, resulted in significant effects from only animal targets, F ( 2, 230.88 ) = 4.18, p <.05. Pairwise comparisons indicate reaction times to high frequ ency animals were significantly ( p <.05) faster than reaction times to medium frequency animals. Frequency effects from neither tools F (1, 22 0.04) <1 nor vehicles, F (2, 204.27) = 1.24, p =.293 was significant.

PAGE 49

49 Table 3 1. Raw mean SRT for younger adults for comparison of neutral, shared features, distinctive features, and combination of shared and distinctive features prime conditions. Asterisks indicate significant pairwise comparisons which are described in parentheses. Category Prime Condition Raw Mean SRT Standard Error Animals Neutral 677.83 235.70 Shared 695.18 541.28 Distinctive* (Faster than neutral primes, p <.05 Faster than shared primes, p =.001) 637.81 308.88 Distinctive & Shared 665.34 331.52 Tools Neutral 775.27 424.63 Shared 783.72 483.57 Distinctive 767.27 466.62 Distinctive & Shared 768.28 354.28 Vehicles Neutral 722.76 245.66 Shared 687.80 259.64 Distinctive 679.26 365.92 Distinctive & Shared 715.57 304.34

PAGE 50

50 Table 3 2. Raw mean SRT for older adults from comparison of neutral, shared features, distinctive features, and combination of shared and distinctive features prime conditions. Asterisks indicate significant pairwise comparisons which are described in parentheses. Category Prime Condition Raw Mean SRT Standard Error Animals Neutral 661.38 207.21 Shared 695.52 272.11 Distinctive* (Faster than shared features, p =.001) 634.34 303.19 Distinctive & Shared 647.25 243.64 Tools Neutral 769.70 303.69 Shared 706.65 276.82 Distinctive* (Faster than neutral primes, p <.05 Faster than shared features, p <.05 Faster than combined features, p <.05) 665.89 254.50 Distinctive & Shared 771.42 360.65 Vehicles Neutral 701.04 236.43 Shared 788.50 450.97 Distinctive* (Faster than shared primes, p =.001) 656.21 307.39 Distinctive & Shared 656.18 227.95

PAGE 51

51 Table 3 3. Raw mean SRT from comparison of shared versus shared shared prime conditions. There were no significant effects of prime condition. Category Prime Condition Raw Mean SRT Standard Error Animals Shared 696.30 270.81 Shared & Shared 694.25 275.83 Tools Shared 713.52 304.50 Shared & Shared 690.18 196.16 Vehicles Shared 774.75 412.43 Shared & Shared 810.19 507.90

PAGE 52

52 Table 3 4. Raw mean SRT from comparison of distinctive and distinctive distinctive prime conditions. There were no significant effects of prime condition. Category Prime Condition Raw Mean SRT Standard Error Animals Distinctive 625.89 289.44 Distinctive & Distinctive 651.35 330.27 Tools Distinctive 669.67 248.42 Distinctive & Distinctive 658.38 267.60 Vehicles Distinctive 671.48 325.08 Distinctive & Distinctive 630.53 275.01

PAGE 53

53 Table 3 5. Raw mean SRT from comparison of 200msec and 600msec ISI for the prime conditions of neutral, shared features, distinctive features, and combination of shared and distinctive features. There were no significant effects of ISI. Category ISI Raw Mean SRT Standard Error Animals 200 679.62 294.65 600 638.87 246.72 Tools 200 713.23 281.64 600 715.41 311.25 Vehicles 200 711.60 330.48 600 695.20 360.63

PAGE 54

54 Table 3 6. Raw mean SRT from comparison of 200msec and 600msec ISI for the shared feature prime conditions. Asterisks indicate significant pairwise comparisons which are described in parentheses. Category ISI Raw Mean SRT Standard Error Animals 200 740.33 303.57 600* (Faster than 200ISI, p <.05) 653.85 232.88 Tools 200 702.39 249.74 600 711.18 304.03 Vehicles 200 799.11 464.01 600 777.35 438.95

PAGE 55

55 Table 3 7. Raw mean SRT from comparison of 200msec and 600msec ISI for the distinctive feature prime conditions. There were no significant effects of ISI. Category ISI Raw Mean SRT Standard Error Animals 200 657.65 325.99 600 609.32 275.77 Tools 200 654.86 210.44 600 673.55 281.56 Vehicles 200 667.80 240.45 600 642.39 372.62

PAGE 56

56 Figure 3 1. Log transformed m ean SRT of young adult responses from comparison of neutral, shared, distinctive and combined prime conditions. Brackets indicate significant pairwise comparisions. Asterisks indicate the prime condition which produced significantly faster SRTs.

PAGE 57

57 Figure 3 2. Log transformed m ean SRT of older adult responses fr om comparison of neutral, shared, distinctive and combined prime conditions Brackets indicate significant pairwise comparisions. Asterisks indicate the prime condition which produced significantly faster SRTs.

PAGE 58

58 Figure 3 3. Log transformed mean SRT of older adult responses from comparison of shared and shared shared prime conditions. There were no significant effects of prime condition.

PAGE 59

59 Figure 3 4. Log transformed mean SRT of older adult responses from comparison of distinctive and distinctive distinctive prime conditions. There were no signifi cant effects of prime condition.

PAGE 60

60 Figure 3 5. Log transformed mean SRT of older adult responses to all prime conditions (collapsed across shared, distinctive, combined, and neutral primes) at 200 ISI versus 600 ISI. There was no significant difference i n SRT between ISI conditions.

PAGE 61

61 Figure 3 6. Log transformed mean SRT of older adult responses to shared primes (collapsed across shared and shared shared) primes at 200 ISI versus 600 ISI. Brackets indicate significant pairwise comparisions. As terisks indicate the prime condition which produced significantly faster SRTs.

PAGE 62

62 Figure 3 7. Log transformed mean SRT of older adult responses to distinctive primes (collapsed across distinctive and distinctive distinctive) primes at 200 ISI versus 600 ISI. There was no significant difference in SRT between ISI conditions.

PAGE 63

63 CHAPTER 4 DISCUSSION The purpose of this study was to investigate the roles of shared and distinctive features, as proposed by the CSA, in conceptual activation. The principles of the CSA could be used to further specify the stimuli used in anomia treatments however, it is unknown if shared and distinctive features can be used to activate concepts because previous studies have used concept to concept activation and not feature to c oncept activation, the route used in anomia treatments. Additionally, there is evidence that more time is necessary to allow convergence of meanings from multiple primes which in turn leads to increased priming and semantic facilitation. Thus, the curren t experiment was conducted at two inter stimulus intervals. Overall, the results showed that indeed, as stated in the CSA, distinctive features have a privileged role in concept activation. However, the proposed differential roles of shared and distinctiv e features in living and nonliving things, was not confirmed. Additionally, the results indicate that more time was not necessary to consolidate the information from two primes. The results are discussed below with respect to theoretical and clinical imp lications. Priming Effects from Related Features and Neutral Primes SRT was compared in priming conditions of related features (shared, distinctive, or combined) and neutral primes ( blank ) to establish the validity of feature to concept priming. Only th e tools category revealed a significant difference in SRTs when primes blank This is most likely due to the difference in frequency between tools and; animals and vehicles. As seen in appendix B animals and vehicles had greater average frequencies (21.39 and 86.16,

PAGE 64

64 and vehicles left little need for feature primes resulting in non significant differences in reduced exposure to tools allowed the feature primes to boost concept activation. Most importantly, there were significant differences in the priming effects of the feature t ypes for all semantic categories. The Effect of Feature Type The first analysis investigated the hypothesized difference between the roles of shared and distinctive features, in the representation of living and nonliving things, by measuring the differen ce in SRT for picture naming following primes of shared or distinctive features or a combination thereof. Based on the CSA (Tyler & Moss, 2001; Taylor et al., 2007), it was predict ed that in living nouns (animals and vehicles), priming with shared feature s would result in shorter SRTs than priming with distinctive features, due to greater spread of activation from strongly correlated shared features and weaker correlation of distinctive features to other features. In the activation of non living nouns (to ols), distinctive features were predicted to boost performance over shared features due to the stronger correlations and greater number of distinctive features than shared features. For all domains of nouns, a combination of shared and disti nctive feature s was predicted to be the optimal condition for performance because the semantic category is activated as well as a feature unique to that concept which will lead to identification of the concept from its category neighbors. Table 1 1 displays these predi ctions. Overall, the predictions of the current study were not confirmed. Analysis of reaction times of animal, tools, and vehicle picture naming revealed that distinctive

PAGE 65

65 feature primes consistently had a greater effect compared to shared feature or c ombination of shared and distinctive feature primes. Although, the CSA proposes that distinctive features play a critical role in concept activation, the level of information about a concept carried by distinctive features is proposed to be different in l iving and nonliving things. The CSA claims that because of weaker correlations between distinctive features and other features, distinctive features are less informative of living things than shared features with stronger correlations to other features; c onversely, distinctive features of nonliving things are more informative and have stronger correlations to other features than shared features of nonliving things. Consequently, it was not surprising that in activation of tools, distinctive features produ ced the fastest SRTs. Although, the same result in living things was surprising it is not unique to this study. Cree, McNorgan and McRae (2006) reported identical results. They used two feature verification tasks, one with concept primes and feature tar gets and importantly one with feature primes and concept targets (as in the current study). Distinctive features consistently led to faster feature verification regardless of the conceptual classification as living or nonliving. Additionally, Cree et al. conducted a computer simulation similar to a feature verification task. The simulation showed that distinctive features activated the target concepts more strongly than did shared features. The findings of Cree et al. and the current study may be attribut ed to the high level of distinctiveness used in both studies. T he distinctive features used in this study, and those used by Cree et al. (2006) were not only distinctive but were unique to a single concept, thus exposure to this information alone (withou t a shared feature) could activate one concept only. For

PAGE 66

66 example, the distinctive features of quacks, cocoon or udders activate only the concepts of duck, butterfly or cow respectively. Likewise, the distinctive features of vehicles used in this study were also unique to each concept (e.g. hovers helicopter, mast sailboat, caboose train ) resulting in only one possible concept. Thus, it appears that when a feature is unique to a concept and not ambiguous, it will activate the concept without need of act ivating the category explicitly, through shared features. Furthermore, some of the shared features used in this study had very high concept production frequencies. As described in the methods, concept production frequency is the count of how many times the feature was listed in the norms by McRae et al. (2005). Shared features with high concept production frequencies may have provided minimal information on category membership. For example large was seen for animals as well as vehicles, consequently, la rge supplied minimal information regarding the target. In comparison, metal a shared feature for tools, at the least indicated the target was a nonliving thing, thereby activating fewer semantic neighbors than large which could refer to a living or nonli ving thing. This raises an important issue which should be addressed in future studies investigating the CSA. It appears that the proposed distinction, between the roles of shared and distinctive features in living and nonliving things, is relative to the level of distinctiveness of the feature. A highly shared feature does little to activate the semantic category or neighborhood, of a living or nonliving thing. On the other hand, a highly distinctive feature, of a living or nonliving thing, can a ctivate the concept directly; or, has stronger connections than proposed, to activate related features and lead to

PAGE 67

67 activation of the concept. Further specifying the effects of different levels of shared and distinctive features is critical to the applicab ility of the CSA. The most surprising finding was the lack of a significant difference in reaction times following, combined shared and distinctive feature primes, compared with distinctive feature primes alone, or shared feature primes. It was hypothesiz ed that a combination of shared and distinctive features would result in the fastest SRT because the participant would have information on category membership and information unique to the target concept prior to seeing the target. The results however, su ggest that shared features did little to activate the concept, whether presented alone or in combination with a distinctive feature. One explanation is that, shared features activated large sets of semantic neighbors which could not be inhibited by the a ctivation of the distinctive feature. In recent work, Mirman and Magnuson (2009) explain this effect at the featural level using the mechanisms of excitation and inhibition, as presented in the semantic neighborhood model. In the process of activating li on, shared features which also belong to tiger are activated (e.g. predator, teeth, wilderness, and circus ) thus, tiger is partially activated. The activation of distinctive features of lion ( mane ) leads to inhibition of tiger while adding to the excitation of lion In regards to the features of the current study, the shared feature prime of four legs would activate most animals to some degree. If the distinctive feature of humps has weak correlations to other features of cam el as suggested by the CSA, it may not provide enough additional activation to signal the system to inhibit all other four legged animals while further exciting camel Thus, the

PAGE 68

6 8 distinctive feature of hump, seen without a shared feature, was just as effe ctive in activating camel The Effect of Number of Features The second analysis investigated the effects of multiple shared features on SRT of picture naming to determine if number of features ha d an effect on concept activation. The findings for anim als, tools, and vehicles were identical; there was no difference in priming effects from one or two shared features, or from one or two distinctive features. For living things (animals and vehicles), it was predicted that multiple shared features would i ncrease priming because these features have strong connections with other features; as such, direct activation of several features should result in activation of several other features. However, it appears that this activation did not reach a measurable l evel. The prediction for nonliving things (tools) was that multiple shared features would not affect priming because these features have weak connections to other features. Results from analyses of research question one showed that shared features with o r without a distinctive feature did not significantly prime picture naming for any category. Therefore it is not surprising that increasing the number of shared features shows no difference. If four legged activated animals the addition of fur does littl e to focus the activation within the semantic neighborhood of animals given so many four legged animals have fur. The CSA model states that shared features reflect category membership only; these results suggest that multiple references to the category do not increase activation of that category. In regards to the effects of multiple distinctive features, it was predicted that for activation of tools, increasing the number of distinctive features would increase priming because these features have strong connections to other features; thus, activating

PAGE 69

69 several features would result in activation of several other features prior to viewing the target. On the other hand, for animals and vehicles it was predicted that increasing the number of distinctive featu res would not increase priming because these features have weak connections to other features. In the present paradigm, there was no significant difference in SRT when the primes were one or two distinctive features for animals, tools, or vehicles. Given the findings in research question one analyses that activation from distinctive features was greater than from shared features or a combination of shared and distinctive, it appears that multiple distinctive features are not necessary to activate a concept in any category. If the distinctive feature is unique to the concept, and therefore, will only activate one concept, additional distinctive features are not necessary to boost activation. The null effects of multiple features have interesting implicatio ns for anomia treatments which will be address in clinical implications below. The Effect of Timing on Multiple Feature Primes The comparison of priming effects at 200 and 600 ISI was motivated by the work of Milberg and colleagues (2003) who found overadditive priming effects at 600 ISI and addivitve effecs at 200 ISI. Additive effects indicate the priming effect from two sequential primes is the same as adding the priming effect of each prime when presented individually; thus, there would be no si gnificant difference in SRTs from conditions with one or two primes. Overadditive effects suggest the priming from two sequential primes is greater than adding the priming effects of each prime presented individually; therefore, there would be a significa nt difference in SRT from conditions with one or two primes. Additionally, multiple primes can result in u nderadditive effects which occur when the facilitation of multiple primes is less than the total facilitation of each prime individually. The curren t findings suggest multiple feature primes produce

PAGE 70

70 underadditive effects, i.e. there is no benefit to presenting multiple feature primes. There is one exception to this finding. There were overadditive effects for naming animals when primed with only sha red features. The implications of these results are discussed below. In all prime conditions, except shared conditions, there was no difference in SRT when ISI was increased from 200msec to 600msec. This result is particularly interesting for the conditi ons containing a distinctive feature because in earlier analyses where ISI was collapsed, distinctive features produced the strongest priming effects. Consequently, it would seem plausible that this effect would increase over time. However, it appears th at distinctive features do not need more time to activate the concept. As discussed earlier, this may be attributed to the high level of distinctiveness of the distinctive feature primes. If the distinctive primes of periscope and underwater can only activate submarine, then more time is not needed; i.e., there are not several concepts which could emerge from the activation patterns of periscope and underwater which would require time to reach a stable state. Consequently, the priming effects from such highly distinctive multiple primes do not change over time. The one significant finding in analysis of ISI was in naming animals when primed with shared features; SRTs from the 600 ISI condition were faster than in the 200 ISI condition. Given t hat earlier analyses suggested the effects of shared features of any semantic category are not measurable in this task, this finding is surprising. The lack of priming effect from shared features in earlier analyses was partially attributed to the highly shared features used in this study (e.g. large or metal which both belong to many concepts). Again, the high shared value of the features may be implicated in the null

PAGE 71

71 effects of ISI. Highly shared features (e.g. legs or large ) may require more time to c onsolidate with other highly shared features (e.g. fur or feathers ) because of the large semantic neighborhood each feature activates. When each feature is activated, related features are partially activated. Highly shared features will have many related features; consequently, a large semantic neighborhood is partially activated. Much of this neighborhood must be inhibited in the process of activating the target. Thus, with more time (600msec versus 200msec) the system is able to inhibit the features u nrelated to the target and increase activation of the overlapping units from the shared primes. The Effect of Word Frequency In general, there was minimal influence of frequency on reaction times. Only SRTs to animal targets were affected by frequency, in the expected direction of faster reaction times following high frequency items. This effect was seen in analyses comparing the prime conditions of shared, distinctive, combined, and neutral; as well as analysis comparing shared and shared shared prime conditions. The lack of frequency effect for vehicles and tools compared to animals is likely due to differences in category size. There are many more animals in semantic memory than vehicles and tools. Word frequency may play a larger role in semantic activation when there is larger semantic neighborhood and therefore, more competitors. However, there were no significant interactions between frequency and prime condition; therefore the significant priming effects are valid regardless of frequency diff erences. Limitations The comparison of only one and two features, may have limited the effects of combining feature types. A difference in priming may be seen with a greater number of features. Specifically, the effects of multiple shared features may be measurable when

PAGE 72

72 comparing one shared feature to three or five shared features. Likewise the comparisons of one feature type to combined features are limited by having used only combinations of one shared and one distinctive. Comparing several shared a nd distinctive features, and even varying the number of each paired together (e.g. 1 shared, distinctive; 2 shared, 3 distinctive), could provide further insights into the level of information carried by each feature type. With regards to the effects of multiple primes over time, the findings are limited to the use of 200msec and 600msec. As mentioned earlier, the aim of neither the Milberg et. al (2003) study nor the present work was the delineation of automatic and controlled processes. Certainly, the mechanisms of feature to concept activation, and the roles of shared and distinctive features would be further clarified by comparing the results from priming paradigms administered under controlled and automatic processes. However, the goal of the curre nt study was to establish the validity of feature to concept activation and test the role of shared and distinctive features in such a paradigm. Additionally, the current study did not employ a timed response. Randall et al. (2004) found slower activati on from distinctive features of living things compared to distinctive features of nonliving things only when there was a response deadline. In an untimed version of the task there was no difference. Consequently, the differential roles of shared and dist inctive features in living and nonliving things may be time dependent. The interaction of time and feature to concept activation is important for the clinical application of the CSA; however, more empirical evidence of an interaction is required. Implic ations for Anomia Treatment Two issues addressed in the current study have implications for anomia treatment: the use of multiple features for activation of a concept and the effects of shared and

PAGE 73

73 distinctive features on concept activation. In regards to the former, the current study would suggest that asking a patient to name several features of a target concept or providing the patient with multiple features of a target is not the optimal condition for concept activation. In fact, the work of Mirman and Magnuson (2008) would further suggest that in the case of shared features, especially, multiple primes will lead to semantic competition. Consequently, semantic feature treatments may be improved in terms of treatment outcomes, if the number of semantic questions per target item were reduced. The second issue regarding clinical treatments which the current findings can address is the feature type used in treatment. Clearly the current findings suggest that distinctive features have the greatest effects on feature to concept activation. The findings by Mirman and Magnuson (2008) regarding inhibitory effects of near neighbors and facilitative effects of distant neighbors, as well as the CATE approach (Thompson et. al, 2003), similarly suggest that distinc tive features are more beneficial for feature to concept activation. Certainly, semantic feature treatments do include activation of distinctive features; however, a more explicit focus on these features may improve treatment outcomes. Future Directions The Conceptual Structure Account (Tyler & Moss, 2001; Taylor et al., 2007) is currently modeled in healthy adults and individuals demonstrating category specific semantic deficits resulting from herpes simplex encephalitis or other degenerative disease. Extending this theory to aphasia may add specificity to the claims of the CSA regarding the structure of the semantic system. Specifically, a comparison of individuals with aphasia and individuals with semantic dementia on a paradigm similar

PAGE 74

74 to the curre nt experiment may be useful in specifying effects of feature correlation on activation. Persons with a degrading semantic system may not be able to use feature correlations as well as someone with lexical semantic access impairment. In addition, several question related to methodology remain. Answering these questions will further specify the claims of the CSA: Are the results of this study different when there is a response deadline? How do the roles of shared and distinctive features compare when used i n automatic versus controlled processing? Does order of prime type (shared distinctive, distinctive shared) affect SRT in a picture naming paradigm? Do other psycholinguistic variables (number of phonemes, phonotactic probability, familiarity etc.) of the target items affect the difference in SRT between prime conditions? Does manipulation of the concept production frequencies (i.e. manipulating the degree to which features are distinctive or shared) change the effect of the prime conditions on SRT? Summary In a series of analyses of speech reaction times following multiple feature primes during a picture naming task, the most consistent finding was greater priming after distinctive feature primes as compared to shared primes, distinctive and shared primes, or neutral primes. One of the most central hypotheses of the Conceptual Structure Account (Tyler & Moss, 2001; Taylor et al., 2007) is the unique and critical role of distinctive features in semantic space and conceptual activation. The current study sup ports this claim and validates it in verbal production, a language behavior not previously used in investigations of the CSA. The lack of support for the postulates of the CSA regarding differences in the distribution and correlations of shared and distin ctive features in living and nonliving things, suggests further research is required to

PAGE 75

75 elucidate these relationships. The unique contribution of the current work is the strong evidence that features prime concepts; suggesting, features can be used to acti vate the semantic network and indeed different feature types activate the network to varying degrees.

PAGE 76

76 APPENDIX A PRIME TARGET STIMULI Table A 1. Animal Stimuli Concept Feature Feature Type CPF butterfly pollinates D 1 cocoon D 1 flies S 46 small S 121 camel spits D 1 two humps D 1 legs S 44 large S 106 chicken clucks D 1 pecks D 1 wings S 44 edible S 78 cow udder D 1 produces milk D 2 four legs S 49 large S 106 duck quacks D 1 waddles D 2 feathers S 38 edible S 78 elephant trunk D 1 tusks D 2 four legs S 49 large S 106 ostrich large eggs D 1 buries head D 1 wings S 44 large S 106 pig squeals D 1 curly tail D 1 four legs S 49 edible S 78 tiger roars D 2 large teeth D 2 ferocious D 2 fur S 22 large S 106 D indicates distinctive features S indicates shared features. CPF is the concept production frequenc y.

PAGE 77

77 Table A 2. Tool Stimuli Concept Feature Feature Type CPF airplane crashes D 1 engine D 2 large S 106 metal S 133 bus many seats D 1 fare D 1 transportation S 33 large S 106 car four doors D 1 steering wheel D 1 wheels S 23 transportation S 33 motorcycle two people D 2 helmet D 2 fast S 33 loud S 34 canoe paddles D 1 tips over D 1 wood S 79 long S 81 helicopter hovers D 1 propellers D 2 loud S 34 metal S 133 sailboat mast D 1 wind D 2 floats S 7 transportation S 33 skateboard board D 1 tricks D 1 wood S 79 long S 81 submarine periscope D 1 underwater D 1 large S 106 metal S 133 train caboose D 1 conductor D 1 fast S 33 transportation S 33 unicycle balance D 1 one wheel D 2 seat S 7 transportation S 33 D indicates distinctive features S indicates shared features. CPF is the concept production frequen cy.

PAGE 78

78 Table A 3. Vehicle Stimuli Concept Feature Feature Type CPF axe metal blade D 2 chopping D 2 heavy S 27 handle S 42 drill bits D 1 makes holes D 1 electrical S 22 loud S 34 hammer hits nails D 1 metal head D 2 heavy S 27 handle S 42 hoe tilling soil D 1 metal bland D 2 handle S 42 long S 81 rake gardening D 1 prongs D 2 long S 81 metal S 133 scissors blades D 1 cuts D 1 sharp S 18 metal S 133 screwdriver tightens D 1 turns D 2 handle S 42 long S 81 shovel digging D 1 scooping D 1 handle S 42 metal S 133 wrench tightens bolts D 1 turns bolts D 2 heavy S 27 metal S 133 D indicates distinctive features S indicates shared features. CPF is the concept production frequenc y.

PAGE 79

79 APPENDIX B TARGET WORD FREQUENCY Table B 1 Taret Word Frequency Category Target Brysbaert and New (2009) Frequency (per million words) Frequency Category Animals Butterfly 5.51 Low Camel 5.02 Low Chicken 61.73 High Cow 25.51 High Duck 24.76 High Elephant 11.37 Medium Ostrich 0.94 Low Pig 39.14 High Tiger 18.53 Medium Average 21.39 Tools Axe 4.88 Low Drill 13.75 Medium Hammer 12.47 Medium Hoe 0.92 Low Rake 2.98 Low Scissors 6.69 Low Screwdriver 0.06 Low Shovel 6.84 Low Wrench 3.96 Low Average 5.96 Vehicles Bus 74.18 High Canoe 3.57 Low Car 483.06 High Helicopter 15.80 Medium Motorcycle 8.92 Low Sailboat 1.61 Low Submarine 7.10 Low Train 95.06 High Average 86.16

PAGE 80

80 LIST OF REFERENCES Balota, D. A., & Paul, S.T. (1996). Summation of activation: Evidence from multiple primes that converge and diverge within semantic memory. Journal of Experimental Psychology Learning, Memory, Cognition, 22 (4), 827 8 45 Bates, E., D'Amico, S., Jacobsen, T., Szkely, A., Andonova, E., Devescovi, A., Herron, D., Lu, C C., Pechmann, T., Plh, C., Wicha, N., Federmeier, K., Gerdjikova, I., Gutierrez, G., Hung, D., Hsu, J., Iyer, G., Kohnert, K., Mehotcheva, T., Orozco Figueroa, A., Tzeng, A., & Tzen g, O. (2003). Timed picture naming in seven languages P sychonomic Bulletin & Review 10 ( 2 ), 344 380 Boyle, M., & Coehlo, C.A. ( 1995 ) Application of semantic features analysis as a treatment for aphasic dysnomia. American Journal of Speech Language Pathology, 4 (4), 94 98 Brysbaert, M., & New, B. (2009). Moving beyond Kucera and Francis: A critical evaluation of current word frequen cy norms and the introduction of a new and improved word frequency measure for American English. Behavior Research Methods, 41 (4), 977 990 Bueno, S., & Frenck Mestre, C. (2008). The activation of semantic memory: Effects of prime exposure, prime target re lationship, and task demands. Memory & Cognition, 36 (4), 882 898 Cree, G. McNorgan, C., & McRae, K. (2006). Distinctive features hold a privileged status in the computation of word meaning: Implications for theories of semantic memory. Journal of Experime ntal Psychology, 32 (4), 643 658 Cree, G., McRae, K., & McNorgan, C. (1999). An attractor model of lexical conceptual processing: Simulating semantic priming. Cognitive Science: A Multidisciplinary Journal, 23 (3), 371 414 de Groot, A. M. (1984). Primed l exical decision: Combined effects of the proportion of related prime target pairs and the stimulus onset asynchrony of prime and target. The Quarterly Journal of Experimental Psychology A: H uman Experimental Psychology, 36 (2): 253 280 de Groot, A.M., Thom assen, A.J., & Hudson, P.T. (1982). Associative facilitation of word recognition as measured from a neutral prime. Memory & Cognition, 10 : 358 370 Folstein M.F., Folstein S.E., & McHugh P.R. (1975). Mini mental state: A practical method for grading the co gnitive state of patients for the clinician. Journal of psychiatric research, 12 (3): 189 98 Grober E., & Sliwinski M. (1991). Development and validation of a model for estimating premorbid verbal intelligence in the elderly. Journal of Clinical and Ex perimental Neuropsychology, 13 933 949

PAGE 81

81 Hampton, J. A. (1979). Polymorphous concepts in semantic memory. Journal of Verbal Learning & Verbal Behavior. 18 (4), 441 461 Howard D, & Patterson K. (1992). Pyramids and palm trees: A test of semantic access from pictures and words Bury St Edmunds, Suffolk: Thames Valley Test Company. Hutchison, K. A. ( 2003 ). Is semantic priming due to association strength or feature overlap? A microanalytic review. Psychonomic Bulletin & Review. 10 (4), 785 813 Kiran, S., & Thompson, C.K. ( 2003 ). The role of semantic complexity in treatment of naming deficits: Training semantic categories in fluent aphasia by controlling exemplar typicality. Journal of Speech, Language, and Hearing Research, 46, 773 787 Larochelle, S., & Pineau H.I. ( 1994 ). Determinants of response times in the semantic verification task. Journal of Memory and Language. 33 (6), 796 823 McNamara, T. (2005). Semantic priming: Perspectives from memory and word recognition Psychology Press. McRae, K., Cree, G.S., Seidenberg, M.S., and McNorgan, C. (2 0 0 5 ) Semantic feature production norms for a large set of living and nonliving things. Behavior Research Methods, Instruments, & Computers, 37, 547 559 McRae, K., Cree, G.S., Westmacott, R., & de Sa, V.R. ( 1999 ) Further evidence for feature correlations in semantic memory. Canadian Journal of Experimental Psychology, 53, 360 373 McRae, K., de Sa, V.R., and Seidenberg, M.S. ( 1997 ) On the nature and scope of featural representations of word meaning. Journal o f Experimental Psychology: Genera, 126, 99 130 Mervis, C., Catlin, J., & Rosch, E. ( 1976 ) Relationships among goodness of example, category norms, and word frequency. Bulletin of the Psychonomic Society 7 (3), 283 284 Meyer, D. E., Schvaneveldt, R.W., Ruddy, M.G ( 1974 ). Functions of graphemic and phonemic codes in visual word recognition Memory & Cognition. 2 (2), 309 321 Milberg, W., Blumstein, S., Sullivan Giovanello, K., & Misiurski, C. ( 2003 ) Summation priming i n aphasia: Evidence for alterations in semantic integration and activation. Brain and Cognition 51 (1), 31 47 Mirman, D., & J. S. Magnuson (2009). Dynamics of activation of semantically similar concepts during spoken word recognition. Memory & Cognition. 37 (7), 1026 1039

PAGE 82

82 Mirman, D., & J.S. Magnuson (2008). Attractor dynamics and semantic neighborhood density: Processing is slowed by near neighbors and speeded by distant neighbors. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34 (1), 65 79. Moss, H., Ostrin, R., Tyler, L., & Marslen Wilson, W. ( 1995 ) Accessing different types of lexical semantic information: Evidence from priming. Journal of Experimental Psychology: Learning, Memory, and Cognition 21 (4), 863 883 Munakata, Y., & Pfa ffly, J. (2004). Hebbian learning and development. Developmental Science 7 (2), 141 148. Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of current findings and theories. In D. Besner & G. Humphreys (Eds.), Basi c processes in reading: Visual word recognition (pp. 264 336). Hillsdale, NJ: Erlbaum. Nelson, H.E. (1982). National Adult Reading Test (NART ). Windsor, Berkshire, England: The NFER NELSON Publishing Company. Nickels, L. (2002). Therapy for naming disorders: Revisiting, revising, and reviewing, Aph asiology 16 (10), 935 979 Plaut, D. C. (1996). Relearning after damage in connectionist networks: Toward a theory of rehabilitation." Brain and Language. 52 (1) 25 82 Randall B., Moss H.E., Rodd J.M., Gr eer M., & Tyler L.K.. (2004) Distinctiveness and correlation in conceptual structure: Behavioral and computational studies. Journal of Experimental Psychology Learning Memory Cognition, 30 393 406 Rosch, E. (1975). Cognitive represen tations of semantic c ategories. Journal of Experimental Psychology: General. 104 (3), 192 233 Seidenberg, M., Waters, G., Sanders, M., & Langer, P. (1984). Pre and postlexical loci of contextual effects on word recognition. Memory & Cognition 12 (4), 315 328 Taylor, K.I., Mo ss, H.E., Tyler, L.K. ( 2007 ) The conceptual structure account: A cognitive model of semantic memory and its neural instantiation. In J. Hart, Jr., & M.A. Kraut (Ed s .), N eural Basis of Semantic Memory, (pp 265 301). Cambridge: Cambridge University Press. T hompson, C.K., Shapiro, L.P., Kiran, S., & Sobeck, J. 2003. The role of syntactic complexity in treatment of sentence deficits in agrammatic aphasia: The complexity account of treatment efficacy (CATE). Journal of Speech, Language, and Hearing Research, 46 591 607 Torgesen, J.K., Wagner, R. K., & Rashotte, C.A. ( 1999 ) Test of Word Reading Efficiency Austin, TX: PRO ED Publishing, Inc.

PAGE 83

83 Tyler, L.K., & Moss, H.E. ( 2001 ) Towards a distributed account of conceptual knowledge. Trends in Cognitive Science s, 5 2, 244 252 Wierenga, C., Perlstein, W., Benjamin, M., Leonard, C., Gonzale z Rothi, L., & Conway, T. ( 2 009 ) Neural substrates of object ide ntification: Functional magnetic resonance imaging evidence that category and visual attribute contribute to semantic knowledge. Journal of the International Neuropsychological Society 15 (2), 169 181

PAGE 84

84 BIOGRAPHICAL SKETCH sciences and disorders from the University of Florida in 2004 and 2006, respectively. She received clinical training as a speech language pathologist at the Malcom Randall Veterans Affairs Medical Center in Gainesville, FL. She has been involved in research at the VA Brain Rehabilitation Research Center since her undergraduate years. She has taught undergraduate and graduate level courses at the University of Florida and the University of Washington. She plans to continue research and teaching in speech language pathology as an academician.