Citation
Choosing Remediation Targets for Naming Deficits in Probable Alzheimer Disease: Does Typicality Matter?

Material Information

Title:
Choosing Remediation Targets for Naming Deficits in Probable Alzheimer Disease: Does Typicality Matter?
Creator:
MORELLI, CLAUDIA A.
Copyright Date:
2008

Subjects

Subjects / Keywords:
Alzheimers disease ( jstor )
Aphasia ( jstor )
Birds ( jstor )
Dementia ( jstor )
Gene therapy ( jstor )
Learning ( jstor )
Memory ( jstor )
Semantics ( jstor )
Vegetables ( jstor )
Words ( jstor )
City of Gainesville ( local )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Claudia A. Morelli. Permission granted to University of Florida to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
8/31/2006
Resource Identifier:
649814467 ( OCLC )

Downloads

This item has the following downloads:

morelli_c ( .pdf )

morelli_c_Page_134.txt

morelli_c_Page_104.txt

morelli_c_Page_103.txt

morelli_c_Page_060.txt

morelli_c_Page_076.txt

morelli_c_Page_066.txt

morelli_c_Page_028.txt

morelli_c_Page_131.txt

morelli_c_Page_077.txt

morelli_c_Page_099.txt

morelli_c_Page_047.txt

morelli_c_Page_136.txt

morelli_c_Page_090.txt

morelli_c_Page_043.txt

morelli_c_Page_057.txt

morelli_c_Page_075.txt

morelli_c_Page_048.txt

morelli_c_Page_023.txt

morelli_c_Page_149.txt

morelli_c_Page_081.txt

morelli_c_Page_054.txt

morelli_c_Page_061.txt

morelli_c_Page_133.txt

morelli_c_Page_045.txt

morelli_c_Page_073.txt

morelli_c_Page_102.txt

morelli_c_Page_087.txt

morelli_c_Page_035.txt

morelli_c_Page_148.txt

morelli_c_Page_005.txt

morelli_c_Page_100.txt

morelli_c_Page_070.txt

morelli_c_Page_022.txt

morelli_c_Page_144.txt

morelli_c_Page_004.txt

morelli_c_Page_069.txt

morelli_c_Page_129.txt

morelli_c_Page_105.txt

morelli_c_Page_120.txt

morelli_c_Page_037.txt

morelli_c_Page_109.txt

morelli_c_Page_093.txt

morelli_c_Page_115.txt

morelli_c_Page_106.txt

morelli_c_Page_013.txt

morelli_c_Page_083.txt

morelli_c_Page_119.txt

morelli_c_Page_008.txt

morelli_c_Page_068.txt

morelli_c_Page_082.txt

morelli_c_Page_071.txt

morelli_c_Page_123.txt

morelli_c_Page_080.txt

morelli_c_Page_143.txt

morelli_c_Page_056.txt

morelli_c_Page_074.txt

morelli_c_Page_137.txt

morelli_c_Page_011.txt

morelli_c_Page_089.txt

morelli_c_Page_025.txt

morelli_c_Page_145.txt

morelli_c_Page_097.txt

morelli_c_Page_114.txt

morelli_c_Page_042.txt

morelli_c_Page_094.txt

morelli_c_Page_122.txt

morelli_c_Page_112.txt

morelli_c_Page_046.txt

morelli_c_Page_101.txt

morelli_c_Page_065.txt

morelli_c_Page_039.txt

morelli_c_Page_139.txt

morelli_c_Page_010.txt

morelli_c_Page_111.txt

morelli_c_Page_026.txt

morelli_c_Page_126.txt

morelli_c_Page_027.txt

morelli_c_Page_118.txt

morelli_c_Page_019.txt

morelli_c_Page_031.txt

morelli_c_Page_085.txt

morelli_c_Page_030.txt

morelli_c_Page_018.txt

morelli_c_Page_014.txt

morelli_c_Page_084.txt

morelli_c_Page_086.txt

morelli_c_Page_053.txt

morelli_c_Page_059.txt

morelli_c_Page_052.txt

morelli_c_Page_142.txt

morelli_c_Page_033.txt

morelli_c_Page_110.txt

morelli_c_Page_009.txt

morelli_c_Page_029.txt

morelli_c_Page_044.txt

morelli_c_Page_124.txt

morelli_c_Page_016.txt

morelli_c_Page_128.txt

morelli_c_Page_092.txt

morelli_c_Page_078.txt

morelli_c_Page_135.txt

morelli_c_Page_032.txt

morelli_c_Page_088.txt

morelli_c_Page_051.txt

morelli_c_Page_064.txt

morelli_c_Page_012.txt

morelli_c_Page_021.txt

morelli_c_Page_141.txt

morelli_c_Page_140.txt

morelli_c_Page_036.txt

morelli_c_Page_001.txt

morelli_c_Page_138.txt

morelli_c_Page_146.txt

morelli_c_Page_096.txt

morelli_c_Page_108.txt

morelli_c_Page_130.txt

morelli_c_Page_050.txt

morelli_c_Page_020.txt

morelli_c_Page_055.txt

morelli_c_Page_038.txt

morelli_c_Page_007.txt

morelli_c_Page_098.txt

morelli_c_Page_058.txt

morelli_c_Page_113.txt

morelli_c_Page_034.txt

morelli_c_Page_079.txt

morelli_c_Page_067.txt

morelli_c_Page_017.txt

morelli_c_Page_095.txt

morelli_c_Page_147.txt

morelli_c_Page_117.txt

morelli_c_Page_049.txt

morelli_c_Page_116.txt

morelli_c_Page_040.txt

morelli_c_Page_006.txt

morelli_c_Page_015.txt

morelli_c_Page_072.txt

morelli_c_Page_121.txt

morelli_c_pdf.txt

morelli_c_Page_003.txt

morelli_c_Page_107.txt

morelli_c_Page_091.txt

morelli_c_Page_063.txt

morelli_c_Page_132.txt

morelli_c_Page_041.txt

morelli_c_Page_062.txt

morelli_c_Page_002.txt

morelli_c_Page_127.txt

morelli_c_Page_125.txt

morelli_c_Page_024.txt


Full Text











CHOOSING REMEDIATION TARGETS FOR NAMING DEFICITS IN
PROBABLE ALZHEIMER DISEASE:
DOES TYPICALITY MATTER?














By

CLAUDIA A. MORELLI


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


2006

































Copyright 2006

by

Claudia A. Morelli


































This dissertation is dedicated to my incredible husband, David Efros, my amazing
mother, Claire Morelli and my sisters; and in memory of my wonderful father, John.
They have been an integral part of my life and their belief in me has been a fortress of
encouragement, hope, and steadfast love.
















ACKNOWLEDGMENTS

First, I thank and acknowledge each participant and their family and/or caregivers

for their time, effort, and for extending themselves and their home to allow me to work

with them to complete this research proj ect. I will always remember them and the

devastating effects of Alzheimer disease, and this will influence the focus of my future

clinical research. I also extend my gratitude and appreciation for the UF Memory

Disorder Program, especially David B. Efros, Dr. Kenneth M. Heilman, Dr. Glen Finney,

Dr. Kimford Meador, Dr. Catherine Price, and Jennifer Rembisz, whose kind and diligent

efforts resulted in referrals for me to recruit these participants.

I also express my deep appreciation for my committee. First, I want to thank my

mentor, Dr. Lori J.P. Altmann, for her vast knowledge of neurolinguistics that provided a

refreshing perspective, especially with regards to research in adults with probable

Alzheimer disease. Working with Dr. Altmann allowed me to broaden my thought

process and expand my approach to this dissertation. These qualities, combined with her

guidance, leadership, and patience have been of great value to me during the exciting and

challenging process of becoming a Ph.D. I also acknowledge Dr. Diane Kendall for her

expert opinion on clinical research in adult neurogenics, and for the many opportunities

to learn more about single subj ect design, especially during my pre-doctoral fellowship

with her at the VA Brain Rehabilitation Research Center. Her outstanding clinical skills

and perspective as a compassionate speech-language pathologist combined with her

insights have had a great influence on me and have motivated my interest in theoretically









driven treatments. I also gratefully acknowledge Dr. Ira Fischler who shared his extensive

knowledge of psychology, which played an important role in this dissertation. Dr.

Kenneth Heilman, with his behavioral neurology expertise combined with his incredible

commitment and compassion for patients, has taught me so much in such inspiring and

meaningful ways. Each of these remarkable individuals has contributed to my

development by their expertise and great role modeling.

I thank the Department of Communication Sciences and Disorders at the University

of Florida, especially Dr. Christine Sapienza, Dr. Samuel Brown, and Dr. Scott Griffiths,

for their support and encouragement, as well as the wonderful opportunities to teach the

undergraduates. Also, I will always remember the incredible opportunities to guest

lecture for Dr. Howard Rothman and Dr. Bonnie Johnson (including the lecture swaps

with Dr. Johnson). I extend my appreciation to the entire faculty and staff for their

professionalism, collegial, and positive approach. In particular, Idella King, Cassie

Mobley, Debbie Butler, and Addie Pons deserve a special thanks. The computer support

from Neal Musson will always be appreciated.

I also want to acknowledge the members of Dr. Altmann' s Language over the

Lifespan Lab. First I thank Alexia Frederiksen, who spent hours developing the stimuli,

coordinating and working with us, and whose attention to details was truly awesome. Her

dedication to the proj ect was sincerely appreciated and always will be. Additional thanks

go to Katie Chiarella, Andrea Holt, and Angel Ziesk for their wonderful help in preparing

the stimuli; and to Maisa Haj Tas, Ashley Mullen, and Deb Gober for lending their

incredible artistic talent. Alexia Frederiksen, Erin Hunt, and Rachel Hogue deserve

acknowledgement for their excellent questions and feedback during the protocol training,









which prepared them for testing the young adults for a related proj ect. Becca Huy also

played an important role by transcribing a large amount of the data and working closely

with me with on the reliability checks. Her devoted attention to the details of accurate

transcription, combined with her enthusiasm, and willingness to help was wonderful.

Also, I thank Charlene Cohen-Deroy for her diligent efforts and commitment to

excellence in coding and scoring the Word Reading subtests, as well as calculating the

phoneme length for the stimulus words. All of these students demonstrated a keen

awareness of the importance of maintaining the integrity of the data. Without the

combined efforts of these wonderful students, this dissertation would have taken longer.

Finally, I thank the collective insights of the members of the Language over the Lifespan

lab, especially those who provided suggestions for the semantic cues. It was their

integrative efforts that helped to improve various aspects of this study.

I also would like to thank the Director and members of the Brain Rehabilitation

Research Center, at the Malcom Randall Veterans Affairs Medical Center, Dr. Leslie

Gonzalez Rothi, Dr. Stephen Nadeau, and Dr. Jay Rosenbek, for the pre-doctoral

fellowship opportunity that Dr. Diane Kendall provided. I also thank Haijing Qin for her

statistical support. I acknowledge other investigators and clinicians, especially Nan

Musson, Susan Leon, and Amy Rodriguez. I appreciated the ongoing support of the staff,

especially Brenda Stidham, Sue Nadeau, Joy McCallum, Sandy Davis, Lisa

DeEmmanuel, and Lynn Dirk. Further appreciation is extended to the incredible VA

library staff, for their expertise, time, and consideration: Marsha H. White, Ellen L.

Umans, Marylyn Gresser, and Harold A. Boyce.










Two Eine clinicians and professors have played an important, yet earlier role in my

career. I extend my thanks and appreciation to Dr. Robert C. Marshall, for inspiring me

years ago in my master' s program, continuing his belief in me, and his incredible

mentoring in a supportive, enthusiastic and gentle manner. If it were not for him, I would

never have started this Ph.D. endeavor. Also, I thank Donna Chadwick for teaching me

the fundamental aspects of being an effective clinician during my undergraduate

program.

My friends have also been a great source of inspiration and hope. Dr. Sally Ann

Giess, whom I met in my first year at the University of Florida, continues to be a great

friend, and role model. I acknowledge my other friends and colleagues from the UF CSD

Department: Dr. Ann Marie Knight, Dr. Amber Hollingsworth, Dr. Judith Wingate, Dr.

Lynnette Bardolf, Susan Leon, and Maisa Haj Tas. I thank The Reverend Jeremy Hole

for his support and encouragement. I will always appreciate my dear friends Ann Case

and Nancy Merrifield for their wonderful presence in my life.

Most importantly, I am forever grateful for my family. Their patience, ongoing

love, support, and belief in me have been truly wonderful. Thus, I thank them: my

husband David, mother, Claire, and my sisters and their families, Jeanmarie; Donna and

Ryan, Buzz and Charlie; Claire and Ross, Martina and Fabiana; as well as Everett, my

father-in-law and Marie, my mother-in-law. Also, I thank my cousins and their families:

Marybeth, Mary Catherine, Luigi, Lou and Joey, and Lee and Mary Anne. Finally, I

would like to acknowledge all those who came before, paying my way through their

loving support and kindness, including my father John; and grandmothers, Florence

McNulty and Anna Efros Kramer.









This dissertation was partially supported by my teaching and research

assistantships from the UF Department of Communication Sciences and Disorders. It was

also partially funded by Dr. Altmann's Advancing Academic-Research Careers award,

which she had received from the American Speech-Language-Hearing Association. I am

grateful for these funding sources.




















TABLE OF CONTENTS


page

ACKNOWLEDGMENT S .............. .................... iv


LIST OF TABLES ............_...... .__ ..............xi....


LIST OF FIGURES .............. ....................xii


AB STRAC T ......__................ ........_._ ........xi


CHAPTER


1 INTRODUCTION ................. ...............1.......... ......


2 REVIEW OF THE LITERATURE 1: NAMING DEFICITS INT PAD ................... .....8


Semantic Memory Deficit in Probable Alzheimer Disease (PAD) .............. ................9
Lexical Access Deficit in PAD ................. ....... .. ..... .... ...............13...
Semantic Impairment and Lexical Access Deficit in PAD .............. ....................16
Naming Treatment Studies in PAD .............. ...............18....

3 LITERATURE REVIEW 2: A BRIEF OVERVIEW OF THE ORGANIZATION
OF SE MANT IC MEMORY ............... ...............26.


A Brief Overview of Semantic Memory and Categorization ................ ................. 26
Family Resemblance/Prototype View .............. ...............30....
The Complexity Account of Treatment Efficacy .............. ...............42....
Connectionist Model and Computer Simulations ................. ......... ................43
Similar Behavioral Studies .............. ...............48....
The Current Study................... .. ......... .... ...............6
Rationale for Applying the Training to the PAD Population ................... ...........60
Research Questions and Predictions............... ..............6

4 M ETHODS .............. ...............70....


Participants .............. ...............71....
Experimental Stimuli ............ _...... ._ ...............72....
Procedure .............. ...............76....
Screening ............ _...... ._ ...............76....
Pretesting ............ _...... ._ ...............76....











Semantic Training .............. ...............78....
P ostte sti ng .........__.. ..... ._ __ ...............79...
Scoring ........._.___..... ._ __ ...............79....
Statistical Analyses ........._._.... ......__ ...............81....
Research Question 1: Control Items ........._._... .. ...___._ ....___ ..........8
Research Question 2: Trained Items in Trained Categories ................... ...............82
Research Question 3: Untrained Items in Trained Categories............... ...............8
Re search Questi on 4: Category Generati on ....._.__._ ..... ... .__. ... .._._.........8

5 RE SULT S .............. ...............85....


Research Question 1: Control Items ......__....._.__._ ......._._. ...........8
Research Question 2: Semantically Trained Items ....._.__._ ........___ ...............86
Research Question 3: Untrained Items in Trained Categories............... ...............8
Re search Questi on 4: Category Generati on ....._.__._ ..... ... .__. ... .._._.........9
Summary of Results............... ...............9

6 DI SCUS SSION ........._.___..... .___ ...............96....


Summary of Findings .............. ... ........ ...........9
Implications for Anomia Treatment in PAD .............. ...............98....
Implications for Methodology .............. ...............103....
Future Studies ........._.___..... .__. ...............107....

APPENDIX


A INFORMED CONSENT FORM ................. ...............111...............

B TABLES WITH INFORMATION ABOUT THE PARTICIPANTS ................... ...113


Participant Demographics and Selected Neuropsychological Test Results .............114
Demographics and Performance Before and After Semantic Training ............_.......115

C TABLES WITH INFORMATION ABOUT THE STIMULI ................. ...............116


Master List of Stimulus Items Used in the Study .......... ................ ...............116
Alternate Names for Stimulus Items Used in the Study ................. ............... .....118

LIST OF REFERENCES ................. ...............120................


BIOGRAPHICAL SKETCH ................. ...............135......... ......

















LIST OF TABLES


Table pg

4-1. Participant Demographics and Scores on Mini Mental Status Exam and Reading
Subtest ................. ...............72.......... .....

4-2. Lists of Training Conditions and Corresponding Training Conditions ................... ..72

4-3. Example from List 1: Train-Typical Vehicles and Train-Atypical Tools. ................75

4-4. Means for Complete List of Atypical and Typical Items. ................ ................ ...76

5-1. Pre and Post Group Accuracy and Response Time Means for Trained Items ..........87

5-2. Group Accuracy and Response Time Means for Trained Items (Train-Atypical;
Train-Typical) .............. ...............87....

5-3. Analysis of Variance for Accuracy Means for Untrained Items in Trained
Categories ................. ...............89.................

5-4. Group Response Time Means (ms) for Untrained Items in Trained Categories.......92

B-1. Participant Demographics and Selected Neuropschological Testing Results from
the PAD Diagnosis ................. ...............114................

B-2. Participant Demographics and Pre and Posttest Category Generation and Picture
Naming Tasks (i.e., Before and After the Semantic Training, day of study
participation in the study) ................. ...............115...............

C-1. Master List of Stimulus Items Used in the Study. Trained Items are Italicized.....116

C-2. Master List of Stimulus Items Used in the Study. Alternate names for stimuli
counted as correct ................. ...............118................

















LIST OF FIGURES

Figure pg

5-1. Accuracy Means for Untrained Items in Trained Categories, Main Effect of Time .90

5-2. Untrained Items in Trained Categories: Three-Way Interaction Among Time..........91

5-3. Category Generation, Number of Items Generated at Pre & Posttest for Trained &
Control Item s. .............. ...............94....
















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

CHOOSING REMEDIATION TARGETS FOR NAMING DEFICITS IN
PROBABLE ALZHEIMER DISEASE:
DOES TYPICALITY MATTER?

By

Claudia A. Morelli

August 2006

Chair: Lori J.P. Altmann
Major Department: Communication Sciences and Disorders

The purpose of this single session exploratory study was to determine if

remediation targets could be selected to improve picture naming and promote

generalization to untrained items in 12 adults with early probable Alzheimer disease

(PAD). Because theories support competing predictions about the relative effects of

training typical versus atypical semantic category members with respect to subsequent

generalization, this study contrasted the effects of training typical and atypical semantic

category exemplars. Specifically, it examined changes in picture naming and category

generation following initial repetition of the items at pretest followed by a semantic

training. Stimuli included 24 items from each of 3 semantic categories, half of the items

were typical and half atypical based on Rosch's norms. Two categories received training,

one using a subset of typical items, one using a subset of atypical items; the third

category remained untrained to track effects of repetition. The untrained category showed









nonsignifieant improvements at post-test. Accuracy scores improved for all trained

categories; however, only trained, typical items were named significantly faster at post-

test. Generalization was found in the untrained typical items in the categories that were

trained with typical items, which were named more accurately at posttest. Neither

atypical items from the categories that were trained with typical items nor any of the

items in categories that were trained with atypical items showed improved accuracy after

training. These Eindings are consistent with those of other researchers investigating the

semantic deterioration in adults with PAD. Several studies have found an advantage for

typical items in this population. These Eindings are attributed to the redundant

connections among features in typical items, which allow them to be more resilient in the

face of progressive damage, allowing them to be more responsive to intensive semantic

training. Atypical items lack the redundancy of connections and, thus, are more vulnerable

to damage. These Eindings are extremely encouraging for the development of principled

strategies for choosing items to encourage generalization in the remediation of anemia, as

well as for the development of lexical-semantic treatment paradigms for individuals with

early PAD.















CHAPTER 1
INTTRODUCTION

Probable Alzheimer disease (PAD), a progressive neurodegenerative disease, is the

most prevalent type of acquired cognitive dysfunction (Thal, 1999) and affects over four

million Americans (Kawas & Katzman, 1999). The prevalence of PAD accounts for 50-

80% of patients with dementia (Chui, 1989) and comprises the greatest percentage of the

dementias in the geriatric population (Chui, 1989; Henderson & Finch, 1989; Sclan &

Kanowski, 2001). Furthermore, PAD is already considered to be a maj or public health

concern, and by 2025, the number of Americans affected by the disease is expected to be

12 million (Rosenberg, 2005) and 14 million by 2050 (Katzman & Fox, 1999). Although

there is no known cure for PAD, and only symptomatic treatment is available at this time,

current research endeavors are aiming to halt the progression of the disease as well as

prevent its occurrence (Petersen et al., 2001; Rosenberg, 2005). Reducing the disability

until this occurs is important (Rothi et al., 2005). Furthermore, when drug therapies are

available that halt the progression of the disease, there will be a demand for linguistic and

cognitive rehabilitation for this population. Therefore, it is incumbent upon speech-

language pathologists to develop and test appropriate treatments for this population now,

so that tested methods are available that address the language difficulties in PAD (ASHA,

in press), which largely revolve around word finding

Word finding difficulties (i.e., anemia), as measured by poor performance on

picture naming tasks have been widely reported in adults with PAD (Barbarotto,

Capitani, Jori, Laiacona, & Molinari, 1998; Bayles, 1982; Bayles & Tomoeda, 1983;









Benson & Geschwind, 1985; Heilman, 2005; Hodges & Patterson, 1995; Hodges,

Patterson, Graham, & Dawson, 1996; Huff, Corkin, & Growdon, 1986; Huff, Mack,

Mahlmann, & Greenberg, 1988; Kirshner, Webb, & Kelly, 1984; Martin & Fedio, 1983;

Smith, Faust, Beeman, Kennedy, & Perry, 1995; Smith, Murdoch, & Chenery, 1989;

Williams, Mack, & Henderson, 1989; Williamson, Adair, Raymer, & Heilman, 1998).

However, to date, there are only three PAD anemia treatment studies (Abrahams &

Camp, 1993; Ousset et al., 2002; Rothi et al., 2005), and only one has reported

generalization to untrained items (Abrahams & Camp, 1993). Consequently, developing

treatments for this population that reduce the word finding difficulty is important (Rothi

et al., 2005), especially those that promote generalization.

Generalization of learning, an essential objective for rehabilitation (Kearns, 1989;

Thompson, 1989), has been described as an observation of an occurrence of a particular

trained behavior in a context that has not been trained (McReynolds, 1989).

Generalization is important for several reasons, including the clinical accountability of

the speech-language pathologist to use methodologies to measure the effectiveness of

treatment (Kearns, 1989). The mechanism underlying generalization has been a long-

standing question (Martin, Laine, & Harley, 2002), and although cognitive models have

been advanced over time, the construct of generalization remains a mystery (Francis,

Clark, & Humphreys, 2002). Perhaps due to this, there is still a lack of generalization

reported in the literature (Francis et al., 2002; Kearns, 1989; Kiran & Thompson, 2003b;

McNeil et al., 1998; McReynolds, 1989; Nickels, 2002; Thompson, 1989; Thompson,

Ballard, & Shapiro, 1998; Thompson et al., 1997; Thompson, Shapiro, Kiran, & Sobecks,










2003). Moreover, attempts to replicate methods that have promoted generalization in

initial studies have not always been successful (McReynolds, 1989; Thompson, 1989).

There are two types of generalization that, according to Thompson (1989), play an

important role in aphasia rehabilitation: response generalization (i.e., when an untrained

response occurs after other responses have been trained (e.g., production of different,

untrained items) and stimulus generalization (i.e., when there is carryover of a trained

behavior to a different, untrained stimulus condition, e.g., outside of the clinic). Without

the former, there would be an inordinate number of responses to train, and without the

latter, trained behaviors would occur only in the clinic setting (Thompson, 1989).

Although both of these are important, the current study focuses only on response

generalization.

Several investigators (Hillis, 1998; Kiran & Thompson, 2003b; Plaut, 1996;

Thompson, 1989) have suggested that treatment studies might be able to influence the

recovery of language if they target factors that are modifiable. A principled selection of

materials for training as well as for testing generalization is just one example of this

(Hillis, 1998; Kiran & Thompson, 2003b; Plaut, 1996; Thompson, 1989; Thompson et

al., 2003). One technique that has successfully induced generalization is to have a

structural relationship among and across the stimuli (Stokes & Baer, 1977; Thompson,

1989; Thompson et al., 1998; Thompson et al., 1997; Thompson et al., 2003). One

approach recommends choosing stimuli that are more versus less complex, a factor that

has been shown to be important for increasing generalization to untrained items (Kiran &

Thompson, 2003b; Plaut, 1996; Thompson et al., 1998; Thompson et al., 1997;

Thompson et al., 2003). This phenomenon has been referred to as the "complexity effect"









and the Complexity Account of Treatment Efficacy (Thompson et al., 1998; Thompson et

al., 1997; Thompson et al., 2003), which has been successfully used in syntactic

treatment studies in the aphasia population.

Kiran and Thompson (2003b) have equated the notion of semantic complexity with

category typicality (i.e., by the degree to which a semantic category exemplar is similar

to the category prototype, as described Rosch, (1975)) and applied this to a semantic

training in adults with aphasia. In this study, Kiran and Thompson (2003b) found

generalization to untrained items after atypical items from a category were trained.

Findings from this study replicated similar results from a connectionist model that were

simulated in a computer experiment of acquired dyslexia (Plaut, 1996). In a similar study,

Stanczak, Waters, and Caplan (2006) reported generalization for one of two participants

with aphasia, after training atypical category exemplars. However, two other aphasia

naming treatment studies (Mayer, Murray, & Karcher, 2004; Stanczak, Waters, &

Caplan, 2005) did not find generalization following training with either atypical and

typical items, although treatment effects were observed. Thus, it is important to further

explore the effectiveness of training typical or atypical category exemplars as a strategy

to encourage generalization (Murray & Clark, 2005).

All of these studies (Kiran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al.,

2005, 2006) have tested the predictions of the Complexity Account of Treatment Efficacy

(Thompson et al., 2003), and contrasted it with an older approach based on the family

resemblance/prototype hypothesis by Rosch (1975). Arguments based on this view center

around the benefits of training qualities or features that are shared among many members

of a semantic category, so that many items in the category might benefit, and thus, it










provides a rationale for training typical items, which contain a larger proportion of shared

features. The structure of a semantic category is an essential element in the rationale for

treatments based on typicality (Plaut, 1996); therefore, a population with preserved

category structure such as individuals with PAD (Flicker, Ferris, Crook, & Bartus, 1987;

Huff et al., 1986; Martin & Fedio, 1983; Nebes, Boller, & Holland, 1986; Salmon,

Butters, & Chan, 1999; Schwartz, Marin, & Saffran, 1979; Schwartz, Kutas, Butters,

Paulsen, & Salmon, 1996; Warrington, 1975) might be optimal for testing these

hypotheses.

Some theories of how semantic representations are affected by PAD have

specifically suggested that items that have many shared or intercorrelated features might

be more preserved in PAD compared those with representations primarily consisting of

distinguishing features, with relatively fewer shared features (Altmann, Kempler, &

Andersen, 2001; Devlin, Gonnerman, Andersen, & Seidenberg, 1998; Gonnerman,

Andersen, Devlin, Kempler, & Seidenberg, 1997). This literature combined with the

evidence that the PAD population has worse performance on atypical items compared to

typical items (Sailor, Antoine, Diaz, Kuslansky, & Kluger, 2004; Smith et al., 1995)

provides further incentive to compare performance by adults with PAD on typical

category exemplars, which have many shared features, and atypical category exemplars,

which have relatively fewer shared features.

This exploratory study extends the small body of research comparing the effects of

training typical and atypical category exemplars in adults with aphasia (Kiran &

Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005, 2006) to those with mild-

moderate PAD using a modified version of the semantic feature generation task (Boyle,










2004; Boyle & Coelho, 1995; Coelho, McHugh, & Boyle, 2000). Since there were no

reports of this type of training with the PAD population, it seemed appropriate to Birst

apply it in a single session to determine its feasibility. Thus, the first aim of the study was

to determine if there is a facilitation effect (Hickin, Best, Herbert, Howard, & Osborne,

2002; Howard, 1985; Howard, Patterson, Franklin, Orchard-Lisle, & Morton, 1985;

Patterson, Purell, & Morton, 1983), for trained items from either item repetition alone

(Kendall, personal communication, 2005, Fuller et al. 2001; Hickin et al., 2002; Mayer et

al., 2004; Patterson et al., 1993; Rothi et al., 2005) or from the repetition plus a semantic

training (Drew & Thompson, 1999; Le Dorze, Boulay, Gaudreau, & Brassard, 1994;

Wiegel-Crump & Koenigsknecht, 1973). The second aim of the study was to determine

whether there was generalization of training to untrained items in the semantically trained

categories. The third aim of the study was to examine generalization to an untrained task,

category generation, using the same semantic categories as the naming and feature

analysis tasks.

The structure of this study is as follows. The next two chapters provide reviews of

the literature. Chapter 2 discusses the naming deficits (i.e., anemia) in adults with PAD,

and the anemia treatment studies for this population. The third chapter provides

background information on semantic theory, the two competing views, the relevant

Endings from the connectionist model (Plaut, 1996) and the treatment studies in adults

with aphasia using semantic category exemplar training with typical and atypical items

(Kiran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005, 2006). This is

followed by a description of the current study, including rationale for applying these

techniques to adults with PAD. Additional findings from the aphasia treatment literature









are identified as they relate to the methodology for our study. Finally, our research

questions and predictions based on the above literature are described. In Chapter 4, the

study methods are explained in detail, followed by the results and discussion in Chapters

5 and 6, respectively. To foreshadow our results, this exploratory study provided good

preliminary evidence that semantic training can have significant effects on the anemia

found in PAD, and, with judicious choice of stimuli, this training may potentially

generalize to other items in the same category.















CHAPTER 2
REVIEW OF THE LITERATURE 1: NAMING DEFICITS IN PAD

It is common for patients in the early stages of PAD to have picture naming deficits

(Appell, Kertesz, & Fisman, 1982; Barker & Lawson, 1968; Bayles & Tomoeda, 1983;

Chertkow, Bub, & Seidenberg, 1989; Flicker et al., 1987; Grossman et al., 2004; Kirshner

et al., 1984; Lipinska & Backman, 1997; Schwartz et al., 1979; Warrington, 1975;

Williams et al., 1989; Williamson et al., 1998). Furthermore, the exact nature of

the naming deficit in PAD is controversial (Bell, Chenery, & Ingram, 2001; Nebes, 1992;

Nebes, Brady, & Huff, 1989) regarding whether the deficit stemmed from impaired

visual-perception, semantic memory, or lexical access (Nebes, 1989). An early theory

suggested that visual misperception is the reason for the naming deficit, and this is based

on the presence of visuo-perceptual errors, presumably due to difficulties perceiving the

object (Barker & Lawson, 1968; Kirshner et al., 1984). These theories and studies

providing evidence are discussed below, beginning with the semantic deficit, then the

lexical access, and followed by the semantic and lexical access deficit. Next, a discussion

of the few naming treatment studies that have targeted this population is provided

(Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005).

These PAD naming therapy studies achieved a naming treatment effect (i.e., acquisition)

but only one also reported generalization (Abrahams & Camp, 1993). This limited

number of studies suggests a need for further research. Insights from all of these studies

and additional research from the aphasia literature provided techniques that were applied









in our exploratory study comparing the effects of training typical and atypical category

exemplars.

Semantic Memory Deficit in Probable Alzheimer Disease (PAD)

Several studies have suggested that there is semantic memory loss or degradation

and, consequently, a loss of information about semantic representations (Alathari, Trinh

Ngo, & Dopkins, 2004; Hodges & Patterson, 1995; Hodges et al., 1996; Hodges, Salmon,

& Butters, 1991, 1992; Huff et al., 1986; Huff et al., 1988; Margolin, Pate, Friedrich, &

Elia, 1990; Martin, 1992; Salmon, Butters et al., 1999; Salmon, Heindel, & Lange, 1999;

Salmon, Shimamura, Butters, & Smith, 1988; Schwartz et al., 1979). Evidence for this

theory includes impaired naming with semantic errors related to the superordinate

category or an associate item within the category (Bayles & Tomoeda, 1983; Martin &

Fedio, 1983; Salmon, Butters et al., 1999); the presence of consistent naming responses at

two different test periods (Henderson, Mack, Freed, Kempler, & Andersen, 1990) an

association between the inability to name an item and the inability to recognize its name

(Flicker et al., 1987; Huff et al., 1986; Huff et al., 1988); and a relationship between the

naming failures and the lack of core information provided about the corresponding item

(Hodges et al., 1996). In addition, deterioration of semantic memory in PAD has been

described based on poor performance on explicit tasks such as picture naming tasks and

category fluency (Bayles & Tomoeda, 1983; Chertkow & Bub, 1990; Flicker et al., 1987;

Huff et al., 1986; Martin & Fedio, 1983); tasks requiring generation of semantic feature

knowledge (Alathari et al., 2004), generation of verbal definitions (Garrard, Lambon

Ralph, Patterson, Pratt, & Hodges, 2005; Hodges et al., 1996); as well as questions

targeting feature knowledge (Chertkow et al., 1989; Giffard et al., 2002). Other tasks that

have showed similar findings include making judgments about semantic relatedness










(Bayles, Tomoeda, & Cruz, 1999); and associating words, defining words, and ranking

associations (Abeysinghe, Bayles, & Trosset, 1990).

There appears to be different interpretations or variants of this theory of semantic

memory loss or degradation in PAD. For example, several investigators have found that,

while attribute knowledge about a specific concept or exemplar in a category is impaired,

superordinate category knowledge is preserved (Chertkow & Bub, 1990; Flicker et al.,

1987; Giffard et al., 2002; Giffard et al., 2001; Hodges et al., 1991; Huff et al., 1986;

Lukatela, Malloy, Jenkins, & Cohen, 1998; Martin & Fedio, 1983; Nebes et al., 1986;

Salmon, Butters et al., 1999; Schwartz et al., 1979; Warrington, 1975). This led some

researcher to assert that the attributes of concepts are degraded in PAD (Chertkow et al.,

1989; Giffard et al., 2002; Giffard et al., 2001; Martin, 1992; Martin & Fedio, 1983).

Martin (1992) suggested that random damage from the pathological process of

PAD resulted in changes to the semantic representations such that they would be

degraded and thus more similar to one another. With progression of the disease, the

ability to distinguish between items in the same category would be diminished. Thus, on

a confrontation naming task, a semantic representation would be activated but it would

lack the specificity needed to correctly name the item and, as a result, several lexical

entries would be activated. Therefore, when the person with PAD sees a picture, the

semantic representation that is activated may be underspecified in terms of the details.

This is due to the absence of the essential attributes knowledge for that item such that the

activated distinguishing features of that obj ect are not strong enough to rule out other

similar items (Altmann et al., 2001; Devlin et al., 1998; Gonnerman et al., 1997; Martin,

1992).










A property verification task has also been used in a group of adults with PAD

(Smith et al., 1995). Participants were asked to verify brief statements containing

information about an item's properties. The items were high or low typicality (and

dominance, which relates to the relevance of the meaning of the item and is correlated

with typicality) category exemplars. The statements were either distinguishing

characteristics (i.e., distinctive) or shared features (i.e., common to the other items in that

category). The results from the accuracy scores and reaction times indicated a

degradation of property level information, particularly regarding both low dominance

typicality items in the category. This was not interpreted as a loss of representations of

the items' properties or a reorganization of relationships among properties of obj ects.

Instead, Smith et al. (1995) offered this as evidence that the representations of category

exemplars that are low-typical and low dominant have been degraded by the Alzheimer

pathology. Furthermore, these investigators hypothesized that task demands might be a

reason for differences in the literature. Smith et al. (1995) suggested that implicit

knowledge allows adults with PAD to have faster verification response times. Thus, in a

category verification task, the demands probably do not need the full semantic

specification. Smith et al. (1995) also suggested that while explicit knowledge is needed

to assess information could be impaired, for example, tasks requiring the participant to

use relevant information in a ranking task. This example comes from Grober, Buschke,

Kawas, and Fuld (1985) who reported that performance on an attribute ranking task

showed that attributes about a concept are preserved, but that the organization of

semantic information is altered by the disease process. Another example from Smith et al.

(1995) is that the naming process involves computation such that an obj ect' s properties









are activated to allow for distinction among other category members. If Alzheimer

disease processes affect both distinguishing and shared features that are low dominant

and low typicality, this could interfere with complete activation of the obj ect

representation (Smith et al., 1995). Consequently, it would reduce the accuracy on a

naming task. Better performance is seen when contextual information is provided, yet

when the task involves full semantic representation without the support, performance

declines (Smith et al., 1995).

While the evidence above for a progressive deterioration of the semantic system

comes largely from tasks requiring explicit access of semantic representations, evidence

for an overall preservation of semantic category knowledge comes from tasks that require

only implicit knowledge of word semantics.

For example, adults with PAD did not show a priming effect on a stem-completion

task and had reduced number of productions for the second semantically related item on a

free association task (Salmon et al., 1988). Hyper-priming has been reported on lexical

decision tasks and found to be associated with degraded semantic representations

(Chertkow et al., 1989; Giffard et al., 2002; Margolin, Pate, & Friedrich, 1996; Martin,

1992). To account for the hyper-priming in PAD, Martin (1992) suggested that changes

in semantic activation are more robust for processing information that is degraded, such

that a degraded semantic network benefits more from a semantic prime in than an intact

semantic network. To account for the difference in performance on tasks requiring

implicit versus explicit access to semantic representations, it has been suggested that

participants with PAD have difficulty performing an intentional search through semantic

memory, but perform relatively normally when relying on the automatic spread of









activation in the semantic network (Chertkow & Bub, 1990; Nebes, 1992; Ober, Shenaut,

& Reed, 1995).

All of these studies with explicit and implicit tasks suggest that at some level, the

semantic system is impaired (due to loss or degradation), and different investigators have

used different tasks and methodologies to address this. However, this theory does not go

unchallenged; an alternative argument suggests that lexical access, specifically, access of

the phonological word form from semantics, is the primary deficit in PAD.

Lexical Access Deficit in PAD

Several researchers have argued that the naming impairment in PAD results from

impaired access to the phonological form of the word (i.e., a retrieval deficit) in the

presence of an overall intact semantic knowledge structure (Albert & Milberg, 1989;

Benson & Geschwind, 1985; Nebes, 1992; Nebes et al., 1986; Nebes et al., 1989; Nebes,

Martin, & Horn, 1984; Neils, Brennan, Cole, Boller, & Gerdeman, 1988; Ober &

Shenaut, 1999; Thompson-Schill, Gabrieli, & Fleischman, 1999). Along these lines, it

has been suggested that the presence of semantic errors is actually an indication that

knowledge about the item is intact, despite the lack of the ability to retrieve it, rather than

in indication of a semantic impairment (Nebes, 1989).

Semantic priming paradigms (as discussed earlier) have been employed to evaluate

the status of semantic memory via an implicit task, in order to reduce the participant' s use

of attentional mechanisms (Ober, 1999). There are reports of normal priming in adults

with PAD (Nebes et al., 1984), as well as reports of hyper-priming which have been

interpreted as being caused by abnormal attentional processes in the presence of

preserved semantic memory (Hartman, 1991; Ober & Shenaut, 1995; Ober et al., 1995).

In a meta-analysis of semantic priming studies, Ober et al. (1995) reported that hyper-










priming was due to attentional mechanisms which co-occurred with large increases in

reaction times found in controlled priming paradigms. These researchers argued that the

hyper-priming represented evidence that automatic spreading activation was occurring

among semantic representations in the semantic priming task (Ober et al., 1995). Further

evidence for this comes from another study by Ober and colleagues (1995) in which

adults with PAD, as well as healthy young and older adults participated in a series of

lexical decision tasks. Results showed that priming effects were equal across the groups.

The PAD group demonstrated longer reaction times on low frequency words before

making lexical decisions, and this was interpreted as an indication that additional time

was needed for reaching activation level threshold for these items. Ober et al. (1995)

argued that these data reflected an intact semantic memory structure in PAD.

Two event-related potential (ERP) studies offered more support for an access

deficit in PAD (Ford et al., 2001; Schwartz et al., 1996). Event related potentials measure

electrical manifestations of particular psychological processes that occur in preparation

for or in response to discrete events (Fabiani, Gratton, & Coles, 2000; Kutas & Hillyard,

1980). The N400 is a negative going deflection occurring at ~400 ms and occurs in

response to anomalous information, specifically, semantic violations; for example it is

larger when the prime is unrelated to the target compared to when it is related (Fabiani et

al., 2000; Kutas & Hillyard, 1980). Schwartz et al. (1996) compared healthy older adults

and adults with PAD to determine whether or not the specificity of a category prime had

a differential effect on the degree of semantic priming in these two populations. Using a

variation of a category verification task, the investigators asked the participants first to

listen to a prime that was delivered auditorily as a name of a category (different levels,










e.g. superordinate and subordinate) and then, to read the written presentation of the target

name (i.e., a member of a category). While the most robust priming effects were found in

the young adults, the smallest priming effect and slowest reaction times were found in the

PAD group. Responsiveness to the category manipulation was revealed in priming effects

that were similar across the groups, not just in the reaction times but also in the ERP

N400 congruity effects being larger on some levels, for example, the subordinate

category level. Thus, category level manipulation affected the PAD group, and this was

interpreted as evidence for an intact semantic network. In contrast, longer response times

and smaller priming effects were attributed to the demands of the task and the necessity

of searching through memory while engaged in online processing required for this and

other similar tasks (Schwartz et al., 1996).

In a more recent ERP study, Ford et al. (2001) examined priming based on age and

dementia, and asked whether the N400 amplitude could be used to show specific

semantic memory deficits for obj ects that could not be named. The participants (adults

with PAD, healthy young controls and healthy older adult controls) completed a pretest

confrontation picture naming task consisting of items from 12 semantic categories. The

following week, the participants completed a picture naming verification task while ERPs

were measured. The participants were instructed to press a button to indicate whether or

not the prime (a picture) matched the target (a word). The consistent finding across the

groups was that for a word that did not match the picture, there was a more negative

N400 amplitude. The ability of the PAD group to correctly name pictures was not

associated with the N400 priming effect or any corresponding scalp distributions. Thus,

despite an inability to access the name of an item, there was evidence from the N400









results that the PAD group had sufficiently intact knowledge for priming responses at the

cortical level (Ford et al., 2001).

It has been suggested that while the structure of semantic memory is not damaged

in PAD, there is a generalized cognitive processing deficit, which is required for

intentionally retrieving and evaluating information, and decision making (Nebes, 1992).

This results in word retrieval or access deficits. Two ERP studies (Ford et al., 2001;

Schwartz et al., 1996) also provide evidence for a fairly preserved semantic structure in

PAD. Additional support for an access deficit in PAD comes from the finding by Ousset

et al., (2002) that initial syllable cues and the sound representing the item were among the

most effective cues during a oral naming to definition task, while providing the semantic

category was the least effective.

Semantic Impairment and Lexical Access Deficit in PAD

The most likely explanation of these disparate findings is that the naming deficit in

PAD is due to a combination of a breakdown in both semantic memory and retrieval

abilities (Bowles, Obler, & Albert, 1987; Huff et al., 1988; Watson, Welsh-Bohmer,

Hoffman, Lowe, & Rubin, 1999; Williamson et al., 1998). This has been suggested based

on results from confrontation naming and fluency tasks.

Williamson et al. (1998) made predictions about performance of healthy older

adults compared to adults with PAD on two confrontation naming tasks: the Boston

Naming Test (Kaplan, Goodglass, & Weintraub, 1983) and the Action Naming Test

(Nicholas, Obler, Albert, & Goodglass, 1985), based on knowledge about the underlying

anatomy and PAD pathology. More specifically, they predicted that, compared to action

naming; obj ect naming would be more impaired in PAD. This was because the disease

process had resulted in more extensive damage to the left temporal area which is needed









for naming obj ects, compared to the frontal areas required for naming actions. Compared

to the control group, the participants with PAD were less accurate on both measures, but

their overall performance was worse on the BNT, on obj ect naming test, compared to the

Action Naming Test. Williamson et al. (1998) suggested that obj ect naming revealed

significant impairments at the semantic specification and lexical levels. However, for

action naming, the impairment was mostly at the semantic specification level, while the

lexical level remained fairly intact. The authors also attributed the presence of more No

Response errors on obj ects than on actions to the corresponding damage to the anatomy

that underlies that function. Williamson et al (1998) concluded that compared to healthy

older control participants, adults with PAD are not as accurate when they name obj ects

and actions. Furthermore, the degree of deficit is more pronounced with obj ects, which

was attributed to impairment not just at the semantic specification level, but also at the

level of lexical access. Based on these findings, we designed our study to address both of

these levels of deficits. The participants engaged in both lexical and semantic tasks to

stimulate the system at both the lexical (e.g., repeating the correct name of the item to

access the word form) and semantic levels (e.g., answering questions about semantic

features of the item).

In summary, the naming impairment in PAD has been attributed to a loss of or

degradation in semantic memory storage and/or organization, disruption of the retrieval

of the information in the network, a combination of these, or other factors related to task

demands. However, there is reason to believe that the semantic structure, at least in terms

of semantic categories, is still fairly preserved in early PAD (Flicker et al., 1987; Huff et

al., 1986; Martin & Fedio, 1983; Nebes et al., 1986; Salmon, Butters et al., 1999;









Schwartz et al., 1979; Warrington, 1975). For example, although Cronin-Golomb et al.

(1992) reported longer reaction times on a category decision task as well as smaller

number of items on a category generation task, the pattern was normal across categories

and items were ranked by typicality in a normal manner. This was interpreted as being a

reflection of intact semantic category organization (Cronin-Golomb et al., 1992).

Naming Treatment Studies in PAD

Although there is a plethora of research on the naming impairment in PAD, there

are only a few naming treatment studies that involved participants with PAD (Abrahams

& Camp, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005). Like the

typicality training studies with the aphasia population (Kiran & Thompson, 2003b; Mayer

et al., 2004; Stanczak et al., 2005, 2006), these three PAD studies also had picture

naming outcomes based on comparisons of performance before and after the intervention

(i.e., a treatment effect), and examined generalization to untrained items. Although these

three PAD treatment studies are very different from each other, they each have important

implications for the future of word finding treatment for this population.

Using a single subj ect design study with two adults with dementia, Abrahams and

Camp (1993) used spaced-retrieval training (SRT). Spaced retrieval training is a

technique that intersperses increasing time intervals between presentation of a target by

the clinician and recall of it by the participant. Participant I was diagnosed with

progressive dementia (16/30O on the Mini Mental Status Exam; MMSE) (Folstein,

Folstein, & McHugh, 1975) and Participant 2 had PAD (13/30 on the MMSE). Each

participant was tested on the Boston Naming Test (Kaplan et al., 1983). From these

results, nine training items from each participant were identified for use as a subset of

training and control items. While two of these items were treated, the other seven were









controls (Abrahams & Camp, 1993). Abrahams and Camp (1993) reported that the results

of two different treatment targets for each participant showed improved performance

(e.g., naming an item after a maximum of 300 seconds). Only Participant I was tested

two weeks later. This showed maintenance effects for one of the two items, despite a six

point drop in her MMSE score (10/30). Abrahams and Camp (1993) described

generalization to untrained exemplars of the trained items: a colored drawing of the

target; a second target; and an actual exemplar. The investigators reported that SRT was

effective in adults with dementia because they not only have word-finding deficits, but

also have cognitive deficits that affect the ability to remember new information. It was

suggested that SRT uses procedural memory, which appears to be intact in this

population, and thus enabled the participants to benefit from the anemia treatment

(Abrahams & Camp, 1993).

Although both participants did not have PAD, the findings from this early anemia

treatment study by Abrahams and Camp (1993) have value because they were the first to

demonstrate a treatment effect after using SRT, in the PAD population, even with

participants with low MMSE scores. Spaced retrieval training does not provide any

semantic information about the item, but uses repetition. It was not clear if the items that

showed generalization actually generalized. Instead, it could be that it was easier for the

participant to name the color drawing.

Ousset et al. (2002) compared the effects of an experimental lexical therapy

targeting both episodic and lexical stimulation in adults with mild PAD who were on an

acetyl cholinesterase inhibitor. The average MMSE score was 21.1. The participants were

divided into two groups of eight: one group received lexical therapy (with narratives and









definitions) and the other, a control group, received occupational therapy (i.e., pottery,

drawing, and conversations). The treatment was provided in 16 sessions, once a week,

with a two week break at the midpoint. The pre and posttest picture naming stimuli

consisted of 120 black and white line drawings from three categories. While 80 of these

items from the three categories were a part of the lexical therapy, the remaining 40 were

not. During each lexical therapy session, the participant first read aloud a narrative from a

computer screen and then listened while the examiner read it. Then, the participant

completed a naming to definition task on the computer. There were 20 definitions, half of

the words were in the narrative to provide both semantic and episodic reinforcement; and

half of the words were not in the narrative, to provide episodic reinforcement. When the

participant did not respond or was incorrect, the computer randomly provided one of five

different types of cues (i.e., semantic category, first syllable, first grapheme, presentation

of the item as a color drawing; or the item' s associated sound). If this did not result in a

correct response, the computer provided the answer. Compared to the control group,

Ousset et al. (2002) reported a significant improvement from pretest to posttest for the

treated items in the lexical therapy group (i.e., a treatment effect) but generalization to

untreated items was not significant. The investigators also commented about the possible

benefit of episodic long-term memory reinforcing the association between the obj ect' s

form and the name of it. Although the narratives (i.e., part of the lexical therapy) were

designed to assist in memorization of the lexical labels, the participants were better at

retrieving the words that were not in the narratives. Ousset et al. (2002) interpreted this

reduced naming as a possible indication that working memory was over-extended by the

semantic context in the narratives.









The analyses from the cues revealed that the presentation of the color drawings and

the initial syllable were the most effective cues, and Ousset et al. (2002) hypothesized

that both of these improved the process of searching the lexicon. The semantic category

cue was the least effective. To account for this, the authors suggested that the naming

deficit in their patients primarily affected their ability to access the phonological form

from semantics. Ousset et al. (2002) questioned if the participants actually had a semantic

deficit and used the training as a semantic intervention. Alternatively, it was hypothesized

that the participants were a subgroup of the PAD population, such that their anemia

might have been due to a retrieval deficit, and the lexical therapy provided rehearsal

targeting episodic lexical information and linking it to obj ects and their corresponding

names (Ousset et al., 2002).

This group study by Ousset et al. (2002) clearly showed treatment effects from the

training and provided an innovative approach to remediate anemia. Trained words were

unrelated to each other. It is unclear why Ousset et al. (2002) did not report

generalization to untrained items, when the training items were from three categories.

Also, further examination of the results indicated that there was generalization to another

ta~sk. The participants showed better naming performance at the posttest compared to the

pretest, but the treatment involved narratives and oral naming to definition, not naming

pictures. Generalization to another task suggests that the treatment strengthened

connections between semantic and phonological representations (Hillis, 1998). As

mentioned earlier, the Eindings from Ousset et al. (2002), which indicated that the

category cue was the least helpful on the naming task, could provide further support for

the notion that semantic category structure in PAD is still intact. The rationale for the use









of the semantic categories as the semantic cue was not addressed. Given the evidence of

preserved superordinate semantic category knowledge (Chertkow & Bub, 1990; Flicker

et al., 1987; Hodges et al., 1991; Huff et al., 1986; Lukatela et al., 1998; Martin, 1992;

Martin & Fedio, 1983; Nebes et al., 1986; Salmon, Butters et al., 1999; Schwartz et al.,

1979; Warrington, 1975) but degraded attribute knowledge (Martin, 1992) it is not

surprising that the category cue had minimal effects. Perhaps a more potent semantic cue

would have been to provide more distinguishing information about the item or possibly a

picture of it with some foils. This might have led to better performance on the oral

naming to definition task.

A recent single subj ect design study (Fuller et al., 2001; Rothi et al., 2005) also

used an acetyl cholinesterases inhibitor, but with an errorless learning paradigm to

improve confrontation naming in six adults with PAD. The scores for the patients on the

MMSE were > 10. In this stud the stimuli consisted of 100 words, evenly divided for

high and low frequency, matched with black and white line drawings from eight semantic

categories representing both natural kinds (e.g., animals) and artifact categories (e.g.,

clothing) with 10 exemplars per category. Three subsets of stimuli based on baseline

performance were developed for each participant using three categories, allowing for

training on two of these, while the third subset was used both for generalization and

experimental control. To establish baseline stability, each participant named the 100

pictures over eight daily probes (i.e., over eight sessions). Sixty minute therapy sessions

were provided four times per week, until criterion was met (90%) or after 20 sessions

(20-3 5 sessions were needed to complete the entire protocol) (Fuller et al., 2001; Rothi et

al., 2005).









The therapy included two similar treatment phases to determine if the

improvements seen in Phase I would show response generalization to performance from

Phase 2, which at that point would be an untrained set of stimuli. There were two

treatment conditions, immediate and delayed repetition. For Phase One, Treatment

Condition One involved simultaneous repetition in which the clinician presented the

picture, stated the correct name of it and had the participant repeat it. For Treatment

Condition Two, delayed repetition was used in which the clinician presented the picture

and if the participant knew the name, he/she stated it. If/when the participant did not

know the name of the item, he/she informed the clinician. Then, the clinician stated it and

had the participant repeat it (Fuller et al., 2001; Rothi et al., 2005). In other words, there

was a delay from the time the participant saw the picture until he/she repeated. The

results revealed that half of the participants (i.e., three of six) showed a treatment effect.

There was no generalization to the untrained items. However, results from the three

month maintenance probes showed that the treatment effect was still evident. The

investigators also conducted a post hoc review of the participant' s records and identified

two other factors that appeared to play a role in the outcomes. The three participants who

responded to the treatment both lived at home (versus in an institution) and were not on

medications that could affect the learning process and brain plasticity (Fuller et al., 2001;

Rothi et al., 2005).

This study (Fuller et al., 2001; Rothi et al., 2005) was the first to demonstrate

successful application of errorless learning (i.e., treatment and maintenance effects) in a

naming treatment in the PAD population. Furthermore, the influence of both the living

status and the medications are important considerations for treatment planning. Perhaps









another reason for the treatment effect was the use of semantic categories (both man-

made and natural kinds) and their exemplars.

These three studies (Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al.,

2002; Rothi et al., 2005) were theoretically motivated, well-designed, and offer different

approaches for improving the naming performance in adults with PAD, including single

subject and group design. More specifically, two of these treatment approaches combined

cholinergic medication with training items from semantic categories, one in a lexical

training with narratives and naming to definition (Ousset et al., 2002) and the other via an

errorless learning paradigm (Fuller et al., 2001; Rothi et al., 2005). Generalization to

untrained items essentially did not occur in any of these studies. However, the finding

that treated behaviors were maintained for one participant two weeks after training

(Abrahams & Camp, 1993) and for three participants three months after training (Rothi et

al., 2005) is remarkable. This reflects the responsiveness to the training and that the effect

was robust despite the MMSE scores. The lack of generalization to untrained items

following anemia treatment in the PAD warrants further investigation.

According to Nadeau and Gonzalez Rothi (2004), the connectionist approach views

anemia (that is caused by a semantic deficit) as a reflection of insufficiently engaged

representations of features that are critical for making distinctions among concepts. When

a network is damaged, a large amount of information still remains in the network, so the

focus should be on refining the damaged network via semantic therapy. In particular, the

network needs to be changed in terms of its connectivity so that there is more reliable

engagement of the distinguishing features, while simultaneously there is relatively a

disengagement of the shared features (Nadeau & Gonzalez Rothi, 2004). These findings










suggest that participants with early PAD might therefore be appropriate for the

semantically based training provided in our study which compared typical and atypical

category exemplars.















CHAPTER 3
LITERATURE REVIEW 2: A BRIEF OVERVIEW OF THE ORGANIZATION OF
SEMANTIC MEMORY

Since this study relied heavily on the predictions of theories about semantic

category structure, it is important to provide a brief overview of some semantic theories

and categorization. This is followed by a discussion of the two views which can be used

to compare training with typical and atypical category exemplars. The family

resemblance/prototype (Rosch, 1975) view provides support for training typical items,

while support for training atypical items to achieve maximum generalization comes from

the Complexity Account of Treatment Efficacy (Thompson et al., 2003), and a

connectionist model of a computer simulation of acquired dyslexia (Plaut, 1996). Next,

there is a brief description of the four studies that have compared these views in the

context of a naming intervention with typical and atypical category exemplars for adults

with aphasia (Kiran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005,

2006). A rationale for applying this training in adults with mild to moderate probable

Alzheimer disease (PAD) is also be provided. This leads to a discussion of the current

study .

A Brief Overview of Semantic Memory and Categorization

Semantic memory has been described as a hierarchical network consisting of

conceptual information and knowledge, storing semantic representations of facts,

knowledge for obj ects and concepts, words and their corresponding meanings, as well as

associations (Au & Bowles, 1991; Bayles, Kasniak, & Tomoeda, 1987; Tulving, 1983).









Within the semantic system, the representations are organized in a hierarchical order and

distributed throughout the association cortices of the brain (Marshall, 1988; McCarthy &

Warrington, 1990).

It is important to briefly describe the networks that allow for this organization at

the word level. The spreading activation theory (Collins & Loftus, 1975; Quillian, 1967)

suggests that there are at least two levels of representations, semantic, which is organized

based on semantic similarity of the meaning of the word, and lexical, which is organized

based on the phonemic similarity (and to an extent, orthographic similarity). Furthermore,

concepts have been hypothesized to have representations as nodes in a network that have

features of the concept that are represented as associated connections to other nodes

(Collins & Loftus, 1975; Quillian, 1967). These links are based on the importance or

relevance of the information to the concept' s meaning such that more important nodes

have shorter links or distances between two concepts, and together the nodes make up the

network (Collins & Loftus, 1975; Quillian, 1967). The similarity allows for information

to spread along pathways in the network (Collins & Loftus, 1975; Quillian, 1967).

Support for the notion of a semantic network and the spreading activation theory

(Collins & Loftus, 1975; Quillian, 1967) comes from the semantic priming effect,

initially described by Meyer and Schvaneveldt (1971), in which participants had faster

reaction times when the prime in a lexical decision task was associated with the target,

but were slower when it was not associated. For example, responses to "butter" were

faster when preceded by "bread" than when preceded by "window." It was theorized

these faster responses (i.e., facilitation) occurred as a result of spreading activation due to

shared semantic contexts. The decision was faster because there was an increase in the









availability of the target as a result of the prime. More specifically, the memory system is

connected to the node for butter, so when one is activated, the activation spreads to

connected nodes (Meyer & Schvaneveldt, 1971). According to Martin (1992), semantic

priming assumes that there should be automatic activation of an obj ect' s representation

upon presentation of the obj ect (picture presentation or word), and related concepts

should be activated by the spread of activation, which increases accessibility to the name

of the object.

Martin (1992) further suggested that these explanations of semantic priming are

based on the assumption that the brain is a distributed neural network that instantiates

semantic representations. For example, presentation of a picture results in activation of a

corresponding neural network, which includes representations of the attributes of the

item. Items that are closely related have many shared attributes and overlapping networks

that represent them (Martin, 1992). This also explains how the semantic network

responds to such information, and when activated above the threshold, semantic

processing allows for selection of the target word instead of others (Boyle, 2004; Boyle

& Coelho, 1995; Coelho et al., 2000). Although there are different theories related to this

topic, it is beyond the focus of our study.

The contents of mental representations have been used to account for a range of

phenomena such as knowledge of correctly recognizing obj ects by their name (i.e.,

word' s label) (Medin & Smith, 1984). There is a large body of research on semantic

categories and how they are represented (Kiran & Thompson, 2003a). This literature

review focuses on the superordinate classification of a category, which is a higher-level









grouping allowing for concrete obj ects to be classified into various divisions (Rosch,

1975).

There are several benefits of categorization, for example, it is both economic and

informative. In terms of economy, categories frankly provide a practical and adaptive

means for classifying similar items in order to avoid the overwhelming cognitive task of

considering all items as if they were unique (Anderson, 1991). Categorizing also allows

for assumptions about information that is not explicitly provided, for example, that the

item will share similarities with other items in that category (Anderson, 1991; Komatsu,

1992).

Historically, different views have been put forth to account for concepts and

categorization. Within some theories of psychology and philosophy the nature of

categories is described as Aristotelian (Rosch, 1973a, 1975): They are logical; have clear

boundaries; and criteria-based membership. Thus, being a category member requires

having a basic set of essential attributes or features such that each item is fully and

equally a member. This approach is known as the classical view which suggests that the

representation of categories is achieved by having a set of defining features (Kiran &

Thompson, 2003a).

The classical view was challenged by Wittgenstein (1953) who argued that the

requirements of formal criteria were not necessary from either a logical or psychological

point of view (Rosch & Mervis, 1975). Despite critique, the classical view was

researched during the 1960s and considered to be adequate as an appropriate way to

describe lexical concepts found in everyday living (Komatsu, 1992). However, by the

1970s, problems with the classical view surfaced, including a lack of evidence for









"defining properties" of obj ects and concepts, identification of categories that did not

have clear-cut boundaries, and evidence suggesting that there is inequality among

category members, and resulted in its decline (Kiran & Thompson, 2003a; Komatsu,

1992; Medin & Smith, 1984; Posner & Keele, 1968; Rosch, 1973a, 1975). This led to the

development of other theories, including the family resemblance/prototype view (Rosch,

1975; Rosch & Mervis, 1975), which has been described as being among the most widely

cited psychological concepts (Hampton, 1995) and is discussed in more detail below.

Family Resemblance/Prototype View

Rosch (1973a; Rosch, 1975) argued that for many natural semantic categories, the

Aristolean approach does not apply. "Family resemblance" is the term that was applied to

highlight the structural principles governing category membership. The notion of family

resemblance means that there is an internal structure to the category, such that it is

organized around a category prototype (Rosch & Mervis, 1975). This allows for a very

general approach to the relationships and for application to categories whether they share

features with other category members or meet formal criteria for membership in the

category. Some of the original work on this topic (Rosch, 1973a) revealed that the

prototype of the category is the "best example" of a category and/or "the clearest case" of

category membership. The prototype occupied the central location within the category,

surrounded by other category members that differed in degree of similarity to the

prototype, and, thus, were not equal in terms of category membership (Rosch, 1973a).

Rosch (1975) reasoned that there were graded representations of what the category

represented signified. Consequently, some items were better representatives of a

category. This approach offered an alternative to the classical view, with both its need for

clear-cut boundaries for categories and criteria of defining features for category









membership. Rosch (Rosch, 1975; Rosch & Mervis, 1975) theorized that categories have

an internal structure such that they are organized by a family resemblance to a

prototypical category member, and it was the degree and type of resemblance to the

category prototype that was the foundation of the category structure.

More specifically, the notion of categories having an internal structure of categories

has been tested by asking healthy participants to use a seven-point scale to rate how well

a category member fits their image of a given category's name (Rosch, 1975; Rosch &

Mervis, 1975). The results showed that some items were more representative than others

of that category and these ratings, because they shared features with other items in that

category. Thus, they had a family resemblance. These were considered good examples,

while others were considered to be poor examples because they did not. These ratings

also have been found to be reliable predictors of category verification task performance

in which the participants are presented with a statement (e.g., a _(i.e., an exemplar) is

a _(i.e., a category)) and decide if it is true or false. Faster reaction times and higher

ratings are found when an item shares features with the category prototype. For example,

the exemplar robin is a good example because it is much closer to the prototype,

compared to a penguin (a poor example); participants were faster and more accurate

verifying that a robin was a bird than that a penguin was a bird (Rips, Shoben, & Smith,

1973; Rosch, 1973a, 1975; Rosch & Mervis, 1975). This is called the typicality effect and

additional reports of it are discussed below. It has been suggested that the organization of

semantic categories that is based on family resemblance to a prototype is relevant to

semantic memory and retrieval (Rips et al., 1973; Rosch, 1973b, 1975).









One of the corollaries of the family resemblance/prototype theory is that the most

prototypical members of a category have the most shared attributes with other items in

that category and, consequently, represent the central tendency of the category (Rosch,

1975; Rosch & Mervis, 1975). Moreover, as a result of being a prototype defining the

category, the prototypical members have the least family resemblance with other

categories (Rosch, 1975; Rosch & Mervis, 1975). Furthermore, the more typical items in

a category are, the more they share features with the prototype and with each other (e.g.,

robins, bluej ays) while the more atypical items share fewer features with the category

prototype and also have more in common with items from other categories (e.g.,

penguins, ostrichs). Therefore, categories have an internal organization somewhat like a

target, such that the prototype is at the center of the bull's-eye, surrounded by typical

category exemplars that have many overlapping features with each other and the

prototype. Atypical items are more distant from the prototype and share fewer features

among themselves and with the prototype (Rosch, 1975; Rosch & Mervis, 1975).

Some of the conclusions from the family resemblance/prototype view (Rosch,

1973a, 1975; Rosch & Mervis, 1975) are applicable to our study. First, there are clear and

pervasive findings suggesting that semantic categories have internal structure that

influence the processing of words. This is inconsistent with the classical view, but

compatible with the spreading activation theory of semantic memory (Collins & Loftus,

1975; Quillian, 1967). Second, faster reaction times to pictures of obj ects than words,

would suggest that the underlying nature of the semantic representation could be closer to

pictures (i.e., more visual than verbal) (Rosch, 1973a, 1975; Rosch & Mervis, 1975).









A wide range of studies have investigated the predictions of the family resemblance

view. The typicality effect, the finding of better performance (e.g., faster reaction times

and better accuracy) for typical compared to atypical category members has been widely

reported in non-brain damaged adults (Hampton, 1995; Kiran & Thompson, 2003a;

Posner & Keele, 1968; Rips et al., 1973; Rosch, 1973a, 1975; Rosch & Mervis, 1975).

The various tasks tend to involve words, and these are discussed first, for example

in category exemplar rating tasks (Rosch, 1975; Uyeda & Mandler, 1980) and priming

paradigms involving category verification (Fujihara, Nageishi, Koyama, & Nakajima,

1998; Kiran & Thompson, 2003a; Rosch, 1975; Rosch & Mervis, 1975; Smith, Shoben,

& Rips, 1974). Patterns have also been used in a task requiring learning items in a

category (Posner & Keele, 1968). A description of the task is provided below, and this is

followed by some studies that suggest that the use of pictures might be different from

words.

In the priming studies that used the category verification task with words, the

superordinate semantic category name was used as the prime, and the target was a typical

or atypical category exemplar from that category. Participants were instructed to decide if

the target item was a member of that category. The results revealed faster reaction times

and more accurate responses for primes that had been rated as good examples, and these

were typical items. Longer reaction times and lower accuracy scores were found for poor

examples, and these were mostly atypical exemplars (Fujihara et al., 1998; Kiran &

Thompson, 2003a; Rosch, 1975; Rosch & Mervis, 1975; Smith et al., 1974).

The typicality studies using categorization tasks offer a comparison of how typical

and atypical items are processed in these tasks (Collins & Loftus, 1975). Typical items









have distinguishing features that reduce a positive decision on the basis of the

superordinate connection (Collins & Loftus, 1975; Quillian, 1967). Also, superodinate

connections have different strengths which relates to accessibility, which is dependent

upon use (Collins & Loftus, 1975; Quillian, 1967). Furthermore, there is a high

correlation between typicality ratings and accessibility (Collins & Loftus, 1975; Quillian,

1967). Thus, the typicality effect provides more evidence that semantic similarity

increases the speed of making a positive decision and reduces the time for making a

negative decision. In contrast, increased reaction times are found for atypical items,

because, based on the overlap of features, they represent only some of the category that

they belong to and might overlap with other categories as well (Collins & Loftus, 1975;

Kiran & Thompson, 2003a; Quillian, 1967; Rosch, 1975; Rosch & Mervis, 1975).

Further support for the typicality effect comes from a study using event-related

potentials (i.e., ERPs) during a category verification task in healthy adults (Fujihara et al.,

1998). Event related potentials measure electrical manifestations of particular

psychological processes that occur in preparation for or in response to discrete events

(Fabiani et al., 2000; Kutas & Hillyard, 1980). The N400 is a negative going deflection

occurring at ~400 ms and occurs in response to anomalous information, specifically,

semantic violations; for example it is larger when the prime is unrelated to the target

compared to when it is related (Fabiani et al., 2000; Kutas & Hillyard, 1980). Fujihara et

al. (1998) found that participants had different responses to typical and atypical words.

The N400 were more negative after atypical words, than they were after typical words. A

typicality effect was found (i.e., fastest and most accurate responses occurred after the

typical words). This led to the conclusion that greater priming occurred during typical









words compared to atypical words. This was interpreted to be consistent with the

hypothesis that the prototype represents the concept of the category, which is the central

tendency for each of the category members (Fujihara et al., 1998).

Pictures have also been used as stimuli in category tasks, and according to

Snodgrass and McCullough (1986) pictures from a category look similar to each other

compared those from different categories, for example, they cite the finding by Rosch

(Rosch, 1977, 1978) that based on visual similarity ratings of items from categories of

fruit, animals, and vehicles, the ratings were higher for the ones that looked similar to

each other. Snodgrass and McCullough (1986) provided more evidence that the time to

decide if a picture belongs to a category could be increased or decreased based on the

similarity of the items in the category. The investigators suggested that people can

categorize a picture and a word via semantic access, but with a picture, people can just

use the visual stimulus or the prototype of the category. It was suggested that both visual

and semantic strategies are used. Snodgrass and McCullough (1986) thus concluded that

picture categorization is not an appropriate task for a comparison of speed of

understanding of words and pictures. Although this study did not compare typical and

atypical items, it might have referred to them in terms of similar (i.e., typical) and

dissimilar (i.e., typical). There is at least one study that provided evidence for the notion

that identification of objects first occurs at a level of abstraction that is not the greatest

degree of general or specific "entry point", but instead varies based on typicality

(Jolicoeur, Gluck, & Kosslyn, 1984). This suggested that identification of typical items

occurred at the "basic level" compared to atypical items, which are subordinate to this

level (Jolicoeur et al., 1984).









The term representativeness has also been used to refer to typicality such that high

representativeness suggests typical items and low representativeness indicates atypical

items (Grossman, Robinson, Biassou, White-Devine, & D'Esposito, 1998). Grossman et

al. (1998) has reported that pictures were used with healthy older adults and they showed

sensitivity to the degree of typicality (i.e., typicality effect). Another study using pictures

found that older adults showed typicality effects with both category judgment tasks and

exemplar judgment tasks (Cobb, 2005).

All of these studies examining the typicality effect have involved healthy, non-

brain damaged adults. The findings have shown that typical items are processed faster

and more accurately. Several of these and other studies were primarily focused on adults

with brain damage, but it was important to reveal the pattern in the normal population

prior to addressing the patient groups. These include adults with probable Alzheimer

disease and other research with adults with aphasia. Some examples are discussed next.

Some studies have found typicality effects in adults with probable Alzheimer

disease (PAD) using category verification tasks (Cobb, 2005; Grossman et al., 1998).

Other variations of this task, for example ranking attributes or ordering the

representativeness of category exemplars (Cronin-Golomb et al., 1992) and some also

included rankings of dominance, which relates to the significance of the feature to the

meaning of the item (Nebes et al., 1986; Nebes & Brady, 1990). In a category fluency

task adults with PAD have shown normal variations in the typicality of the items

produced (Ober, Dronkers, Koss, Delis, & Friedland, 1986). Grossman et al. (1998)

found normal typicality effects on a category judgment task, but this was only in the










subgroup of PAD participants who did not have a semantic impairment, but had naming

deficits.

In contrast, Grossman et al. (1998) reported that the other subgroup of their PAD

participants had a semantic deficit, demonstrated naming deficits and insensitivity to the

representativeness. These participants did not differentiate among the levels (low,

moderate, and high) and this was attributed to compromised ability to make similarity

judgments (Grossman, 1981; 1998). Other PAD studies have shown no typicality effects

or otherwise differences in processing atypical and typical item in this population, for

example impaired performance on ranking attributes (Grober et al., 1985). Sailor et al.

(2004) compared the responses of healthy older adults and PAD participants on a

category generation task. The analyses revealed that compared to healthy older adults, the

PAD group produced significantly fewer atypical items. This was the trend across the

categories of fruit, vegetables, animals, and footwear. It was concluded that the PAD

group had a slower search rate through semantic memory compared to the healthy older

adults (Sailor et al., 2004). Also, as mentioned in the previous chapter, a property

verification task revealed worse performance on category generation low typical items

(Smith et al., 1995).

Although most of these studies indicated that the PAD population has deficits on

atypical items, to the best of our knowledge, there is at least one study that showed better

performance on atypical items (Cobb, 2005). The purpose of this study by Cobb (2005)

was to attempt to understand if typical items were affected by the pathology of Alzheimer

disease. Using the bottom-up deterioration hypothesis in PAD, Cobb (2005) theorized

that the vulnerability of typical items compared to atypical items could be described as:









Generalized (equal); Strength (not as vulnerable); Distinctiveness (more vulnerable).

Category judgment tasks and exemplar judgment tasks were used in which pictures were

shown and the participants verified them. The results indicated that like the healthy older

adults, the PAD group showed a typicality effect on the category judgments (i.e., faster

and more accurate responses were found on the typical items). In contrast, on the other

task of verifying the exemplars, the PAD group showed deficits on the typical items

compared to healthy older adults. Cobb (2005) suggested that these Eindings provide

some evidence that the typical items were more vulnerable to the deterioration, and thus

provided support for the distinctiveness model. However, Cobb (2005) also cautioned

about generalizing the results because of this preliminary study due to some issues related

to methodology.

Research on the typicality effect has also been conducted in the aphasia population.

In a category generation task, Grossman (1981) compared the number and typicality of

the exemplars generated (i.e., typical and atypical) between a group of participants with

nonfluent aphasia and another group with fluent aphasia. The majority of the items

produced by the nonfluent group were typical items. Grossman (1981) suggested that this

occurred because these participants used the central tendency of the category as a guide

and compared the target item to it. In contrast, the fluent aphasia group produced a more

even distribution of typical and atypical exemplars, starting out with more typical items

(like the nonfluent group), but then producing items that were as close to the central part

of the category and even from other categories. Moreover, participants with fluent

aphasia actually produced many responses that were not members of the specified

category. To account for these Eindings, Grossman (1981) suggested that these










participants were aware of the central tendency of the category but had reduced mental

representations related to the category's border. Thus, they violated category

boundaries(Grossman et al., 1998). Grober et al. (1980) used a category verification task

in two modalities (i.e., pictures and words) to determine if performance on these tasks

could distinguish groups of patients with anterior (i.e., nonfluent) aphasia and posterior

(i.e., fluent) aphasia. There were no differences between performance with pictures and

words for any of the patient groups. Furthermore, across groups, the typicality effect was

realized as both faster response times and better accuracy for the typical exemplars.

However, while the nonfluent patients were fairly accurate in their classifications of

atypical category exemplars, this was not the case for the fluent patients. When the item

clearly belonged to a given category, the fluent group was accurate, but when the item

was at the category boundary, they were less accurate. Grober et al. (1980) suggested this

difficulty with atypical items was due to a disruption of the underlying semantic structure

of semantic categories in fluent aphasia. Although this task was different from the

category generation task used by Grossman (1981), the findings for the patients with

fluent aphasia are compatible.

In a category verification study, Kiran and Thompson (2003a) compared the

performance of two groups of adults with fluent and nonfluent aphasia to young and older

adults without brain damage. Typicality effects were expected to occur in all of the

groups. However, based on previous research (Grober et al., 1985; Grossman, 1981), the

pattern of activation for the fluent aphasia group was expected to be different. Kiran and

Thompson (2003a) used words from the categories of vegetables, birds, and fish

typicality ratings that they developed based on the seven-point scale used by Rosch










(1975). Participants first read a category name and decided if the next word was a

member of that category or not. Furthermore, the overall results were consistent with the

predictions: typicality effects were found in all the groups, and the fluent group displayed

a different pattern of performance from the other groups. The fluent group had the

smallest effect of typicality and the greatest number of errors. Although this group was

more accurate on the typical items compared to the atypical items, their reaction times for

accepting the correct typical items were not significantly faster than the correct atypical

items. Kiran and Thompson (2003a) interpreted this as two different manifestations of a

semantic impairment from the fluent aphasia. The first was that weakened boundaries of

the target category led to increased numbers of errors. The second difficulty was that

there was a decrease in the ability to access the prototype and this caused a deficit when

comparing the target with the prototype of the category. They concluded that the findings

of longer reaction times and more errors with atypical exemplars suggested impoverished

associations linking atypical exemplars with their category's prototype. Kiran and

Thompson (2003a) theorized that these findings could offer more information about the

nature of atypical category exemplars as having differential representation in semantic

memory. Furthermore, the greater number of dissimilarities among atypical items in a

category allows them to convey a broader range of variation across the category, and thus

they were considered to be "more complex" (Kiran & Thompson, 2003a).

Thus, in the aphasia literature, there is at least one study suggesting that fluent

aphasia affects the processing of atypical category exemplars in a category generation

task (Grossman, 1981). This is understandable given that category generation requires

searching semantic memory for an item within a category when only provided with the









superordinate category label. Some category verification task studies have shown

typicality effects in adults with nonfluent aphasia (Grober et al., 1980; Kiran &

Thompson, 2003a; Stanczak et al., 2005), including transcortical motor aphasia (Stanczak

et al., 2006), and fluent aphasia (Stanczak et al., 2005), including conduction aphasia

(Stanczak et al., 2006). In others studies, the typicality effect was found but was

somewhat disrupted in adults with fluent aphasia (Grober et al., 1980; Grossman, 1981;

Kiran & Thompson, 2003a), as shown by more difficulty (e.g., slower or less accurate)

on atypical items. Although the typicality effect indicated better performance with typical

items, at least one study did not find that typical items were verified faster than atypical

items that were correct (i.e., member of the given category) by the adults with fluent

aphasia (Kiran & Thompson, 2003a).

In summary, the family resemblance/prototype view (Rosch, 1975; Rosch &

Mervis, 1975) has had a strong influence on natural language research in category

representation (Barr & Caplan, 1987). Based on the family resemblance/prototype

hypothesis, there are several reasons for training typical category exemplars and they

could have an advantage over training atypical category exemplars. Typical items have

richer representations, and people have greater knowledge of them so they are used more,

and therefore accessed faster (Heilman, 2005, personal communication). Evidence from

adults with brain damage suggests that knowledge about atypical items is disrupted

following strokes to the posterior portion of the left hemisphere (Grober et al., 1985;

Grossman, 1981; Kiran & Thompson, 2003a), and also as a result of PAD (Sailor et al.,

2004; Smith et al., 1995).









The Complexity Account of Treatment Efficacy

As described above, atypical items have been hypothesized to be more complex

(Kiran & Thompson, 2003a). The Complexity Account of Treatment Efficacy (CATE)

was developed by Thompson et al. (2003) based on their study in which adults with

agrammatic aphasia were trained to use sentences that were syntactically complex. As a

result, participants showed generalization to less complex sentences whose structures

shared processes with trained items. Several studies by Thompson and colleagues

(Thompson et al., 1998; Thompson et al., 1997; Thompson et al., 2003) have reported

that adults with aphasia have shown generalization following treatments in which

complex syntactic structures were trained. This was because when more complex items

were trained, this included variables that are relevant to items that are simpler. This

promoted greater degrees of access to untrained items compared to the training of simple

items (Thompson et al., 1998; Thompson et al., 1997; Thompson et al., 2003). Thompson

et al. (2003) concluded that these findings were compatible with other studies in the

literature and suggested that enhancement of performance is found under the following

two conditions: when there is related linguistic structure and when treatment moves from

greater degrees of complexity to lesser degrees of complexity. In addition to acquired

language disorders, complexity has been used by Gierut (2001) as a guide for stimuli to

promote generalization with children who have phonological disorders. The findings also

have been positive and indicate that with greater degrees of target complexity, there are

greater degrees of gains in phonological skills; these behaviors have been shown to

generalize to untreated sounds (Gierut, 2001). These principles are borrowed from motor

skill learning in adults and conditions under which the behavior is practiced (Schmidt &

Lee, 1999). Generalization to untrained items in a picture naming study has also been









found to be successful when the focus has been on atypical items as a means for stimuli

selection because they have a hierarchical complexity (Kiran & Thompson, 2003b). This

and other behavioral studies are discussed below. However, it is important to first

understand the initial work in this area by Plaut (1996) in which he compared the training

of atypical and typical items in a connectionist model of acquired dyslexia.

Connectionist Model and Computer Simulations

Prior to describing the computer simulations by Plaut (1996), a brief overview of

the connectionist or parallel distributed processing (PDP) model is provided. The PDP is

a computational framework that allows for exploration of cognitive processes of normal

and damaged systems. Simulations are conducted to learn more about the processes and

to contribute to the development of potential rehabilitation programs (Plaut, 1996).

Parallel processing is the term used to describe how individual processing units (similar

to neurons), in a large array, are connected, but each performs a basic function and

represents a component of the modeled entity (Nadeau, 2000; Plaut, 1996; Rumelhart,

McClelland, & PDP Research Group, 1986). The units are linked by connections and the

strength of these connections reflects the knowledge of the network (Nadeau, 2000;

Nadeau & Gonzalez Rothi, 2004). Large patterns of activation, based on excitatory and

inhibitory signals, represent a concept which is referred to as a distributed representation

(Nadeau, 2000; Plaut, 1996; Rumelhart et al., 1986).

Plaut (1996) tested the family resemblance theory using a connectionist computer

model in which he simulated acquired dyslexia in three experiments. Only the first two

are discussed here. First, Plaut (1996) trained the model, then lesioned it to simulate the

effects of stroke. Finally, he retrained the model to simulate therapy after stroke in order

to better understand the degree and speed of recovery; how generalization occurs from









treated to untreated items, and the way this generalization might be maximized via

selection of the treatment items.

In Plaut' s (1996) first experiment, he programmed the network and trained it to

generate semantic information of about 40 words from five semantic categories (natural

kinds and man-made) based on orthographic input. The training was done to allow

mapping from the written words to their corresponding meanings. Then, Plaut (1996)

damaged the network by random selection and removal of some of the units representing

semantics and others representing orthography. Plaut (1996) randomly identified items

for retraining: half of the words that were trained correctly and half that were not trained

correctly. The remaining words were not retrained. Plaut (1996) reported that, compared

to initial learning, retraining was faster because there was a re-establishment of

consistency that was relevant among words and thus untreated words improved. Plaut

(1996) suggested that location of the damage was important. When the semantic level

was damaged, there was fast relearning and generalization was substantial. However,

when the orthography level sustained damage, relearning was slower and generalization

did not occur. Thus, there was better generalization at the semantic level because the

words had relationships based on their meanings. This was based on the structure of the

semantic organization of the set of words. Plaut (1996) hypothesized that the part of the

system that was damaged could influence recovery. Furthermore, Plaut (1996) suggested

that generalization had improved because the semantic categories from which the words

were derived had a substantial amount of overlap in their semantic representation. This

experiment provided support for using items from semantic categories in therapy because

the consistency inherent in the items in the category allowed for efficient processing and









generalization. These findings also offer support for the family resemblance/prototype

hypothesis (Rosch, 1975; Rosch & Mervis, 1975) for achieving a treatment effect.

However, in terms of maximizing generalization, Plaut (1996) provided evidence for the

benefit of training atypical items, and this was discovered in Experiment 2.

Plaut (1996) theorized that generalization could be obtained if the nature of the

semantic representation of the items was a good estimate of semantic structure relevant to

the whole set of words in the training stimuli. Selection of words for training were based

on the word' s semantic representations (Plaut, 1996). Acknowledging that nouns are

categorically organized, Plaut (1996) also used typicality, a semantic variable that is

essential because it reflects the proximity to the central tendency of a category (Rosch,

1975). Plaut (1996) hypothesized that the extent of generalization from retraining is

affected by the relative typicality of the treated words (i.e., typical versus atypical).

The training items that Plaut (1996) chose were 100 artificial "words" from one

semantic category. These words were evenly divided into two sets in which the meaning

was generated by varying the degree of featural overlap between a prototype item and

other items in the category. Thus, the semantic features for the typical words were closer

to the prototype of the category; whereas the features for the atypical words were further

from the prototype. Training the network involved iterations that allowed for generation

of the correct semantic representation for each word when the corresponding

orthographic form was presented. Then, to lesion the network, Plaut (1996) randomly

selected and deleted connections. Next he retrained 25 of the typical words and 25

atypical words in different networks. This allowed for evaluation of the impact of

typicality on generalization to untrained typical items and untrained atypical items. The









results were consistent with Plaut's (1996) predictions. The lesion rendered typical words

more impaired compared to atypical words. Thus, Plaut (1996) suggested that typical

words were more vulnerable to damage because they had more competitors that were

close, due to the overlap. Thus, distinguishing typical words from other typical words

(compared to distinguishing atypical words) required better accuracy by the network. In

contrast, there were not as many competitors for atypical words, and distinguishing

among atypical words was not as difficult as it was with typical words. However, as a

result of retraining, trained typical words had significantly better performance compared

to trained atypical words (Plaut, 1996).

Although these findings are relevant to recovery, Plaut (1996) argued that the more

critical finding is the fate of the untreated words, because this related to choosing stimuli

that could influence generalization, an important component of rehabilitation that

clinicians could control. When the training was with either typical or atypical words,

there was overall substantial improvement in the untreated typical words. However,

atypical words only improved when the network was trained with atypical words. When

atypical words were retrained there was more generalization to both untrained atypical

and typical words compared to when typical words were trained. In contrast, when the

network was trained with typical words, this had a negative impact on performance of

untreated atypical words. To account for these results, Plaut (1996) hypothesized that

typical words just focus on the central tendency of its category, without indicating the

semantic features that can differ from the prototype Thus, the typical words offer a good

estimate about the degree of variation in the semantic features that is shared by other

typical words, but not the information about the variation for the atypical words. In









contrast, atypical words indicate the amount of variation that can exist across members of

a category while also approximating the category's central tendency. Thus, Plaut (1996)

hypothesized that atypical words collectively estimate the semantic representations that

span more of the features required by the full set of words compared to representations of

typical words. Consequently, training atypical words should result in generalization to

typical words, because the semantic representations of the atypical words included

information about the degree of variability within the category structure as well as

information about the central tendency of the category (Plaut, 1996). While Plaut (1996)

recommended using this approach, he cautioned that the atypical items should not be

extremely atypical. The semantic dimensions should be relevant and thus allow for the

average effects to be near the category's central tendency. Plaut (1996) recommended

that future research endeavors should involve training atypical items from several

categories at the same time.

In summary, the PDP model has been described as operating in a manner consistent

with the brain (Nadeau, 2000; Plaut, 1996; Rumelhart et al., 1986). Overall findings from

Plaut (1996) provided a rationale for training items within a semantic category. Also,

while training typical items resulted in better performance (i.e., acquisition or a training

effect), training atypical yielded better generalization to untrained items. The results from

Plaut (1996) are very positive and have great implications for clinical application. In

particular, both category structure and atypical items are important in promoting

generalization because they relate to the semantic representations of the items. However,

it is important to remember that computer simulations involve isolated information and

events, in which the experimenter programs all of the inputs and thus has complete









control over the information in the model. For example, the second experiment used

artificial words from one category and the typical and atypical items were created by the

network. This differs from the wide range of variation in humans in terms of semantic

memory, learning, and the effects of brain damage and behavioral manifestations of

different disorders. Nevertheless, as discussed below, studies with humans have been

conducted and some of the findings are consistent with Plaut (1996).

Similar Behavioral Studies

A series of experiments on category learning in non-brain damaged college

students was conducted by Posner and Keele (1968). Participants looked at patterns with

varying degrees of distortions. These distortions ranged from high (i.e., the distance from

the central tendency of the category was greater) to low (i.e., closer to the central

tendency). The participants pressed a button to specify their selection for the given

pattern, and received feedback about accuracy. This was repeated over several trials to

allow for learning. Results showed that the participants had better generalization from

learned patterns to new ones after training with highly variable patterns compared to

when the training emphasized low variability. To account for this, Posner and Keele

(1968) suggested that variability is an important part of the learning process. Although

this study did not involve naming or a patient population, and did not use the term typical

and atypical, the patterns with higher variability might be considered atypical, whereas

those with much lower variability might be considered typical. Furthermore, the

theoretical underpinnings are along the same lines, and can be considered support for

training atypical items (Plaut, 1996).

More recently, three different groups of investigators (Kiran & Thompson, 2003b;

Mayer et al., 2004; Stanczak et al., 2005, 2006) examined and compared the effects of









training typical and atypical category exemplars, and thus compared the two opposing

views of typicality discussed above (Plaut, 1996; Rosch, 1975; Rosch & Mervis, 1975).

Each of these therapy studies involved adults with aphasia in the context of a single-

subj ect design targeting confrontation picture naming, and showed consistent evidence of

treatment effects. However, the results were mixed, especially in terms of generalization.

A brief description of these three studies is provided below. This is followed by a brief

critique pertaining to both their strengths and limitations, highlighting how they offer

unique contributions to the literature and have implications for the current study.

Kiran and Thompson (2003b) predicted that when atypical items were trained, this

process would result in emphasizing both how much the features in the category vary a~s

well as the prototype's features. This, in turn, would result in a strengthening of the

associations among the typical items, increased access to these typical items, and finally

generalization to untrained typical items. In contrast, they argued that training typical

category exemplars would have more limited effects, because typical items have limited

variation, focusing only on the central tendency of the category (Plaut, 1996). Kiran and

Thompson (2003b) argued that, compared to typical items, the atypical items were more

complex because their semantic features have a greater degree of dissimilarity with each

other, and thus they offer more diversified information about the whole category, as a

collective unit (Plaut, 1996; Thompson et al., 1998; Thompson et al., 2003).

To test these predictions, Kiran and Thompson (2003b) provided therapy for four

adults with fluent aphasia on a naming task. Baseline naming performance was assessed

using 48 items consisting of 24 exemplars from each of two semantic categories (birds

and vegetables). The study design allowed for each participant to be separately trained on









one category at a time, starting with either eight atypical category exemplars or eight

typical items from each category. After the participant met the criteria the next category

was trained with the opposite typicality. The participants attended two hour therapy

sessions twice a week until either criteria of 7/8 correct was achieved over two

consecutive sessions or after a maximum of 20 training sessions had been completed

(Kiran & Thompson, 2003b). During training, participants performed a variety of

semantic tasks tapping both comprehension and production skills: picture naming,

category sorting, feature selection, answering yes/no questions; and picture naming again.

Participant I was first trained on typical birds for seven weeks and this led to a

treatment effect (i.e., acquisition of the trained items) but there was no generalization to

untrained intermediate or atypical items. Subsequent training of atypical birds resulted in

improved performance on those items. At that time, there were no changes on

performance of the control category (vegetables). Then Participant 1 was trained on

atypical vegetables, and in 8 weeks, acquisition criterion was observed as was

generalization to untrained intermediate and typical birds. Participant 2 was first trained

with atypical birds and in 11 weeks, met criteria and showed generalized naming on

intermediate and typical items. At that time, performance on the vegetable category

remained the same. Treatment of birds led to no change in performance on the untreated

category, vegetables, so the participant was subsequently trained on atypical vegetables.

This resulted in a treatment effect and generalization to untrained intermediate and

typical items. Participant 3 was first trained with typical vegetables and although criteria

was met, there was no generalization to intermediate or atypical vegetables. Next,

Participant 3 was trained on atypical vegetables that resulted in improved performance on









the trained atypical vegetables; no further treatment was provided with the bird category

because of the extended time for training on the first (28 weeks). Participant 4 was

trained only on atypical items and showed a pattern similar to Participant 2: a treatment

effect after 6 weeks of training on vegetables and after nine weeks on birds, and

generalization to both untrained intermediate and atypical items.

The patients in the Kiran and Thompson (2003b) study acquired atypical items

faster acquisition than typical items. Furthermore, naming of untrained typical and

intermediate items improved following training of atypical items, thus showing

generalization. However, when the typical items were trained, there was no

generalization to exemplars that were atypical or of intermediate typicality. Kiran and

Thompson (2003b) suggested that the training for the atypical items emphasized how the

semantic features in the category varied; whereas the training for the typical items was

limited to a small number of features that were shared among the other typical items.

Moreover, maintenance of the treatment effects was found 6-10 weeks after the training.

Kiran and Thompson (2003b) conducted an error analysis to gain some insights

into the effects of the treatment. Before the intervention, the errors were primarily

general, as characterized by superordinate labels, no responses, and neologisms. This was

attributed to the participants failing to retrieve the specified semantic and/or phonological

details of the target name. However, after the treatment, the participants showed more

accurate naming on trained items and on untrained items, reflecting better semantic and

phonological access to the representation. The shift in the type of errors, from general at

baseline to semantic and phonemic after the training was attributed to a greater impact of

excitation at the level of semantics and phonology that had occurred when the









participants attempted to name the items. However, this interfered with the ability to

accurately select the target name from others that were also activated. Kiran and

Thompson (2003b) suggested that although the treatment had a positive impact that was

illustrated in enhanced spread of activation to targets within a category that were related,

the intervention was not completely successful in eliminating the interference caused by

many category exemplars being activated at the semantic/phonemic level during naming.

Kiran and Thompson (2003b) suggested that there was improved performance because

the treatment was based on semantics and highlighted the underlying elements of

semantic representations and processing. The investigators suggested that this was

compatible with the literature using the semantic feature approach (Boyle, 2004; Boyle &

Coelho, 1995; Coelho et al., 2000; Drew & Thompson, 1999). Kiran and Thompson

(2003b) concluded that their findings provided more support for improving generalization

to untrained items based on using more complex training items, and thus providing more

evidence for the Complexity Approach to Treatment Efficacy (Thompson et al., 1998;

Thompson et al., 2003) as well as for training atypical items to achieve better

generalization (Plaut, 1996).

In this study, Kiran and Thompson (2003b) were the first to show that the computer

simulations of acquired dyslexia by Plaut (1996) could be successfully applied to adults

with fluent aphasia. The findings illustrated the value of training atypical category

exemplars in the context of a training highlighting semantic category structure. This

study was well designed, including the development of typicality ratings and semantic

attributes, intentionally selecting low frequency items to avoid effects of word frequency,

the use of both comprehension and production tasks to activate the entire word










representation, and a training regimen that focused on just one category at a time. Two of

the training tasks, feature selection and yes/no questions provided comprehensive

information about the items, including the written form. However, specific information

was provided by the clinician, and the participant had to select it. It is possible that the

training might have been even more robust if the participants were required to generate

the information. More importantly, there appears to be a lack of generalization probes for

untrained atypical items and untrained typical items that have the same and the opposite

typicality of the items that were trained (i.e., trained atypical and trained typical items).

Our study included a measurement of untrained atypical and typical items for the same

and opposite typicality of the trained items (i.e., that were trained with atypical and

typical items; e.g., Training on typical items and untrained typical items), and a more

active role for the participants on the training tasks. These are discussed later.

Mayer et al. (2004) attempted to replicate the findings of Kiran and Thompson

(2003b) in three participants with severe aphasia and to determine if other semantic

categories might yield potent typicality effects. Each participant had a different type of

aphasia. Participant I had global aphasia. Participant 2 was considered borderline fluent,

(i.e., fairly intact auditory comprehension and reading ability, with fluent speech and

neologisms including some word approximations). Participant 3 had mixed nonfluent

aphasia (i.e., intact auditory comprehension for basic information, and although

nonverbal, he spontaneously gestured and used facial expressions and drawings).

A total of six categories were used by Mayer et al. (2004): animals, clothing,

furniture, sports, tools, and vehicles. Each participant was trained separately on 10

atypical items from one category, and then on 10 typical items from another category.









Each category was treated for nine weeks, for a total of 18 weeks. Two other categories

were used as treatment probes (before and after treatment) and generalization probes. To

maximize support in a structured way using phonological and semantic stimulation,

training utilized a response-contingent hierarchy (Bandur & Shewan, 2001). The clinician

presented a picture and guided the participant through a semantic and phonological

cueing hierarchy (starting with the least effective and moving to the most effective). This

was used until the participant named the item. Then, the participant was asked to repeat

the correct name of the item five times (Pring, Hamilton, Harwood, & MacBride, 1993),

before completing a confrontation naming task. The results indicated that there was a

treatment effect for each participant, but this did not generalize to any untrained items

that were probed. Mayer et al (2004) reported that one participant initially showed

progress and typicality effects that were described as subtle: generalized naming to

untrained typical sports items after training on atypical sports items. However, because

there were unanticipated improvements in not just in trained categories but also in

untrained categories, Mayer et al. (2004) indicated that this could not be considered

generalization.

Mayer et al. (2004) identified three factors to explain the lack of generalization.

First, they considered it possible that the severity of the aphasia might have interfered

with the cognitive processes needed for rehabilitation. Second, Mayer et al. (2004)

suggested that the training tasks were structured to compensate for the severity of the

deficits, and, thus, perhaps did not have the metalinguistic requirements of the tasks used

by Kiran and Thompson (2003b). Third, Mayer et al. (2004) discussed Plaut' s (1996)

assumption pertaining to the retraining and the mapping; this might not have been










applicable to their participants. According to Plaut (1996) generalization might require a

good estimate of the relevant semantic information of the training set overall and this

needs to be in alignment with the consistency of the mapping that the damaged part of the

network is responsible for carrying out. Mayer et al. (2004) reasoned that the severity of

the brain damage in their participants negatively affected the integrity of their lexical-

semantic network, rendering it incapable of retrieving the semantic features in order to

demonstrate generalization.

This study by Mayer et al. (2004) was carefully designed and included various

nuances that are important in therapy, including development of typicality norms,

selecting six categories that were based on participants' interests when possible. In

addition they included repetition of the target as a training task, to minimize phonological

access deficits.

Stanczak et al. (2005) also sought to understand the relationship among

performance on a category verification task (i.e., a measure of online typicality), training

typical and atypical category exemplars for naming production, and generalization. Two

participants with anomic aphasia, one fluent and the other nonfluent participated in the

study. First, prior to the baseline picture naming probes, the participants completed a

category verification task (with stimuli from the Kiran and Thompson (2003b) study).

The purpose of this was to determine if they would show typicality effects (i.e., faster

reaction times and more accurate responses on typical versus atypical items) in response

to the written category primes and their written exemplar targets. Then the participants

completed baseline probes. This was followed by a naming therapy, in which each

participant was simultaneously trained on two categories, birds and vegetables, one with









typical category exemplars and the other with atypical. The training tasks were naming

the target picture, verification of semantic attributes related to it, and naming the picture

again. The results of the category verification task that was completed at the onset of the

study indicated that the participants were faster to verify typical items compared to

atypical items (i.e., this was described as "online typicality effects"). The naming

treatment had differential effects. The participant with fluent aphasia only demonstrated

acquisition of trained typical birds, but not atypical vegetables. In contrast, the participant

with nonfluent aphasia demonstrated treatment effects for the atypical birds only (not

typical vegetables). There was no generalization to untrained items in either participant.

The finding of a typicality effect on the category verification task was interpreted by

Stanczak et al. (2005) as activation occurring in a bottom-up fashion (i.e., the semantic

retrieval was automatic). The differences in these two participants' responses to the

typical and atypical training was attributed to their lesion site and fluency (anterior and

nonfluent versus posterior and fluent) interacting with the semantic processes involved

with typical versus atypical items. Specifically, it was hypothesized that the controlled

semantic processing in the fluent participant was relatively intact, allowing for better

performance with the typical words, despite their similarity with other items and higher

degrees of semantic interference. However, learning the atypical items was considered to

be more difficult. In contrast, the nonfluent participant, Stanczak et al. (2005), theorized

that the demands for atypical items were less than the typical items (i.e., fewer

competitors), and thus yielded better performance. This study by Stanczak et al. (2005)

provided an intriguing interpretation of the results by linking them to neural substrates.









This is feasible. However, one limitation was that the investigators did not elaborate

about the difficulty related to atypical words.

Very recently, Stanczak, Waters, and Caplan (2006) further extended this research

by applying the same protocol (Stanczak et al., 2005) to two different participants with

anemia from different etiologies. The purpose of this study was to determine if

generalization would be affected by the type of deficit and the type of treatment provided.

Participant I had conduction aphasia, and her deficits were primarily phonological; her

training was with atypical vegetables and typical birds. These were trained over a 20

week period. Participant 2 had transcortical motor aphasia and had phonological and

semantic impairments; his training was on atypical birds and typical vegetables. The

results indicated that compared to atypical items, typical items were learned faster by

both participants. Although both participants showed significant treatment effects and

better learning of birds compared to vegetables, the patterns of responsiveness to the

training varied for each participant. Participant 1, who had a primary phonological

impairment, displayed greater learning when trained with typical items compared to

atypical items. In contrast, Participant 2, who exhibited both semantic and phonological

deficits, did not initially show a difference between learning typical versus atypical items,

but did show better learning of atypical items as the treatment continued. Participant 2,

who had been trained on atypical birds and typical vegetables, also had statistically

significant generalization to untrained typical birds, and only marginal generalization to

atypical vegetables.

To account for these different findings, Stanczak et al. (2006) offered

interpretations which again were based on neural substrates. It was not the phonological









impairment in Participant 1 that resulted in the lack of leaming atypical items, but the

greater degree of semantic difficulty inherent in atypical items. However, for participants

with damage to the left prefrontal area, as in Participant 2 (with both semantic and

phonological impairments), the typical items initially should be learned better because of

the high degree of similarity in the features but, then over time, this overlap leads to

confusion. Stanczak et al. (2006) hypothesized that learning atypical items, which have

greater variation, should result in better performance because the degree of competition is

less. Despite the better performance on birds, the investigators refuted the possibility that

this was a category specific learning (i.e., that it was easier to learn birds compared to

vegetables), and provided a rationale for the effects of typicality on learning (i.e. both

participants were responsive to being trained with typical items but only one responded to

the training with atypical items). Stanczak et al. (2006) concluded that while their

Endings were broadly consistent with Plaut (1996) and Thompson et al. (2003), the

differences from Eindings by Kiran and Thompson (2003b) might be attributable to

differences in the training (i.e., training two categories at that same time) or the

participants (degree of semantic impairment).

These four studies (Kiran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al.,

2005, 2006) were theoretically motivated and well planned, including programming for

generalization. The training tasks included at least some information about the target item

in terms of semantics and the phonological form, as well as confrontation naming. Some

potential limitations of these studies, with the exception of Mayer et al. (2004) included

the use of only two animate categories without a third control category (although the

second category was used as a control) and a lack of information about the clinician's









response to a failed naming attempt by the participant (Kiran & Thompson, 2003b;

Stanczak et al., 2005, 2006). The clinician' s response to the participant's naming failure

is important because it could affect the outcome. This is because it provides information

about the response so that the patient can better understand the information and can learn

to self-evaluate (Bandur & Shewan, 2001).

Also, in some of the studies it was not clear if generalization probes to untrained

items that had the same, different, or both types of typicality as the items that were

trained (e.g., train atypical items, probe untrained typical items in the trained category). It

appears that the generalization probes were not measured on both in some of the studies.

Other broader methodological concerns about these four studies include the lack of

opportunity to or ability to generate semantic information about the item in a more active

manner (i.e., beyond yes/no questions). Although yes/no tasks, sorting and other similar

tasks are used in the literature and have value, they require the participant to play a less

active role in the therapy. Mayer et al. (2004) and Stanczak et al. (2006) identified

concerns related to this in terms of the severity of their patient population and how it

might have affected generalization. Another possibility that Mayer et al. (2004) identified

was that the semantic category structure in their patients was impaired such that they

were unable to benefit from the training which was built around the semantic

representations of the category (Plaut, 1996). Another important consideration about

achieving a generalization effect has been raised by Murray and Clark (2005): it is not

known if it is the combination of training typical and atypical category exemplars in the

context of semantic training requiring task requiring processing of semantic

representations (Kiran & Thompson, 2003b), or simply the use of atypical and typical










category exemplars. We attempted to address some of these issues in the methodology of

the current study.

The Current Study

Rationale for Applying the Training to the PAD Population

Brain damage can cause a language impairment, and obj ect naming is a sensitive

marker for this (Benson, 1979). There is some evidence that the semantic system is

preserved in adults with nonfluent and fluent aphasia based on similar performance on a

lexical decision task compared to non-brain damaged adults (Gerratt & Jones, 1987).

However, several of the studies discussed above examining adults with aphasia suggested

there was a disruption or disorganization of the semantic system in adults with fluent

aphasia (Grober et al., 1985; Grossman, 1981; Kiran & Thompson, 2003a). There are

other reports of a semantic level impairment in adults with aphasia based on impaired

performance on property verification and category verification tasks (Koemeda-Lutz,

Cohen, & Meier, 1987); and lexical-semantic discrimination tasks (Chieff6, Carlomagno,

Silveri, & Gainotti, 1989; Gainotti, 1981). More specifically, in transcortical sensory

aphasia and anomic aphasia, disturbed semantic processing has been identified as the

underlying mechanism causing the language deficits (Raymer et al., 1997; Raymer &

Gonzalez Rothi, 2001). It is possible that these deficits involving the semantic category

structure might prevent generalization to untrained items and perhaps account for such

Endings in the literature (Mayer et al., 2004; Stanczak et al., 2005, 2006).

The current study proposed that individuals with early PAD provide an alternative

population to study the effects of typical and atypical exemplar training from semantic

categories. The rationale for this is based on the disease process, which results in diffuse

damage (as opposed to focal damage from a stroke) and evidence that to some degree, the









semantic system's structure is preserved in PAD (Albert & Milberg, 1989; Benson &

Geschwind, 1985; Hartman, 1991; Nebes, 1992; Ober et al., 1995; Schwartz et al., 1996).

This assertion is supported by findings that show knowledge about semantic categories is

preserved in PAD (Flicker et al., 1987; Huff et al., 1986; Martin & Fedio, 1983; Nebes et

al., 1986; Salmon, Butters et al., 1999; Schwartz et al., 1979; Warrington, 1975).

The goal of our exploratory study was to determine if choosing remediation targets

for anemia based on typicality (i.e., atypical and typical) would result in generalization to

untrained items. This involved a comparison of the family resemblance/prototype (Rosch,

1975) favoring typical category exemplars and the Complexity Account of Treatment

Efficacy (Thompson et al., 2003), as well as the connectionist model of a computer

simulation of acquired dyslexia (Plaut, 1996) favoring training atypical category

exemplars for achieving generalization. We provided both lexical phonological

information about the item via repetition as well as a semantic feature training with

atypical and typical items for adults with PAD. In designing our study, relevant findings

were culled from the PAD literature, particularly the three naming treatment studies

(Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005), the

five aphasia studies comparing typical and atypical category exemplars (Kiran &

Thompson, 2003b; Mayer et al., 2004; Plaut, 1996; Stanczak et al., 2005, 2006), and

other relevant findings from the aphasia treatment literature.

According to Nadeau and Gonzalez Rothi (2004), the connectionist approach views

anemia (that is caused by a semantic deficit) as a reflection of insufficiently engaged

representations of features that are critical for making distinctions among concepts. When

a network is damaged, a large amount of information still remains in the network, so the









focus should be on refining the damaged network via semantic therapy. In particular, the

network needs to be changed in terms of its connectivity so that there is more reliable

engagement of distinguishing features simultaneously while there is relatively a

disengagement of the shared features (Nadeau & Gonzalez Rothi, 2004). These Eindings

suggest that participants with early PAD might therefore be appropriate for the

semantically based training provided in our study which compared typical and atypical

category exemplars.

As noted previously from the aphasia literature, a facilitation effect refers to

improved performance on trained items from pretest to posttest in a single session

(Howard, 1985; Patterson et al., 1983). Our study used a single session for different

reasons, including the lack of picture naming treatment studies in the PAD literature, and

to determine if the lexical-semantic system affected by PAD could be stimulated in a

short period. This information would be useful to know prior to development of an

intervention for naming deficits in adults with PAD.

For the purpose of this study, the term "response generalization" is adapted from

Thompson (1989) to indicate that a participant has improved naming of untrained items

based on training of other items. Thompson (1989) suggested that for response

generalization to occur, the training and the probes should allow for sampling of

responses that have a similarity to the trained items, either within a defined set or across

it. The methodology for our study allowed for a measure of generalization to untrained

atypical and untrained typical items regardless of the type of training that category

received. (e.g., in the Train-Typical vehicles condition, there were untrained typical

vehicles and untrained atypical vehicles). The aphasia literature suggests criteria to









measure generalization; some examples include an increase in performance from baseline

by 40% (Kiran & Thompson, 2003b) or 50-80% (Thompson, 1989). Another approach

has been an increase of baseline performance as measured by three or more probe items

(Boyle, 2004). In our exploratory study, we chose to measure generalization based on an

increase from pretest that was statistically significant at the alpha .05 level.

The typicality ratings from Rosch (1975) provided categories and their exemplars

for this study. Using 24 items per semantic category allowed for an adequate number of

training and generalization items (Kiran & Thompson, 2003b; Rothi et al., 2005). Two

categories were used for the training (seven atypical and seven typical items) and the

other was a control category (Mayer et al., 2004; Rothi et al., 2005). While three of the

four aphasia treatment studies comparing typical and atypical category exemplar training

(Kiran & Thompson, 2003b; Stanczak et al., 2005, 2006) used the categories of birds and

vegetables, two of them reported generalization just one used a combination of natural

kinds and man-made artifacts (Mayer et al., 2004).

Our study involved three artifact categories: vehicles, clothing and tools. Although

at least one PAD study has suggested that there was no significant difference in

performance on natural kinds and artifacts (Tippett, Grossman, & Farah, 1996), others

have found there is evidence that adults with PAD have shown lower levels of

performance on natural kinds (Chertkow, Bub, & Caplan, 1992; Montanes, Goldblum, &

Boiler, 1995; Warrington, 1975), or that the profile changes with the progression of the

disease (Gonnerman et al., 1997). To avoid this controversy, the current study used only

artifact categories.









It has been suggested that providing feedback is an important part of the learning

process (Bandur & Shewan, 2001). Repetition has been used as a training task, and found

to be effective in obtaining a treatment effect (Fuller et al., 2001; Mayer et al., 2004;

Patterson et al., 1983; Pring et al., 1993; Rothi et al., 2005). Hickin et al. (2002)

suggested that presenting a picture stimuli and repeating the name of the word actually

elicits semantic processing. Our study combined these three components. This was

followed by the semantic training requiring generation of semantic information about the

target item. The rationale for this comes from studies that suggest that the naming deficit

in PAD was both due to semantic specification and lexical access deficits (Bowles et al.,

1987; Williamson et al., 1998). Our study used repetition so that if the PAD participants

have a deficit in both the semantic system and lexical access, the form of the word was

provided and being reinforced, and the semantic training targeted the semantic part of the

deficit.

Evidence of a treatment effect has been reported after training with a combination

of semantic and lexical (i.e., word form) tasks (Drew & Thompson, 1999; Le Dorze et al.,

1994; Wiegel-Crump & Koenigsknecht, 1973). Two other anemia treatment approaches

have also been found to result in treatment and generalization effects in adults with

aphasia are semantic distinction training (Hillis, 1998); and semantic feature analysis

(SFA) (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000). Both of these

approaches combined specific training of semantic features with presentation of the

phonological form. As noted above, Kiran and Thompson (2003b) applied a similar

process and suggested it was an important aspect of the study. The three other typicality









studies (Mayer et al., 2004; Stanczak et al., 2005, 2006) also used a more simplified

version of this.

Our study attempted to achieve some of the benefits of SFA (Boyle, 2004; Boyle

& Coelho, 1995; Coelho et al., 2000), for example, to facilitate a threshold level of

activation of the semantic network that surrounds the targeted item. Furthermore, like

SFA, we aimed for systematic activation of the distinguishing features of the item rather

than shared features (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000). This

process is based on the spreading activation theory, which posits that selection of the

target item among competitors is based on the highest activation achieved (Collins &

Loftus, 1975; Dell & O'Seaghdha, 1992; Levelt, Roelofs, & Meyer, 1999). Once the

concept is selected, phonological information is activated so that the target item is

produced (Collins & Loftus, 1975; Dell & O'Seaghdha, 1992; Levelt et al., 1999).

Finally, SFA has also been shown to result in both a treatment effect and generalization

to untrained items (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000). More

specifically, our training protocol was a modified version of SFA (Boyle, 2004; Boyle &

Coelho, 1995; Coelho et al., 2000). Our PAD participants answered questions about the

target item and then named it. We also included an aspect of the semantic distinction

training (Hillis, 1998), by asking them about how an item differs from others like it.

When the participant was unable to respond correctly, the clinician provided the

information and had the participant repeat it, based on Coelho et al. (2000).

Other studies also required active production of information, including the use of

antonyms and synonyms (McNeil et al., 1998), personalized cueing (Freed, Marshall, &

Nippold, 1995; Lowell, Beeson, & Holland, 1995; Marshall, Freed, & Karow, 2001); and










the participant using circumlocution (i.e., describing the item) (Francis et al., 2002).

Some of these studies also reported generalization to untrained items (Francis et al.,

2002; Lowell et al., 1995). In conclusion, we used information from the three PAD

naming treatment studies (Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al.,

2002; Rothi et al., 2005), studies comparing training atypical and typical category

exemplars (Kiran & Thompson, 2003b; Mayer et al., 2004; Plaut, 1996; Stanczak et al.,

2005, 2006) and other aphasia studies that showed treatment and generalization effects

(Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000; Davis & Pring, 1991; Francis

et al., 2002; Hickin, Best, Herbert, Howard, & Osborne, 2001; Hillis, 1998; Howard et

al., 1985; Pring et al., 1993; Raymer, Thompson, Jacobs, & le Grand, 1993). Many of

these studies also provided rationale for predictions about our research questions, which

are discussed below.

Research Questions and Predictions

The first four research questions focused on performance (i.e., accuracy and

reaction times) on picture naming. While the first two research questions addressed the

effects of repetition and the training (i.e., facilitation effects), the third question addressed

changes in performance on untrained items (i.e., generalization or practice effects). The

fourth question also addressed generalization, but to another task, category generation.

These questions are listed below with their corresponding predictions. The type of

training (i.e., training type) always refers to the typicality of trained items in a particular

trained category, and are always referred to as Train-Atypical, to indicate that atypical

items were trained, and Train-Typical, to indicate that typical items were trained.

Research Question 1: Does repetition of the control items (i.e., not with semantic

training) impact performance at posttest?









Prediction: There will be better performance (higher accuracy and faster reaction

times) on repeated items compared to items that were not repeated (Fuller et al.,

2001; Patterson et al., 1983; Pring et al., 1993; Rothi et al., 2005)

Research Question 2. What are the effects of semantic training on trained items?

Research Question 2a: Do accuracy scores and the reaction times for trained items

change from pretest to posttest?

Prediction: Accuracy and reaction times for trained items will show a facilitation

effect after training: Accuracy scores will be higher and reaction times will be

faster on the trained items at the posttest (Drew & Thompson, 1999; Howard, 1985;

Le Dorze et al., 1994).

Research Question 2b: Does the type of training (i.e., Train-Atypical or Train-Typical)

impact the performance on trained items at posttest?

Prediction 2b. 1: Performance will improve after both types of training, but there

will be a greater improvement after training typical items (Plaut, 1996).

Prediction 2b.2: Performance will only improve after training typical items (Rosch,

1975).

Research Question 3a: Will there be an overall change in performance on untrained items

from trained categories?

Prediction 3a: Accuracy will increase and reaction times will decrease in untrained

items in trained categories (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al.,

2000; Hillis, 1998; Plaut, 1996; Wiegel-Crump & Koenigsknecht, 1973).

Research Question 3b: Does the type of training (i.e., Train-Atypical/Train-Typical)

affect performance on untrained items in trained categories (i.e., generalization)?









Prediction 3b.1: Training atypical items (Train-Atypical) will result in greater

generalization as shown by higher accuracy scores and faster reaction times (Kiran

& Thompson, 2003b; Plaut, 1996; Stanczak et al., 2006; Thompson et al., 2003).

Prediction 3b.2: Training typical exemplars will result in greater generalization:

higher accuracy scores and faster reaction times for typical exemplars within the

same category (Rosch, 1975).

Prediction 3b.3: Training typical exemplars will result in greater generalization to

both untrained typical and atypical items, because typical items have richer

representations and share more features with all other category members (Heilman,

personal communication, 2005).

Research Question 3c: Is there an interaction between time and item typicality (i.e.,

atypical items and typical items)?

Prediction 3c: Untrained typical items from trained categories will show more

improvement than atypical items from trained categories after training.

Research Question 3d: Is there an interaction among time, training type, and item type?

Prediction 3d. 1: Training atypical items will result in generalization to untrained

typical items (Kiran & Thompson, 2003b; Plaut, 1996; Stanczak et al., 2006) and

to untrained atypical items (Plaut, 1996; Stanczak et al., 2006).

Prediction 3d.2: Training typical items will have a positive impact on untrained

typical items from trained categories and a negative impact on the performance

with atypical items from trained categories (Plaut, 1996).

Research Question 4: Is there generalization to category generation?






69


Prediction 4: There will be an increase in the number of items named in the

trained categories, and this will be greater for the trained categories compared to

the untrained categories (Ousset et al., 2002).















CHAPTER 4
METHOD S

The purpose of this exploratory study was to extend previous research on the

effects of training typical and atypical category exemplars on generalization (Kiran &

Thompson, 2003b; Mayer et al., 2004; Plaut, 1996; Stanczak et al., 2005, 2006) while

comparing two views of typicality (Plaut, 1996; Rosch, 1975; Rosch & Mervis, 1975).

Participants with early probable Alzheimer disease participated in this picture naming

training study. The primary goal was to determine if remediation targets could be chosen

based on typicality, to maximize generalization to untrained items (Kiran & Thompson,

2003b; Plaut, 1996; Stanczak et al., 2006). This was achieved by training atypical (i.e.,

Train-Atypical) and typical (i.e., Train-Typical) category exemplars from two semantic

categories and leaving untrained atypical and typical items in each category for both the

same item typicality that was trained (e.g., Train-Atypical, untrained atypical items), and

the opposite (e.g., Train-Atypical, untrained typical items). Using a repeated measures

(i.e., within subj ect) design, our study provided each participant with a modified semantic

feature based training (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000) for each

category. This was completed in a single session and the results were analyzed for a

facilitation effect (Hickin et al., 2002; Howard, 1985; Howard et al., 1985; Patterson et

al., 1983) and generalization to untrained items. While the outcome measures (i.e.,

dependent variables) for the picture naming were accuracy and response time means (in

milliseconds), the number of items generated was the outcome measure for the category

generation task.









Participants

Following approval from the UF Institutional Review Board (Appendix A), the

participants with PAD were recruited and enrolled into the study. Recruitment of

participants was accomplished with the help of the University of Florida Memory and

Cognitive Disorder Clinics. Additional recruitment efforts included presenting

information and flyers about the study at the Alzheimer's Association, and Al'z Place, an

adult day care facility for individuals with Alzheimer disease and other dementias.

Participants were compensated $20 for their time.

Diagnosis of the participants with mild-moderate PAD was made by a team of

medical professionals including neurologists and neuropsychologists based on criteria

from the National Institute of Neurological and Communicative Disorders & Stroke

(NINCDS), and the Alzheimer's Disease and Related Disorders Association (ADRDA)

(McKhann et al., 1984). Individuals were not eligible for the study if they had any history

of brain or head injury, psychiatric hospitalization, alcohol or drug abuse; any chronic

medical or psychiatric conditions, or any developmental learning disability. Participants

were pre-screened at the UF Memory Disorders Clinics for PAD. Some of the results

from the neuropsychological testing are shown in Table B-1, Appendix B.

A phone screen was used to confirm eligibility and to collect demographic

information. Twelve adults (7 women and 5 men) with newly diagnosed early PAD

completed the study. The average age was 77.6 (SD = 9.6) and the average number of

years of education was 15.8 (SD = 3.8). The mean score for the Mini Mental Status Exam

(Folstein et al., 1975) was 22.6 (SD = 3.4) with a range of 17-27. More specific

demographic information and performance scores for each participant are shown in Table

4-1.










Table 4-1. Participant Demographics and Scores on Mini Mental Status Exam and Reading
Subtest
ID # Age Years of Gender Mini Mental Status Reading Subtest
Education Exam WRAT-3 (percentile)
1001 72 17 F 21 87
1002 77 13 F 24 50
1003 76 10 AI 17 8
1004 92 24 M 24 91
1005 82 12 F 19 10
1006 78 12 F 22 50
1007 55 20 M 27 90
1008 84 15 F 25 75
1009 69 18 NJ 20 50
1010 86 16 F 19 63
1011 75 16 M 27 81
1012 85 16 F 26 87

Mean 77.6 15.8 22.6 61.8
Notes: Mini Mental Status Exam (Folstein et al., 1975); WRAT-3=Wide Range Achievement
Test-3 Reading subtest (Wilkinson, 1993).


Experimental Stimuli

There were six training lists (1-6) that were counter balanced using a Latin Square

for training conditions (i.e., Train-Atypical, Train-Typical, Control) and semantic

category (i.e., vehicles, clothing, and tools). As depicted in Table 4-2, each list had two

training conditions and one control condition. Participants were assigned to these lists

sequentially, as they were enrolled into the study. For example, the first participant

(#1001) had List 1, and this process continued with the remaining participants (e.g.,

participant eight (#1008) had list 2).

Table 4-2. Lists of Training. Conditions and Corresponding. Training. Conditions
List # Vehicles Clothing Tools
1 Train-Typical Control Train-Atypical
2 Train-Atyical Train-Tyical Control
3 Control Train-Atypical Train-Typical
4 Train-Typical Train-Atypical Control
5 Control Train-Typical Train-Atypical
6 Train-Atypical Control Train-Typical










The pre and posttesting picture naming stimuli consisted of 72 black and white line

drawings from three categories: vehicles, clothing, and tools. (Table C-1, in Appendix C,

shows the list of 72 items used in the study). Since there are several variables being

addressed, it is important to first clarify the terminology for the items in the pre and

posttest naming tasks, then the details related to the training are discussed. All items were

designated as either typical or atypical, based on the ratings in the Rosch norms (Rosch,

1975). Items were chosen from these three categories based on the seven-point typicality

ratings from Rosch (1975), such that half the items in each category were atypical and the

other half were typical. Typicality ratings for items used in this study ranged from 1.02

(the most typical) to 5.36 (the most atypical). For each category, the typical and atypical

items were selected at about a midpoint range so that half would be typical and the other

half would be atypical. The mean typicality score for the atypical items 3.5 (SE = .12),

and the mean typicality score for the typical items was 1.6 (SE = .08). An independent t-

test indicated that these groups of items were significantly different, t(70)=12.967, p =

.00.

The picture stimuli were gleaned from a variety of sources, with the maj ority from

the Florida Semantic Battery (Raymer et al., 1990). Name agreement was determined by

piloting the naming task on 10 University of Florida (UF) undergraduates in the

Language over the Lifespan Lab.

Each semantic category was designated as either the control category (i.e., no items

received semantic training), Train-Typical (i.e., only typical items received semantic

training), or Train Atypical (i.e., only atypical items received semantic training). There

were several reasons for the choice and number of categories. All artifact categories were









chosen to avoid the possible confound of mixing biological and artifact categories.

Furthermore, it was important to prevent any overlap from other semantic categories

(e.g., vehicles and sports). Also, previous research on picture naming has included these

categories (Grossman, 1981; Mayer et al., 2004), although other studies of typicality

effects have only included natural kinds (Fuller et al., 2001; Kiran & Thompson, 2003b;

Rothi et al., 2005). There were 24 exemplars in each category (Kiran and Thompson,

2003b) 12 typical and 12 atypical.

From these 72 items, the semantic training stimuli were selected. For each

category, seven atypical items were designated as the training set for when the category

appeared in the Train-Atypical condition, and 7 typical items were designated as the

training set for when that category appeared in Train-Typical condition. Thus, in a

category that appeared in the Train-Typical condition, for example, the 7 typical items in

the training set would receive semantic training, leaving 5 untrained typical items and 12

untrained atypical items from that category to analyze for generalization. Untrained items

were those items that received no semantic training, but were members of the trained

category. Therefore, for any given item, it was necessary to specify whether it was typical

or atypical, trained or untrained, and what type of training, if any, that category received.

The control items were untrained items in the untrained category. Control items were

named at both pretest and posttest. Because all items from the trained categories were

repeated, half of the items from the control category were also repeated to determine

what effect simple repetition without semantic training had on lexical access. An example

based on list 1 is found in Table 4-3.









Table 4-3. Example from List 1: Train-Typical Vehicles and Train-Atypical Tools.
Trained Untrained (same as trained) Untrained (opposite of trained)
7 Typical Vehicles 5 Typical Vehicles 12 Atypical Vehicles
7 Atypical Tools 5 Atypical Tools 12 Typical Tools
Note: The control category was comprised of 24 items of clothing.

The list of phrases or cues for the semantic training was developed by polling

volunteers (ranging from a Ph.D. investigator, a certified speech-language pathologist,

and several undergraduate research assistants) who are members of the UF Language

over the Lifespan Lab in the Department of Communication Sciences and Disorders.

These cues were compiled and selected based on the most salient ones.

In a series of post hoc analyses of the stimulus words from these pictures, we

examined a possible influence from four variables on the atypical and typical items:

typicality (Rosch, 1975), which we expected to show a significant difference, written

word frequency (Frances & Kucera, 1982) and familiarity (Wilson, 1987), and phoneme

length. We used an independent sample t-test to compare the complete list of 72 atypical

and typical items on the four variables. For typicality and phoneme length, ratings were

available for each item. The independent t-test for the typicality ratings had a p- value of

.000, indicating a significant difference in terms of typicality between the atypical and

typical items. The independent t-test for phoneme length was not significant, t(70)=2.528,

p = .69, indicating there was no difference between the atypical and typical items. In

other words, the phoneme length for the atypical and typical items did not influence the

results.

The independent t-test comparing typical and atypical items from the Complete List

was not significant for frequency, t(70) =-1.3 89, p =. 169, suggesting that there was no

difference in the frequency ratings between the atypical and typical items. However,

although the independent t-test for familiarity was significant, t(3 8) =-3.625, p = .00, this










analysis was only computed on 40/72 items (i.e., many of the items did not have

familiarity ratings). Thus, the results must be interpreted with caution. Table 4-4 depicts

the means of the four variables discussed above.

Table 4-4. Means for Complete List of Atypical and Typical Items.
Train Type Typicality Score Frequency Familiarity Phoneme Length
Atypical 3.6 15.7 480 4.8
Typical 1.7 25.1 551 4.9
Notes: Typicality (Rosch, 1975); Freuency (Frances and Kucera, 1982); Familiarity (Wilson,
1987).


Procedure

Prior to participation in the study, written informed consent was obtained from each

participant. As needed, the clinician allowed time for the participant to read the informed

consent form. Then, the clinician answered any questions and assisted the participant in

interpreting the information.

Screening

Two measures were used for the screening: The Mini-Mental Status Exam

(Folstein et al., 1975) and the reading subtest of the Wide Range Achievement Test

(WRAT-3) (Wilkinson, 1993). The clinician administered the MMSE using the standard

procedure (Folstein et al., 1975). Next, the clinician presented a computerized version of

the reading subtest (Wilkinson, 1993) in which the participant read aloud the individual

words from the screen. These responses were digitally recorded on an Olympus DM-10

digital recorder (Olympus, 2002) and response times (RTs) were collected via DirectRT

(Jarvis, 2004) for later scoring and analysis.

Pretesting

The pretest consisted of two tasks, Category Generation (Benton, 1968) and Picture

Naming. For the Category Generation task, the clinician provided instructions for the










participant to name as many items as possible in one minute for a given category (i.e.,

vehicles, clothing, and tools). The categories were presented in a pre-determined random

order. The clinician wrote the responses verbatim on the pretest form, and responses were

also digitally recorded for checking reliability.

The Picture Naming task consisted of the complete set of 72 line drawings

randomly presented on the computer via DirectRT (Jarvis, 2004). At the onset of the task,

the clinician read aloud the set of instructions that appeared on the screen and confirmed

comprehension of the task. The clinician controlled the pace of the session by using the

mouse button presses on a wireless mouse. Each picture stayed on the screen until the

clinician advanced to the next picture (i.e., the picture did not disappear or time out).

There were four practice items from the category of fruit. After the participant attempted

to name a presented picture, the clinician informed the participant if he/she was correct or

not, and then asked the participant to repeat the correct name of the item three times. If

the participant produced a correct alternative name of the item (e.g., "j et" for "airplane")

their response was accepted as correct, and this word was repeated (Appendix C includes

a list of acceptable alternative names for items). The clinician then advanced to the next

trial. As stated above, twelve items from the control category were not repeated (half

were typical and half were atypical). A prompt appeared on the computer screen to alert

the clinician as to whether this was a repeated item or not. This procedure was used for

all of the items in the two trained categories, and for half of the control items (i.e., repeat

or no repeat, without semantic training).

More specifically, after the participant named an item, the word "repeat" appeared

on the lower center section of the screen. Next, the clinician provided verbal feedback









and directed the participant to repeat the name of it three times (i.e., "That' s correct, now

you say it three times or It' s supposed to be a _, now you say it three

times, _,_ ") (Kendall, personal communication, 2005; Fuller et al. 2001; Rothi

et al. 2005). When an unrepeated item was presented, the participant named the item,

then the words "Go on" appeared in place of the word "Repeat," and the clinician

advanced to the next item. Thus, on these control items the clinician did not provide

feedback about the participant' s response, and the item was not repeated. The clinician

continued this process with each of the remaining pictures until all 72 items had been

named. At this point, the participant took a short break while the clinician set up for the

semantic training. All of the naming responses were digitally recorded for later

transcription and analysis, and response times were collected via DirectR (Jarvis, 2004).

Semantic Training

This was a guided, feature generation task, loosely based on semantic feature

analysis (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000) and semantic

distinction training (Hillis, 1998). The computer presented the pictures from one of the

training categories (i.e., seven items) in random order and the clinician asked the

participant to answer each of the following questions about the item: 1. "What is its

function?"; 2. "Who uses it?"; and 3. "How is it different from other _?" (i.e., things like

it). After the participant answered the three questions, the clinician asked the participant

to name the item. Whenever a participant was unable to correctly answer a question or

provided ambiguous information, the clinician provided cues from a pre-determined

typed list. Whenever the participant was unable to name the item, the clinician stated it

and asked the participant to repeat it. The clinician continued this process with the

remaining six training items for that category. Each participant was trained on the two









categories from their designated list. The trained categories (e.g., Train-Atypical Tools,

and Train-Typical Vehicles) were randomly presented in blocks, but the order of the

items presented within each category was random. This allowed each participant to be

trained on each category and the corresponding training type (i.e., Train-Atypical, Train-

Typical) for a total of six practice sets (i.e., three presentations for each category). Then,

the participant took a break while the clinician prepared for the posttesting. The responses

from the training were also digitally recorded.

Posttesting

The posttesting consisted of the same two tasks and instructions as the pre-testing:

Category Generation and Picture Naming. During the Category Generation task, the

participant was given the same instructions and the same random order for generating as

many items as possible for each of the three categories (vehicles, clothing, and tools).

The clinician wrote down participants' responses verbatim and also digitally recorded

them in order to check reliability.

For the Picture Naming task, the same 72 item picture set was presented by

computer running Direct-RT (Jarvis, 2004). Items were presented in a different random

order than at pretest. In this version, the participant only named the pictures aloud,

without feedback or repetition.

Scoring

The scoring for the screening tasks was completed as follows. The Mini Mental

Status Exam was scored using the standard scoring outlined by Folstein et al. (1975).

Digital recordings of the reading subtest from the WRAT-3 (Wilkinson, 1993) were used

to code the data for accuracy of production by a trained research assistant. The clinician

scored the responses and calculated raw and standard scores and converted them to the










percentiles, based on instructions from the WRAT-3 Manual (Wilkinson, 1993). It is

important to note that normative data for adults with PAD is not provided in the WRAT-3

Manual (Wilkinson, 1993), and that most of our PAD participants were beyond the age

ranges provided (i.e., 55-64. 11; and 65-74. 11). Thus, these findings must be interpreted

with caution.

The pre and posttest picture naming performance for accuracy and response times

was collected by Direct RT (Jarvis, 2004). Accuracy and response times were tracked by

the clinician's mouse clicks (i.e., left for correct and right for incorrect) based on the

participant' s response and coded these as ones and zeros. The list of acceptable alternate

names is displayed in Appendix C. The clinician used the same procedure for the

response times (i.e., left for usable and right for unusable). For the response times, only

fluent, correct responses were computed in the analyses (e.g., response times were

excluded if the participant said, "uh airplane", but the response would be scored as

accurate/correct). Furthermore, response times for each participant were adjusted for

outliers by replacing any value that was greater than three standard deviations above the

mean, with the mean value.

The responses for the Category Generation task were transcribed independently by

a trained research assistant. The clinician and research assistant then compared these

results to the clinician's original on-line written documentation from the session and

discrepancies were resolved by referring back to the corresponding voice files until

agreement was reached. Informal reliability measures were carried out with ~ 50% of the

data, using the voice files. Agreement was ~98% and disagreements about scoring of










responses were resolved by consensus during the Language over the Lifespan Lab

meetings.

Responses from the Picture Naming task were also transcribed by a trained research

assistant. The clinician used these transcripts to verify coding of 100% of the picture

naming scoring.

Statistical Analyses

Our study used a repeated measures design. The analyses compared pre and post-

test category generation and picture naming performance using both accuracy and

response times where possible. Analyses evaluated the effects of repetition and practice

from data in the control condition, the effects of semantic training, and generalization

effects to untrained items in trained categories. The effects of training atypical and typical

category exemplars were also examined.

The pre and posttest picture naming data were analyzed using t-tests, repeated

measures analysis of variances (ANOVAs) and follow-up post hoc t-tests where

necessary. It is important to point out that the pretest naming data reflects the

participants' initial attempt at naming the pictures; this was followed by the repetition

and semantic training. Thus, the effect of repetition and training can only be evaluated by

comparing pre and posttest performance. Due to an error in the programming related to

the control items, four participants repeated all items. Consequently, data from only eight

of the twelve participants could be included in the analysis of repetition effects.

Since inspection of the data for some of the analyses showed pronounced

differences at pretest, difference scores were calculated from the pre and posttest data and

then analyzed. The dependent variables for picture naming were means from the accuracy

and response time data. The independent variables varied according to the questions, but









included time (pre /posttest), repetition (no repeat/repeat), training type (i.e., the Train-

Atypical/Train-Typical), item typicality (i.e., item type, atypical/typical). For

consistency, whenever possible the term atypical is always discussed before typical, this

convention is alphabetical and used in order to reduce confusion with the terminology

(there is no intentional bias).

The dependent variables for the pre and posttest category generation data were the

number of items generated per category, and the independent variables were time and

training type (i.e., Control category, Train-Atypical, Train-Typical). The data was

analyzed with repeated measures ANOVAs.

Research Question 1: Control Items

Does repetition of the control items (i.e., not with semantic training) impact

performance at posttest? This question addressed the possibility of a repetition or practice

effect. Only accuracy and response time means from the untrained items in untrained

categories (i.e., the control condition) made up the dependent variable. Half of these were

in the repetition condition and half were not (i.e., not repeated). The independent

variables were time (pre/post) and repetition (yes/no). A (2) Time (pre/post) X (2)

Repetition (yes/no) repeated measures ANOVA was used to answer this question.

Research Question 2: Trained Items in Trained Categories

This two-part question addressed the possibility of a facilitation effect in

semantically trained items (i.e., Train-Atypical and Train-Typical). The data consisted of

scores from the 14 trained items (i.e., the seven atypical and seven typical items from the

two trained categories). This within category measure was assessed as a comparison of

pretest and posttest performance on the confrontation naming task









Research Question 2a: Do accuracy scores and the response times for trained items

change from pretest to posttest? The independent variable was time (i.e., change from

pretest to post test).

Research Question 2b: Does the type of training (i.e., Train-Atypical or Train-

Typical) impact the performance at posttest? The independent variables were time and

training type.

To answer both of these questions, (2) Time (pre/post) x (2) Training type (Train-

Atypical/Train-Atypical) repeated measures ANOVAs were computed.

Research Question 3: Untrained Items in Trained Categories

Question 3 had several subcomponents, but overall addressed the possibility of a

primary generalization effect. If there was a facilitation effect in question 2, the purpose

of questions 3a, b, c, and d was to determine if the facilitation effect generalized to

untrained items within the same trained category.

Research Question 3a: Is there generalization to untrained items in the trained

category? The independent variable was time (pre/post).

Research Question 3b: Does the type of training (i.e., Train-Atypical/Train-

Typical) affect generalization to untrained items in trained categories? The independent

variables were training type and time.

Research Question 3c: Does the typicality of the untrained item (i.e., atypical items

and typical items) determine whether it will benefit from generalization? The

independent variables were time and item type.

Research Question 3d: Do time, training type, and item type interact to promote

generalization limited to a particular item type following a particular training type? The

independent variables were time, training type, and item type.









The analyses for Question 3 used the 34 untrained items in the two trained

categories (i.e., 17 untrained items in each trained category). For the accuracy means, we

used a (2) Time x (2) Training Type (Train-Atypical/Train-Typical) x (2) Item Type

(Atypical/Typical) repeated measures ANOVA.

The response time means were analyzed differently. For Question 3a, a t-test was

used to compare pre and posttest performance. To answer Questions b, c, and d

difference scores were computed between pre- and posttest response times and analyzed

via a (2) Training Type x (2) Item type repeated measures ANOVA.

Research Question 4: Category Generation

Research Question 4: Is there generalization from picture naming to category

generation? Does the type of training the category receive affect whether there is

generalization to category generation. The independent variables were time and training

type. The dependent variable was total number of items generated per category and these

were analyzed with a (3) Training Type (Train-Atypical, Train-Typical and Control) x (2)

Time repeated measures ANOVA.















CHAPTER 5
RESULTS

In this study we compared the performance of adults with probable Alzheimer

disease (PAD) on picture naming and category generation tasks before and after a

semantic training using atypical and typical category exemplars. The accuracy means and

response times (in milliseconds, ms) from pre and posttest picture naming, and number of

items generated per category on a category generation task were analyzed with t-tests and

repeated measures analysis of variances (ANOVAs). To control for some of the

differences in the pretest scores across conditions, difference scores were computed for

several of the questions and then analyzed to compare direction and magnitude of change.

An a priori decision was made to use a .05 level of significance for all of the analyses.

The two training conditions are always referred to as Train-Atypical (i.e., training

atypical items) and Train-Typical (i.e., training typical items).

Research Question 1: Control Items

Question 1 asked if repetition of the control items (i.e., not with semantic training)

impacted the performance of picture naming at posttest. The dependent variables were

the accuracy and response times, and the independent variables were time and the

repetition and no repetition conditions. The data for these analyses were from the control

categories (i.e., untrained items in untrained categories) in which half of the items were

repeated (i.e., repetition condition) and the other half were not (i.e., no repetition

condition).









A (2) Time (pre/posttest) x (2) Repetition (yes/no) repeated measures ANOVA was

used to examine the data from eight participants. The results showed that there were no

significant main effects or interactions for the accuracy means. Although the items in the

repetition condition were somewhat higher (Mean (M)1 = .89, SE = .04) compared to items

in the no repetition condition (M~= .80; SE = .04), this was not significant, F(1,7) = 3.691;

p = .096; r12= .35. There were no other main effects or interactions with the accuracy

means for the control items. Thus, repetition did not have a significant effect on the

accuracy means.

The response time means showed a similar pattern from the (2) Time x (2)

Repetition repeated measures ANOVA. There were no significant main effects to suggest

that repetition alone impacted performance from pretest to posttest. There were decreases

in the response time means, indicating faster responses, but these were not significant.

One was for time, F(1,7) = 3.702; p = .096; r12 = .346; 2Mre = 2456 ms (SE = 521); 2Most

= 1858 ms (SE = 243). The other was for repetition, F(1,7) = 4.416; p = .074; r12 = .387;

M~REP = 1979 ms (SE=-321 ms), M~NoRep = 2334 ms (SE = 440 ms). The interaction was not

significant, F(1,7) =1.566; p =.251; r12 = .183.

The items in the control condition did not receive any semantic training, but half of

them were repeated at pretest. Although the accuracy in the control condition means were

slightly higher, this was not significant. The response times decreased somewhat, but not

significantly. Thus, repeating items in the control condition did not result in statistically

significant changes in accuracy or response times.

Research Question 2: Semantically Trained Items

Questions 2a and b asked if performance on the semantically trained items (i.e.,

Train-Atypical and Train-Typical) changed from pre to posttest, and, if so, whether there




Full Text

PAGE 1

CHOOSING REMEDIATION TARGETS FOR NAMING DEFICITS IN PROBABLE ALZHEIMER DISEASE: DOES TYPICALITY MATTER? By CLAUDIA A. MORELLI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Claudia A. Morelli

PAGE 3

This dissertation is dedicated to my in credible husband, Davi d Efros, my amazing mother, Claire Morelli and my sisters; and in memory of my wonderful father, John. They have been an integral part of my life and their belief in me has been a fortress of encouragement, hope, and steadfast love.

PAGE 4

iv ACKNOWLEDGMENTS First, I thank and acknowledge each participant and their family and/or caregivers for their time, effort, and for extending themselves and their home to allow me to work with them to complete this research proj ect. I will always remember them and the devastating effects of Alzheimer disease, and this will influence the focus of my future clinical research. I also extend my grat itude and appreciation for the UF Memory Disorder Program, especially David B. Efro s, Dr. Kenneth M. Heilman, Dr. Glen Finney, Dr. Kimford Meador, Dr. Catherine Price, a nd Jennifer Rembisz, whose kind and diligent efforts resulted in referrals for me to recruit these participants. I also express my deep appreciation for my committee. First, I want to thank my mentor, Dr. Lori J.P. Altma nn, for her vast knowledge of ne urolinguistics that provided a refreshing perspective, especially with rega rds to research in adults with probable Alzheimer disease. Working with Dr. Altmann allowed me to broaden my thought process and expand my approach to this diss ertation. These qualities, combined with her guidance, leadership, and patien ce have been of great value to me during the exciting and challenging process of becoming a Ph.D. I also acknowledge Dr. Diane Kendall for her expert opinion on clinical re search in adult neurogenics, and for the many opportunities to learn more about single subject design, es pecially during my pre-doctoral fellowship with her at the VA Brain Rehabilitation Resear ch Center. Her outstandi ng clinical skills and perspective as a compassionate speech-l anguage pathologist combined with her insights have had a great influence on me and have motivated my interest in theoretically

PAGE 5

v driven treatments. I also grat efully acknowledge Dr. Ira Fischl er who shared his extensive knowledge of psychology, which played an important role in this dissertation. Dr. Kenneth Heilman, with his behavioral neurol ogy expertise combined with his incredible commitment and compassion for patients, has ta ught me so much in such inspiring and meaningful ways. Each of these remarkab le individuals has contributed to my development by their expertise and great role modeling. I thank the Department of Communication Sc iences and Disorders at the University of Florida, especially Dr. Christine Sapienza, Dr. Samuel Brown, and Dr. Scott Griffiths, for their support and encouragement, as we ll as the wonderful opportunities to teach the undergraduates. Also, I will always rememb er the incredible opportunities to guest lecture for Dr. Howard Rothman and Dr. Bonnie Johnson (includi ng the lecture swaps with Dr. Johnson). I extend my appreciation to the entire faculty and staff for their professionalism, collegial, and positive appr oach. In particular, Idella King, Cassie Mobley, Debbie Butler, and Addie Pons deserv e a special thanks. Th e computer support from Neal Musson will al ways be appreciated. I also want to acknowledge the members of Dr. Altmanns Language over the Lifespan Lab. First I thank Alexia Frederikse n, who spent hours developing the stimuli, coordinating and working with us, and whose a ttention to details was truly awesome. Her dedication to the project was sincerely appreciated and always will be. Additional thanks go to Katie Chiarella, Andrea Holt, and Angel Ziesk for their wonderful help in preparing the stimuli; and to Maisa Haj Tas, Ashley Mullen, and Deb Gober for lending their incredible artistic talent. Alexia Frederik sen, Erin Hunt, and Rachel Hogue deserve acknowledgement for their excelle nt questions and feedback during the protocol training,

PAGE 6

vi which prepared them for testing the young adul ts for a related project. Becca Huy also played an important role by transcribing a la rge amount of the data and working closely with me with on the reliability checks. Her devoted attention to the details of accurate transcription, combined with her enthusiasm, and willingness to help was wonderful. Also, I thank Charlene Cohen-Deroy for her diligent efforts and commitment to excellence in coding and scori ng the Word Reading subtests as well as calculating the phoneme length for the stimulus words. All of these students demonstrated a keen awareness of the importance of maintaini ng the integrity of the data. Without the combined efforts of these wonderful students, this dissertation woul d have taken longer. Finally, I thank the colle ctive insights of the members of the Language over the Lifespan lab, especially those who provided suggesti ons for the semantic cues. It was their integrative efforts that helped to im prove various aspects of this study. I also would like to thank the Director and members of the Brain Rehabilitation Research Center, at the Malcom Randall Vete rans Affairs Medical Center, Dr. Leslie Gonzalez Rothi, Dr. Stephen Nadeau, a nd Dr. Jay Rosenbek, for the pre-doctoral fellowship opportunity that Dr. Diane Kendall provided. I also thank Haijing Qin for her statistical support. I acknowledge other inve stigators and clinicia ns, especially Nan Musson, Susan Leon, and Amy Rodriguez. I appr eciated the ongoing support of the staff, especially Brenda Stidham, Sue Nad eau, Joy McCallum, Sandy Davis, Lisa DeEmmanuel, and Lynn Dirk. Further appreci ation is extended to the incredible VA library staff, for their expertise, time, a nd consideration: Marsha H. White, Ellen L. Umans, Marylyn Gresser, and Harold A. Boyce.

PAGE 7

vii Two fine clinicians and professors have pl ayed an important, yet earlier role in my career. I extend my thanks and appreciation to Dr. Robert C. Marshall, for inspiring me years ago in my masters program, conti nuing his belief in me, and his incredible mentoring in a supportive, enthusiastic and gent le manner. If it were not for him, I would never have started this Ph.D. endeavor. Al so, I thank Donna Chadwick for teaching me the fundamental aspects of being an eff ective clinician during my undergraduate program. My friends have also been a great sour ce of inspiration and hope. Dr. Sally Ann Giess, whom I met in my first year at the Un iversity of Florida, continues to be a great friend, and role model. I acknowledge my ot her friends and colleagues from the UF CSD Department: Dr. Ann Marie Knight, Dr. Am ber Hollingsworth, Dr. Judith Wingate, Dr. Lynnette Bardolf, Susan Leon, and Maisa Haj Tas. I thank The Reverend Jeremy Hole for his support and encouragement. I will always appreciate my dear friends Ann Case and Nancy Merrifield for their wonderful presence in my life. Most importantly, I am forever grateful for my family. Their patience, ongoing love, support, and belief in me have been truly wonderful. Thus, I thank them: my husband David, mother, Claire, and my sisters and their families, Jeanmarie; Donna and Ryan, Buzz and Charlie; Claire and Ross, Mar tina and Fabiana; as well as Everett, my father-in-law and Marie, my mother-in-law. Also, I thank my cousins and their families: Marybeth, Mary Catherine, Luigi, Lou a nd Joey, and Lee and Mary Anne. Finally, I would like to acknowledge all those who ca me before, paving my way through their loving support and kindness, including my father John; and grandmothers, Florence McNulty and Anna Efros Kramer.

PAGE 8

viii This dissertation was partially suppo rted by my teaching and research assistantships from the UF Department of Communication Sciences a nd Disorders. It was also partially funded by Dr. Altmanns Adva ncing Academic-Research Careers award, which she had received from the American Speech-Language-Hearing Association. I am grateful for these funding sources.

PAGE 9

ix TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES.............................................................................................................xi LIST OF FIGURES..........................................................................................................xii ABSTRACT.....................................................................................................................xi ii CHAPTER 1 INTRODUCTION........................................................................................................1 2 REVIEW OF THE LITERATURE 1: NAMING DEFICITS IN PAD........................8 Semantic Memory Deficit in Probable Alzheimer Disease (PAD)..............................9 Lexical Access Deficit in PAD...................................................................................13 Semantic Impairment and Lexical Access Deficit in PAD........................................16 Naming Treatment Studies in PAD............................................................................18 3 LITERATURE REVIEW 2: A BRIEF OVERVIEW OF THE ORGANIZATION OF SEMANTIC MEMORY.......................................................................................26 A Brief Overview of Semantic Memory and Categorization.....................................26 Family Resemblance/Prototype View........................................................................30 The Complexity Account of Treatment Efficacy.......................................................42 Connectionist Model and Computer Simulations.......................................................43 Similar Behavioral Studies.........................................................................................48 The Current Study.......................................................................................................60 Rationale for Applying the Trai ning to the PAD Population..............................60 Research Questions and Predictions....................................................................66 4 METHODS.................................................................................................................70 Participants.................................................................................................................71 Experimental Stimuli..................................................................................................72 Procedure....................................................................................................................76 Screening.............................................................................................................76 Pretesting.............................................................................................................76

PAGE 10

x Semantic Training...............................................................................................78 Posttesting............................................................................................................79 Scoring.................................................................................................................79 Statistical Analyses.....................................................................................................81 Research Question 1: Control Items...........................................................................82 Research Question 2: Trained Items in Trained Categories.......................................82 Research Question 3: Untrained Items in Trained Categories....................................83 Research Question 4: Category Generation................................................................84 5 RESULTS...................................................................................................................85 Research Question 1: Control Items...........................................................................85 Research Question 2: Sema ntically Trained Items.....................................................86 Research Question 3: Untrained Items in Trained Categories....................................88 Research Question 4: Category Generation................................................................93 Summary of Results....................................................................................................94 6 DISCUSSION.............................................................................................................96 Summary of Findings.................................................................................................97 Implications for Anomia Treatment in PAD..............................................................98 Implications for Methodology..................................................................................103 Future Studies...........................................................................................................107 APPENDIX A INFORMED CONSENT FORM..............................................................................111 B TABLES WITH INFORMATI ON ABOUT THE PARTICIPANTS......................113 Participant Demographics and Select ed Neuropsychological Test Results.............114 Demographics and Performance Before and After Semantic Training....................115 C TABLES WITH INFORMAT ION ABOUT THE STIMULI..................................116 Master List of Stimulus Items Used in the Study.....................................................116 Alternate Names for Stimulus Items Used in the Study...........................................118 LIST OF REFERENCES.................................................................................................120 BIOGRAPHICAL SKETCH...........................................................................................135

PAGE 11

xi LIST OF TABLES Table page 4-1. Participant Demographics and Scores on Mini Mental Status Exam and Reading Subtest......................................................................................................................72 4-2. Lists of Training Conditions an d Corresponding Training Conditions.....................72 4-3. Example from List 1: Train-Typi cal Vehicles and Train-Atypical Tools.................75 4-4. Means for Complete List of Atypical and Typical Items..........................................76 5-1. Pre and Post Group Accuracy and Re sponse Time Means for Trained Items..........87 5-2. Group Accuracy and Response Time M eans for Trained Items (Train-Atypical; Train-Typical)..........................................................................................................87 5-3. Analysis of Variance for Accuracy Means for Untrained Items in Trained Categories.................................................................................................................89 5-4. Group Response Time Means (ms) for Untrained Items in Trained Categories.......92 B-1. Participant Demographics and Selected Neuropschological Te sting Results from the PAD Diagnosis.................................................................................................114 B-2. Participant Demographics and Pre and Posttest Category Generation and Picture Naming Tasks (i.e., Before and After the Semantic Training, day of study participation in the study).......................................................................................115 C-1. Master List of Stimul us Items Used in the Study. Trained Items are Italicized .....116 C-2. Master List of Stimulus Items Used in the Study. Alternate names for stimuli counted as correct...................................................................................................118

PAGE 12

xii LIST OF FIGURES Figure page 5-1. Accuracy Means for Untrained Items in Tr ained Categories, Main Effect of Time .90 5-2. Untrained Items in Trained Categorie s: Three-Way Interaction Among Time..........91 5-3. Category Generation, Number of Items Ge nerated at Pre & Posttest for Trained & Control Items............................................................................................................94

PAGE 13

xiii Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy CHOOSING REMEDIATION TARGETS FOR NAMING DEFICITS IN PROBABLE ALZHEIMER DISEASE: DOES TYPICALITY MATTER? By Claudia A. Morelli August 2006 Chair: Lori J.P. Altmann Major Department: Communicatio n Sciences and Disorders The purpose of this single session expl oratory study was to determine if remediation targets could be selected to improve picture naming and promote generalization to untrained items in 12 adu lts with early probable Alzheimer disease (PAD). Because theories support competing pr edictions about the relative effects of training typical versus atypical semantic cate gory members with respect to subsequent generalization, this study contrasted the effect s of training typical a nd atypical semantic category exemplars. Specifically, it examined changes in picture naming and category generation following initial repetition of the items at pretest followed by a semantic training. Stimuli included 24 items from each of 3 semantic categories, half of the items were typical and half atypical based on Ro schs norms. Two categories received training, one using a subset of typical items, one us ing a subset of atypical items; the third category remained untrained to track effects of repetition. The untrain ed category showed

PAGE 14

xiv nonsignificant improvements at post-test. A ccuracy scores improved for all trained categories; however, only trai ned, typical items were named significantly faster at posttest. Generalization was found in the untrained typical items in the ca tegories that were trained with typical items, which were na med more accurately at posttest. Neither atypical items from the categories that were trained with typical items nor any of the items in categories that were trained with atypical items s howed improved accuracy after training. These findings are cons istent with those of other researchers inve stigating the semantic deterioration in adults with PAD. Several studies have found an advantage for typical items in this population. These fi ndings are attributed to the redundant connections among features in t ypical items, which allow them to be more resilient in the face of progressive damage, allowing them to be more responsive to intensive semantic training. Atypical item s lack the redundancy of connections and, thus, are more vulnerable to damage. These findings are extremely enc ouraging for the development of principled strategies for choosing items to encourage gene ralization in the remediation of anomia, as well as for the development of lexical-semantic treatment paradigms for individuals with early PAD.

PAGE 15

1 CHAPTER 1 INTRODUCTION Probable Alzheimer disease (PAD), a progressi ve neurodegenerative disease, is the most prevalent type of acquired cognitive dys function (Thal, 1999) and affects over four million Americans (Kawas & Katzman, 1999). The prevalence of PAD accounts for 5080% of patients with dementia (Chui, 1989) and comprises the greatest percentage of the dementias in the geriatric population (C hui, 1989; Henderson & Fi nch, 1989; Sclan & Kanowski, 2001). Furthermore, PAD is already considered to be a major public health concern, and by 2025, the number of Americans aff ected by the disease is expected to be 12 million (Rosenberg, 2005) and 14 milli on by 2050 (Katzman & Fox, 1999). Although there is no known cure for PAD, and only sympto matic treatment is available at this time, current research endeavors are aiming to halt the progression of the disease as well as prevent its occurrence (Petersen et al., 2001; Rosenberg, 2005). Reducing the disability until this occurs is important (Rothi et al ., 2005). Furthermore, when drug therapies are available that halt the progression of the dis ease, there will be a demand for linguistic and cognitive rehabilitation for this populati on. Therefore, it is incumbent upon speechlanguage pathologists to develop and test appropriate treatmen ts for this population now, so that tested methods are available that address the language difficulties in PAD (ASHA, in press), which largely revolve around word finding Word finding difficulties (i.e., anomia), as measured by poor performance on picture naming tasks have been widely re ported in adults with PAD (Barbarotto, Capitani, Jori, Laiacona, & Molinari, 1998; Bayles, 1982; Bayles & Tomoeda, 1983;

PAGE 16

2 Benson & Geschwind, 1985; Heilman, 2005; Hodges & Patterson, 1995; Hodges, Patterson, Graham, & Dawson, 1996; Huff, Corkin, & Growdon, 1986; Huff, Mack, Mahlmann, & Greenberg, 1988; Kirshner, Webb & Kelly, 1984; Martin & Fedio, 1983; Smith, Faust, Beeman, Kennedy, & Perry, 1995; Smith, Murdoch, & Chenery, 1989; Williams, Mack, & Henderson, 1989; Williamson, Adair, Raymer, & Heilman, 1998). However, to date, there are only three P AD anomia treatment studies (Abrahams & Camp, 1993; Ousset et al., 2002; Rothi et al., 2005), and only one has reported generalization to untrained items (Abraham s & Camp, 1993). Consequently, developing treatments for this population that reduce the word finding di fficulty is important (Rothi et al., 2005), especially those that promote generalization. Generalization of learning, an essentia l objective for rehabi litation (Kearns, 1989; Thompson, 1989), has been described as an obser vation of an occurrence of a particular trained behavior in a context that has not been traine d (McReynolds, 1989). Generalization is important for several reas ons, including the clinical accountability of the speech-language pathologist to use met hodologies to measure the effectiveness of treatment (Kearns, 1989). The mechanism unde rlying generalization has been a longstanding question (Martin, Laine, & Harley, 2002), and although cognitive models have been advanced over time, the construct of generalization remains a mystery (Francis, Clark, & Humphreys, 2002). Perhaps due to this there is still a l ack of generalization reported in the literature (Francis et al., 2002; Kearns, 1989; Kiran & Thompson, 2003b; McNeil et al., 1998; McReynolds, 1989; Ni ckels, 2002; Thompson, 1989; Thompson, Ballard, & Shapiro, 1998; Thompson et al., 1997; Thompson, Shapiro, Kiran, & Sobecks,

PAGE 17

3 2003). Moreover, attempts to replicate methods that have promoted generalization in initial studies have not always been successful (McReynolds, 1989; Thompson, 1989). There are two types of gene ralization that, according to Thompson (1989), play an important role in aphasia rehabilitation: re sponse generalization (i.e ., when an untrained response occurs after other re sponses have been trained (e.g., production of different, untrained items) and stimulus generalization (i.e., when there is carryover of a trained behavior to a different, untra ined stimulus condition, e.g., outs ide of the clinic). Without the former, there would be an inordinate num ber of responses to train, and without the latter, trained behaviors would occur only in the clinic setting (Thompson, 1989). Although both of these are important, th e current study focuses only on response generalization. Several investigators (Hillis, 1998; Kiran & Thompson, 2003b; Plaut, 1996; Thompson, 1989) have suggested that treatment studies might be able to influence the recovery of language if they ta rget factors that are modifiab le. A principled selection of materials for training as well as for testing generalization is just one example of this (Hillis, 1998; Kiran & Thompson, 2003b; Pl aut, 1996; Thompson, 1989; Thompson et al., 2003). One technique that has successfu lly induced generalization is to have a structural relationship among and across the stimuli (Stokes & Baer, 1977; Thompson, 1989; Thompson et al., 1998; Thompson et al., 1997; Thompson et al., 2003). One approach recommends choosing s timuli that are more versus less complex, a factor that has been shown to be important for increasi ng generalization to untra ined items (Kiran & Thompson, 2003b; Plaut, 1996; Thompson et al., 1998; Thompson et al., 1997; Thompson et al., 2003). This phenomenon has been referred to as the complexity effect

PAGE 18

4 and the Complexity Account of Treatment Efficacy (Thompson et al., 1998; Thompson et al., 1997; Thompson et al., 2003), which has been successfully used in syntactic treatment studies in the aphasia population. Kiran and Thompson (2003b) have equated th e notion of semantic complexity with category typicality (i.e., by the degree to whic h a semantic category exemplar is similar to the category prototype, as described Ro sch, (1975)) and applied this to a semantic training in adults with aphasia. In this study, Kiran and Thompson (2003b) found generalization to untrained items after atyp ical items from a category were trained. Findings from this study replicated similar re sults from a connectionist model that were simulated in a computer experiment of ac quired dyslexia (Plaut, 1996). In a similar study, Stanczak, Waters, and Caplan (2006) reported ge neralization for one of two participants with aphasia, after training atypical categor y exemplars. However, two other aphasia naming treatment studies (Mayer, Murra y, & Karcher, 2004; Stanczak, Waters, & Caplan, 2005) did not find generalization fo llowing trainings with either atypical and typical items, although treatment effects were observed. Thus it is important to further explore the effectiveness of training typical or atypical category exemplars as a strategy to encourage generalizati on (Murray & Clark, 2005). All of these studies (Kiran & Thompson, 2003b; Mayer et al ., 2004; Stanczak et al., 2005, 2006) have tested the predictions of the Complexity Account of Treatment Efficacy (Thompson et al., 2003), and cont rasted it with an older approach based on the family resemblance/prototype hypothesis by Rosch (1975). Arguments based on this view center around the benefits of training qualities or f eatures that are shared among many members of a semantic category, so that many items in the category might benefit, and thus, it

PAGE 19

5 provides a rationale for training typical items, which contain a larger proportion of shared features. The structure of a semantic category is an essential element in the rationale for treatments based on typicality (Plaut, 1996) ; therefore, a population with preserved category structure such as individuals with PAD (Flicker, Ferris, Crook, & Bartus, 1987; Huff et al., 1986; Martin & Fedio, 19 83; Nebes, Boller, & Holland, 1986; Salmon, Butters, & Chan, 1999; Schwartz, Marin, & Saffran, 1979; Schwartz, Kutas, Butters, Paulsen, & Salmon, 1996; Warrington, 1975) mi ght be optimal for testing these hypotheses. Some theories of how semantic repr esentations are affected by PAD have specifically suggested that item s that have many shared or in tercorrelated features might be more preserved in PAD compared those w ith representations primarily consisting of distinguishing features, with relatively fewer shared f eatures (Altmann, Kempler, & Andersen, 2001; Devlin, Gonnerman, A ndersen, & Seidenberg, 1998; Gonnerman, Andersen, Devlin, Kempler, & Seidenberg, 1997). This literature combined with the evidence that the PAD population has worse pe rformance on atypical items compared to typical items (Sailor, Antoine, Diaz, Ku slansky, & Kluger, 2004; Smith et al., 1995) provides further incentive to compare perf ormance by adults with PAD on typical category exemplars, which have many shared features, and atypical category exemplars, which have relatively fewer shared features. This exploratory study extends the small body of research comparing the effects of training typical and atypical category exempl ars in adults with aphasia (Kiran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005, 2006) to those with mildmoderate PAD using a modified version of th e semantic feature ge neration task (Boyle,

PAGE 20

6 2004; Boyle & Coelho, 1995; Coelho, McHugh, & Boyle, 2000). Since there were no reports of this type of training with the PAD population, it seemed appropriate to first apply it in a single session to determine its feasibility. Thus, the fi rst aim of the study was to determine if there is a facilitation eff ect (Hickin, Best, Herbert, Howard, & Osborne, 2002; Howard, 1985; Howard, Patterson, Fra nklin, Orchard-Lisle, & Morton, 1985; Patterson, Purell, & Morton, 1983), for trained items from either item repetition alone (Kendall, personal communication, 2005, Fuller et al. 2001; Hickin et al., 2002; Mayer et al., 2004; Patterson et al., 1993; Rothi et al., 2005 ) or from the repetition plus a semantic training (Drew & Thompson, 1999; Le Dor ze, Boulay, Gaudreau, & Brassard, 1994; Wiegel-Crump & Koenigsknecht, 1973). The second aim of the study was to determine whether there was generalization of training to untrained items in the semantically trained categories. The third aim of the study was to examine generalization to an untrained task, category generation, using the same seman tic categories as the naming and feature analysis tasks. The structure of this study is as follows. The next two chapters provide reviews of the literature. Chapter 2 discu sses the naming deficits (i.e., an omia) in adults with PAD, and the anomia treatment studies for th is population. The third chapter provides background information on semantic theory, the two competing views, the relevant findings from the connectionist model (Plaut, 1996) and the treatment studies in adults with aphasia using semantic category exempl ar training with typi cal and atypical items (Kiran & Thompson, 2003b; Mayer et al., 2004 ; Stanczak et al., 2005, 2006). This is followed by a description of the current study, including rationale for applying these techniques to adults with PAD. Additional findings from the aphasia treatment literature

PAGE 21

7 are identified as they relate to the me thodology for our study. Finally, our research questions and predictions based on the above literature are descri bed. In Chapter 4, the study methods are explained in detail, followe d by the results and discussion in Chapters 5 and 6, respectively. To foreshadow our re sults, this exploratory study provided good preliminary evidence that semantic training ca n have significant effects on the anomia found in PAD, and, with judicious choice of stimuli, this training may potentially generalize to other items in the same category.

PAGE 22

8 CHAPTER 2 REVIEW OF THE LITERATURE 1: NAMING DEFICITS IN PAD It is common for patients in the early stag es of PAD to have picture naming deficits (Appell, Kertesz, & Fisman, 1982; Barker & Lawson, 1968; Bayles & Tomoeda, 1983; Chertkow, Bub, & Seidenberg, 1989; Flicker et al., 1987; Grossman et al., 2004; Kirshner et al., 1984; Lipinska & B ackman, 1997; Schwartz et al., 1979; Warrington, 1975; Williams et al., 1989; Williamson et al., 1998). Furthermore, the exact nature of the naming deficit in PAD is controversial (B ell, Chenery, & Ingram 2001; Nebes, 1992; Nebes, Brady, & Huff, 1989) regarding whet her the deficit stemmed from impaired visual-perception, semantic memory, or le xical access (Nebes, 1989). An early theory suggested that visual mispercep tion is the reason for the naming deficit, and this is based on the presence of visuo-perceptual errors, presumably due to difficulties perceiving the object (Barker & Lawson, 1968; Kirshner et al., 1984). These theories and studies providing evidence are discussed below, beginning with the se mantic deficit, then the lexical access, and followed by the semantic a nd lexical access deficit. Next, a discussion of the few naming treatment studies that ha ve targeted this population is provided (Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005). These PAD naming therapy studies achieved a naming treatment effect (i.e., acquisition) but only one also reported generaliza tion (Abrahams & Camp, 1993). This limited number of studies suggests a need for further research. Insights from all of these studies and additional research from the aphasia lit erature provided technique s that were applied

PAGE 23

9 in our exploratory study compar ing the effects of training ty pical and atypical category exemplars. Semantic Memory Deficit in Prob able Alzheimer Disease (PAD) Several studies have suggested that there is semantic memory loss or degradation and, consequently, a loss of information abou t semantic representati ons (Alathari, Trinh Ngo, & Dopkins, 2004; Hodges & Patterson, 19 95; Hodges et al., 1996; Hodges, Salmon, & Butters, 1991, 1992; Huff et al., 1986; Huff et al., 1988; Margolin, Pate, Friedrich, & Elia, 1990; Martin, 1992; Salmon, Butters et al., 1999; Salmon, Heindel, & Lange, 1999; Salmon, Shimamura, Butters, & Smith, 1988; Sc hwartz et al., 1979). Evidence for this theory includes impaired naming with semantic errors related to the superordinate category or an associate item within the category (Bayles & Tomoeda, 1983; Martin & Fedio, 1983; Salmon, Butters et al ., 1999); the presence of consistent naming responses at two different test periods (Henderson, M ack, Freed, Kempler, & Andersen, 1990) an association between the inability to name an item and the inability to recognize its name (Flicker et al., 1987; Huff et al., 1986; Huff et al., 1988); and a relationship between the naming failures and the lack of core info rmation provided about the corresponding item (Hodges et al., 1996). In addition, deteriorati on of semantic memory in PAD has been described based on poor performance on explic it tasks such as picture naming tasks and category fluency (Bayles & Tomoeda, 1983; Ch ertkow & Bub, 1990; Flicker et al., 1987; Huff et al., 1986; Martin & Fedio, 1983); tasks requiring generation of semantic feature knowledge (Alathari et al., 2004), generation of verbal definitions (Garrard, Lambon Ralph, Patterson, Pratt, & Hodges, 2005; Hodge s et al., 1996); as well as questions targeting feature knowledge (Chertkow et al., 19 89; Giffard et al., 2002 ). Other tasks that have showed similar findings include maki ng judgments about semantic relatedness

PAGE 24

10 (Bayles, Tomoeda, & Cruz, 1999); and associ ating words, defining words, and ranking associations (Abeysinghe, Ba yles, & Trosset, 1990). There appears to be different interpretations or variants of this theory of semantic memory loss or degradation in PAD. For exam ple, several investigators have found that, while attribute knowledge about a specific concep t or exemplar in a category is impaired, superordinate category knowle dge is preserved (Chertkow & Bub, 1990; Flicker et al., 1987; Giffard et al., 2002; Giffard et al ., 2001; Hodges et al., 1991; Huff et al., 1986; Lukatela, Malloy, Jenkins, & Cohen, 1998; Ma rtin & Fedio, 1983; Nebes et al., 1986; Salmon, Butters et al., 1999; Schwartz et al., 1979; Warrington, 1975). This led some researcher to assert that the attributes of concepts are degr aded in PAD (Chertkow et al., 1989; Giffard et al., 2002; Giffard et al., 2001; Martin, 1992; Ma rtin & Fedio, 1983). Martin (1992) suggested th at random damage from th e pathological process of PAD resulted in changes to the semantic re presentations such that they would be degraded and thus more similar to one anot her. With progression of the disease, the ability to distinguish between items in the same category would be diminished. Thus, on a confrontation naming task, a semantic repr esentation would be activated but it would lack the specificity needed to correctly na me the item and, as a result, several lexical entries would be activated. Therefore, when the person with PAD sees a picture, the semantic representation that is activated may be underspecified in terms of the details. This is due to the absence of the essential attributes knowledg e for that item such that the activated distinguishing featur es of that object are not st rong enough to rule out other similar items (Altmann et al., 2001; Devlin et al., 1998; G onnerman et al., 1997; Martin, 1992).

PAGE 25

11 A property verification task has also been used in a group of adults with PAD (Smith et al., 1995). Participants were as ked to verify brief statements containing information about an items properties. Th e items were high or low typicality (and dominance, which relates to th e relevance of the meaning of the item and is correlated with typicality) category exemplars. The statements were e ither distinguishing characteristics (i.e., distinctive) or shared f eatures (i.e., common to the other items in that category). The results from the accuracy scores and reaction times indicated a degradation of property level information, particularly regarding both low dominance typicality items in the category. This was not interpreted as a loss of representations of the items properties or a re organization of relationships among properties of objects. Instead, Smith et al. (1995) offered this as evidence that the representations of category exemplars that are low-typical and low domina nt have been degraded by the Alzheimer pathology. Furthermore, these investigators hypo thesized that task demands might be a reason for differences in the literature. Smith et al. ( 1995) suggested that implicit knowledge allows adults with PAD to have fa ster verification response times. Thus, in a category verification task, the demands probably do not need the full semantic specification. Smith et al. (1995) also suggest ed that while explic it knowledge is needed to assess information could be impaired, fo r example, tasks requiri ng the participant to use relevant information in a ranking tas k. This example comes from Grober, Buschke, Kawas, and Fuld (1985) who reported that performance on an attribute ranking task showed that attributes about a c oncept are preserved, but that the organization of semantic information is altered by the disease process. Another example from Smith et al. (1995) is that the naming process involves co mputation such that an objects properties

PAGE 26

12 are activated to allow for distinction among other category members. If Alzheimer disease processes affect both distinguishing and shared feat ures that are low dominant and low typicality, this could interfere with complete activation of the object representation (Smith et al ., 1995). Consequently, it woul d reduce the accuracy on a naming task. Better performance is seen when contextual information is provided, yet when the task involves full semantic repr esentation without the support, performance declines (Smith et al., 1995). While the evidence above for a progressive deterioration of the semantic system comes largely from tasks requiring explicit ac cess of semantic repr esentations, evidence for an overall preservation of semantic categ ory knowledge comes from tasks that require only implicit knowledge of word semantics. For example, adults with PAD did not s how a priming effect on a stem-completion task and had reduced number of productions fo r the second semantically related item on a free association task (Salmon et al., 1988). Hyper-priming has been reported on lexical decision tasks and found to be associated with degraded semantic representations (Chertkow et al., 1989; Giffard et al., 2002; Margolin, Pate, & Frie drich, 1996; Martin, 1992). To account for the hyper-priming in PAD, Martin (1992) sugge sted that changes in semantic activation are more robust for pr ocessing information that is degraded, such that a degraded semantic network benefits more from a semantic prime in than an intact semantic network. To account for the diffe rence in performance on tasks requiring implicit versus explicit access to semantic re presentations, it has been suggested that participants with PAD have di fficulty performing an inten tional search through semantic memory, but perform relatively normally wh en relying on the au tomatic spread of

PAGE 27

13 activation in the semantic network (Chertkow & Bub, 1990; Nebes, 1992; Ober, Shenaut, & Reed, 1995). All of these studies with e xplicit and implicit tasks suggest that at some level, the semantic system is impaired (due to loss or degradation), and differe nt investigators have used different tasks and methodologies to addr ess this. However, this theory does not go unchallenged; an alternative argument suggests th at lexical access, specifically, access of the phonological word form from semantics, is the primary deficit in PAD. Lexical Access Deficit in PAD Several researchers have ar gued that the naming impairment in PAD results from impaired access to the phonologi cal form of the word (i.e., a retrieval deficit) in the presence of an overall intact semantic knowledge structure (A lbert & Milberg, 1989; Benson & Geschwind, 1985; Nebe s, 1992; Nebes et al., 1986; Ne bes et al., 1989; Nebes, Martin, & Horn, 1984; Neils, Brennan, Co le, Boller, & Gerdeman, 1988; Ober & Shenaut, 1999; Thompson-Schill, Gabrieli, & Fleischman, 1999). Along these lines, it has been suggested that the presence of seman tic errors is actually an indication that knowledge about the item is intact, despite the lack of the ability to retrieve it, rather than in indication of a semantic impairment (Nebes, 1989). Semantic priming paradigms (as discussed earlier) have been employed to evaluate the status of semantic memory via an implicit task, in order to reduce the participants use of attentional mechanisms (Ober, 1999). Ther e are reports of norma l priming in adults with PAD (Nebes et al., 1984), as well as reports of hyper-priming which have been interpreted as being caused by abnormal a ttentional processes in the presence of preserved semantic memory (Hartman, 1991; Ober & Shenaut, 1995; Ober et al., 1995). In a meta-analysis of semantic priming studi es, Ober et al. (1995) reported that hyper-

PAGE 28

14 priming was due to attentional mechanisms which co-occurred with large increases in reaction times found in controlled priming para digms. These researchers argued that the hyper-priming represented evidence that auto matic spreading activation was occurring among semantic representations in the semantic priming task (Ober et al., 1995). Further evidence for this comes from another st udy by Ober and colleagues (1995) in which adults with PAD, as well as healthy young and older adults participated in a series of lexical decision tasks. Results showed that priming effects were equal across the groups. The PAD group demonstrated longer reacti on times on low frequency words before making lexical decisions, and this was interpreted as an indication that additional time was needed for reaching activation level thre shold for these items. Ober et al. (1995) argued that these data reflected an intact semantic memory structure in PAD. Two event-related potential (ERP) studies offered more support for an access deficit in PAD (Ford et al., 2001; Schwartz et al., 1996). Even t related potentials measure electrical manifestations of particular psychological proce sses that occur in preparation for or in response to discrete events (Fab iani, Gratton, & Coles, 2000; Kutas & Hillyard, 1980). The N400 is a negative going deflecti on occurring at ~400 ms and occurs in response to anomalous information, specificall y, semantic violations; for example it is larger when the prime is unrelated to the target compared to when it is related (Fabiani et al., 2000; Kutas & Hillyard, 1980). Schwartz et al. (1996) compared h ealthy older adults and adults with PAD to determine whether or not the specificity of a category prime had a differential effect on the degree of seman tic priming in these two populations. Using a variation of a category verifica tion task, the investigators asked the participants first to listen to a prime that was delivered auditorily as a name of a category (different levels,

PAGE 29

15 e.g. superordinate and subordinate) and then, to read the written presen tation of the target name (i.e., a member of a category). While th e most robust priming effects were found in the young adults, the smallest pr iming effect and slowest reac tion times were found in the PAD group. Responsiveness to the category mani pulation was revealed in priming effects that were similar across the groups, not just in the reaction times but also in the ERP N400 congruity effects being larger on some levels, for example, the subordinate category level. Thus, category level manipul ation affected the PAD group, and this was interpreted as evidence for an intact seman tic network. In contrast, longer response times and smaller priming effects were attributed to the demands of the task and the necessity of searching through memory while engaged in online processing required for this and other similar tasks (Schwartz et al., 1996). In a more recent ERP study, Ford et al. ( 2001) examined priming based on age and dementia, and asked whether the N400 amplitude could be used to show specific semantic memory deficits for objects that co uld not be named. The participants (adults with PAD, healthy young controls and healt hy older adult controls ) completed a pretest confrontation picture naming task consisting of items from 12 sema ntic categories. The following week, the participants completed a picture naming verification task while ERPs were measured. The participants were instruct ed to press a button to indicate whether or not the prime (a picture) matched the target (a word). The consistent finding across the groups was that for a word that did not matc h the picture, there was a more negative N400 amplitude. The ability of the PAD gr oup to correctly name pictures was not associated with the N400 priming effect or any corresponding scalp distributions. Thus, despite an inability to access the name of an item, there was evidence from the N400

PAGE 30

16 results that the PAD group had sufficiently in tact knowledge for priming responses at the cortical level (Ford et al., 2001). It has been suggested that while the stru cture of semantic memory is not damaged in PAD, there is a generalized cognitive pr ocessing deficit, which is required for intentionally retrie ving and evaluating information, and decision making (Nebes, 1992). This results in word retrieval or access de ficits. Two ERP studies (Ford et al., 2001; Schwartz et al., 1996) also provi de evidence for a fairly pres erved semantic structure in PAD. Additional support for an access defic it in PAD comes from the finding by Ousset et al., (2002) that initial sylla ble cues and the sound represen ting the item were among the most effective cues during a oral naming to definition task, while providing the semantic category was the least effective. Semantic Impairment and Le xical Access Deficit in PAD The most likely explanation of these disparat e findings is that the naming deficit in PAD is due to a combination of a breakdown in both semantic memory and retrieval abilities (Bowles, Obler, & Albert, 19 87; Huff et al., 1988; Watson, Welsh-Bohmer, Hoffman, Lowe, & Rubin, 1999; Williamson et al., 1998). This has been suggested based on results from confrontation naming and fluency tasks. Williamson et al. (1998) made predictions about performance of healthy older adults compared to adults with PAD on two confrontation naming tasks: the Boston Naming Test (Kaplan, Goodglass, & Wein traub, 1983) and the Action Naming Test (Nicholas, Obler, Albert, & Goodglass, 1985) based on knowledge about the underlying anatomy and PAD pathology. More specifically, they predicted that, compared to action naming; object naming would be more impaired in PAD. This was because the disease process had resulted in more extensive damage to the left temporal area which is needed

PAGE 31

17 for naming objects, compared to the frontal areas required for naming actions. Compared to the control group, the partic ipants with PAD were less ac curate on both measures, but their overall performance was worse on the BN T, on object naming test, compared to the Action Naming Test. Williamson et al. (1998) suggested that object naming revealed significant impairments at the semantic speci fication and lexical levels. However, for action naming, the impairment was mostly at the semantic specification level, while the lexical level remained fairly intact. The author s also attributed the presence of more No Response errors on objects than on actions to the corresponding damage to the anatomy that underlies that function. W illiamson et al (1998) concluded that compared to healthy older control participants, adults with PAD are not as accurate when they name objects and actions. Furthermore, the degree of defi cit is more pronounced with objects, which was attributed to impairment not just at the semantic specification le vel, but also at the level of lexical access. Based on these findings we designed our study to address both of these levels of deficits. The participants e ngaged in both lexical a nd semantic tasks to stimulate the system at both the lexical (e.g., repeating the correct name of the item to access the word form) and semantic levels (e.g., answering questions about semantic features of the item). In summary, the naming impairment in P AD has been attributed to a loss of or degradation in semantic memory storage and/ or organization, disrupt ion of the retrieval of the information in the network, a combination of these, or other factors related to task demands. However, there is reason to believe that the semantic structure, at least in terms of semantic categories, is sti ll fairly preserved in early PAD (Flicker et al., 1987; Huff et al., 1986; Martin & Fedio, 1983; Nebes et al., 1986; Salmon, Butters et al., 1999;

PAGE 32

18 Schwartz et al., 1979; Warring ton, 1975). For example, although Cronin-Golomb et al. (1992) reported longer reaction times on a cat egory decision task as well as smaller number of items on a category generation ta sk, the pattern was normal across categories and items were ranked by typicality in a normal manner. This was interpreted as being a reflection of intact semantic category organization (Cronin-Golomb et al., 1992). Naming Treatment Studies in PAD Although there is a plethora of research on the naming impairment in PAD, there are only a few naming treatment studies that involved participants with PAD (Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005). Like the typicality training studies with the aphasia population (K iran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005, 2006), thes e three PAD studies also had picture naming outcomes based on comparisons of perf ormance before and after the intervention (i.e., a treatment effect), and examined gene ralization to untrained items. Although these three PAD treatment studies are very different from each other, they each have important implications for the future of word finding treatment for this population. Using a single subject design study with tw o adults with dementia, Abrahams and Camp (1993) used spaced-retrieval traini ng (SRT). Spaced retrieval training is a technique that intersperses increasing time inte rvals between presentation of a target by the clinician and recall of it by the partic ipant. Participant 1 was diagnosed with progressive dementia (16/30 on the Mini Mental Status Exam ; MMSE) (Folstein, Folstein, & McHugh, 1975) and Participant 2 had PAD (13/30 on the MMSE). Each participant was tested on the Boston Na ming Test (Kaplan et al., 1983). From these results, nine training items from each particip ant were identified for use as a subset of training and control items. Wh ile two of these items were treated, the other seven were

PAGE 33

19 controls (Abrahams & Camp, 1993). Abrahams and Camp (1993) reported that the results of two different treatment targets for each participant showed improved performance (e.g., naming an item after a maximum of 300 seconds). Only Participant 1 was tested two weeks later. This showed maintenance eff ects for one of the two items, despite a six point drop in her MMSE score (10/30). Abrahams and Camp (1993) described generalization to untrained exemplars of th e trained items: a colored drawing of the target; a second target; and an actual exemplar. The inves tigators reported that SRT was effective in adults with dementia because th ey not only have word-finding deficits, but also have cognitive deficits that affect the ability to remember new information. It was suggested that SRT uses procedural memor y, which appears to be intact in this population, and thus enabled the participants to benefit from the anomia treatment (Abrahams & Camp, 1993). Although both participants did not have PAD, the findings from this early anomia treatment study by Abrahams and Camp (1993) ha ve value because they were the first to demonstrate a treatment effect after usi ng SRT, in the PAD population, even with participants with low MMSE scores. Spaced retrieval training does not provide any semantic information about the item, but uses repetition. It was not clear if the items that showed generalization actually generalized. In stead, it could be that it was easier for the participant to name the color drawing. Ousset et al. (2002) compared the effect s of an experimental lexical therapy targeting both episodic and lexical stimulation in adults with mild PAD who were on an acetyl cholinesterase inhibitor. The averag e MMSE score was 21.1. The participants were divided into two groups of ei ght: one group received lexical therapy (with narratives and

PAGE 34

20 definitions) and the other, a control group, re ceived occupational therapy (i.e., pottery, drawing, and conversations). The treatmen t was provided in 16 sessions, once a week, with a two week break at the midpoint. Th e pre and posttest picture naming stimuli consisted of 120 black and white line drawings from three categories. While 80 of these items from the three categories were a part of the lexical therapy, the remaining 40 were not. During each lexical therapy session, the pa rticipant first read aloud a narrative from a computer screen and then listened while th e examiner read it. Then, the participant completed a naming to definition task on the computer. There were 20 definitions, half of the words were in the narrative to provide both semantic and episodic reinforcement; and half of the words were not in the narrative, to provide episodic reinforcement. When the participant did not respond or was incorrect, the computer randomly provided one of five different types of cues (i.e., semantic categor y, first syllable, first grapheme, presentation of the item as a color drawing; or the items associated sound). If th is did not result in a correct response, the comput er provided the answer. Co mpared to the control group, Ousset et al. (2002) reported a significant impr ovement from pretest to posttest for the treated items in the lexical therapy group (i.e ., a treatment effect) but generalization to untreated items was not signifi cant. The investigators also commented about the possible benefit of episodic long-term memory reinfo rcing the association between the objects form and the name of it. Alt hough the narratives (i.e ., part of the lexical therapy) were designed to assist in memorizat ion of the lexical labels, the participants were better at retrieving the words that were not in the narr atives. Ousset et al. ( 2002) interpreted this reduced naming as a possible indication that working memory was over-extended by the semantic context in the narratives.

PAGE 35

21 The analyses from the cues revealed that the presentation of the color drawings and the initial syllable were the most effective cues, and Ousse t et al. (2002) hypothesized that both of these improved the process of searching the lexicon. The semantic category cue was the least effective. To account for this, the authors suggested that the naming deficit in their patients primarily affected their ability to a ccess the phonological form from semantics. Ousset et al. (2002) questioned if the participants ac tually had a semantic deficit and used the training as a semantic intervention. Alternatively, it was hypothesized that the participants were a subgroup of the PAD population, such that their anomia might have been due to a retrieval deficit, and the lexical ther apy provided rehearsal targeting episodic lexical information and linking it to objects and their corresponding names (Ousset et al., 2002). This group study by Ousset et al. (2002) clea rly showed treatmen t effects from the training and provided an innovati ve approach to remediate anomia. Trained words were unrelated to each other. It is unclear why Ousset et al. (2002) did not report generalization to untrained items, when the tr aining items were from three categories. Also, further examination of the results in dicated that there was generalization to another task The participants showed better naming performance at the posttest compared to the pretest, but the treatment involved narra tives and oral nami ng to definition, not naming pictures. Generalization to another task suggests that the treatment strengthened connections between semantic and phonologi cal representations (Hillis, 1998). As mentioned earlier, the findings from Ousset et al. (2002), which indicated that the category cue was the least helpful on the na ming task, could provide further support for the notion that semantic category structure in PAD is still intact. The rationale for the use

PAGE 36

22 of the semantic categories as the semantic cue was not addressed. Given the evidence of preserved superordinate semantic categor y knowledge (Chertkow & Bub, 1990; Flicker et al., 1987; Hodges et al., 1991; Huff et al ., 1986; Lukatela et al ., 1998; Martin, 1992; Martin & Fedio, 1983; Nebes et al., 1986; Salm on, Butters et al., 1999; Schwartz et al., 1979; Warrington, 1975) but degraded attribut e knowledge (Martin, 1992) it is not surprising that the category cue had minimal e ffects. Perhaps a more potent semantic cue would have been to provide more distinguish ing information about the item or possibly a picture of it with some foils. This might ha ve led to better performance on the oral naming to definition task. A recent single subject design study (Fulle r et al., 2001; Rothi et al., 2005) also used an acetyl cholinesterases inhibitor, but with an errorless learning paradigm to improve confrontation naming in six adults wi th PAD. The scores for the patients on the MMSE were > 10. In this study, the stimuli consiste d of 100 words, evenly divided for high and low frequency, matched with black and white line drawings from eight semantic categories representing both natural kinds (e.g., animals) and artifact categories (e.g., clothing) with 10 exemplars per category. Th ree subsets of stimuli based on baseline performance were developed for each partic ipant using three categories, allowing for training on two of these, while the third s ubset was used both for generalization and experimental control. To establish baseline stability, each participant named the 100 pictures over eight daily probe s (i.e., over eight sessions). Si xty minute therapy sessions were provided four times per week, until crit erion was met (90%) or after 20 sessions (20-35 sessions were needed to complete the entire protocol) (Fulle r et al., 2001; Rothi et al., 2005).

PAGE 37

23 The therapy included two similar treatment phases to determine if the improvements seen in Phase 1 would show re sponse generalization to performance from Phase 2, which at that point would be an untrained set of stimuli. There were two treatment conditions, immediate and delaye d repetition. For Phase One, Treatment Condition One involved simultaneous repetiti on in which the clinician presented the picture, stated the correct name of it and had the participant repeat it. For Treatment Condition Two, delayed repetition was used in which the clinician presented the picture and if the participant knew th e name, he/she stated it. If /when the participant did not know the name of the item, he/she informed th e clinician. Then, the clinician stated it and had the participant repeat it (Fuller et al., 2001; Rothi et al., 2005). In other words, there was a delay from the time the participant sa w the picture until he/she repeated. The results revealed that half of the participants (i.e., three of six) showed a treatment effect. There was no generalization to the untrained items. However, results from the three month maintenance probes showed that the treatment effect was still evident. The investigators also conducted a pos t hoc review of the particip ants records and identified two other factors that appeared to play a role in the outcome s. The three participants who responded to the treatment both lived at home (versus in an institution) and were not on medications that could affect the learning pro cess and brain plasticity (Fuller et al., 2001; Rothi et al., 2005). This study (Fuller et al., 2001; Rothi et al., 2005) was the first to demonstrate successful application of errorless learning (i .e., treatment and maintenance effects) in a naming treatment in the PAD population. Furthe rmore, the influen ce of both the living status and the medications are important co nsiderations for treatment planning. Perhaps

PAGE 38

24 another reason for the treatment effect was the use of semantic categories (both manmade and natural kinds) and their exemplars. These three studies (Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005) were th eoretically motivated, well-d esigned, and offer different approaches for improving the naming performan ce in adults with PAD, including single subject and group design. More specifically, two of these tr eatment approaches combined cholinergic medication with training items fr om semantic categories, one in a lexical training with narratives and nami ng to definition (Ousset et al., 2002) and the other via an errorless learning paradigm (Fuller et al ., 2001; Rothi et al., 2005). Generalization to untrained items essentially did not occur in any of these studies. However, the finding that treated behaviors were maintained fo r one participant two weeks after training (Abrahams & Camp, 1993) and for three particip ants three months afte r training (Rothi et al., 2005) is remarkable. This reflects the res ponsiveness to the traini ng and that the effect was robust despite the MMSE scores. The l ack of generalization to untrained items following anomia treatmen t in the PAD warrants further investigation. According to Nadeau and Gonzalez Rothi ( 2004), the connectionist approach views anomia (that is caused by a semantic defici t) as a reflection of insufficiently engaged representations of features th at are critical for making distinctions among concepts. When a network is damaged, a large amount of inform ation still remains in the network, so the focus should be on refining the damaged networ k via semantic therapy. In particular, the network needs to be changed in terms of its connectivity so that th ere is more reliable engagement of the distinguishing features, while simultaneously there is relatively a disengagement of the shared features (N adeau & Gonzalez Rothi, 2004). These findings

PAGE 39

25 suggest that participants with early PAD might theref ore be appropriate for the semantically based training provided in our study which compared typical and atypical category exemplars.

PAGE 40

26 CHAPTER 3 LITERATURE REVIEW 2: A BRIEF OVE RVIEW OF THE ORGANIZATION OF SEMANTIC MEMORY Since this study relied heav ily on the predictions of theories about semantic category structure, it is importa nt to provide a brief overview of some semantic theories and categorization. This is followed by a disc ussion of the two views which can be used to compare training with typical and at ypical category exemplars. The family resemblance/prototype (Rosch, 1975) view provides support for training typical items, while support for training atypical items to achieve maximum generalization comes from the Complexity Account of Treatment Efficacy (Thompson et al., 2003), and a connectionist model of a computer simula tion of acquired dyslexi a (Plaut, 1996). Next, there is a brief description of the four studies that have compared these views in the context of a naming intervention with typica l and atypical category exemplars for adults with aphasia (Kiran & Thompson, 2003b; Ma yer et al., 2004; Stanczak et al., 2005, 2006). A rationale for applying this training in adults with mild to moderate probable Alzheimer disease (PAD) is also be provided. This leads to a discussion of the current study. A Brief Overview of Semantic Memory and Categorization Semantic memory has been described as a hierarchical network consisting of conceptual information and knowledge, stori ng semantic representations of facts, knowledge for objects and concepts, words a nd their corresponding meanings, as well as associations (Au & Bowles, 1991; Bayles Kasniak, & Tomoeda, 1987; Tulving, 1983).

PAGE 41

27 Within the semantic system, the representations are organized in a hi erarchical order and distributed throughout the association cortices of the brain (Marshall, 1988; McCarthy & Warrington, 1990). It is important to briefly describe the ne tworks that allow for this organization at the word level. The spreading activation th eory (Collins & Loftus, 1975; Quillian, 1967) suggests that there are at leas t two levels of representations semantic, which is organized based on semantic similarity of the meaning of the word, and lexical, which is organized based on the phonemic similarity (and to an ex tent, orthographic similarity). Furthermore, concepts have been hypothesized to have repr esentations as nodes in a network that have features of the concept that are represente d as associated connections to other nodes (Collins & Loftus, 1975; Quillian, 1967). Thes e links are based on the importance or relevance of the information to the concep ts meaning such that more important nodes have shorter links or distances between two c oncepts, and together the nodes make up the network (Collins & Loftus, 1975; Quillian, 1967). The similarity allows for information to spread along pathways in the netw ork (Collins & Loftus, 1975; Quillian, 1967). Support for the notion of a semantic netw ork and the spreading activation theory (Collins & Loftus, 1975; Quillian, 1967) comes from the semantic priming effect, initially described by Meyer and Schvaneveldt (1971), in which participants had faster reaction times when the prime in a lexical deci sion task was associated with the target, but were slower when it was not associated. For example, responses to butter were faster when preceded by bread than when preceded by window. It was theorized these faster responses (i.e., facilitation) occurr ed as a result of spr eading activation due to shared semantic contexts. The decision was faster because there was an increase in the

PAGE 42

28 availability of the target as a result of the pr ime. More specifically, the memory system is connected to the node for butter, so when one is activated, the activation spreads to connected nodes (Meyer & Schvaneveldt, 1971). According to Martin (1992), semantic priming assumes that there should be automa tic activation of an objects representation upon presentation of the object (picture presentation or wo rd), and related concepts should be activated by the spread of activati on, which increases accessi bility to the name of the object. Martin (1992) further suggest ed that these explanations of semantic priming are based on the assumption that the brain is a di stributed neural netw ork that instantiates semantic representations. For example, presenta tion of a picture results in activation of a corresponding neural network, wh ich includes representations of the attributes of the item. Items that are closely related have ma ny shared attributes and overlapping networks that represent them (Martin, 1992). This also explains how the semantic network responds to such information, and when activated above the threshold, semantic processing allows for selection of the target word instead of others (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000). Although ther e are different theories related to this topic, it is beyond the focus of our study. The contents of mental representations ha ve been used to account for a range of phenomena such as knowledge of correctly recognizing objects by their name (i.e., words label) (Medin & Smith, 1984). There is a large body of research on semantic categories and how they are represented (Kiran & Thom pson, 2003a). This literature review focuses on the superordinate classifi cation of a category, which is a higher-level

PAGE 43

29 grouping allowing for concrete objects to be classified into vari ous divisions (Rosch, 1975). There are several benefits of categoriza tion, for example, it is both economic and informative. In terms of economy, categorie s frankly provide a pr actical and adaptive means for classifying similar items in order to avoid the overwhelming cognitive task of considering all items as if they were uni que (Anderson, 1991). Cate gorizing also allows for assumptions about information that is no t explicitly provided, for example, that the item will share similarities with other item s in that category (Anderson, 1991; Komatsu, 1992). Historically, different views have been put forth to account for concepts and categorization. Within some theories of psychology and philosophy the nature of categories is described as Aristotelian (R osch, 1973a, 1975): They are logical; have clear boundaries; and criteria-based membership. Thus, being a category member requires having a basic set of essential attributes or features such that each item is fully and equally a member. This approach is known as the classical view wh ich suggests that the representation of categories is achieved by having a set of defining features (Kiran & Thompson, 2003a). The classical view was challenged by Wittgenstein (1953) who argued that the requirements of formal criteria were not neces sary from either a logical or psychological point of view (Rosch & Mervis, 1975). De spite critique, the classical view was researched during the 1960s and considered to be adequate as an appropriate way to describe lexical concepts found in everyda y living (Komatsu, 1992). However, by the 1970s, problems with the classical view su rfaced, including a lack of evidence for

PAGE 44

30 defining properties of object s and concepts, identification of categories that did not have clear-cut boundaries, and evidence sugge sting that there is inequality among category members, and resulted in its decline (Kiran & Thompson, 2003a; Komatsu, 1992; Medin & Smith, 1984; Posner & Keele, 1968; Rosch, 1973a, 1975). This led to the development of other theories, including the family resemblance/prototype view (Rosch, 1975; Rosch & Mervis, 1975), which has been described as being among the most widely cited psychological concepts (Hampton, 1995) an d is discussed in more detail below. Family Resemblance/Prototype View Rosch (1973a; Rosch, 1975) argued that for many natural semantic categories, the Aristolean approach does not apply. Family re semblance is the term that was applied to highlight the structural prin ciples governing category membership. The notion of family resemblance means that there is an internal structure to the category such that it is organized around a category prototype (Rosch & Mervis, 1975). This allows for a very general approach to the relationships and for application to categories whether they share features with other category members or m eet formal criteria for membership in the category Some of the original work on this topic (Rosch, 1973a) re vealed that the prototype of the category is th e best example of a category a nd/or the clea rest case of category membership. The prototype occupied the central location within the category, surrounded by other category members that diffe red in degree of similarity to the prototype, and, thus, were not equal in te rms of category membership (Rosch, 1973a). Rosch (1975) reasoned that there were grad ed representations of what the category represented signified. Consequently, some items were better representatives of a category. This approach offered an alternative to the classical view, with both its need for clear-cut boundaries for categ ories and criteria of defi ning features for category

PAGE 45

31 membership. Rosch (Rosch, 1975; Rosch & Merv is, 1975) theorized that categories have an internal structure such that they ar e organized by a family resemblance to a prototypical category member and it was the degree and t ype of resemblance to the category prototype that was the foundation of the category structure. More specifically, the notion of categories having an in ternal structure of categories has been tested by asking healthy participants to use a seven-point scale to rate how well a category member fits their image of a given categorys name (Rosch, 1975; Rosch & Mervis, 1975). The results showed that some ite ms were more representative than others of that category and these rati ngs, because they shared features with other items in that category. Thus, they had a family resemblance. These were considered good examples, while others were considered to be poor examples because they did not. These ratings also have been found to be reliable predicto rs of category verificat ion task performance in which the participants are presented with a statement (e .g., a ___ (i.e., an exemplar) is a ___ (i.e., a category)) and decide if it is true or false. Fa ster reaction times and higher ratings are found when an item shares features with the category prototype. For example, the exemplar robin is a good example becau se it is much closer to the prototype, compared to a penguin (a poor example); par ticipants were faster and more accurate verifying that a robin was a bird than that a penguin was a bird (Rips, Shoben, & Smith, 1973; Rosch, 1973a, 1975; Rosch & Mervis, 1975). Th is is called the typicality effect and additional reports of it are disc ussed below. It has been sugge sted that the organization of semantic categories that is based on family resemblance to a prototype is relevant to semantic memory and retrieval (Rips et al., 1973; Rosch, 1973b, 1975).

PAGE 46

32 One of the corollaries of the family resemblance/prototype theory is that the most prototypical members of a categ ory have the most shared attributes with other items in that category and, consequently, represent the central tendency of the category (Rosch, 1975; Rosch & Mervis, 1975). Moreover, as a result of being a pr ototype defining the category, the prototypical members have th e least family resemblance with other categories (Rosch, 1975; Rosch & Mervis, 1975). Furthermore, the more typical items in a category are, the more they share features with the prototype and with each other (e.g., robins, bluejays) while the more atypical ite ms share fewer features with the category prototype and also have more in comm on with items from other categories (e.g., penguins, ostrichs). Therefore, categories have an internal organization somewhat like a target, such that the protot ype is at the center of the bulls-eye, surrounded by typical category exemplars that have many overla pping features with each other and the prototype. Atypical items are more distant fr om the prototype and share fewer features among themselves and with the prototype (Rosch, 1975; Rosch & Mervis, 1975). Some of the conclusions from the fam ily resemblance/prototype view (Rosch, 1973a, 1975; Rosch & Mervis, 1975) are applicable to our study. First, there are clear and pervasive findings suggesting that semantic categories have internal structure that influence the processing of words. This is inconsistent with th e classical view, but compatible with the spreading activation theo ry of semantic memory (Collins & Loftus, 1975; Quillian, 1967). Second, faster reaction time s to pictures of objects than words, would suggest that the underlying nature of th e semantic representation could be closer to pictures (i.e., more visual than verbal ) (Rosch, 1973a, 1975; Rosch & Mervis, 1975).

PAGE 47

33 A wide range of studies have investigated the predictions of the family resemblance view. The typicality effect, the finding of better performance (e.g., faster reaction times and better accuracy) for typical compared to atypical category members has been widely reported in non-brain damaged adults (Hampton, 1995; Kiran & Thompson, 2003a; Posner & Keele, 1968; Rips et al., 1973; Rosch, 1973a, 1975; Rosch & Mervis, 1975). The various tasks tend to involve words, a nd these are discussed first, for example in category exemplar rating tasks (Rosc h, 1975; Uyeda & Mandler, 1980) and priming paradigms involving category verification (F ujihara, Nageishi, Koyama, & Nakajima, 1998; Kiran & Thompson, 2003a; Rosch, 1975; Rosch & Mervis, 1975; Smith, Shoben, & Rips, 1974). Patterns have also been used in a task requiring learning items in a category (Posner & Keele, 1968). A description of the task is provided below, and this is followed by some studies that suggest that th e use of pictures might be different from words. In the priming studies that used the ca tegory verification task with words, the superordinate semantic category name was used as the prime, and the target was a typical or atypical category exem plar from that category. Participan ts were instructed to decide if the target item was a member of that category. The results re vealed faster reaction times and more accurate responses for primes that had been rated as good examples, and these were typical items. Longer reaction times and lower accuracy scores were found for poor examples, and these were mostly atypical exemplars (Fujihara et al., 1998; Kiran & Thompson, 2003a; Rosch, 1975; Rosch & Me rvis, 1975; Smith et al., 1974). The typicality studies using categorizati on tasks offer a compar ison of how typical and atypical items are processed in these tasks (Collins & Loftus, 1975). Typical items

PAGE 48

34 have distinguishing features that reduce a positive decision on the basis of the superordinate connection (Collins & Loftus 1975; Quillian, 1967). Also, superodinate connections have different st rengths which relates to acce ssibility, which is dependent upon use (Collins & Loftus, 1975; Quillian, 1967). Furthermore, there is a high correlation between typicality ratings and accessibility (Collins & Loftus, 1975; Quillian, 1967). Thus, the typicality effect provides mo re evidence that semantic similarity increases the speed of maki ng a positive decision and reduc es the time for making a negative decision. In contrast, increased r eaction times are found for atypical items, because, based on the overlap of features, they represent only some of the category that they belong to and might overlap with othe r categories as well (Collins & Loftus, 1975; Kiran & Thompson, 2003a; Quillian, 1967; Rosch, 1975; Rosch & Mervis, 1975). Further support for the typi cality effect comes from a study using event-related potentials (i.e., ERPs) during a ca tegory verification task in h ealthy adults (Fujihara et al., 1998). Event related potentials measure el ectrical manifestations of particular psychological processes that occu r in preparation for or in response to discrete events (Fabiani et al., 2000; Kutas & Hillyard, 1980). The N400 is a ne gative going deflection occurring at ~400 ms and occurs in res ponse to anomalous information, specifically, semantic violations; for example it is larger when the prime is unrel ated to the target compared to when it is related (Fabiani et al., 2000; Kutas & Hillyard, 1980). Fujihara et al. (1998) found that participants had different responses to typical and atypical words. The N400 were more negative after atypical word s, than they were af ter typical words. A typicality effect was found (i.e., fastest a nd most accurate responses occurred after the typical words). This led to the conclusion th at greater priming occurred during typical

PAGE 49

35 words compared to atypical words. This wa s interpreted to be consistent with the hypothesis that the prototype re presents the concept of the category, which is the central tendency for each of the category members (Fujihara et al., 1998). Pictures have also been used as stim uli in category tasks, and according to Snodgrass and McCullough (1986) pi ctures from a category look similar to each other compared those from different categories, for example, they cite the finding by Rosch (Rosch, 1977, 1978) that based on visual similari ty ratings of items from categories of fruit, animals, and vehicles, the ratings were higher for the ones that looked similar to each other. Snodgrass and McCullough (1986) pr ovided more evidence that the time to decide if a picture belongs to a category could be increased or decreased based on the similarity of the items in the category. Th e investigators suggested that people can categorize a picture and a word via semantic access, but with a picture, people can just use the visual stimulus or the prototype of the category. It was sugge sted that both visual and semantic strategies are used. Snodgrass and McCullough (1986) t hus concluded that picture categorization is not an appropriate task for a comparison of speed of understanding of words and pictures. Alt hough this study did not compare typical and atypical items, it might have referred to th em in terms of similar (i.e., typical) and dissimilar (i.e., typical). There is at least one study that provided evidence for the notion that identification of objects fi rst occurs at a level of abstraction that is not the greatest degree of general or specific entry point but instead varies based on typicality (Jolicoeur, Gluck, & Kosslyn, 1984). This sugges ted that identificati on of typical items occurred at the basic level compared to at ypical items, which ar e subordinate to this level (Jolicoeur et al., 1984).

PAGE 50

36 The term representativeness has also been us ed to refer to typicality such that high representativeness suggests typical items a nd low representativeness indicates atypical items (Grossman, Robinson, Biassou, White-D evine, & D'Esposito, 1998). Grossman et al. (1998) has reported that pictures were used with healthy older adults and they showed sensitivity to the degree of t ypicality (i.e., typical ity effect). Another study using pictures found that older adults showed typicality effects with both category judgment tasks and exemplar judgment tasks (Cobb, 2005). All of these studies examining the typi cality effect have involved healthy, nonbrain damaged adults. The findings have show n that typical items are processed faster and more accurately. Several of these and othe r studies were primarily focused on adults with brain damage, but it was important to reveal the pattern in the normal population prior to addressing the patient groups. Thes e include adults with probable Alzheimer disease and other research with adults with aphasia. Some examples are discussed next. Some studies have found typicality effect s in adults with probable Alzheimer disease (PAD) using category verification tasks (Cobb, 2005; Grossman et al., 1998). Other variations of this task, for exam ple ranking attribut es or ordering the representativeness of categor y exemplars (Cronin-Golomb et al., 1992) and some also included rankings of dominance, which relates to the significance of the feature to the meaning of the item (Nebes et al., 1986; Nebes & Brady, 1990). In a category fluency task adults with PAD have shown normal va riations in the typicality of the items produced (Ober, Dronkers, Koss, Delis, & Friedland, 1986). Grossman et al. (1998) found normal typicality effects on a category ju dgment task, but this was only in the

PAGE 51

37 subgroup of PAD participants who did not ha ve a semantic impairment, but had naming deficits. In contrast, Grossman et al. (1998) reported that the other subgroup of their PAD participants had a semantic deficit, demonstr ated naming deficits and insensitivity to the representativeness. These pa rticipants did not differen tiate among the levels (low, moderate, and high) and this was attributed to compromised ability to make similarity judgments (Grossman, 1981; 1998). Other PAD studies have shown no typicality effects or otherwise differences in pr ocessing atypical and typical item in this population, for example impaired performance on ranking attrib utes (Grober et al., 1985). Sailor et al. (2004) compared the responses of healthy older adults and PAD participants on a category generation task. The analyses revealed that compared to healthy older adults, the PAD group produced significantly fewer atypical items. This was the trend across the categories of fruit, vegetables, animals, a nd footwear. It was concluded that the PAD group had a slower search rate through semantic memory co mpared to the healthy older adults (Sailor et al., 2004). Also, as men tioned in the previous chapter, a property verification task revealed worse performance on category generation low typical items (Smith et al., 1995). Although most of these studies indicated that the PAD population has deficits on atypical items, to the best of our knowledge, th ere is at least one study that showed better performance on atypical items (Cobb, 2005). The purpose of this study by Cobb (2005) was to attempt to understand if typical items were affected by the pathology of Alzheimer disease. Using the bottom-up deterioration hypothesis in PAD, Cobb (2005) theorized that the vulnerability of typical items compar ed to atypical items could be described as:

PAGE 52

38 Generalized (equal); Strength (not as vulnerable); Distin ctiveness (more vulnerable). Category judgment tasks and exemplar judgment tasks were used in which pictures were shown and the participants verified them. The results indicated that like the healthy older adults, the PAD group showed a typicality eff ect on the category judgments (i.e., faster and more accurate responses were found on the typical items). In contrast, on the other task of verifying the exemplars, the PAD group showed deficits on the typical items compared to healthy older adults. Cobb ( 2005) suggested that th ese findings provide some evidence that the typical items were mo re vulnerable to the de terioration, and thus provided support for the distinctiveness m odel. However, Cobb (2005) also cautioned about generalizing the results because of this preliminary study due to some issues related to methodology. Research on the typicality effect has also been conducted in th e aphasia population. In a category generation task, Grossman (1981) compared the number and typicality of the exemplars generated (i.e., typical and at ypical) between a group of participants with nonfluent aphasia and another group with fluent aphasia. The majority of the items produced by the nonfluent group were typical it ems. Grossman (1981) suggested that this occurred because these particip ants used the central tendenc y of the category as a guide and compared the target item to it. In cont rast, the fluent aphasi a group produced a more even distribution of typical and atypical exem plars, starting out with more typical items (like the nonfluent group), but then producing items that were as close to the central part of the category and even from other categor ies. Moreover, participants with fluent aphasia actually produced many responses th at were not members of the specified category. To account for these findings, Grossman (1981) suggested that these

PAGE 53

39 participants were aware of the central tendency of the category but had reduced mental representations related to the categorys border. Thus, they violated category boundaries(Grossman et al., 1998). Grober et al. (1980) used a category verification task in two modalities (i.e., pictures and words) to determine if performance on these tasks could distinguish groups of pa tients with anterior (i.e., nonf luent) aphasia and posterior (i.e., fluent) aphasia. There were no differences between performance with pictures and words for any of the patient groups. Furthermor e, across groups, the typicality effect was realized as both faster response times and better accuracy for the typical exemplars. However, while the nonfluent patients were fa irly accurate in thei r classifications of atypical category exemplars, this was not the case for the fluent patients. When the item clearly belonged to a given category, the flue nt group was accurate, but when the item was at the category boundary, they were less accurate. Grober et al. (1980) suggested this difficulty with atypical items was due to a disruption of the underlying semantic structure of semantic categories in fl uent aphasia. Although this task was different from the category generation task used by Grossman (1981), the findings for the patients with fluent aphasia are compatible. In a category verification study, Kira n and Thompson (2003a) compared the performance of two groups of adults with fl uent and nonfluent aphasia to young and older adults without brain damage. Typicality eff ects were expected to occur in all of the groups. However, based on previous research (Grober et al., 1985; Grossman, 1981), the pattern of activation for the fl uent aphasia group was expected to be different. Kiran and Thompson (2003a) used words from the cate gories of vegetables, birds, and fish typicality ratings that they developed ba sed on the seven-point scale used by Rosch

PAGE 54

40 (1975). Participants first read a category name and decided if the next word was a member of that category or not. Furthermore, the overall results were consistent with the predictions: typicality effects were found in all the groups, and the fluent group displayed a different pattern of performance from th e other groups. The fluent group had the smallest effect of typicality and the gr eatest number of errors. Although this group was more accurate on the typical items compared to the atypical items, their reaction times for accepting the correct typical items were not signi ficantly faster than the correct atypical items. Kiran and Thompson (2003a) interpreted th is as two different manifestations of a semantic impairment from the fluent aphasi a. The first was that weakened boundaries of the target category le d to increased numbers of errors. The second difficulty was that there was a decrease in the ab ility to access the prototype and this caused a deficit when comparing the target with the prototype of th e category. They concluded that the findings of longer reaction times and more errors with atypical exemplars suggested impoverished associations linking atypical exemplars with their categorys pr ototype. Kiran and Thompson (2003a) theorized that these findings could offer more information about the nature of atypical category exemplars as ha ving differential representation in semantic memory. Furthermore, the greater number of dissimilarities among atypical items in a category allows them to convey a broader range of variation across the category, and thus they were considered to be mor e complex (Kiran & Thompson, 2003a). Thus, in the aphasia literature, there is at least one study sugg esting that fluent aphasia affects the processing of atypical category exempl ars in a category generation task (Grossman, 1981). This is understandabl e given that category generation requires searching semantic memory fo r an item within a category when only provided with the

PAGE 55

41 superordinate category label. Some categor y verification task studies have shown typicality effects in adults with nonflu ent aphasia (Grober et al., 1980; Kiran & Thompson, 2003a; Stanczak et al., 2005), includi ng transcortical motor aphasia (Stanczak et al., 2006), and fluent aphasia (Stanczak et al., 2005), including conduction aphasia (Stanczak et al., 2006). In others studies, the typicality effect was found but was somewhat disrupted in adults with fluent aphasia (Grober et al., 1980; Grossman, 1981; Kiran & Thompson, 2003a), as shown by more di fficulty (e.g., slower or less accurate) on atypical items. Although the typicality effect indicated better perf ormance with typical items, at least one study did not find that typical items were ve rified faster than atypical items that were correct (i.e., member of the given category) by the adults with fluent aphasia (Kiran & Thompson, 2003a). In summary, the family resemblance/prototype view (Rosch, 1975; Rosch & Mervis, 1975) has had a strong influence on natural language research in category representation (Barr & Capla n, 1987). Based on the family resemblance/prototype hypothesis, there are several reasons for tr aining typical category exemplars and they could have an advantage over training atypi cal category exemplars. Typical items have richer representations, and people have greater knowledge of them so they are used more, and therefore accessed faster (Heilman, 2005, personal communication). Evidence from adults with brain damage suggests that know ledge about atypical items is disrupted following strokes to the posterior portion of the left hemisphere (Grober et al., 1985; Grossman, 1981; Kiran & Thomps on, 2003a), and also as a resu lt of PAD (Sailor et al., 2004; Smith et al., 1995).

PAGE 56

42 The Complexity Account of Treatment Efficacy As described above, atypical items have been hypothesized to be more complex (Kiran & Thompson, 2003a). The Complexity Account of Treatment Efficacy (CATE) was developed by Thompson et al. (2003) base d on their study in which adults with agrammatic aphasia were trained to use sent ences that were syntactically complex. As a result, participants showed generalization to less comple x sentences whose structures shared processes with trained items. Several studies by Thompson and colleagues (Thompson et al., 1998; Thompson et al., 1997 ; Thompson et al., 2003 ) have reported that adults with aphasia have shown generalization following treatments in which complex syntactic structures were trained. This was because when more complex items were trained, this included variables that ar e relevant to items that are simpler. This promoted greater degrees of access to untrained items compared to the training of simple items (Thompson et al., 1998; Thompson et al., 1997; Thompson et al., 2003). Thompson et al. (2003) concluded that th ese findings were compatible with other studies in the literature and suggested that enhancement of performance is found under the following two conditions: when there is related linguistic structure a nd when treatment moves from greater degrees of complexity to lesser degr ees of complexity. In addition to acquired language disorders, complexity has been used by Gierut (2001) as a guide for stimuli to promote generalization with children who ha ve phonological disorders. The findings also have been positive and indicate that with grea ter degrees of target complexity, there are greater degrees of gains in phonological skills; these beha viors have been shown to generalize to untreated sounds (Gierut, 2001). These principles are borrowed from motor skill learning in adults and conditions under wh ich the behavior is practiced (Schmidt & Lee, 1999). Generalization to untrained items in a picture naming study has also been

PAGE 57

43 found to be successful when the focus has been on atypical items as a means for stimuli selection because they have a hierarchical complexity (Kiran & Thompson, 2003b). This and other behavioral studies are discussed below. However, it is important to first understand the initial work in this area by Pl aut (1996) in which he compared the training of atypical and typical items in a conn ectionist model of acquired dyslexia. Connectionist Model and Computer Simulations Prior to describing the computer simula tions by Plaut (1996), a brief overview of the connectionist or parallel distributed pro cessing (PDP) model is provided. The PDP is a computational framework that allows for exploration of cognitive processes of normal and damaged systems. Simulations are conducte d to learn more about the processes and to contribute to the development of poten tial rehabilitation programs (Plaut, 1996). Parallel processing is the term used to desc ribe how individual processing units (similar to neurons), in a large array, are connecte d, but each performs a basic function and represents a component of the modeled en tity (Nadeau, 2000; Plaut, 1996; Rumelhart, McClelland, & PDP Research Group, 1986). The units are linked by connections and the strength of these connections reflects the knowledge of the network (Nadeau, 2000; Nadeau & Gonzalez Rothi, 2004). Large pattern s of activation, based on excitatory and inhibitory signals, represent a concept which is referred to as a distributed representation (Nadeau, 2000; Plaut, 1996; Rumelhart et al., 1986). Plaut (1996) tested the family resemblan ce theory using a conn ectionist computer model in which he simulated acquired dyslexia in three experiments. Only the first two are discussed here. First, Plaut (1996) trained the model, then lesion ed it to simulate the effects of stroke. Finally, he retrained the model to simulate therapy after stroke in order to better understand the degree and speed of recovery; how generalization occurs from

PAGE 58

44 treated to untreated items, and the way th is generalization might be maximized via selection of the treatment items. In Plauts (1996) first experiment, he programmed the network and trained it to generate semantic information of about 40 wo rds from five semantic categories (natural kinds and man-made) based on orthographic input. The training was done to allow mapping from the written words to their corresponding meanings. Then, Plaut (1996) damaged the network by random selection and re moval of some of the units representing semantics and others representing orthogr aphy. Plaut (1996) randomly identified items for retraining: half of the words that were tr ained correctly and half that were not trained correctly. The remaining words were not retr ained. Plaut (1996) reported that, compared to initial learning, retraini ng was faster because there was a re-establishment of consistency that was relevant among words and thus untreated words improved. Plaut (1996) suggested that location of the damage was important. When the semantic level was damaged, there was fast relearning and generalization was substantial. However, when the orthography level sustained damage relearning was slower and generalization did not occur Thus, there was better generalization at the semantic level because the words had relationships based on their meani ngs. This was based on the structure of the semantic organization of the set of words. Pl aut (1996) hypothesized that the part of the system that was damaged could influence re covery. Furthermore, Plaut (1996) suggested that generalization had improved because the semantic categories from which the words were derived had a substantial amount of overl ap in their semantic representation. This experiment provided support for using items fr om semantic categories in therapy because the consistency inherent in the items in the category allowed for efficient processing and

PAGE 59

45 generalization. These findings also offer s upport for the family resemblance/prototype hypothesis (Rosch, 1975; Rosch & Mervis, 197 5) for achieving a treatment effect. However, in terms of maximizing generaliza tion, Plaut (1996) provided evidence for the benefit of training atypical items, and this was discovered in Experiment 2. Plaut (1996) theorized that generalization could be obtained if the nature of the semantic representation of the items was a good estimate of semantic structure relevant to the whole set of words in the training stimuli. Selection of words for training were based on the words semantic representations (Plaut, 1996). Acknowledging that nouns are categorically organized, Plaut (1996) also us ed typicality, a semantic variable that is essential because it reflects the proximity to the central tendency of a category (Rosch, 1975). Plaut (1996) hypothesized that the exte nt of generalization from retraining is affected by the relative typicality of the trea ted words (i.e., typical versus atypical). The training items that Plaut (1996) chos e were 100 artificial words from one semantic category. These words were evenly divided into two sets in which the meaning was generated by varying the de gree of featural overlap be tween a prototype item and other items in the category. Thus, the semantic features for the typical words were closer to the prototype of the category; whereas the features for the atypical words were further from the prototype. Training the network invol ved iterations that allowed for generation of the correct semantic representation for each word when the corresponding orthographic form was presented. Then, to lesion the network, Plaut (1996) randomly selected and deleted connections. Next he retrained 25 of the typical words and 25 atypical words in different networks. This allowed for evaluation of the impact of typicality on generalization to untrained typical items and untrained atypical items. The

PAGE 60

46 results were consistent with Plauts (1996) predictions. Th e lesion rendered typical words more impaired compared to atypical words. Thus, Plaut (1996) s uggested that typical words were more vulnerable to damage becau se they had more competitors that were close, due to the overlap. Thus, distinguish ing typical words from other typical words (compared to distinguishing atypical words) required better accuracy by the network. In contrast, there were not as many competitors for atypical words, and distinguishing among atypical words was not as difficult as it was with typical words. However, as a result of retraining, trained typical words had significantly better performance compared to trained atypical words (Plaut, 1996). Although these findings are rele vant to recovery, Plaut ( 1996) argued that the more critical finding is the fate of the untreated word s, because this rela ted to choosing stimuli that could influence generalization, an im portant component of rehabilitation that clinicians could control. When the training was with either typical or atypical words, there was overall substantial improvement in the untreated typical words. However, atypical words only improved when the network was trained with atypical words. When atypical words were retrained there was more generalization to bot h untrained atypical and typical words compared to when typical words were trained. In contrast, when the network was trained with typical words, th is had a negative impact on performance of untreated atypical words. To account for th ese results, Plaut ( 1996) hypothesized that typical words just focus on the central tendency of its category, w ithout indicating the semantic features that can differ from the prototype Thus, the typical words offer a good estimate about the degree of variation in the semantic features that is shared by other typical words, but not the in formation about the variation for the atypical words. In

PAGE 61

47 contrast, atypical words indicate the amount of variation that can exist across members of a category while also approximating the categorys central tendency. Thus, Plaut (1996) hypothesized that atypical words collectively es timate the semantic representations that span more of the features required by the full set of words compared to representations of typical words. Consequently, training atypical words should result in generalization to typical words, because the semantic repres entations of the atypical words included information about the degree of variability within the category st ructure as well as information about the central tendency of th e category (Plaut, 1996). While Plaut (1996) recommended using this approach, he cauti oned that the atypical items should not be extremely atypical. The semantic dimensions should be relevant and thus allow for the average effects to be near the category s central tendency. Pl aut (1996) recommended that future research endeavors should i nvolve training atypical items from several categories at the same time. In summary, the PDP model has been desc ribed as operating in a manner consistent with the brain (Nadeau, 2000; Plaut, 1996; Ru melhart et al., 1986). Overall findings from Plaut (1996) provided a rationa le for training items within a semantic category. Also, while training typical items resulted in better performance (i.e., acquisition or a training effect), training atypical yielde d better generalization to untra ined items. The results from Plaut (1996) are very positive and have great implications for clinical application. In particular, both category st ructure and atypical items ar e important in promoting generalization because they relate to the sema ntic representations of the items. However, it is important to remember that computer simulations involve isol ated information and events, in which the experimenter programs all of the inputs and thus has complete

PAGE 62

48 control over the information in the model. For example, the second experiment used artificial words from one category and the ty pical and atypical items were created by the network. This differs from the wide range of variation in humans in terms of semantic memory, learning, and the effects of brain da mage and behavioral manifestations of different disorders. Neverthe less, as discussed below, st udies with humans have been conducted and some of the findings are consistent with Plaut (1996). Similar Behavioral Studies A series of experiments on category l earning in non-brain damaged college students was conducted by Posner and Keele (19 68). Participants looked at patterns with varying degrees of distortions These distortions ranged from high (i.e., the distance from the central tendency of the category was great er) to low (i.e., closer to the central tendency). The participants pressed a button to specify their selection for the given pattern, and received feedback about accuracy. This was repe ated over several trials to allow for learning. Results showed that the participants had better generalization from learned patterns to new ones af ter training with highly variab le patterns compared to when the training emphasized low variabilit y. To account for this, Posner and Keele (1968) suggested that va riability is an important part of the learning pr ocess. Although this study did not involve naming or a patient population, and did not use the term typical and atypical, the patterns with higher variability might be considered atypical, whereas those with much lower variability might be considered typical. Furthermore, the theoretical underpinnings are along the same lines, and can be considered support for training atypical items (Plaut, 1996). More recently, three different groups of investigators (Kiran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005, 2006) examined and compared the effects of

PAGE 63

49 training typical and atypical category exempl ars, and thus compared the two opposing views of typicality discussed above (Pla ut, 1996; Rosch, 1975; Rosch & Mervis, 1975). Each of these therapy studies involved adu lts with aphasia in th e context of a singlesubject design targeting confr ontation picture naming, and show ed consistent evidence of treatment effects. However, th e results were mixed, especially in terms of generalization. A brief description of these three studies is provided below. This is followed by a brief critique pertaining to both th eir strengths and limitations, highlighting how they offer unique contributions to the literature a nd have implications for the current study. Kiran and Thompson (2003b) predicted that when atypical items were trained, this process would result in emphasizing both how much the features in the category vary as well as the prototypes features. This, in turn, would result in a strengthening of the associations among the typical items, increased access to these typical items, and finally generalization to untrained typical items. In contrast, they argued that training typical category exemplars would have more limited e ffects, because typical items have limited variation, focusing only on the cen tral tendency of the categ ory (Plaut, 1996). Kiran and Thompson (2003b) argued that, compared to t ypical items, the atypi cal items were more complex because their semantic features have a greater degree of di ssimilarity with each other, and thus they offer more diversifie d information about the whole category, as a collective unit (Plaut, 1996; Thompson et al., 1998; Thompson et al., 2003). To test these predictions, Kiran and Th ompson (2003b) provided therapy for four adults with fluent aphasia on a naming tas k. Baseline naming performance was assessed using 48 items consisting of 24 exemplars from each of two semantic categories (birds and vegetables). The study design allowed for each participant to be separately trained on

PAGE 64

50 one category at a time, starting with either eight atypical category exemplars or eight typical items from each category. After the par ticipant met the criteria the next category was trained with the opposite typicality. The participants atte nded two hour therapy sessions twice a week until either criteri a of 7/8 correct was achieved over two consecutive sessions or after a maximum of 20 training sessions had been completed (Kiran & Thompson, 2003b). During training, participants performed a variety of semantic tasks tapping both comprehension and production skills: picture naming, category sorting, feature sele ction, answering yes/no questi ons; and picture naming again. Participant 1 was first trained on typical birds for seven weeks and this led to a treatment effect (i.e., acquisi tion of the trained items) but there was no generalization to untrained intermediate or atypical items. Subs equent training of atypi cal birds resulted in improved performance on those items. At that time, there were no changes on performance of the control category (veget ables). Then Participant 1 was trained on atypical vegetables, and in 8 weeks, ac quisition criterion wa s observed as was generalization to untrained inte rmediate and typical birds. Pa rticipant 2 was first trained with atypical birds and in 11 weeks, me t criteria and showed generalized naming on intermediate and typical items. At that time, performance on the vegetable category remained the same. Treatment of birds led to no change in performa nce on the untreated category, vegetables, so the participant was subsequently trained on atypical vegetables. This resulted in a treatment effect and ge neralization to untrained intermediate and typical items. Participant 3 was first traine d with typical vegetabl es and although criteria was met, there was no generalization to in termediate or atypical vegetables. Next, Participant 3 was trained on at ypical vegetables that result ed in improved performance on

PAGE 65

51 the trained atypical vegetables; no further treat ment was provided with the bird category because of the extended time for training on the first (28 weeks). Participant 4 was trained only on atypical items and showed a patte rn similar to Participant 2: a treatment effect after 6 weeks of training on vegeta bles and after nine weeks on birds, and generalization to both untrained in termediate and atypical items. The patients in the Kiran and Thomps on (2003b) study acquired atypical items faster acquisition than typical items. Furt hermore, naming of unt rained typical and intermediate items improved following tr aining of atypical items, thus showing generalization. However, when the typical items were trained, there was no generalization to exemplars that were atypical or of interm ediate typicality. Kiran and Thompson (2003b) suggested that the training for the atypical items emphasized how the semantic features in the category varied; wh ereas the training for the typical items was limited to a small number of features that were shared among the other typical items. Moreover, maintenance of the treatment eff ects was found 6-10 weeks after the training. Kiran and Thompson (2003b) conducted an e rror analysis to gain some insights into the effects of the treatment. Before the intervention, the errors were primarily general, as characterized by superordinate labels, no responses, and neologisms. This was attributed to the participants failing to retrieve the specifi ed semantic and/or phonological details of the target name. However, after the treatment, the participants showed more accurate naming on trained items and on untrained items, reflecting better semantic and phonological access to the representa tion. The shift in the type of errors, from general at baseline to semantic and phonemic after the trai ning was attributed to a greater impact of excitation at the level of semantics and phonology that had occurred when the

PAGE 66

52 participants attempted to name the items. Ho wever, this interfered with the ability to accurately select the target name from ot hers that were also activated. Kiran and Thompson (2003b) suggested that although the treatment had a positive impact that was illustrated in enhanced spread of activation to targets within a category that were related, the intervention was not completely successful in eliminating the interference caused by many category exemplars being activated at the semantic/phonemic level during naming. Kiran and Thompson (2003b) suggested that there was improved performance because the treatment was based on semantics and highlighted the underlying elements of semantic representations and processing. Th e investigators suggested that this was compatible with the literatur e using the semantic feature approach (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000; Drew & Thompson, 1999). Kiran and Thompson (2003b) concluded that their findings provided more support for improving generalization to untrained items based on using more comp lex training items, and thus providing more evidence for the Complexity Approach to Treatment Efficacy (Thompson et al., 1998; Thompson et al., 2003) as well as for trai ning atypical items to achieve better generalization (Plaut, 1996). In this study, Kiran and Thompson (2003b) were the first to show that the computer simulations of acquired dyslexi a by Plaut (1996) could be suc cessfully applied to adults with fluent aphasia. The findings illustra ted the value of training atypical category exemplars in the context of a training high lighting semantic category structure. This study was well designed, including the developm ent of typicality ra tings and semantic attributes, intentionally selec ting low frequency items to avoid effects of word frequency, the use of both comprehension and producti on tasks to activate the entire word

PAGE 67

53 representation, and a training regimen that focu sed on just one category at a time. Two of the training tasks, feature selection an d yes/no questions provided comprehensive information about the items, including the wri tten form. However, specific information was provided by the clinician, and the participant had to select it. It is possible that the training might have been even more robust if the participants were required to generate the information. More importantly, there appear s to be a lack of generalization probes for untrained atypical items and untrained typica l items that have the same and the opposite typicality of the items that were trained (i .e., trained atypical and trained typical items). Our study included a measurement of untrained atypical and typical items for the same and opposite typicality of the trained items (i .e., that were traine d with atypical and typical items; e.g., Training on typical items and untrained typical items), and a more active role for the participants on the tr aining tasks. These are discussed later. Mayer et al. (2004) attempted to replic ate the findings of Kiran and Thompson (2003b) in three participants with severe a phasia and to determine if other semantic categories might yield potent t ypicality effects. Each partic ipant had a different type of aphasia. Participant 1 had global aphasia. Part icipant 2 was considered borderline fluent, (i.e., fairly intact auditory comprehension a nd reading ability, with fluent speech and neologisms including some word approximati ons). Participant 3 had mixed nonfluent aphasia (i.e., intact auditory compre hension for basic information, and although nonverbal, he spontaneously gestured and us ed facial expressions and drawings). A total of six categories were used by Mayer et al. (2004): animals, clothing, furniture, sports, tools, and vehicles. Each participant was trained separately on 10 atypical items from one category, and then on 10 typical items from another category.

PAGE 68

54 Each category was treated for nine weeks, fo r a total of 18 weeks. Two other categories were used as treatment probes (before and af ter treatment) and generalization probes. To maximize support in a structured way us ing phonological and semantic stimulation, training utilized a response-c ontingent hierarchy (Bandur & Shewan, 2001). The clinician presented a picture and guided the partic ipant through a seman tic and phonological cueing hierarchy (starting with the least effec tive and moving to the most effective). This was used until the participant named the item. Then, the participant was asked to repeat the correct name of the item five times (Pring, Hamilton, Harwood, & MacBride, 1993), before completing a confrontation naming tas k. The results indicated that there was a treatment effect for each partic ipant, but this did not genera lize to any untrained items that were probed. Mayer et al (2004) reported that one pa rticipant initi ally showed progress and typicality effects that were described as subtle: generalized naming to untrained typical sports item s after training on atypical spor ts items. However, because there were unanticipated improvements in not just in trained categories but also in untrained categories, Mayer et al. (2004) indicated that th is could not be considered generalization. Mayer et al. (2004) identified three factor s to explain the lack of generalization. First, they considered it possible that the se verity of the aphasia might have interfered with the cognitive processes needed for re habilitation. Second, Mayer et al. (2004) suggested that the training ta sks were structured to compensate for the severity of the deficits, and, thus, perhaps did not have the metalinguistic requirements of the tasks used by Kiran and Thompson (2003b). Third, Mayer et al. (2004) discussed Plauts (1996) assumption pertaining to the retraining a nd the mapping; this might not have been

PAGE 69

55 applicable to their participan ts. According to Plaut (1996) generalization might require a good estimate of the relevant semantic inform ation of the training set overall and this needs to be in alignment with the consistenc y of the mapping that the damaged part of the network is responsible for carrying out. Mayer et al. (2004) reasoned th at the severity of the brain damage in their participants negati vely affected the integrity of their lexicalsemantic network, rendering it incapable of retr ieving the semantic features in order to demonstrate generalization. This study by Mayer et al. (2004) was carefully designed and included various nuances that are important in therapy, including development of typicality norms, selecting six categories that were based on participants in terests when possible. In addition they included repetition of the target as a training task, to minimize phonological access deficits. Stanczak et al. (2005) also sought to understand the relationship among performance on a category verification task (i.e ., a measure of online typicality), training typical and atypical category exemplars fo r naming production, and generalization. Two participants with anomic apha sia, one fluent and the other nonfluent participated in the study. First, prior to the base line picture naming probes, th e participants completed a category verification task (with stimuli from the Kiran and Thompson (2003b) study). The purpose of this was to determine if they would show typicality effects (i.e., faster reaction times and more accurate responses on typical versus atypical items) in response to the written category primes and their written exemplar targets. Then the participants completed baseline probes. This was followed by a naming therapy, in which each participant was simultaneously trained on two categories, birds and vegetables, one with

PAGE 70

56 typical category exemplars and the other w ith atypical. The training tasks were naming the target picture, verification of semantic attributes relate d to it, and naming the picture again. The results of the category verification ta sk that was completed at the onset of the study indicated that the particip ants were faster to verify typical items compared to atypical items (i.e., this was described as online typicality e ffects). The naming treatment had differential effects. The particip ant with fluent aphasia only demonstrated acquisition of trained typical bi rds, but not atypical vegetables In contrast, the participant with nonfluent aphasia demonstrated treatmen t effects for the atypical birds only (not typical vegetables). There was no generalization to untrained items in either participant. The finding of a typicality effect on the cat egory verification task was interpreted by Stanczak et al. (2005) as activation occurri ng in a bottom-up fashion (i.e., the semantic retrieval was automatic). The differences in these two participants responses to the typical and atypical training was attributed to their lesion site and fluency (anterior and nonfluent versus posterior and fluent) inter acting with the semantic processes involved with typical versus atypical items. Specifi cally, it was hypothesized that the controlled semantic processing in the fluent participant was relatively intact, allowing for better performance with the typical words, despite th eir similarity with other items and higher degrees of semantic interference. However, learning the atypical items was considered to be more difficult. In contrast, the nonfluent participant, Stanczak et al. (2005), theorized that the demands for atypical items were less than the typical items (i.e., fewer competitors), and thus yielde d better performance. This study by Stanczak et al. (2005) provided an intriguing interpre tation of the results by linking them to neural substrates.

PAGE 71

57 This is feasible. However, one limitation wa s that the investigat ors did not elaborate about the difficulty related to atypical words. Very recently, Stanczak, Waters, and Caplan (2006) further extended this research by applying the same protocol (S tanczak et al., 2005) to two different participants with anomia from different etiologies. The pur pose of this study was to determine if generalization would be affected by the type of deficit and the type of treatment provided. Participant 1 had conduction a phasia, and her deficits we re primarily phonological; her training was with atypical vegetables and t ypical birds. These were trained over a 20 week period. Participant 2 had transcortical motor a phasia and had phonological and semantic impairments; his training was on at ypical birds and typi cal vegetables. The results indicated that compared to atypical items, typical items were learned faster by both participants. Although both participants showed significant tr eatment effects and better learning of birds compared to vegetabl es, the patterns of responsiveness to the training varied for each participant. Pa rticipant 1, who had a primary phonological impairment, displayed greater learning when trained with typical items compared to atypical items. In contrast, Participant 2, who exhibited both semantic and phonological deficits, did not initially show a difference between learning typical versus atypical items, but did show better learning of atypical items as the treatment continued. Participant 2, who had been trained on atypical birds and t ypical vegetables, also had statistically significant generalization to unt rained typical birds, and onl y marginal generalization to atypical vegetables. To account for these different findin gs, Stanczak et al. (2006) offered interpretations which again were based on ne ural substrates. It was not the phonological

PAGE 72

58 impairment in Participant 1 that resulted in the lack of learning atypical items, but the greater degree of semantic diffi culty inherent in atypical item s. However, for participants with damage to the left prefrontal area, as in Participant 2 (with both semantic and phonological impairments), the typical items initi ally should be learned better because of the high degree of similarity in the features but, then over time, this overlap leads to confusion. Stanczak et al. (2006) hypothesized that learning atypical items, which have greater variation, should result in better performance because the degree of competition is less. Despite the better performance on birds, the investigators refuted the possibility that this was a category specific l earning (i.e., that it was easier to learn birds compared to vegetables), and provided a rationale for the effects of typicality on learning (i.e. both participants were responsive to being trained with typical items but only one responded to the training with atypical items). Stanczak et al. (2006) concluded that while their findings were broadly consis tent with Plaut (1996) and Thompson et al. (2003), the differences from findings by Kiran and Thom pson (2003b) might be attributable to differences in the training (i.e., training two categories at that same time) or the participants (degree of semantic impairment). These four studies (Kiran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005, 2006) were theoretically motivated and well planned, includi ng programming for generalization. The training task s included at least some info rmation about the target item in terms of semantics and the phonological fo rm, as well as confrontation naming. Some potential limitations of these st udies, with the exception of Mayer et al. (2004) included the use of only two animate categories wi thout a third control category (although the second category was used as a control) and a lack of information about the clinicians

PAGE 73

59 response to a failed naming attempt by th e participant (Kiran & Thompson, 2003b; Stanczak et al., 2005, 2006). The clinicians re sponse to the participants naming failure is important because it could affect the outco me. This is because it provides information about the response so that the patient can be tter understand the info rmation and can learn to self-evaluate (B andur & Shewan, 2001). Also, in some of the studies it was not clear if generalization probes to untrained items that had the same, different, or both t ypes of typicality as the items that were trained (e.g., train atypical items, probe untrain ed typical items in th e trained category). It appears that the generalization probes were not measured on both in some of the studies. Other broader methodological c oncerns about these four st udies include the lack of opportunity to or ability to generate semantic information about the item in a more active manner (i.e., beyond yes/no questions). Although yes/no tasks, sorting and other similar tasks are used in the literature and have value, they require th e participant to play a less active role in the therapy. Mayer et al. (2004) and Stanczak et al. (2006) identified concerns related to this in terms of the severity of their patient population and how it might have affected generalization. Another possi bility that Mayer et al. (2004) identified was that the semantic category structure in their patients wa s impaired such that they were unable to benefit from the training which was built around the semantic representations of the category (Plaut, 1996) Another important consideration about achieving a generalization effect has been ra ised by Murray and Clark (2005): it is not known if it is the combination of training t ypical and atypical categ ory exemplars in the context of semantic training requiring ta sk requiring processing of semantic representations (Kiran & Thompson, 2003b), or simply the use of atypical and typical

PAGE 74

60 category exemplars. We attempted to address some of these issues in the methodology of the current study. The Current Study Rationale for Applying the Training to the PAD Population Brain damage can cause a language impairment, and object naming is a sensitive marker for this (Benson, 1979). There is so me evidence that the semantic system is preserved in adults with nonfluent and flue nt aphasia based on similar performance on a lexical decision task compared to non-brain damaged adults (Gerratt & Jones, 1987). However, several of the studies discussed a bove examining adults w ith aphasia suggested there was a disruption or disorg anization of the semantic syst em in adults with fluent aphasia (Grober et al., 1 985; Grossman, 1981; Kiran & Thompson, 2003a). There are other reports of a semantic level impairment in adults with aphasia based on impaired performance on property verification and cat egory verification tasks (Koemeda-Lutz, Cohen, & Meier, 1987); and le xical-semantic discrimination tasks (Chieffi, Carlomagno, Silveri, & Gainotti, 1989; Gai notti, 1981). More specificall y, in transcortical sensory aphasia and anomic aphasia, disturbed semantic processing has been identified as the underlying mechanism causing the language de ficits (Raymer et al., 1997; Raymer & Gonzalez Rothi, 2001). It is possible that th ese deficits involving the semantic category structure might prevent generalization to unt rained items and perhaps account for such findings in the literature (Mayer et al., 2004; Stanczak et al., 2005, 2006). The current study proposed that individuals with early PAD provi de an alternative population to study the effects of typical and atypical exemplar training from semantic categories. The rationale for th is is based on the disease pro cess, which results in diffuse damage (as opposed to focal damage from a st roke) and evidence that to some degree, the

PAGE 75

61 semantic systems structure is preserve d in PAD (Albert & Milberg, 1989; Benson & Geschwind, 1985; Hartman, 1991; Nebes, 1992; Ob er et al., 1995; Schw artz et al., 1996). This assertion is supported by findings that show knowledge about semantic categories is preserved in PAD (Flicker et al., 1987; Huff et al., 1986; Ma rtin & Fedio, 1983; Nebes et al., 1986; Salmon, Butters et al., 1999; Sc hwartz et al., 1979 ; Warrington, 1975). The goal of our exploratory study was to de termine if choosing remediation targets for anomia based on typicality (i.e., atypical a nd typical) would result in generalization to untrained items. This involved a comparison of the family resemblance/prototype (Rosch, 1975) favoring typical category exemplars a nd the Complexity Account of Treatment Efficacy (Thompson et al., 2003), as well as the connectionist model of a computer simulation of acquired dyslexia (Plaut 1996) favoring training atypical category exemplars for achieving generalizati on. We provided both lexical phonological information about the item via repetition as well as a semantic feature training with atypical and typical items for adults with P AD. In designing our st udy, relevant findings were culled from the PAD literature, partic ularly the three naming treatment studies (Abrahams & Camp, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005), the five aphasia studies compar ing typical and atypical ca tegory exemplars (Kiran & Thompson, 2003b; Mayer et al., 2004; Plaut, 1996; Stanczak et al., 2005, 2006), and other relevant findings from th e aphasia treatment literature. According to Nadeau and Gonzalez Rothi ( 2004), the connectionist approach views anomia (that is caused by a semantic defici t) as a reflection of insufficiently engaged representations of features th at are critical for making distinctions among concepts. When a network is damaged, a large amount of inform ation still remains in the network, so the

PAGE 76

62 focus should be on refining the damaged networ k via semantic therapy. In particular, the network needs to be changed in terms of its connectivity so that th ere is more reliable engagement of distinguishing features si multaneously while there is relatively a disengagement of the shared features (N adeau & Gonzalez Rothi, 2004). These findings suggest that participants with early PAD might theref ore be appropriate for the semantically based training provided in our study which compared typical and atypical category exemplars. As noted previously from the aphasia lite rature, a facilitation effect refers to improved performance on trained items from pretest to posttest in a single session (Howard, 1985; Patterson et al., 1983). Our st udy used a single session for different reasons, including the lack of picture naming treatment studies in the PAD literature, and to determine if the lexical-semantic system affected by PAD could be stimulated in a short period. This information would be us eful to know prior to development of an intervention for naming deficits in adults with PAD. For the purpose of this study, the term r esponse generalization is adapted from Thompson (1989) to indicate that a particip ant has improved naming of untrained items based on training of other items. Thom pson (1989) suggested that for response generalization to occur, the training and the probes should allow for sampling of responses that have a similarity to the traine d items, either within a defined set or across it. The methodology for our study allowed for a measure of generalization to untrained atypical and untrained typical items regardle ss of the type of training that category received. (e.g., in the Train-Typical vehicl es condition, there were untrained typical vehicles and untrained atypical vehicles). The aphasia literature suggests criteria to

PAGE 77

63 measure generalization; some examples include an increase in performance from baseline by 40% (Kiran & Thompson, 2003b) or 50-80% (Thompson, 1989). A nother approach has been an increase of baseline performan ce as measured by three or more probe items (Boyle, 2004). In our exploratory study, we chose to measure generalization based on an increase from pretest that was statistically signifi cant at the alpha .05 level. The typicality ratings from Rosch (1975) provided categories and their exemplars for this study. Using 24 items per semantic category allowed for an adequate number of training and generalization items (Kiran & Thompson, 2003b; Rothi et al., 2005). Two categories were used for the training (seven atypical and seven typical items) and the other was a control category (M ayer et al., 2004; Rothi et al ., 2005). While three of the four aphasia treatment studies comparing ty pical and atypical categ ory exemplar training (Kiran & Thompson, 2003b; Stanczak et al., 20 05, 2006) used the categories of birds and vegetables, two of them reported generalizati on just one used a combination of natural kinds and man-made artifacts (Mayer et al., 2004). Our study involved three artif act categories: vehicles, cl othing and tools. Although at least one PAD study has suggested that there was no significant difference in performance on natural kinds and artifacts (Tippett, Grossman, & Farah, 1996), others have found there is evidence that adults with PAD have shown lower levels of performance on natural kinds (Chertkow, B ub, & Caplan, 1992; Montanes, Goldblum, & Boller, 1995; Warrington, 1975), or that the pr ofile changes with the progression of the disease (Gonnerman et al., 1997). To avoid this controversy, the current study used only artifact categories.

PAGE 78

64 It has been suggested that providing feedback is an impor tant part of the learning process (Bandur & Shewan, 2001). Repetition has been used as a training task, and found to be effective in obtaining a treatment eff ect (Fuller et al., 20 01; Mayer et al., 2004; Patterson et al., 1983; Pring et al., 1993; Rothi et al., 2005). Hickin et al. (2002) suggested that presenting a picture stimuli a nd repeating the name of the word actually elicits semantic processing. Our study comb ined these three components. This was followed by the semantic training requiring ge neration of semantic information about the target item. The rationale for this comes from studies that suggest that the naming deficit in PAD was both due to semantic specification and lexical access defi cits (Bowles et al., 1987; Williamson et al., 1998). Our study used repe tition so that if the PAD participants have a deficit in both the semantic system a nd lexical access, the form of the word was provided and being reinforced, and the semantic training targeted the semantic part of the deficit. Evidence of a treatment effect has been re ported after training with a combination of semantic and lexical (i.e., word form) tasks (Drew & Thompson, 1999; Le Dorze et al., 1994; Wiegel-Crump & Koenigsknecht, 1973). Two other anomia treatment approaches have also been found to result in treatmen t and generalization eff ects in adults with aphasia are semantic distinction training (H illis, 1998); and semantic feature analysis (SFA) (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000). Both of these approaches combined specific training of se mantic features with presentation of the phonological form. As noted above, Kiran and Thompson (2003b) applied a similar process and suggested it was an important aspe ct of the study. The three other typicality

PAGE 79

65 studies (Mayer et al., 2004; Stanczak et al., 2005, 2006) also used a more simplified version of this. Our study attempted to achieve some of the benefits of SFA (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000), for exam ple, to facilitate a threshold level of activation of the semantic network that su rrounds the targeted item. Furthermore, like SFA, we aimed for systematic activation of th e distinguishing features of the item rather than shared features (Boyle, 2004; Boyl e & Coelho, 1995; Coelho et al., 2000). This process is based on the spreading activation theory, which posits th at selection of the target item among competitors is based on the highest activation achieved (Collins & Loftus, 1975; Dell & O'Seaghdha, 1992; Leve lt, Roelofs, & Meyer, 1999). Once the concept is selected, phonologica l information is activated so that the target item is produced (Collins & Loftus, 1975; Dell & O'Seaghdha, 1992; Levelt et al., 1999). Finally, SFA has also been shown to result in both a treatment effect and generalization to untrained items (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000). More specifically, our training protoc ol was a modified version of SFA (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000). Our PAD part icipants answered questions about the target item and then named it. We also incl uded an aspect of the semantic distinction training (Hillis, 1998), by asking them about how an item differs from others like it. When the participant was unable to res pond correctly, the clinician provided the information and had the participant rep eat it, based on Coel ho et al. (2000). Other studies also required active producti on of information, including the use of antonyms and synonyms (McNeil et al., 1998), personalized cueing (F reed, Marshall, & Nippold, 1995; Lowell, Beeson, & Holland, 1995; Marshall, Freed, & Karow, 2001); and

PAGE 80

66 the participant using circumlocution (i.e., describing the item) (F rancis et al., 2002). Some of these studies also reported generalization to untra ined items (Francis et al., 2002; Lowell et al., 1995). In conclusion, we used information from the three PAD naming treatment studies (Abrahams & Cam p, 1993; Fuller et al., 2001; Ousset et al., 2002; Rothi et al., 2005), studies compari ng training atypical and typical category exemplars (Kiran & Thompson, 2003b; Mayer et al., 2004; Plaut, 1996; Stanczak et al., 2005, 2006) and other aphasia studies that show ed treatment and ge neralization effects (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000; Davis & Pring, 1991; Francis et al., 2002; Hickin, Best, Herbert, Howard, & Osborne, 2001; Hillis, 1998; Howard et al., 1985; Pring et al., 1993; Raymer, Th ompson, Jacobs, & le Grand, 1993). Many of these studies also provided ra tionale for predictions about our research questions, which are discussed below. Research Questions and Predictions The first four research questions fo cused on performance (i.e., accuracy and reaction times) on picture naming. While the fi rst two research quest ions addressed the effects of repetition and the tr aining (i.e., facilitation effects) the third question addressed changes in performance on untrained items (i .e., generalization or practice effects). The fourth question also addresse d generalization, but to another task, category generation. These questions are listed below with th eir corresponding predictions. The type of training (i.e., training type) alwa ys refers to the typicality of trained items in a particular trained category, and are always referred to as Train-Atypical, to in dicate that atypical items were trained, and TrainTypical, to indicate that typical items were trained. Research Question 1: Does repetition of the control items (i.e., not with semantic training) impact performance at posttest?

PAGE 81

67 Prediction: There will be better performance (higher accuracy and faster reaction times) on repeated items compared to items that were not repeated (Fuller et al., 2001; Patterson et al., 1983; Pring et al., 1993; Rothi et al., 2005) Research Question 2. What are the effects of sema ntic training on trained items? Research Question 2a: Do accuracy scores and the r eaction times for trained items change from pretest to posttest? Prediction: Accuracy and reaction times for tr ained items will show a facilitation effect after training: Accuracy scores will be higher and reaction times will be faster on the trained items at the pos ttest (Drew & Thompson, 1999; Howard, 1985; Le Dorze et al., 1994). Research Question 2b: Does the type of training (i.e ., Train-Atypical or Train-Typical) impact the performance on trained items at posttest? Prediction 2b.1: Performance will improve after both types of training, but there will be a greater improvement after training typical items (Plaut, 1996). Prediction 2b.2: Performance will only improve af ter training typical items (Rosch, 1975). Research Question 3a : Will there be an overall change in performance on untrained items from trained categories? Prediction 3a: Accuracy will increase and reactio n times will decrease in untrained items in trained categories (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000; Hillis, 1998; Plaut, 1996; Wiegel-Crump & Koenigsknecht, 1973). Research Question 3b: Does the type of training (i .e., Train-Atypical/ Train-Typical) affect performance on untrained items in trained categories (i .e., generalization)?

PAGE 82

68 Prediction 3b.1: Training atypical items (Train-Atypical) will result in greater generalization as shown by higher accuracy scores and faster reaction times (Kiran & Thompson, 2003b; Plaut, 1996; Stanczak et al., 2006; Thompson et al., 2003). Prediction 3b.2: Training typical exemplars will result in greater generalization: higher accuracy scores and faster reacti on times for typical exemplars within the same category (Rosch, 1975). Prediction 3b.3 : Training typical exemplars will result in greater generalization to both untrained typical and atypical items because typical items have richer representations and share more features w ith all other category members (Heilman, personal communication, 2005). Research Question 3c: Is there an interaction betw een time and item typicality (i.e., atypical items and typical items)? Prediction 3c: Untrained typical items from trained categories will show more improvement than atypical items from trained categories after training. Research Question 3d: Is there an intera ction among time, training type, and item type? Prediction 3d.1: Training atypical items will result in generalization to untrained typical items (Kiran & Th ompson, 2003b; Plaut, 1996; Stanczak et al., 2006) and to untrained atypical items (Pla ut, 1996; Stanczak et al., 2006). Prediction 3d.2: Training typical items will have a positive impact on untrained typical items from trained categories a nd a negative impact on the performance with atypical items from tr ained categories (Plaut, 1996). Research Question 4: Is there generalization to category generation?

PAGE 83

69 Prediction 4: There will be an increase in th e number of items named in the trained categories, and this will be greater for the trained categories compared to the untrained categories (Ousset et al., 2002).

PAGE 84

70 CHAPTER 4 METHODS The purpose of this exploratory study wa s to extend previous research on the effects of training typical and atypical cat egory exemplars on generalization (Kiran & Thompson, 2003b; Mayer et al., 2004; Plaut, 1996; Stanczak et al., 2005, 2006) while comparing two views of typicality (Pla ut, 1996; Rosch, 1975; Rosch & Mervis, 1975). Participants with early probable Alzheimer di sease participated in this picture naming training study. The primary goal was to determin e if remediation targ ets could be chosen based on typicality, to maximize generalizati on to untrained items (Kiran & Thompson, 2003b; Plaut, 1996; Stanczak et al., 2006). Th is was achieved by training atypical (i.e., Train-Atypical) and typical (i .e., Train-Typical) category ex emplars from two semantic categories and leaving untrained atypical and typical items in each category for both the same item typicality that was trained (e.g., Train-Atypical, untr ained atypical items), and the opposite (e.g., Train-Atypical, untrained typical items). Us ing a repeated measures (i.e., within subject) design, our study provided each participant with a modified semantic feature based training (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000) for each category. This was completed in a single se ssion and the results were analyzed for a facilitation effect (Hickin et al., 2002; Howard, 1985; Howard et al., 1985; Patterson et al., 1983) and generalization to untrained it ems. While the outcome measures (i.e., dependent variables) for the picture naming were accuracy and res ponse time means (in milliseconds), the number of items generated was the outcome measure for the category generation task.

PAGE 85

71 Participants Following approval from the UF Institutional Review Board (Appendix A), the participants with PAD were recruited and enrolled into the study. Recruitment of participants was accomplished with the help of the University of Florida Memory and Cognitive Disorder Clinics. Additional recruitment efforts included presenting information and flyers about the study at the Alzheimers Association, and Alz Place, an adult day care facil ity for individuals with Alzheime r disease and other dementias. Participants were compensated $20 for their time. Diagnosis of the participants with m ild-moderate PAD was made by a team of medical professionals includi ng neurologists and neuropsyc hologists based on criteria from the National Institute of Neurologi cal and Communicative Disorders & Stroke (NINCDS), and the Alzheimers Disease a nd Related Disorders Association (ADRDA) (McKhann et al., 1984). Individual s were not eligible for the study if they had any history of brain or head injury, psychiatric hospita lization, alcohol or drug abuse; any chronic medical or psychiatric conditi ons, or any developmental lear ning disability. Participants were pre-screened at the UF Memory Disord ers Clinics for PAD. Some of the results from the neuropsychological testing ar e shown in Table B-1, Appendix B. A phone screen was used to confirm e ligibility and to collect demographic information. Twelve adults (7 women and 5 men) with newly diagnosed early PAD completed the study. The average age was 77.6 ( SD = 9.6) and the average number of years of education was 15.8 ( SD = 3.8). The mean score for the Mini Mental Status Exam (Folstein et al., 1975) was 22.6 ( SD = 3.4) with a range of 17-27. More specific demographic information and performance scor es for each participant are shown in Table 4-1.

PAGE 86

72 Table 4-1. Participant Demographics and Scor es on Mini Mental Status Exam and Reading Subtest ID # Age Years of Education Gender Mini Mental Status Exam Reading Subtest WRAT-3 (percentile) 1001 72 17 F 21 87 1002 77 13 F 24 50 1003 76 10 M 17 8 1004 92 24 M 24 91 1005 82 12 F 19 10 1006 78 12 F 22 50 1007 55 20 M 27 90 1008 84 15 F 25 75 1009 69 18 M 20 50 1010 86 16 F 19 63 1011 75 16 M 27 81 1012 85 16 F 26 87 Mean 77.6 15.8 22.6 61.8 Notes: Mini Mental Status Exam (Folstein et al., 1975); WRAT-3=Wide Range Achievement Test-3 Reading subtest (Wilkinson, 1993). Experimental Stimuli There were six training lists (1-6) that we re counter balanced using a Latin Square for training conditions (i.e., Train-Atypical, Train-Typical, Control) and semantic category (i.e., vehicles, clothi ng, and tools). As depicted in Table 4-2, each list had two training conditions and one cont rol condition. Participants were assigned to these lists sequentially, as they were enrolled into the study. For example, the first participant (#1001) had List 1, and this process conti nued with the remaining participants (e.g., participant eight ( #1008) had list 2). Table 4-2. Lists of Training Conditions and Corresponding Training Conditions List # Vehicles Clothing Tools 1 Train-Typical Control Train-Atypical 2 Train-Atypical Train-Typical Control 3 Control Train-Atypical Train-Typical 4 Train-Typical Train-Atypical Control 5 Control Train-Typical Train-Atypical 6 Train-Atypical Control Train-Typical

PAGE 87

73 The pre and posttesting picture naming stimu li consisted of 72 black and white line drawings from three categories: vehicles, clot hing, and tools. (Table C-1, in Appendix C, shows the list of 72 items used in the st udy). Since there are several variables being addressed, it is important to first clarify the terminology for the items in the pre and posttest naming tasks, then the details related to the training are discussed. All items were designated as either typical or atypical, based on the ratings in the Rosch norms (Rosch, 1975). Items were chosen from these three categ ories based on the seven-point typicality ratings from Rosch (1975), such that half th e items in each category were atypical and the other half were typical. Typicality ratings fo r items used in this study ranged from 1.02 (the most typical) to 5.36 (the most atypica l). For each category, the typical and atypical items were selected at about a midpoint range so that half would be typical and the other half would be atypical. The mean typical ity score for the atypical items 3.5 ( SE = .12), and the mean typicality score for the typical items was 1.6 ( SE = .08). An independent ttest indicated that these groups of items were significantly different, t (70)=12.967 p = .00. The picture stimuli were gleaned from a va riety of sources, with the majority from the Florida Semantic Battery (Raymer et al ., 1990). Name agreement was determined by piloting the naming task on 10 University of Florida (UF) undergraduates in the Language over the Lifespan Lab. Each semantic category was designated as e ither the control category (i.e., no items received semantic training), Train-Typical (i.e., only typical items received semantic training), or Train Atypical (i .e., only atypical items received semantic training). There were several reasons for the choice and number of categories. All artifact categories were

PAGE 88

74 chosen to avoid the possible confound of mi xing biological and artifact categories. Furthermore, it was important to prevent a ny overlap from other semantic categories (e.g., vehicles and sports). Also, previous re search on picture nami ng has included these categories (Grossman, 1981; Ma yer et al., 2004), although othe r studies of typicality effects have only included natural kinds (Fuller et al., 2001; Kiran & Thompson, 2003b; Rothi et al., 2005). There were 24 exemplar s in each category (Kiran and Thompson, 2003b) 12 typical and 12 atypical. From these 72 items, the semantic training stimuli were selected. For each category, seven atypical items were designated as the training set for when the category appeared in the Train-Atypical condition, and 7 typical items were designated as the training set for when that category appear ed in Train-Typical condition. Thus, in a category that appeared in the Train-Typical condition, for example, the 7 typical items in the training set would receive semantic training, leaving 5 un trained typical items and 12 untrained atypical items from that category to analyze for generalization. Untrained items were those items that received no semantic training, but were members of the trained category. Therefore, for any given item, it was necessary to specify whether it was typical or atypical, trained or untrain ed, and what type of training, if any, that category received. The control items were untrained items in the untrained category. Control items were named at both pretest and posttest. Because a ll items from the trai ned categories were repeated, half of the items from the contro l category were also repeated to determine what effect simple repetition without semantic training had on lexical access. An example based on list 1 is found in Table 4-3.

PAGE 89

75 Table 4-3. Example from List 1: Train-Typical Vehicles and Train-Atypical Tools. Trained Untrained (same as trained) Untrained (opposite of trained) 7 Typical Vehicles 5 Typical Vehicles 12 Atypical Vehicles 7 Atypical Tools 5 Atypical Tools 12 Typical Tools Note: The control category was comprised of 24 items of clothing. The list of phrases or cues for the se mantic training was developed by polling volunteers (ranging from a Ph.D. investigator, a certified speech-la nguage pathologist, and several undergraduate res earch assistants) who are me mbers of the UF Language over the Lifespan Lab in the Department of Communication Scienc es and Disorders. These cues were compiled and selected based on the most salient ones. In a series of post hoc analyses of the stimulus words from these pictures, we examined a possible influence from four va riables on the atypical and typical items: typicality (Rosch, 1975), which we expected to show a significant difference; written word frequency (Frances & Kucera, 1982) and familiarity (Wilson, 1987), and phoneme length. We used an independent sample t-test to compare the complete list of 72 atypical and typical items on the four variables. For typicality and phoneme length, ratings were available for each item. The independent ttest for the typica lity ratings had a p value of .000, indicating a significant difference in te rms of typicality between the atypical and typical items. The independent t-test for phoneme length was not significant, t (70)=2.528 p = .69 indicating there was no difference betwee n the atypical and typical items. In other words, the phoneme length for the atypic al and typical items did not influence the results. The independent t-test compar ing typical and atypical items from the Complete List was not significant for frequency, t (70) =-1.389, p =.169, suggesti ng that there was no difference in the frequency ratings between the atypical and typical items. However, although the independent t-test fo r familiarity was significant, t (38) =-3.625, p = .00, this

PAGE 90

76 analysis was only computed on 40/72 items (i.e., many of the items did not have familiarity ratings). Thus, the results must be interpreted with caution. Table 4-4 depicts the means of the four variables discussed above. Table 4-4. Means for Complete List of Atypical and Typical Items. Train Type Typicality Score Freque ncy Familiarity Phoneme Length Atypical 3.6 15.7 480 4.8 Typical 1.7 25.1 551 4.9 Notes: Typicality (Rosch, 1975); Freuency (Frances and Kucera, 1982); Familiarity (Wilson, 1987). Procedure Prior to participation in the study, written informed consent was obtained from each participant. As needed, the clinician allowed ti me for the participant to read the informed consent form. Then, the clinician answered a ny questions and assisted the participant in interpreting the information. Screening Two measures were used for the scre ening: The Mini-Mental Status Exam (Folstein et al., 1975) and the reading subt est of the Wide Range Achievement Test (WRAT-3) (Wilkinson, 1993). The clinician ad ministered the MMSE using the standard procedure (Folstein et al., 1975). Next, the clin ician presented a com puterized version of the reading subtest (Wilkinson, 1993) in whic h the participant read aloud the individual words from the screen. These responses were digitally recorded on an Olympus DM-10 digital recorder (Olympus, 2002) and response times (RTs) were collected via DirectRT (Jarvis, 2004) for later scoring and analysis. Pretesting The pretest consisted of two tasks, Ca tegory Generation (Benton, 1968) and Picture Naming. For the Category Generation task, th e clinician provided instructions for the

PAGE 91

77 participant to name as many items as possibl e in one minute for a given category (i.e., vehicles, clothing, and tools). The categories were presented in a pre-determined random order. The clinician wrote the responses verb atim on the pretest form, and responses were also digitally recorded for checking reliability. The Picture Naming task consisted of th e complete set of 72 line drawings randomly presented on the computer via DirectRT (Jarvis, 2004). At the onset of the task, the clinician read aloud the set of instructions that appeared on the screen and confirmed comprehension of the task. The clinician controlled the pace of the session by using the mouse button presses on a wire less mouse. Each picture stayed on the screen until the clinician advanced to the next picture (i.e., the picture di d not disappear or time out). There were four practice items from the categor y of fruit. After the participant attempted to name a presented picture, the clinician informed the participant if he/she was correct or not, and then asked the participant to repeat the correct name of th e item three times. If the participant produced a corr ect alternative name of the item (e.g., jet for airplane) their response was accepted as correct, and this word was repeated (Appendix C includes a list of acceptable alternative names for items). The clinician then advanced to the next trial. As stated above, twelve items from the control category were not repeated (half were typical and half were atypical). A prom pt appeared on the computer screen to alert the clinician as to whether this was a repeat ed item or not. This procedure was used for all of the items in the two trained categories, and for half of the control items (i.e., repeat or no repeat, without semantic training). More specifically, after the participant na med an item, the word repeat appeared on the lower center section of the screen. Ne xt, the clinician provi ded verbal feedback

PAGE 92

78 and directed the participant to repeat the name of it three times (i.e., Thats correct, now you say it three times ___, ___, ___ or Its s upposed to be a ___, now you say it three times, ___, ___, ___) (Kendall, personal communi cation, 2005; Fuller et al. 2001; Rothi et al. 2005). When an unrepeated item was presented, the participant named the item, then the words Go on appeared in place of the word Repeat, and the clinician advanced to the next item. Thus, on these control items the clinician did not provide feedback about the participants response, and the item was not re peated. The clinician continued this process with each of the re maining pictures until all 72 items had been named. At this point, the participant took a s hort break while the clinician set up for the semantic training. All of the naming respons es were digitally recorded for later transcription and analysis, and response times were collected via DirectR (Jarvis, 2004). Semantic Training This was a guided, feature generation ta sk, loosely based on semantic feature analysis (Boyle, 2004; Boyle & Coelho, 1995; Coelho et al., 2000) and semantic distinction training (Hillis, 1998). The computer presented the pictures from one of the training categories (i.e., seve n items) in random order and the clinician asked the participant to answer each of the followi ng questions about the item: 1. What is its function?; 2. Who uses it?; and 3. How is it different from other __? (i.e., things like it). After the participant answered the three questions, the clinician asked the participant to name the item. Whenever a participant wa s unable to correctly answer a question or provided ambiguous information, the clinician provided cues from a pre-determined typed list. Whenever the participant was unabl e to name the item, the clinician stated it and asked the participant to repeat it. The clinician con tinued this process with the remaining six training items for that categor y. Each participant was trained on the two

PAGE 93

79 categories from their designate d list. The trained categorie s (e.g., Train-Atypical Tools, and Train-Typical Vehicles) we re randomly presented in bl ocks, but the order of the items presented within each category was ra ndom. This allowed each participant to be trained on each category and the corresponding training type (i.e., Tr ain-Atypical, TrainTypical) for a total of six practice sets (i.e ., three presentations fo r each category). Then, the participant took a break while the clinicia n prepared for the postt esting. The responses from the training were also digitally recorded. Posttesting The posttesting consisted of th e same two tasks and instructions as the pre-testing: Category Generation and Pict ure Naming. During the Catego ry Generation task, the participant was given the same instructions and the same random order for generating as many items as possible for each of the three categories (vehicles, clothing, and tools). The clinician wrote down partic ipants responses verbatim an d also digitally recorded them in order to check reliability. For the Picture Naming task, the same 72 item picture set was presented by computer running Direct-RT (Jarvis, 2004). It ems were presented in a different random order than at pretest. In this version, the participant only named the pictures aloud, without feedback or repetition. Scoring The scoring for the screening tasks was completed as follows. The Mini Mental Status Exam was scored using the standard scoring outlined by Folstein et al. (1975). Digital recordings of the r eading subtest from the WRAT-3 (Wilkinson, 1993) were used to code the data for accuracy of production by a trained resear ch assistant. The clinician scored the responses and calcu lated raw and standard scores and converted them to the

PAGE 94

80 percentiles, based on instructions from the WRAT-3 Manual (Wilkinson, 1993). It is important to note that normative data for adu lts with PAD is not provided in the WRAT-3 Manual (Wilkinson, 1993), and that most of our PAD participants were beyond the age ranges provided (i.e., 55-64.11; and 65-74.11). Th us, these findings must be interpreted with caution. The pre and posttest picture naming performance for accuracy and response times was collected by Direct RT (J arvis, 2004). Accuracy and re sponse times were tracked by the clinicians mouse clicks (i.e., left for correct and right for incorrect) based on the participants response and coded these as one s and zeros. The list of acceptable alternate names is displayed in Appendix C. The clinician used the same procedure for the response times (i.e., left for usable and right for unusable). For the response times, only fluent, correct responses were computed in the analyses (e.g., response times were excluded if the participant sa id, uh airplane, but the re sponse would be scored as accurate/correct). Furthermore, response times for each participant were adjusted for outliers by replacing any value th at was greater than three st andard deviations above the mean, with the mean value. The responses for the Category Generation task were transcribed independently by a trained research assistant. The clinician and research assistant then compared these results to the clinicians original on-line written documentation from the session and discrepancies were resolved by referring back to the corres ponding voice files until agreement was reached. Informal reliability m easures were carried out with ~ 50% of the data, using the voice files. Agreement was ~98% and disagreements about scoring of

PAGE 95

81 responses were resolved by consensus dur ing the Language over the Lifespan Lab meetings. Responses from the Picture Naming task were also transcribed by a trained research assistant. The clinician used these transcri pts to verify coding of 100% of the picture naming scoring. Statistical Analyses Our study used a repeated measures desi gn. The analyses compared pre and posttest category generation a nd picture naming performance using both accuracy and response times where possible. Analyses eval uated the effects of repetition and practice from data in the control condition, the eff ects of semantic training, and generalization effects to untrained items in trained categorie s. The effects of trai ning atypical and typical category exemplars were also examined. The pre and posttest picture naming data we re analyzed using t-tests, repeated measures analysis of variances (ANOVAs ) and follow-up post hoc t-tests where necessary. It is important to point out that the pretest naming data reflects the participants initial attempt at naming the pictures; this was followed by the repetition and semantic training. Thus, the effect of re petition and training can only be evaluated by comparing pre and posttest performance. Due to an error in the programming related to the control items, four partic ipants repeated all items. Cons equently, data from only eight of the twelve participants could be included in the anal ysis of repetition effects. Since inspection of the data for some of the analyses showed pronounced differences at pretest, difference scores were calculated from the pr e and posttest data and then analyzed. The dependent variables for picture naming were means from the accuracy and response time data. The independent variab les varied according to the questions, but

PAGE 96

82 included time (pre /posttest), repetition (no re peat/repeat), training type (i.e., the TrainAtypical/Train-Typical), item typicality (i .e., item type, atypical/typical). For consistency, whenever possible the term atypical is always discussed before typical, this convention is alphabetical and used in orde r to reduce confusion with the terminology (there is no intentional bias). The dependent variables for the pre and pos ttest category genera tion data were the number of items generated per category, a nd the independent variables were time and training type (i.e., Control category, TrainAtypical, Train-Typical). The data was analyzed with repeated measures ANOVAs. Research Question 1: Control Items Does repetition of the control items (i.e ., not with semantic training) impact performance at posttest? This question addresse d the possibility of a re petition or practice effect. Only accuracy and res ponse time means from the unt rained items in untrained categories (i.e., the control condition) made up the dependent variable. Half of these were in the repetition condition and half were not (i.e., not repeated ). The independent variables were time (pre/pos t) and repetition (yes/no). A (2) Time (pre/post) X (2) Repetition (yes/no) repeated measures ANOVA was used to an swer this question. Research Question 2: Trained Items in Trained Categories This two-part question addressed the po ssibility of a facilitation effect in semantically trained items (i.e., Train-Atypica l and Train-Typical). Th e data consisted of scores from the 14 trained items (i.e., the se ven atypical and seven t ypical items from the two trained categories). This within categor y measure was assessed as a comparison of pretest and posttest performance on the confrontation naming task

PAGE 97

83 Research Question 2a: Do accuracy scores and the response times for trained items change from pretest to posttest? The indepe ndent variable was time (i.e., change from pretest to post test). Research Question 2b: Does the type of training (i.e., Trai n-Atypical or TrainTypical) impact the performance at posttest ? The independent variables were time and training type. To answer both of these questions, (2) Time (pre/post) x (2) Training type (TrainAtypical/Train-Atypical) repeated measures ANOVAs were computed. Research Question 3: Untrained Items in Trained Categories Question 3 had several subcomponents, but overall addressed the possibility of a primary generalization effect. If th ere was a facilitation effect in question 2, the purpose of questions 3a, b, c, and d was to determin e if the facilitation effect generalized to untrained items within the same trained category. Research Question 3a: Is there generalization to untrained items in the trained category? The independent vari able was time (pre/post). Research Question 3b: Does the type of training (i.e., Train-Atypical/TrainTypical) affect generalization to untrained it ems in trained categories? The independent variables were training type and time. Research Question 3c: Does the typicality of the untrained item (i.e., atypical items and typical items) determine whether it will benefit from generalization? The independent variables were time and item type. Research Question 3d: Do time, training t ype, and item type interact to promote generalization limited to a particular item t ype following a particul ar training type? The independent variables were time, training type, and item type.

PAGE 98

84 The analyses for Question 3 used the 34 untrained items in the two trained categories (i.e., 17 untrained items in each trai ned category). For the accuracy means, we used a (2) Time x (2) Training Type (Tra in-Atypical/Train-Typical) x (2) Item Type (Atypical/Typical) repeated measures ANOVA. The response time means were analyzed di fferently. For Question 3a, a t-test was used to compare pre and posttest perform ance. To answer Questions b, c, and d difference scores were comput ed between preand posttest response times and analyzed via a (2) Training Type x (2) Item type repeated measures ANOVA. Research Question 4: Category Generation Research Question 4: Is there generaliz ation from picture naming to category generation? Does the type of training th e category receive affe ct whether there is generalization to category generation. The inde pendent variables were time and training type. The dependent variable was total numbe r of items generated per category and these were analyzed with a (3) Training Type (Tra in-Atypical, Train-Typical and Control) x (2) Time repeated measures ANOVA.

PAGE 99

85 CHAPTER 5 RESULTS In this study we compared the performa nce of adults with probable Alzheimer disease (PAD) on picture naming and categor y generation tasks before and after a semantic training using atypical and typical category exemplars. The accuracy means and response times (in milliseconds, ms) from pre and posttest picture naming, and number of items generated per category on a category generation task were analyzed with t-tests and repeated measures analysis of variances (ANOVAs). To control for some of the differences in the pretest scores across c onditions, difference scores were computed for several of the questions and th en analyzed to compare direct ion and magnitude of change. An a priori decision was made to use a .05 le vel of significance for all of the analyses. The two training conditions are always refe rred to as Train-At ypical (i.e., training atypical items) and Train-Typica l (i.e., training typical items). Research Question 1: Control Items Question 1 asked if repetition of the control items (i.e., no t with semantic training) impacted the performance of picture naming at posttest. The dependent variables were the accuracy and response times, and the independent variables were time and the repetition and no repetition conditions. The data for these analyses were from the control categories (i.e., untrained items in untrained ca tegories) in which half of the items were repeated (i.e., repetition c ondition) and the other half were not (i.e., no repetition condition).

PAGE 100

86 A (2) Time (pre/posttest) x (2) Repetiti on (yes/no) repeated measures ANOVA was used to examine the data from eight particip ants. The results showed that there were no significant main effects or inte ractions for the accuracy m eans. Although the items in the repetition condition were somewhat higher (Mean ( M ) = .89, SE = .04) compared to items in the no repetition condition ( M = .80; SE = .04), this was not significant, F (1,7) = 3.691; p = .096; 2= .35. There were no other main effects or interactions with the accuracy means for the control items. Thus, repetiti on did not have a significant effect on the accuracy means. The response time means showed a simila r pattern from the (2) Time x (2) Repetition repeated measures ANOVA. There were no significant main effects to suggest that repetition alone impacted performance fr om pretest to posttest There were decreases in the response time means, indicating faster responses, but these were not significant. One was for time, F (1,7) = 3.702; p = .096; 2 = .346; MPre = 2456 ms ( SE = 521); MPost = 1858 ms ( SE = 243). The other was for repetition, F (1,7) = 4.416; p = .074; 2 = .387; MREP = 1979 ms ( SE =321 ms), MNoRep = 2334 ms ( SE = 440 ms). The interaction was not significant, F (1,7) = 1.566; p = .251; 2 = .183. The items in the control condition did not re ceive any semantic tr aining, but half of them were repeated at pretest. Although th e accuracy in the control condition means were slightly higher, this was not significant. The response times decreased somewhat, but not significantly. Thus, repeating ite ms in the control condition di d not result in statistically significant changes in accuracy or response times. Research Question 2: Semantically Trained Items Questions 2a and b asked if performance on the semantically trained items (i.e., Train-Atypical and Train-Typica l) changed from pre to posttest, and, if so, whether there

PAGE 101

87 was a difference between the two types of training (i.e., Train-Atypical versus TrainTypical). The dependent variables for both que stions were accuracy and response times from the 14 trained items (i.e., seven atypi cal and seven typical items) from the 12 participants. Table 5-1 provides an overview of the accuracy and response time means. These are discussed separately below. Table 5-1. Pre and Post Group Accuracy and Response Time Means for Trained Items Accuracy Response times Pretest Posttest p value Pretest Posttest p value .76 (.04) .86 (.03) .004 1719 ms (183) 1362 ms (96) p < .06 Note: The numbers in parentheses are standard errors ( SE ). To determine the effects of semantic trai ning on performance, a (2) Time (pre/post) x (2) Training type (Train-Atypical/Train-Ty pical) repeated measures ANOVA was used to compare pre and posttest accuracy means. Th ese showed a significant main effect of time, F (1,11) = 12.780; p < .01; 2 = .54. The trained items were named significantly more accurately at posttest compared to the pretest ( MPre = .76, SE =.04; MPost = .86, SE =.03). There were no other main effects or interactions for the accuracy means. Although trained items were more accurate after th e training, there was no difference between training typical versus atypical items. Table 5-2 shows the re sults for the accuracy and response times. Table 5-2. Group Accuracy and Response Time Me ans for Trained Items (Train-Atypical; TrainTypical) Accuracy Response Times Pretest Posttest Pretest Posttest Difference Scores* TrainAtypical .71 (.07) .84 (.05) 1506 ms (168) 1397 ms (154) -109 (217) TrainTypical .81 (.06) .87 (.05) 1931 ms (243) 1326 ms (101) -605 (186) Notes: ( SE ).* Difference scores for the response time means were significant ( p =.04).

PAGE 102

88 Response times were examined to determ ine if semantic training impacted the speed of lexical access. A t-test of pre a nd posttest performance showed that response times were 357 ms faster after the training; these results approached significance, t (11) = 2.12; p = .057; MPre = 1719, SE = 183 ms; MPost = 1362 ms, SE = 96 ms. Since the pretest response times for the two training conditions varied, difference scores were calculated for Question 2b to dete rmine if the type of training (i.e., TrainAtypical versus Train-Typical) affected the response times. The difference scores were computed by subtracting the posttest response times from pr etest response times for each training type. Next, a paired t-test was com puted, which showed that the results were significant, t (11) = 2.28; p = .04. As shown in Table 5-2, ther e was a larger difference for the Train-Typical items ( M = -605, SE = 217 ms), which were the slowest at pretest, but fastest at posttest compared to Train-Atypical items ( M = -109, SE = 186 ms). Thus, the semantic training had a facilitati on effect on lexical access in individuals with mild-moderate probable Alzheimer di sease (PAD), increasing accuracy on picture naming. Although the change in accuracy for trained items was similar regardless of training type, response times for Train-Typi cal items benefited more from semantic training. Research Question 3: Untrained Items in Trained Categories Question 3 addressed various aspects of ge neralization to the 17 untrained items in the 2 trained categories. For each trained cat egory, these 17 items comprised 5 untrained items with the same typicality as the items trained and 12 items of the opposite typicality. For example, if Typical Vehicles were tr ained (i.e., Train-Typical), untrained items would consist of 5 untrained typical vehicl es and 12 untrained atypical vehicles. The dependent variables were accuracy and re sponse times. The different components of

PAGE 103

89 Question 3 were addressed using differen t repeated measures ANOVAs. The accuracy scores are discussed together, answering Qu estions 3,a, b, c, and d, and is followed by a similar procedure for the response times. The purpose of Question 3a was to dete rmine if there was generalization to untrained items in the trained category. Ques tion 3b aimed to determine if the type of training (Train-Atypical versus Train-Typical) affected ge neralization within a category at posttest. The independent variables were ti me and training type. Questions 3c, d, and e asked about the possibility of specific intera ctions between time and training type, time and item type (i.e., item typicality, atypical or typical); and among time, training type, and item type. The independent variables were time, training type, and item type, and the dependent variable was accuracy means. To a ddress these questions systematically, a (2) Time x (2) Training type (Train-Aty pical/Train-Typical) x (2) Item type (Atypical/Typical) repeated measures ANOV A was computed. Table 5-3 shows the results. Table 5-3. Analysis of Variance for Accuracy Means for Untrained Items in Trained Categories. Effect df F 2 p value Time 1 5.811 .346 .035* Train 1 4.968 .311 .048* Item Type 1 1.902 .147 .195 Time x Train 1 2.681 .196 .130 Time x Item Type 1 6.597 .375 .026* Train x Item Type 1 3.717 .253 .080 Time x Train x Item Type 1 5.987 .352 .032* Notes: = significant at .05 alpha level. The accuracy results for Question 3a show ed that there was a significant main effect of time, F (1,11) = 5.811; p < .04; 2 = .35. As depicted in Figure 5-1, the accuracy means were significantly higher at posttest compared to pretest ( MPre .71, versus MPost =.77; SEs = .04).

PAGE 104

90 0.77 0.71 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Pre-testPost-test Figure 5-1. Accuracy Means for Untrained Items in Trained Categories, Main Effect of Time ( p = .035; the numbers in the figure are proportions). There was also a main effect of training type, F (1,11) = 4.968; p =.048; 2 = .31, but this was secondary to the three-way interaction, discussed below. There was no interaction between time and training type that the category had received. However, a significant two-way interaction between item type and time, F (1,11) = 6.597; p = .03; 2 =.38 was found for Question 3d. This revealed that, at pretest, the accuracy means for the atypical items were at .76 ( SE = .06) but the typical items were only .66 ( SE = .03). At posttest, the atypical it ems were at .78 ( SE =.047) and the typical items were at .77 ( SE = .04). Thus, the typical items showed a larger increase in accuracy than atypical items. A three-way interaction among tr aining type, item type and time was also significant F (1,11) = 5.987; p = .03; 2 = .35. As can be seen in Figure 5-2, significant increases in accuracy only occurred for untra ined typical items in categor ies that received training on typical items (i.e., untrained typical it ems in the Train-Typical condition). p = .035

PAGE 105

91 0.81 0.75 0.76 0.76 0.56 0.76 0.75 0.79 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Atypical itemsTypical itemsAtypical items Typical items Pre Post Trained AtypicalTrained Typical Figure 5-2. Untrained Items in Trained Categori es: Three-Way Interaction Among Time (Pretest versus Posttest), Training Type (Trained Atypical versus Trained Typical) and Item Type (Untrained Atypical Items and Untrained Typical Items); p = .03; the numbers in the figure are proportions. Only typical items in categories that were Trained Typical increased in accuracy. Due to variability in the pr etest accuracy data, the 3-wa y interaction was explored using difference scores to examine change due to training. We computed pre to post difference scores for each of the four condi tions. Then, we used these four difference scores in a (2) Training type x (2) Item t ype repeated measures ANOVA. This showed a main effect of item typicality, F (1,11) = 6.597; p = .03; 2 = .38, indicating a larger difference for typical items ( M = .113) compared to atypical items ( M = .02). A significant interaction between training type and item typicality was also found F (1,11) = 5.987; p = .03; 2 = .35. This confirmed that the bi ggest difference for the accuracy means ( M =.20) was in the typical items from cat egories that were tr ained with typical items. Table 5-4 shows the mean response time s for the untrained items in trained categories for the 12 participants.

PAGE 106

92 Table 5-4. Group Response Time Means (ms) for Untrained Items in Trained Categories. Train-Atypical Train-Typical Pre Post Pre Post Atypical Items 1957 ms (235) 1575 ms (180) 1668 ms (156) 1255 ms (60) Typical Items 1837 ms (194) 1756 ms (212) 1551 ms (101) 1535 ms (199) Notes: ( SE response times are in ms). The pre and posttest response time means fo r Question 3a were analyzed with a ttest to determine if there was generalizati on to untrained items in trained categories. These results were not significant, t (7) = 1.77; p = .12; ( MPre = 1596 ms, SE = 154 ms; MPost = 1430, SE = 156 ms. It is important to note th at for this analysis, only eight participants had usable response times for bot h pre and posttest, and these were the only data included in the analysis. As mentioned ab ove, the criteria for usable response times were a correct response for the item name and a fluent response that was not preceded by other words, such as Its a __. Consequentl y, there was missing data in four of the cells at pretest, and thus a reduc tion in the number of availabl e data points for the response time analyses. This could have influenced the results. Response time means for Question 3b, c, and d were analyzed with pre to post test difference scores calculated for each item type within each Training Type (e.g., for untrained typical items and untrained atypi cal items in each trained category). While Question 3b asked if there was a main eff ect of training type (e.g., Train-Typical), Question 3c focused on a main effect of ite m type (e.g., atypical items). Question 3c addressed the possibility of a two way interaction between Training type and Item type. To answer these questions, a (2) Training ty pe x (2) Item type repeated measures ANOVA was computed. There were no significant main effect s: for the Training type, F (1,7) = .170; p = .69; 2 = .02; or for Item type, F (1,7) = 2.294; p = 17; 2 = .25. There

PAGE 107

93 was no interaction between Tr aining type and Item type, F (1,7) = .72; p = .43; 2 = .09. This lack of significance coul d be due to missing data as mentioned above or due to a lack of effect of training and item type on response times. Overall, participants showed significant improvements in accuracy on the untrained items in trained categories, and this was evid ent in the main effect of time: there was higher accuracy after the training. Also, ther e were two interacti ons. The interaction between time and item type showed that th ere was more improvement for the untrained typical items compared to the untrained atypi cal items. From the three-way interaction among time, training type and item type, it was evident that at posttest, accuracy improved significantly in the untrained typica l items from the Train-Typical category. Thus, at posttest, our participants had improved naming accuracy on untrained items in trained categories; especially on the typical it ems, in general, and more specifically, when they were in the category that had been trained with typical items. However, there was no concomitant generalization of training to response times. Research Question 4: Category Generation Is there generalization to category gene ration? Question 4 as ked if there was generalization to another task, category genera tion, and if the type of training influenced the outcome. While the dependent variable was total number of items generated per category, the independent variables were tim e and training type. A (3) Training type (Train-Atypical/Tra in-Typical/Control) x (2 ) Time (per/post) re peated measures ANOVA was calculated. There was a significant main effect of time, F (1,11) = 9.968; p < .01; 2 = .48. Figure 4-3 shows that a greater number of items were generated at posttest compared to at pretest ( MPre = 7.16, SE = .72, versus MPost = 8.61, SE = .83). There were no other significant findings. Training type had no significant main effect, F (2,10) = .578; p =

PAGE 108

94 .578; 2 = .104, nor was the interaction between training type and time significant, F (2,10) = .186; p = .833; 2 = .036. These results suggested th at the type of training did not influence the category generation perf ormance at posttest. Figure 5-3 shows the number of items generated at pretest a nd posttest for the picture naming training conditions (i.e., Train-Atypical, Train-Typical, and Controls). 6.7 7.6 7 7.9 9.1 8.9 4 5 6 7 8 9 10 Train ATTrain TYControlsNumber of Items Generated p = .833 Pre Post Figure 5-3. Category Generation, Number of Item s Generated at Pre & Posttest for Trained & Control Items. Train-AT=Train-Atypical, Train-TY =Train Typical. p= 833. Summary of Results In summary, the group of adults with PAD performed picture naming and category generation tasks after participating in a se mantic training by answering questions about typical and atypical items and then naming th em. The participants repeated the correct name of most of the items after an initial attempt to name them during the pretest. Although the control items showed slight increases in accuracy and faster response times, these changes were not significant. The accu racy scores for trained items showed a facilitation effect from the semantic traini ng (i.e., combined Trai n-Atypical and Train-

PAGE 109

95 Typical conditions). Although the responses were faster for the trained items, this only approached significance. However, the differenc e scores for the response times did show a significant reduction for the Train-Typical items compared to the Train-Atypical items. The accuracy scores showed generalized naming to untrained items in trained categories, and a three-way interaction s uggested that the participants also demonstrated better performance on the typical items, especially when they were in the Train-Typical condition, but this was only significant for th e accuracy scores. In the category generation task, the PAD group also demonstrated a signi ficant improvement in the number of the items generated from pre to posttest, but th e type of training did not have an impact. These results are reviewed and interpreted in the next chapter.

PAGE 110

96 CHAPTER 6 DISCUSSION In this exploratory study, 12 adults recently diagnosed with early probable Alzheimer disease (PAD) individually co mpleted a three phase session involving pretesting, semantic training, and postesting. The first phase involved category generation and picture naming. After an in itial attempt at naming the picture, participants received performance feedback and repeated the name of all of the items, except for half of the items in the control conditi on. In the second phase, 14 items were trained, 7 atypical items (i.e., Train-Atypical) from one seman tic category and seven typical items (i.e., Train-Typical) from another semantic category. Participants were required to answer questions about the target item and then name it. In the third phas e, category generation and picture naming tasks were used again, ex cept there was no feedback or repetition. Outcome measures were accuracy scores a nd response times for the picture naming and number of items produced for the category generation task. The research questions addressed th e effects of repetition, training, and generalization, as well as typicality. The tw o views were compared, one favoring typical items, based on the family resemblance vi ew (Rosch, 1975) and the other favoring atypical items, based on the Complexity Account of Treatment Efficacy (Thompson et al., 2003) and a connectionist model of acquire d dyslexia (Plaut, 1996), as well as adults with aphasia (Kiran & Thompson, 2003b; St anczak et al., 2006) who had better generalization after training with atypical items. Interpretations of our data are related to the predictions from these views and theori es of semantic representations in PAD

PAGE 111

97 (Altmann et al., 2001; Devlin et al., 1998; Gonnerman et al., 1997), as well as PAD anomia therapy studies (Abrahams & Camp, 19 93; Fuller et al., 2001; Rothi et al., 2005) and some of the anomia treatment studies from the aphasia literature. It is important to note that there were many differences in thes e studies, including i ssues related to the patient population (i.e., PAD ve rsus aphasia; aphasia clas sification; severity, and corresponding language deficits) and methodol ogy (i.e., group versus single subject design; accuracy and response time measures versus just accuracy). Summary of Findings In this exploratory study, adults with PAD showed that repeating items in the control condition led to slight changes (i.e., increase s in the accuracy scores and decreases in the response times) but these were not significant. However, the participants showed a facilitation effect in this single semantic trai ning using typical and atypical semantic category exemplars in which they we re required to generate information about the target item and than name it. Thus, at posttest, they showed a facilitation effect characterized by both significantly improved ac curacy scores and faster response times on picture naming. Although accuracy and resp onse times for both Train-Atypical and Train-Typical items changed, only the change in response times for the Train-Typical condition was significant. In terms of gene ralization to untrained items in trained categories, changes in accuracy scores were significant; spec ifically, there was significant improvement in the accuracy scores for the untr ained typical items that were in the TrainTypical category. For the category generation ta sk, the participants produced significantly more items in the category generation at postt est; however, this was not due to semantic training, since all categories were affected.

PAGE 112

98 It is important to note that several comparisons showed relatively high effect sizes but non-significant results. This is likely due both to the small sample size and to the high variability among participants, which gave this study little power to detect subtle differences. Therefore, while the results are suggestive, a larger sample size is recommended for future studies. In the next section, implications of the findings are discussed. Implications for Anomia Treatment in PAD The findings from our group of adults with PAD are consistent with the family resemblance view (Rosch, 1975) and the hypothesi s that training typical items results in generalization (Heilman, pe rsonal communication, 2005). Furthermore, the PAD groups performance is compatible with some (Plaut 1996; Stanczak et al ., 2006) but not all of the treatment effects from typicality traini ng studies with aphasic speakers and some accounts of PAD (Altmann et al., 2001; Dev lin et al., 1998; Gonnerman et al., 1997; Martin, 1992; Sailor et al., 2004; Smith et al., 1995). Previous research has suggested possible reasons to account for better performance with typical items including the timeframe re lated to the initial phase of the training (Stanczak et al., 2006). These are discussed br iefly and compared with our findings. This is followed by an alternative account of why our PAD group di d better with the typical items. To account for the performance of one of th eir participants with transcortical motor aphasia and semantic and phonological deficits Stanczak et al. (2006) proposed that for such participants, typical and atypical items might be best treated in sequence. This was because over time, training the typical it ems may lead to confusion due to the competition among related typical items, at whic h point the atypical items, which provide

PAGE 113

99 more variation and thus have fewer compe titors, are learned better compared to the typical items (Stanczak et al., 2006). A similar phenomenon was also noticed in the training of a connectionist ne twork. Plaut (1996) reported that his computer simulation of acquired dyslexia had more difficulty lear ning to distinguish among highly typical category exemplars although ultimately the mode l learned the typical items better than the atypical items. The results of the current study support th is hypothesis to some extent. We found that at pretest, responses to typica l items were more accurate but much slower than responses to atypical items, which is consistent with there being some difficulty distinguishing between items. It is certainl y possible that continued intense training of typical items could lead to confusion among th em (Stanczak et al., 2006). This has also been described as a semantic blocking effect or interference: repeatedly accessed words with similar semantics become more difficult to access the more frequently they must be named (Schnur, Schwartz, Brecher, & H odgson, 2006). At that point, the greater interitem distinguishability among the atypical items might benef it our participants (Stanczak et al., 2006). However, this is an empirical question that requires investigation using a more standard tr eatment study methodology. It is important to note that these typi cality training studies were conducted in systems that had damage due to focal lesions either acquired dyslexi a in a connectionist model (Plaut, 1996) or aphasia in adults (K iran & Thompson, 2003b; Mayer et al., 2004; Stanczak et al., 2005, 2006). Gonnerman et al (1997) reminds us that PAD does not result in exclusive damage to an area, inst ead, it is a diffuse pattern of damage. Thus, it might be unreasonable to expect similar out comes from training typical and atypical category exemplars in adults with PAD. I ndeed, the results of the current study are

PAGE 114

100 consistent with previous studi es examining the effects of PAD on semantic tasks, and can be accounted for by existing hypotheses about th e effects of PAD on the semantic system (Altmann et al., 2001; Devlin et al ., 1998; Gonnerman et al., 1997). The primary findings of the current study we re that training effects were strongest among trained typical items, mani festing as faster response times, and only typical items from categories in which typical items we re named more accurately, a generalization effect. Further, the training en couraged participants to answer questions that emphasized distinguishing features of the target items, as suggested by Hillis (1998) and Nadeau and Rothi (2004). Our findings are consistent with those reported in Sailor et al. (2004) who found that individuals with PAD generated fewe r atypical than typical items. Smith et al. (1995) reported that a group of adults w ith PAD performed poorly when verifying features of low typical (and low dominance, a similar construct related to importance), suggesting that information about the low typi cal items had deteriorated. However, Smith et al. (1995) found that both shared and distinguishing features were affected. The differential effects of PAD on the accessi bility of shared versus distinguishing features can account for the findings in this st udy. Earlier, typical items were described as sharing more features with others in the category compared to atypical items (Rosch, 1975; Rosch & Mervis, 1975). When sets of features co-occur in many items in a category, they are frequently activated t ogether, and, consequently develop strong connections with each other (Collins & Loftus, 1975; Hebb, 1949). This highly interconnected network provides strong re dundancy in the channels through which activation can flow among these connected feat ures, which allows alternate pathways for activation of features, even when there is damage to some of these connections. Thus,

PAGE 115

101 when there is damage to connections am ong the shared features, the behavioral manifestations are not severe because activ ation can still be achieved by the existing conditions, although it might not be as efficient (Altmann et al., 2001; Devlin et al., 1998; Gonnerman et al., 1997; McRae, de Sa, & Se idenberg, 1997). The difficulty with items whose representations consist primarily of sh ared features, however, comes from trying to distinguish one from the other. Distinguish ing features are, by de finition outside the set core of shared features of a category and serve to distinguish one item from another. Distinguishing features occur in only one or possibly a few items within a category, and, thus, have relatively fewer connections th at link them to the shared feature set. Consequently, they are not highly interconne cted with other features and have no redundant or alternative pathways for activat ion to compensate for damage (Altmann et al., 2001; Devlin et al., 1998; Gonnerman et al., 1997). As a result, damage to the connections to distinguishing feat ures increases the possibility that it will be inaccessible, whereas similar damage to connections between shared features is more likely to be compensated for (Altmann et al., 2001; De vlin et al., 1998; G onnerman et al., 1997; Martin, 1992; Sailor et al ., 2004; Smith et al., 1995). In the earlier discussion of the semantic re presentations of atypical items, they were described as having a lower proportion of sh ared features and higher proportion of distinguishing features than t ypical items. As just discus sed, this would make atypical items particularly vulnerable to the diffuse damage of P AD. Furthermore, due to the relative lack of connect ions to distinguishing features and to the lack of connections between distinguishing features atypical items might be mo re difficult to remediate.

PAGE 116

102 In the current study, we suspect that the semantic training used, as well as the emphasis on training distinguishing features, f acilitated access to other typical items by raising the overall activation levels of the core semantic features in the category. Due to the constant stimulation of the connections between the shared features, it likely strengthened their c onnectivity by raising their activa tion thresholds (Altmann et al., 2001; Devlin et al., 1998; Gonnerman et al., 1997). Moreover, we also emphasized distinguishing features for the typical items, allowing participants to better discriminate among them. Atypical items might not have benefited as much as the typical items for a number of reasons. According to the hypothesis, the semantic representations of atypical items contain fewer of the shared features of th e category, and there is no guarantee that the shared features included in the representati on of atypical item 1 are the same shared features included in the repr esentation of atypical item 2. Thus, the degree of feature overlap among atypical items might have been minimal, so there could be no benefit of spreading activation from other trained items in the category. Alternatively, it is possible that the connections to the distinguishing feat ures were not merely just below threshold activation but were damaged and inaccessible, possibly due to infrequency of use as suggested in Smith et al. (1995). This discussion has shown that the findings of the current study are consistent with previous findings on the performance of indi viduals with PAD on semantic tasks, and can be well accounted for by existing theories of semantic representations and their degradation in PAD (Altmann et al., 2001; Devl in et al., 1998; Gonnerman et al., 1997; Martin, 1992; Sailor et al., 2004; Smith et al., 1995). Extending these theories to the

PAGE 117

103 creation of a possible intervention for anom ia in PAD has led to promising results supporting the use of a combination of semantic treatment with word repetition for this population. Whether atypical items will be re mediable in this population, or whether there may even be an advantage for atypical items after further treatment is an open question to be addressed in future studies. Implications for Methodology There are several methodologica l factors that might have influenced the outcomes of our study relative to other st udies investigating the effect s of training typical versus atypical items. As discussed above, some of these include the population being trained (e.g., connectionist networks, adul ts with different types of a phasia or PAD) and type and severity of the damage (Mayer et al., 2004; Stanczak et al., 2005, 2006), as well as the typicality of the items (K iran & Thompson, 2003b; Stanczak et al., 2005, 2006) and the timeframe (Stanczak et al., 2006). Other differences include the type of study (e.g., facilitation versus treatment) and the study design (singlesubject, versus group), as well as the procedure and training tasks. The group de sign used here is very different from the majority of PAD and aphasia studies discu ssed above. There are num erous advantages of the single-subject design, such as, the car eful monitoring of treatment outcomes in relationship to the targeted behavior (K earns, 2000). Group data does not capture the nuances of individual subject performance as well as a sing le subject design. However, the goal of the current study was to determ ine if a facilitation effect (Howard, 1985) could be elicited after this single training, which was a que stion more suited to group design. Knowing that a facilita tion effect is possible in individuals with PAD should encourage researchers to begin te sting more in-depth treatments.

PAGE 118

104 The combination of the repetition (i.e., lexical information) and the semantic training appeared to be bene ficial for our PAD group. Repe tition over a short period was not effective by itself. Yet a treatment eff ect in the PAD anomia study by Fuller et al. (2001) and Rothi et al. (2005) was found afte r many training sessions with repetition in an errorless learning paradigm Similarly, Abrahams and Ca mp (1993) reported that the spaced retrieval technique, which also involve d repetition, resulted in a treatment effect. The possibility of generation effects (Slam ecka & Graf, 1978) has been raised. According to Mitchell et al. (1986) the generation effect has been at tributed to persistence of activation of previously accessed information within semantic memory. Generation effects have been documented in some PAD studies (Fleischman et al., 1997; Multhaup & Balota, 1997) but not in others (Dick, K ean, & Sands, 1989; Mitche ll et al., 1986). Our study found somewhat increased accuracy and faster response times in the unrepeated control items, but these we re not significant. These findings support the use of interventions for anomia in PAD that include both a repetition element and a semantic element to address the two common loci of word finding difficulties in this population. The semantic training task could have also been a factor influencing the performance of our PAD group. The questions were specific to the particular categories and effective but they might not have been as powerful as the feature selection task used for the fluent participants in the Kiran and Thompson (2003b) study. These participants saw a board containing a large number of sema ntic features pertaining to the category. While the set of defining features were shar ed by all of the typical category exemplars from that category, another set was comprised of characteristic features, such that some were more relevant to typical items while ot her features covered a broader spectrum of

PAGE 119

105 information pertinent to atypical items. Cons equently, when participants were asked to choose appropriate features they were exposed to a large set of featur es appropriate to an item. The task demands included choosing not on ly the features that were appropriate to the item, but also rejecting those that were not relevant to the item. It is possible that this, combined with the other trai ning tasks, and the atypical items themselves, allowed the essential differences between training atypical and typical items to be revealed and have their impact on items in a measurable way, in favor of atypical items. This variation of a semantic training task has only been used by Kiran and Thompson (2003b), and it differs considerably from the other semantic trai ning methods used in the other typicality training studies (Mayer et al., 2004; Plaut, 1996; Stanczak et al., 2005, 2006). It is also different from having to self-generate char acteristics of an item, as in our study. We suspect that our training method di d not provide as robust a comparison between items to allow participants to hone in on the broad distinct ions between atypical and typical items (Kiran & Thompson, 2003b) Instead, our method may have put the focus on the individual items themselves. The PAD population might benefit from using the feature selection task, but modified to present fewer pieces of information to avoid overstimulation and confusion from the simu ltaneously presented printed information. Another factor that might have influen ced the performance of our PAD group is the choice of semantic categories. At least tw o studies have shown generalization following training with natural kinds (vegetables a nd birds) (Kiran & Thompson, 2003b; Stanczak et al., 2006). These studies de monstrated evidence for bette r generalization after training atypical items. However, we used artifact ca tegories for training, which has been shown to have more functional information than perceptual features found in natural kinds

PAGE 120

106 (Warrington & Shallice, 1984). Thus, artifact s might have very different internal structures (e.g., for vehicles, there are many subtypes, e.g., 2-wheel, 4-wheel, floating, flying) than the more hierarchically orga nized natural kind categories, like birds and vegetables (Devlin et al., 1998; Gonnerman et al., 1997). Clearly, future studies should contrast the effects of traini ng natural kinds and artifacts. It is difficult to know if training two cat egories simultaneously with two different training types (i.e., Train-Atypi cal, Train-Typical) provided an advantage or not. Both Stanczak et al. (2006) and Mayer et al. (2004) provided traini ng for two categories at the same time and found a treatment effect. Ma yer et al (2004) used a combination of artifacts and natural kinds; they attributed th e lack of generalizati on of training in their participants to the severity of their aphasias. Stanczak et al. (2006) trained birds and vegetables simultaneously, in blocks, and f ound treatment effects, but only one of two participants showed generaliza tion (i.e., to untrained typica l birds after being trained on atypical birds and marginally generalized nami ng of untrained atypical vegetables after typical vegetables were trained). Howe ver, Kiran and Thompson (2003b) showed successful outcomes (i.e., treatment effect and generalization after being trained on atypical items) with adults with fluent apha sia when only training one category (i.e., birds or vegetables) at a time with atypical or with typical exemplar s. Considering these variable results across studies, it is difficult to determine if the number and type of trained categories accounted for any di fferences in the outcomes. A lack of power was certainly an issue, especially for the it ems in the control condition (i.e., untrained items in untrained cat egories). The effects of the small sample size were likely exaggerated by variability among the participants in variables such as

PAGE 121

107 age, education, and severity of impairment (despite being recen tly diagnosed). This variability is causally related to the cha llenges of identifying and recruiting the participants with mild-moderate PAD a nd completing data collection (e.g., the 12 participants with PAD took about four mont hs). There might also have been effects related to other factors such as mental fa tigue, learning curves, a nd practice schedules. Although the clinician offered opportunities for breaks between the pretesting and training, and before the posttesting, it is po ssible that more time was needed for these breaks in order to fully benefit from the trai ning. Anecdotally, severa l of the participants indicated that they were tired at the posttest yet they still enthusiastically expressed interest in and completed thei r participation in the study with out incident. The wide range of differences among our study and the othe r typicality studies warrant caution in generalizing the findings. However, despite th ese methodological challenges, the data has provided meaningful information. Future Studies The findings from our study provide a good foundation for further development of this protocol into an anomia treatment program for PAD. Various issues have been raised and will be considered in our future anomia treatment studies. First, generalization should be programmed prior to starting the study (M cNeil et al., 1998). We also recommend continuing the use of generalization probes fo r untrained atypical items and untrained typical items so that the type of training can be compared with the item type. Again, performance should be compared across condition s, but with a category in which there is no repetition. This would allow for a comp arison between repetition with semantic training versus just semantic training.

PAGE 122

108 In addition, sessions should be scheduled for the screening, and pre and posttesting as well as a follow-up maintenance session. The testing should continue to use the Mini Mental Status Exam (Folstein et al., 1975) but perhaps with mo re strict criteria (18-24). Additional cognitive-linguistic assessments should be used: the Boston Naming Test (Kaplan et al. 1983, 2001), for a standardized measure of confrontation picture naming, the Hopkins Verbal Learning Test (Brandt, 1991) to measur e memory and new learning, and perhaps the Western Aphasia Battery (K ertesz, 1982) for an overall standardized language screening. A pretest picture naming w ith the 72 items should also be used as a screening to determine eligibility for the study perhaps with a cutoff score to identify treatment items based on failed naming attempts at pretest. At the onset of the study, a category verification task could be conduc ted (Stanczak et al., 2005, 2006). Perhaps a few categories and their corresponding exemplar s that are not part of the naming task could be used. The training should be extended into a tr eatment in which the participant attends several sessions per week, fo r several weeks, for example three one hour sessions per week for three weeks. To insure that the participant can correctly name the item, the treatment protocol could st art with the errorless nami ng (i.e., immediate repetition) paradigm (Fuller et al., 2001; Rothi et al., 2005). This could be followed by the current training (i.e., answer the three questions a nd name it). Additional suggestions are to include categories that are m eaningful to the participant (Mayer et al., 2004). Natural kinds should also be included to further examine some of the hypotheses related to artifacts (Warrington & Sha llice, 1984) and intercorrela ted features compared to distinguishing features in PAD (Altmann et al., 2001; Devlin et al., 1998; Gonnerman et

PAGE 123

109 al., 1997). A modified semantic feature ta sk based on Kiran and Thompson (2003b) should be considered, to emphasize the br oad differences between the typical and atypical items in the categories. Semantic featur e distinction with the target items that the participant had difficulty with could also be incorporated into the intervention (Hillis, 1998) to compare and contrast the features Family members or caregivers could be involved and trained to assist with a home program. Perhap s with additional training and over a longer period of time, some of these unresolved questions will be answered and the current protocol will be transformed in to a more potent vehicle for improving naming deficits in adults with PAD. In summary, in a period of less than thr ee hours, we found that adults with early PAD respond well and quickly to a semantic training program. Following training, the PAD participants showed an advantage for t ypical category exemplars, first in faster response times (i.e., a facilitation effect from trained items) and then in accuracy scores (i.e., generalization to untrained items categor ies trained with typi cal exemplars). These preliminary findings suggest training typica l items produced the best results, although this advantage of typical over atypical it ems may change over time (Stanczak et al., 2006). Overall, our findings not only provide su pport for the family resemblance (Rosch, 1975) view, and the prediction by Heilman (per sonal communication, 2005) that typical items would show better generalization, but al so suggest that in PAD, Hebbian learning (Hebb, 1949) is possible. Another importa nt finding is that that there was no generalization to untrained categories, which is highly important for clinicians involved in semantic training of all populations. Thes e findings are extremely encouraging for the development of principled strategies for c hoosing items to encourage generalization in

PAGE 124

110 the remediation of anomia, as well as for th e development of lexical-semantic treatment paradigms for individuals with early PAD.

PAGE 125

111 APPENDIX A INFORMED CONSENT FORM Informed Consent Choosing Remediation Targets for Nami ng Deficits: Effects of Typicality Please read this consent document car efully before you decide to participate in this study. Purpose of the research project The purpose of this study is to test two potential treatments for the naming impairment in Alzheimer disease. Both tr eatments are methodolog ically identical, varying only in the ty pe of stimuli used. What you will be asked to do in the study In this study, you will be asked to comp lete two tasks for the screening phase by answering general knowledge questions, and reading words aloud. For the pretesting there will be two tasks. First, t he examiner will name each of the three semantic categories and ask you to name as many items as you can that belong to each. Second, the exam iner will present pictures and ask you to name them, and then to repeat the correct name of the picture. For the training phase the examiner will present the training pictures and ask you three questions about the item. Then you will be asked to name the pi cture. For the post-testing, you will be asked to do the same tasks as you did for the pre-testing (name as many items as you can in three categories, and name pictures but without repeating the correct name of the pictur e). Responses on this task will be digitally recorded and transcribed for later analysis. During the time that you are participating in the study, we will offer brief breaks, and you can ask for a longer break at any time. All testing will be carried out by the princi pal investigator, Claudia A. Morelli, Dr. Altmann, or trained research assistants. Time required We anticipate it will take 2-3 hours for t he participants with Alzheimer disease to complete the protocol and about 2 hours for the healthy participants.

PAGE 126

112 Risks and Benefits Participating in this study may help pe ople who have difficulties with word finding The risk is minimal, but it is possib le that participants will become fatigued or bored, during the course of the testin g and training. These will be minimized by offering brea ks as needed. Compensation You will be compensated for participating in this study ($20.00 for people with Alzheimer disease, and $15.00 for older adults). Confidentiality Your identity will be kept confidential to the extent provided by law. Your information will be assigned a code number. The list connecting your name to this number will be kept in a locked file in the Language over the Lifespan lab. When the study is completed and the dat a have been analyzed, the list will be destroyed. Your name will not be used in any report. Voluntary participation Your participation in this study is comple tely voluntary. There is no penalty for not participating. Right to withdraw from the study You have the right to withdraw from the study at any time. Whom to contact if you have questions about the study Claudia A. Morelli, M.S., CCC-SLP, Ph.D Candidate, Communication Sciences and Disorders Department, P.O. Box 11742 0, 336 Dauer Hall, University of Florida, Gainesville, FL 32611-7420. cmorelli@ufl.edu 352-392-2113 X232. OR Lori J. Altmann, Ph.D., Communication Sciences and Disorders Department, P.O. Box 117420, 336 Dauer Hall, University of Florida, Gainesville, FL 326117420 laltmann@ufl.edu 352-392-2113 x279 Whom to contact about your rights as a research participant. UFIRB Office, box 112250, University of Fl orida, Gainesville, FL 32611-2250 352-392-0433. Agreement I have read the procedure described above. I voluntarily agree to participate in the procedure and have received a copy of this description. Participant: Date: Principal Investigator: Date:

PAGE 127

APPENDIX B TABLES WITH INFORMATION ABOUT THE PARTICIPANTS

PAGE 128

114 Participant Demographics and Select ed Neuropsychological Test Results Table B-1. Participant Demographics and Selected Ne uropschological Testing Results from the PAD Diagnosis. ID # Age Years of Education Gender MMSE Total Score (Max 30) MMSE Recall (Max 3 items) Boston Naming Test Hopkins Verbal Learning Test: Free Recall (Max 36) Hopkins Verbal Learning Test: Delayed Recall (Max 12) 1001 72 17 F N/A N/A N/A N/A N/A 1002 77 13 F 20/30 N/A N/A N/A N/A 1003 76 10 M 18/30 0/3 10/30 9/36 0/12 1004 92 24 M 19/30 0/3 N/A 9/36 0/12 1005 82 12 F 19/30 0/3 11/19 4/36 0/12 1006 78 12 F 22/30 0/3 14/15 15/36 0/12 1007 55 20 M 25/30 0/3 15/15 20/36 0/12 1008 84 15 F 23/30 0/3 15/15 16/36 0/12 1009 69 18 M 24/30 2/3 15/15 12/36 N/A 1010 86 16 F 21/30 0/3 11/15 10/36 0/12 1011 75 16 M N/A N/A N/A N/A N/A 1012 85 16 F 24/30 2/3 24/30 12/36 0/12 Notes: N/A: Scores not available; MMSE= Mini Mental Status Ex am (Folstein et al., 1975); Hopkins Verbal Learning Test (Brandt, 1991); Boston Naming Test (Kaplan et al., 1983, 2001).

PAGE 129

115Demographics and Performance Before and After Semantic Training Table B-2. Participant Demographics and Pre and Posttest Categor y Generation and Picture Naming Tasks (i.e., Before and After t he Semantic Training, Day of Participation in the Study). ID Number Age Years of Education Gender Pretest Category Generation Posttest Category Generation Pretest Picture Naming (Proportion) Pretest Picture Naming (Max=72) Posttest Picture Naming (Proportion) Posttest Picture Naming (Max=72) 1001 72 17 F 25 25 .74 53/72 .83 60/72 1002 77 13 F 23 23 .81 58/72 .83 60/72 1003 76 10 M 15 15 .71 51/72 .78 56/72 1004 92 24 M 14 14 .51 37/72 .57 41/72 1005 82 12 F 18 17 .60 43/72 .58 42/72 1006 78 12 F 30 36 .81 58/72 .81 58/72 1007 55 20 M 24 34 .89 64/72 .94 68/72 1008 84 15 F 27 37 .83 60/72 .86 60/72 1009 69 18 M 20 26 .81 58/72 .83 60/72 1010 86 16 F 23 33 .76 55/72 .82 59/72 1011 75 16 M 24 31 .83 60/72 .96 69/72 1012 85 16 F 19 22 .60 43/72 .72 52/72 Group Mean 77.6 15.8 21.8 26.1 .74 53/72 .79 57/72 Notes: Category Generation is the total across the 3 categories (vehicles, clothing and tools); Picture Naming is proportion co rrect, and Total Correct, Maximum=72.

PAGE 130

116 APPENDIX C TABLES WITH INFORMATION ABOUT THE STIMULI Master List of Stimulus Items Used in the Study Table C-1. Master List of Stimulus Items Used in the Study. Trained Items are Italicized Number Item Name Typicality Rank Typicality Score Written Word Frequency Familiarity Rating Phoneme Length TY 1 Automobile 1 1.02 74 456 8 2 station wagon 2 1.14 0 11 3 truck 3 1.17 80 620 4 4 bus 5.5 1.27 42 3 5 taxi 5.5 1.27 19 5 6 jeep 7 1.35 16 564 3 7 motorcycle 9 1.65 0 8 8 van 11 1.95 2 542 3 9 train 14 2.15 86 548 4 10 bicycle 16 2.51 7 6 11 airplane 18 2.64 21 6 12 boat 20 2.75 123 584 3 AT 13 ship 22 2.82 126 553 3 14 scooter 23 3.24 1 468 5 15 tractor 24 3.3 31 518 6 16 wagon 25 3.31 72 443 5 17 wheelchair 29 3.68 0 6 18 tank 31 3.84 30 511 4 19 rowboat 33 3.92 0 5 20 tricycle 35 4 0 436 7 21 canoe 36 4.01 8 441 4 22 raft 37 4.37 5 483 4 23 submarine 38 4.51 35 450 7 24 blimp 42 4.81 1 5 TY 25 pants 1 1.12 9 575 5 26 shirt 2.5 1.14 29 612 3 27 dress 2.5 1.14 63 588 4 28 skirt 4 1.21 22 551 4 29 blouse 5 1.27 2 562 4 30 suit 6 1.49 64 543 3 31 jacket 8 1.68 39 596 5 32 coat 9 1.88 52 610 3 33 sweater 10 1.89 18 5 34 socks 16 2.08 7 4 35 parka 17 2.19 0 5 36 pajamas 18 2.25 3 7 AT 37 bathing suit 22.5 2.44 0 8 38 bath robe 24 2.65 3 6 39 shoes 27 2.73 44 3 40 vest 30 2.81 4 472 4 41 boots 33 3.42 1 4

PAGE 131

117 Table C-1. Continued Number Item Name Typicality Rank Typicality Score Written Word Frequency Familiarity Rating Phoneme Length 42 sandals 34 3.56 5 6 43 tie 35 3.71 27 559 2 44 belt 37 3.93 36 550 4 45 scarf 38 3.96 4 5 46 mittens 39 3.99 2 5 47 hat 40 4.08 71 580 3 48 gloves 43 4.53 6 5 TY 49 saw 1 1.04 8 552 2 50 hammer 2 1.34 6 515 4 51 ruler 3 1.48 13 571 4 52 screwdriver 4 1.56 1 544 9 53 drill 5 1.59 21 473 4 54 nails 6 1.67 20 4 55 tape measure 7 1.69 0 7 56 sawhorse 8 1.77 1 6 57 level 11 1.82 50 504 4 58 toolbox 14 2.12 1 7 59 T-square 15 2.22 0 7 60 chisel 16 2.26 5 469 4 AT 61 pliers 22 2.59 1 499 5 62 wrench 23 2.6 1 4 63 ladder 24 2.64 19 507 4 64 vise 25 2.76 1 368 3 65 screws 26 2.77 10 5 66 awl 33 3.09 0 257 2 67 crowbar 34.5 3.12 0 6 68 bolts 40 3.63 1 5 69 paintbrush 45 3.81 1 8 70 stapler 48 4.21 0 6 71 axe 53 4.53 19 461 3 72 scissors 59 5.36 1 559 5 Notes: TY=Typical items; AT=Typical items. Typicality Rank and Score ; Written Word Frequency ; Familiarity

PAGE 132

118 Alternate Names for Stimulus Items Used in the Study Table C-2. Master List of Stimulus Items Used in the Study. Alternate names for stimuli counted as correct. Number Item Name Alternate Names Counted as Correct TY 1 Automobile Car, sedan 2 station wagon Volvo 3 truck Pick-up, semi, 18 wheeler; semi-truck, tractor trailer 4 bus 5 taxi Taxi cab, cab 6 jeep 7 motorcycle 8 van 9 train 10 bicycle Bike 11 airplane Plane, jet 12 boat Tugboat AT 13 ship Cruiseliner, oc eanliner, steamship 14 scooter Moped 15 tractor 16 wagon 17 wheelchair 18 tank 19 rowboat 20 tricycle 21 canoe 22 raft 23 submarine 24 blimp Derigible, zeppelin TY 25 pants 26 shirt Dress shirt 27 dress 28 skirt 29 blouse 30 suit 31 jacket 32 coat Winter coat 33 sweater Turtleneck sweater, turtleneck 34 socks 35 parka 36 pajamas AT 37 bathing suit Swimsuit 38 bath robe Robe 39 shoes 40 vest 41 boots 42 sandals 43 tie Necktie 44 belt 45 scarf 46 mittens 47 hat Fedora 48 gloves

PAGE 133

119 Table C-2. Continued Number Item Name Alternate Names Counted as Correct TY 49 saw 50 hammer 51 ruler 52 screwdriver 53 drill Electric drill 54 nails 55 tape measure Measuring tape 56 sawhorse 57 level 58 toolbox Toolkit 59 T-square 60 chisel AT 61 pliers 62 wrench 63 ladder 64 vise 65 screws 66 awl 67 crowbar 68 bolts 69 paintbrush 70 stapler 71 axe 72 scissors Notes: TY=Typical items; AT=Typical items.

PAGE 134

120 LIST OF REFERENCES Abeysinghe, S. C., Bayles, K. A., & Tro sset, M. W. (1990). Semantic memory deterioration in Alzheimer's subjects: Evidence from word association, definition, and associate ranking tasks. Journal of Speech and Hearing Research, 33 (3), 574582. Abrahams, J. P., & Camp, C. J. (1993). Main tenance and generalization of object naming training in anomia associated with degenerative dementia. Clinical Gerontologist, 12 (3), 57-72. Alathari, L., Trinh Ngo, C., & Dopkins, S. (2004). Loss of distinctive features and a broader pattern of priming in Alzheimer's Disease. Neuropsychology, 18 (4), 603612. Albert, M. S., & Milberg, W. (1989). Seman tic processing in patients with Alzheimer's disease. Brain and Language, 37 (1), 163-171. Altmann, L. J., Kempler, D., & Andersen, E. S. (2001). Speech errors in Alzheimer's disease: Reevaluating mo rphosyntactic preservation. Journal of Speech, Language, and Hearing Research, 44 (5), 1069-1082. Anderson, J. (1991). The adaptive nature of human categorization. Psychological Review, 98 (3), 409-429. Appell, J., Kertesz, A., & Fisman, M. (1982). A study of language functioning in Alzheimer patients. Brain and Language, 17 (1), 73-91. ASHA. (American Speech-Language-Hearing Association). (In press). Roles of speechlanguage pathologists in the identification, diagnosis, and treatment of individuals with cognitive-co mmunication disorders: Po stion statement. ASHA Supplement, 25 Au, R., & Bowles, N. (1991). Memory influences on language in normal aging. In D. Ripich (Ed.), Handbook of geriatric communication disorders (pp. 293-305). Austin: Pro-Ed, Inc. Bandur, D. L., & Shewan, C. M. ( 2001). Language-Oriented Treatment: A psycholinguistic approach to aphasia. In R. Chapey (Ed.), Language intervention strategies in aphasia and related neurogenic communication disorders (pp. 629662). Philadelphia: Lippincott, Williams & Wilkins.

PAGE 135

121 Barbarotto, R., Capitani, E., Jori, T., Laiacona, M., & Moli nari, S. (1998). Picture naming and progression of Alzheimer's diseas e: An analysis of error types. Neuropsychologia, 36 (5), 397-405. Barker, M. G., & Lawson, J. S. ( 1968). Nominal aphasia in dementia. British Journal of Psychiatry, 114 (516), 1351-1356. Barr, R. A., & Caplan, L. J. (1987). Categor y representations and their implications for category structure. Memory and Cognition, 15 (5), 397-418. Bayles, K. A. (1982). Language function in senile dementia. Brain and Language, 16 (2), 265-280. Bayles, K. A., Kasniak, A. W., & Tomoeda, C. K. (1987). Communication and cognition in normal aging and dementia Austin, TX: Pro-Ed, Inc. Bayles, K. A., & Tomoeda, C. K. (1983). C onfrontation naming impairment in dementia. Brain and Language, 19 (1), 98-114. Bayles, K. A., Tomoeda, C. K., & Cruz, R. F. (1999). Performance of Alzheimer's disease patients in judging word relatedness. Journal of the International Neuropsychological Society, 5 (7), 668-675. Bell, E. E., Chenery, H. J., & Ingram, J. C. (2001). Semantic priming in Alzheimer's dementia: Evidence for dissociation of automatic and attentional processes. Brain and Language, 76 (2), 130-144. Benson, D. F. (1979). Neurologic correlates of anomia. In H. Whitaker & H. A. Whitaker (Eds.), Studies in Neurolinguistics (Vol. 4). New York: Academic Press. Benson, D. F., & Geschwind, N. (1985). Apha sia and related disorders: A clinical approach. In M. M. Mesulam (Ed.), Principles of Behavioral Neurology (pp. 193238). Philadelphia: F.A. Davis. Benton, A. L. (1968). Differential behavior al effects in frontal lobe disease. Neuropsychologia, 6 53. Bowles, N. L., Obler, L. K., & Albert, M. L. (1987). Naming errors in healthy aging and dementia of the Alzheimer type. Cortex, 23 (3), 519-524. Boyle, M. (2004). Semantic feature analysis treatment for anomia in two fluent aphasia syndromes. American Journal of Speech -Language Pathology, 13 (3), 236-249. Boyle, M., & Coelho, C. A. (1995). Applicat ion of semantic feature analysis as a treatment for aphasic dysnomia. American Journal of Speech-Language Pathology, 4 94-98.

PAGE 136

122 Brandt, J. (1991). The Hopkins Verbal Lear ning Test: Development of a new memory test with six equivalent forms. The Clinical Neuropsychologist, 5 125-142. Chertkow, H., & Bub, D. (1990). Semantic memo ry loss in dementia of Alzheimer's type. What do various measures measure? Brain, 113 397-417. Chertkow, H., Bub, D., & Caplan, D. (1992). C onstraining theories of semantic memory processing: Evidence from dementia. Cognitive Neuropsychology, 9 (4), 327-365. Chertkow, H., Bub, D. N., & Seidenberg, M. S. (1989). Priming and semantic memory loss in Alzheimer's disease. Brain and Language, 36 (3), 420-446. Chieffi, S., Carlomagno, S., Silveri, M. C ., & Gainotti, G. (1989). The influence of semantic and perceptual factors on lexi cal comprehension in aphasic and right brain-damaged patients. Cortex, 25 (4), 591-598. Chui, H. C. (1989). Dementia. A review em phasizing clinicopathologic correlation and brain-behavior relationships. Archive of Neurology, 46 (7), 806-814. Cobb, R. E. (2005). The bottom-up deterioration hypothe sis of the semantic impairment associated with Alzheimer's disease: The effect of typicality on object verification. University Microfilms International. Coelho, C. A., McHugh, R. E., & Boyle, M. (2000). Semantic feature analysis as a treatment for aphasic dysnomia: A replication. Aphasiology, 14 (2), 133-142. Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82 (6), 407-428. Cronin-Golomb, A., Keane, M. M., Kokodis, A., Corkin, S., & Growdon, J. H. (1992). Category knowledge in Alzheimer's diseas e: Normal organization and a general retrieval deficit. Psychology and Aging, 7 (3), 359-366. Davis, A., & Pring, T. (1991). Therapy for word -finding deficits: More on the effects of semantic and phonological approaches to treatment with dysphasic patients. Neuropsychological Rehabilitation, 1 (2), 135-145. Dell, G. S., & O'Seaghdha, P. G. (1992). Stag es of lexical access in language production. Cognition, 42 (1-3), 287-314. Devlin, J. T., Gonnerman, L. M., Andersen, E. S., & Seidenberg, M. S. (1998). Categoryspecific semantic deficits in focal and widespread brain damage: A computational account. Journal of Cognitive Neuroscience, 10 (1), 77-94. Dick, M. B., Kean, M. L., & Sands, D. (1989). Memory for internally generated words in Alzheimer-type dementia: Breakdown in encoding and semantic memory. Brain & Cognition, 9 (1), 88-108.

PAGE 137

123 Drew, R. L., & Thompson, C. K. (1999). M odel-based semantic treatment for naming deficits in aphasia. Journal of Speech, Language, and Hearing Research, 42 (4), 972-989. Fabiani, M., Gratton, G., & Coles, M. G. H. (2000). Event-related brain potentials methods, theory, and applications. In J. T. Cacioppo, L. G. Tassinary & G. G. Berntson (Eds.), Handbook of Psychophysiology (2nd ed.). Cambridge: Cambridge University Press. Fleischman, D. A., Gabrieli, J. D., Rinaldi, J. A., Reminger, S. L., Grinnell, E. R., Lange, K. L., et al. (1997). Word-stem comp letion priming for perceptually and conceptually encoded words in patients with Alzheimer's disease. Neuropsychologia, 35 (1), 25-35. Flicker, C., Ferris, S. H., Crook, T., & Bartus R. T. (1987). Implications of memory and language dysfunction in the naming deficit of senile dementia. Brain and Language, 31 (2), 187-200. Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). "Mini-mental state". A practical method for grading the cognitive st ate of patients for the clinician. Journal of Psychiatric Research, 12 (3), 189-198. Ford, J. M., Askari, N., Mathalon, D. H., Menon, V., Gabrieli, J. D., Tinklenberg, J. R., et al. (2001). Event-related brain poten tial evidence of spared knowledge in Alzheimer's disease. Psychology and Aging, 16 (1), 161-176. Frances, N., & Kucera, H. (1982). Frequency analysis of English usage Boston: Houghton Mifflin. Francis, D. R., Clark, N., & Humphreys, G. W. (2002). Circumlocution-induced naming (CIN): A treatment for effecting generalisation in anomia?, Aphasiology (Vol. 16, pp. 243-259): Psychology Press (T&F). Freed, D. B., Marshall, R. C., & Nippold, M. A. (1995). Comparison of personalized cueing and provided cueing on the facil itation of verbal labeling by aphasic subjects. Journal of Speech and Hearing Research, 38 (5), 1081-1090. Fujihara, N., Nageishi, Y., Koyama, S., & Nakajima, Y. (1998). Electrophysiological evidence for the typicality effect of human cognitive categorization. International Journal of Psychophysiology, 29 (1), 65-75. Fuller, R. H., Kendall, D. L., Nadeau, S. E., Spevack, A. A., Heilman, K., & Gonzalez Rothi, L. J. (2001). Cognitive-cholinergi c therapy of anomia in Alzheimer's disease. Neurology, 56 (Supplement 3), A57. Gainotti, G. (1981). The relati onship between type of naming error and semantic-lexical discrimination in aphasic patients. Cortex, 17 (3), 401-409.

PAGE 138

124 Garrard, P., Lambon Ralph, M. A., Patterson, K ., Pratt, K. H., & Hodges, J. R. (2005). Semantic feature knowledge and picture naming in dementia of Alzheimer's type: A new approach. Brain and Language, 93 (1), 79-94. Gerratt, B. R., & Jones, D. (1987). Apha sic performance on a lexical decision task: Multiple meanings and word frequency. Brain and Language, 30 (1), 106-115. Gierut, J. A. (2001). Complexity in P honological Treatment: Clinical Factors. Language, Speech, & Hearing Services in Schools, 32 (4), 229. Giffard, B., Desgranges, B., Nore-Mary, F., Lale vee, C., Beaunieux, H., de la Sayette, V., et al. (2002). The dynamic time course of semantic memory impairment in Alzheimer's disease: Clues from hyperpriming and hypopriming effects. Brain, 125 (Pt 9), 2044-2057. Giffard, B., Desgranges, B., Nore-Mary, F., Lalev ee, C., de la Sayette, V., Pasquier, F., et al. (2001). The nature of semantic memory deficits in Alzhei mer's disease: New insights from hyperpriming effects. Brain, 124 (Pt 8), 1522-1532. Gonnerman, L. M., Andersen, E. S., Devlin, J. T., Kempler, D., & Seidenberg, M. S. (1997). Double dissociation of semantic categories in Alzheimer's disease. Brain and Language, 57 (2), 254-279. Grober, E., Buschke, H., Kawas, C., & Fuld, P. (1985). Impaired ranking of semantic attributes in dementia. Brain and Language, 26 (2), 276-286. Grober, E., Perecman, E., Kellar, L., & Brow n, J. (1980). Lexical knowledge in anterior and posterior aphasics. Brain and Language, 10 (2), 318-330. Grossman, M. (1981). A bird is a bird is a bird: Making referen ce within and without superordinate categories. Brain and Language, 12 (2), 313-331. Grossman, M., McMillan, C., Moore, P., Ding, L., Glosser, G., Wor k, M., et al. (2004). What's in a name: Voxel-based morpho metric analyses of MRI and naming difficulty in Alzheimer's disease, fron totemporal dementia and corticobasal degeneration. Brain, 127 628-649. Grossman, M., Robinson, K., Biassou, N., Wh ite-Devine, T., & D'Esposito, M. (1998). Semantic memory in Alzheimer's disease: Representativeness, ontologic category, and material. Neuropsychology, 12 (1), 34-42. Hampton, J. A. (1995). Testing the Prototype Theory of concepts. Journal of Memory & Language, 34 (5), 686-708. Hartman, M. (1991). The use of semantic know ledge in Alzheimer's disease: Evidence for impairments of attention. Neuropsychologia, 29 (3), 213-228.

PAGE 139

125 Hebb, D. O. (1949). The organization of behavior: A neuropsychological theory New York: John Wiley & Sons, Inc. Heilman, K. M. (2005). Aphasic disorders associated with dementia. Paper presented at the 31st Annual Course in Behavioral Neurology and Neur olopsychology, Lake Buena Vista, FL. Henderson, V. W., & Finch, C. E. (1989). Th e neurobiology of Alzheimer's disease. Journal of Neurosurgery, 70 (3), 335-353. Henderson, V. W., Mack, W., Freed, D. M., Kempler, D., & Andersen, E. S. (1990). Naming consistency in Alzheimer's disease. Brain and Language, 39 (4), 530-538. Hickin, J., Best, W., Herbert, R., Howard, D., & Osborne, F. (2001). Treatment of word retrieval in aphasia: Generali zation to conversational speech, International Journal of Language & Communication Disorders (Vol. 36, pp. 13): Royal College of Speech & Language Therapists. Hickin, J., Best, W., Herbert, R., Howa rd, D., & Osborne, F. (2002). Phonological therapy for word-finding difficulties: A re-evaluation. Aphasiology, 16 (10/11), 981-999. Hillis, A. E. (1998). Treatment of naming disord ers: New issues regarding old therapies. Journal of the International Neuropsychological Society, 4 648-660. Hodges, J. R., & Patterson, K. (1995). Is sema ntic memory consisten tly impaired early in the course of Alzheimer's disease? Neur oanatomical and diagnostic implications. Neuropsychologia, 33 (4), 441-459. Hodges, J. R., Patterson, K., Graham, N., & Dawson, K. (1996). Naming and knowing in dementia of the Alzheimer's type. Brain and Language, 54 (2), 302-325. Hodges, J. R., Salmon, D. P., & Butters, N. ( 1991). The nature of the naming deficit in Alzheimer's and Huntington's disease. Brain, 114 1547-1558. Hodges, J. R., Salmon, D. P., & Butters, N. (1992). Semantic memory impairment in Alzheimer's disease: Failure of access or degraded knowledge? Neuropsychologia, 30 (4), 301-314. Howard, D. (1985). The facilitation of picture naming in aphasia. Cognitive Neuropsychology, 2 (1), 49-80. Howard, D., Patterson, K., Franklin, S., Orchard-Lisle, V., & Morton, J. (1985). Treatment of word retrieval deficits in aphasia. A comparison of two therapy methods. Brain, 108 (4), 817-829. Huff, F. J., Corkin, S., & Growdon, J. H. (1986). Semantic impairment and anomia in Alzheimer's disease. Brain and Language, 28 (2), 235-249.

PAGE 140

126 Huff, F. J., Mack, L., Mahlmann, J., & Gree nberg, S. (1988). A comparison of lexicalsemantic impairments in left hemisphere stroke and Alzheimer's disease. Brain and Language, 34 (2), 262-278. Jarvis, B. (2004). DirectRT. New York: Empirisoft. Jolicoeur, P., Gluck, M. A., & Kosslyn, S. M. (1984). Pictures and names: Making the connection. Cognitive Psychology, 16 (2), 243-275. Kaplan, E. F., Goodglass, H., & Wein traub, S. (1983). Boston Naming Test. Philadelphia: Lea and Febiger. Katzman, R., & Fox, P. J. (1999). The world-wi de impact of dementia: Projections of prevalence and costs. In R. Mayeux & Y. Christen (Eds.), Epidemiology of Alzheimer's disease: From gene to prevention (pp. 1-17). Heidelberg: SpringerVerlag. Kawas, C. H., & Katzman, R. (1999). Epidemio logy of dementia and Alzheimer disease. In R. D. Terry, R. Katzman, K. L. Bick & S. S. Sangram (Eds.), Alzheimer Disease (2nd ed.). Philadelphia: Lippincott Williams & Wilkins. Kearns, K. P. (1989). Generali zation issues in the treatmen t of communication disorders. In L. V. McReynolds & J. E. Spradlin (Eds.), Methodologies for studying generalization (pp. 13-30). Toronto, Canada: B.C. Decker. Kearns, K. P. (2000). Single-subject experimental designs in aphasia. In S. E. Nadeau, L. J. Gonzalez Rothi & B. Crosson (Eds.), Aphasia and language (pp. 421-441). New York: Guilford. Kertesz, A. (1982). The Western Aphasi a Battery. New York: Grune & Stratton. Kiran, S., & Thompson, C. K. (2003a). Effect of typicality on online category verification of animate category exemplars in aphasia. Brain and Language, 85 (3), 441-450. Kiran, S., & Thompson, C. K. (2003b). The role of semantic complexity in treatment of naming deficits: Training semantic cate gories in fluent aphasia by controlling exemplar typicality. Journal of Speech, Language, and Hearing Research, 46 (3), 608-622. Kirshner, H. S., Webb, W. G., & Kelly, M. P. (1984). The naming disorder of dementia. Neuropsychologia, 22 (1), 23-30. Koemeda-Lutz, M., Cohen, R., & Meier, E. (1987). Organization of and access to semantic memory in aphasia. Brain and Language, 30 (2), 321-337. Komatsu, L. K. (1992). Recent views of conceptual structure. Psychological Bulletin, 112 (3), 500-526.

PAGE 141

127 Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207 (4427), 203-205. Le Dorze, G., Boulay, N., Gaudreau, J., & Bra ssard, C. (1994). The contrasting effects of a semantic versus a formal-semantic techni que for the facilitation of naming in a case of anomia. Aphasiology, 8 (2), 127-141. Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. Behavioral and Brain Science, 22 (1), 1-38. Lipinska, B., & Backman, L. (1997). Encoding-re trieval interactions in mild Alzheimer's disease: The role of access to categorical information. Brain and Cognition, 34 (2), 274-286. Lowell, S., Beeson, P. M., & Holland, A. L. (1995). The efficacy of a semantic cueing procedure on naming performance of adults with aphasia. American Journal of Speech-Language Pathology, 4 109-114. Lukatela, K., Malloy, P., Jenkins, M., & Cohe n, R. (1998). The naming deficit in early Alzheimer's and vascular dementia. Neuropsychology, 12 (4), 565-572. Margolin, D. I., Pate, D. S., & Friedrich, F. J. (1996). Lexical priming by pictures and words in normal aging and in dementia of the Alzheimer's type. Brain and Language, 54 (2), 275-301. Margolin, D. I., Pate, D. S., Fr iedrich, F. J., & Elia, E. (1990). Dysnomia in dementia and in stroke patients: Different underlying cognitive deficits. Journal of Clinical and Experimental Neuropsychology, 12 (4), 597-612. Marshall, J. C. (1988). Cognitive neuropsychology. Sensation and semantics. Nature, 334 (6181), 378. Marshall, R. C., Freed, D. B., & Karow, C. M. (2001). Learning of subordinate category names by aphasic subjects: A comparison of deep and surface-level training methods. Aphasiology, 15 (6), 585-598. Martin, A. (1992). Degraded knowledge repres entations in patients with Alzheimer's disease: Implications for models of se mantic and repetition priming. In L. R. Squire & N. Butters (Eds.), Neuropsychology of memory (2nd ed., pp. 220-232). New York: Guilford. Martin, A., & Fedio, P. (1983). Word produc tion and comprehension in Alzheimer's disease: The breakdown of semantic knowledge. Brain and Language, 19 (1), 124141.

PAGE 142

128 Martin, N., Laine, M., & Harley, T. A. ( 2002). How can connectionist cognitive models of language inform models of language re habilitation? In A. E. Hillis (Ed.), The handbook of adult language disorders: In tegrating cognitive neuropsychology, neurology, and rehabilitation (pp. 375-396). New York: Psychology Press. Mayer, J. F., Murray, L. L., & Karcher, L. (2004, June, 2004). Treatment of anomia in severe aphasia. Paper presented at the Clinic al Aphasiology Conference, Park City, UT. McCarthy, R., & Warrington, E. K. (1990) The dissolution of semantics. Nature, 343 (6259), 599. McKhann, G., Drachman, D., Folstein, M., Ka tzman, R., Price, D., & Stadlan, E. M. (1984). Clinical diagnosis of Alzheime r's disease: Report of the NINCDSADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology, 34 (7), 939-944. McNeil, M. R., Doyle, P. J., Spencer, K., G oda, A. J., Flores, D., & Small, S. (1998). Effects of training multiple form cl asses on acquisition, generalization and maintenance of word retrieval in a single subject. Aphasiology, 12 (7/8), 575-585. McRae, K., de Sa, V. R., & Seidenberg, M. S. (1997). On the nature and scope of featural representations of word meaning. Journal of Experimental Psychology: General, 126 (2), 99-130. McReynolds, L. V. (1989). Generalization i ssues in the treatm ent of communication disorders. In L. V. McReynolds & J. E. Spradlin (Eds.), Generalization strategies in the treatment of communication disorders (pp. 1-12). Toronto, Canada: B.C. Decker. Medin, D. L., & Smith, E. E. (1984). Concepts and concept formation. Annual Review of Psychology, 35 113-138. Meyer, D. E., & Schvaneveldt, R. W. (1971). Facilitation in recogni zing pairs of words: Evidence of a dependence be tween retrieval operations. Journal of Experimental Psychology, 90 (2), 227-234. Mitchell, D. B., Hunt, R. R., & Schmitt, F. A. (1986). The generation effect and reality monitoring: Evidence from dementia and normal aging. Journal of Gerontology, 41 (1), 79-84. Montanes, P., Goldblum, M. C., & Boller, F. (1995). The naming impairment of living and nonliving items in Alzheimer's disease. Journal of the International Neuropsychological Society, 1 (1), 39-48. Multhaup, K. S., & Balota, D. A. (1997). Generation effects and source memory in healthy older adults and in adults wi th dementia of the Alzheimer type. Neuropsychology, 11 (3), 382-391.

PAGE 143

129 Murray, L. L., & Clark, H. M. (2005). Neurogenic disorders of language: Theory driven clinical practice Clifton Park, NY: Thompson Delmar Learning. Nadeau, S. E. (2000). Connectionist models a nd language. In S. E. Nadeau, L. J. Rothi & B. Crosson (Eds.), Aphasia and language: Theory to practice (pp. 299-347). New York: Guilford Press. Nadeau, S. E., & Gonzalez Rothi, L. J. (2004). Rehabilitation of language disorders. In J. Ponsford (Ed.), Cognitive and Behavioral Rehabilitation: From Neurobiology to Clinical Practice (pp. 129-174). New York: The Guilford Press. Nebes, R. D. (1989). Semantic memory in Alzheimer's disease. Psychological Bulletin, 106 (3), 377-394. Nebes, R. D. (1992). Semantic memory dysfunc tion in Alzheimer's disease: Disruption of semantic knowledge or information proces sing limitation? In L. R. Squire & N. Butters (Eds.), Neuropsychology of memory (2nd ed., pp. 233-240). New York: Guilford Press. Nebes, R. D., Boller, F., & Holland, A. (1986). Use of semantic context by patients with Alzheimer's disease. Psychology and Aging, 1 (3), 261-269. Nebes, R. D., & Brady, C. B. (1990). Preserve d organization of semantic attributes in Alzheimer's disease. Psychology and Aging, 5 (4), 574-579. Nebes, R. D., Brady, C. B., & Huff, F. J. (1989). Automatic and attentional mechanisms of semantic priming in Alzheimer's disease. Journal of Clinical and Experimental Neuropsychology, 11 (2), 219-230. Nebes, R. D., Martin, D. C., & Horn, L. C. (1984). Sparing of semantic memory in Alzheimer's disease. Journal of Abnormal Psychology, 93 (3), 321-330. Neils, J., Brennan, M. M., Cole, M., Boller, F., & Gerdeman, B. (1988). The use of phonemic cueing with Alzheimer's disease patients. Neuropsychologia, 26 (2), 351-354. Nicholas, M., Obler, L., Albert, M., & Goodglas s, H. (1985). Lexical retrieval in healthy aging. Cortex, 21 (4), 595-606. Nickels, L. (2002). Therapy for naming disord ers: Revisiting, revi sing, and reviewing. Aphasiology, 16 (10/11), 935-979. Ober, B. A. (1999). Semantic memory of Alzheimer's disease: Loss of knowledge or deficits in retrieval? Introducti on from the symposium organizer. Journal of the International Neuropsyc hological Society, 5 (7), 623-625.

PAGE 144

130 Ober, B. A., Dronkers, N. F., Koss, E., Delis, D. C., & Friedland, R. P. (1986). Retrieval from semantic memory in Alzheimer-type dementia. Journal of Clinical and Experimental Neuropsychology, 8 (1), 75-92. Ober, B. A., & Shenaut, G. K. (1995). Seman tic priming in Alzheimer's disease: Metaanalysis and theoretical evaluation. In P. A. Allen & T. R. Bashore (Eds.), Age differences in word and language processing (pp. 247-271). Amsterdam, Netherlands: Elsevier. Ober, B. A., & Shenaut, G. K. (1999). Well-organized conceptual domains in Alzheimer's disease. Journal of the International Neuropsychological Society, 5 (7), 676-684. Ober, B. A., Shenaut, G. K., & Reed, B. R. (1995). Assessment of asso ciative relations in Alzheimer's disease: Evidence for preservation of semantic memory. Aging and Cognition, 2 (4), 254-267. Olympus. (2002). DM-10 [Digital Voice Recorder]: Olympus Corporation. Ousset, P. J., Viallard, G., Puel, M., Celsis P., Demonet, J. F., & Cardebat, D. (2002). Lexical therapy and episodic word learni ng in dementia of the Alzheimer type. Brain and Language, 80 (1), 14-20. Patterson, K., Purell, C., & Morton, J. (1983). The facilitation of word retrieval in aphasia. In C. Code & D. J. Muller (Eds.), Aphasia therapy London: Edward Arnold. Petersen, R. C., Stevens, J. C., Ganguli, M., Tangalos, E. G., Cummings, J. L., & DeKosky, S. T. (2001). Practice parameter: Early detection of dementia: Mild cognitive impairment (an evidence-based review). Report of the Quality Standards Subcommittee of the American Academy of Neurology. Neurology, 56 (9), 1133-1142. Plaut, D. C. (1996). Relearning after damage in connectionist networks: Toward a theory of rehabilitation. Brain and Language, 52 (1), 25-82. Posner, M. I., & Keele, S. W. (1968). On the genesis of abstract ideas. Journal of Experimental Psychology, 77 (3), 353-363. Pring, T., Hamilton, A., Harwood, A., & MacBri de, L. (1993). Generalization of naming after picture/word matching: Only it ems appearing in therapy benefit. Aphasiology, 7 (4), 383-394. Quillian, M. R. (1967). Word concepts: A theory and simulation of some basic semantic capabilities. Behavioral Science, 12 (5), 410-430.

PAGE 145

131 Raymer, A. M., Foundas, A. L., Maher, L. M., Greenwald, M. L., Morris, M., Rothi, L. J., et al. (1997). Cognitive neuropsyc hological analysis and neuroanatomic correlates in a case of acute anomia. Brain and Language, 58 (1), 137-156. Raymer, A. M., & Gonzalez Rothi, L. J. (2001). The semantic system. In S. E. Nadeau & L. J. Gonzalez Rothi (Eds.), Language and Aphasia Raymer, A. M., Maher, L. M., Greenwald, M. L., Morris, M., Gonzalez-Rothi, L., & Heilman, K. M. (1990). The Florida Se mantic Battery: Experimental Edition. Gainesville, Florida. Raymer, A. M., Thompson, C. K., Jacobs, B., & le Grand, H. R. (1993). Phonological treatment of naming deficits in aphasi a: Model-based generalization analysis. Aphasiology, 7 (1), 27-53. Rips, L. J., Shoben, E. J., & Smith, E. E. ( 1973). Semantic distance and the verification of semantic relations. Journal of Verbal Learning & Verbal Behavior, Vol. 12 (1), 120. Rosch, E. (1973a). On the internal structure of perceptual and semantic categories. In T. E. Moore (Ed.), Cognitive development and the acquisition of language New York: Academic Press. Rosch, E. (1973b). On the internal structure of perceptual and semantic categories. Journal of Experimental Psychology: General, 104 ( 192-233). Rosch, E. (1975). Cognitive represen tations of semantic categories. Journal of Experimental Psychol ogy: General, 104 (3), 192-233. Rosch, E. (1977). Human categoriz ation. In N. Warren (Ed.), Advances in cross-cultural psychology (Vol. 1, pp. 1-72). London: Academic Press Ltd. Rosch, E. (1978). Principles of categoriza tion. In E. Rosch & B. B. Llyod (Eds.), Cognition and categorization (pp. 27-48). Hillsdale, NJ: Erlbaum. Rosch, E., & Mervis, C. B. (1975). Family rese mblances: Studies in the internal structure of categories. Cognitive Psychology, 7 (4), 573-605. Rosenberg, R. N. (2005). Translational resear ch on the way to effective therapy for Alzheimer disease. Archives of General Psychiatry, 62 (11), 1186-1192. Rothi, L. J., Fuller, R. H., Kendall, D. L., Nadeau, S., Leon, S., Wu, S., et al. (2005). Naming treatment using an errorless lear ning technique: A possible adjuvant to donepezil for language dysfunction in patients with Alzheimer's disease. Submitted for publication

PAGE 146

132 Rumelhart, D. E., McClelland, J. L., & PDP Research Group. (1986). Parallel distributed processing: Explorations in th e microstructure of cognition Cambridge, Mass.: MIT Press. Sailor, K., Antoine, M., Diaz, M., Kuslans ky, G., & Kluger, A. (2004). The effects of Alzheimer's disease on item output in verbal fluency tasks. Neuropsychology, 18 (2), 306-314. Salmon, D. P., Butters, N., & Chan, A. S. ( 1999). The deterioration of semantic memory in Alzheimer's disease. Canadian Journal of Expe rimental Psychology, 53 (1), 108-116. Salmon, D. P., Heindel, W. C., & Lange, K. L. (1999). Differential decline in word generation from phonemic and semantic categories during the course of Alzheimer's disease: Implications fo r the integrity of semantic memory. Journal of the International Neur opsychological Society, 5 (7), 692-703. Salmon, D. P., Shimamura, A. P., Butters, N ., & Smith, S. (1988). Lexical and semantic priming deficits in patients with Alzheimer's disease. Journal of Clinical and Experimental Neuropsychology, 10 (4), 477-494. Schmidt, A. M., & Lee, T. D. (1999). Motor control and learning: A behavioral emphasis Champaign, IL: Human Kinetics. Schnur, T. T., Schwartz, M. F., Brech er, A., & Hodgson, C. (2006). Semantic interference during blocked-cyclic naming: Evidence from aphasia. Journal of Memory & Language, 54 (2), 199-227. Schwartz, M. F., Marin, O. S., & Saffran, E. M. (1979). Dissociations of language function in dementia: A case study. Brain and Language, 7 (3), 277-306. Schwartz, T. J., Kutas, M., Butters, N., Paulsen, J. S., & Salmon, D. P. (1996). Electrophysiological insights into the nature of the semantic deficit in Alzheimer's disease. Neuropsychologia, 34 (8), 827-841. Sclan, S. G., & Kanowski, S. (2001). Alzhei mer disease: Stage-related interventions. Lippincott's Case Management, 6 (2), 48-63. Slamecka, N. J., & Graf, P. (1978). The gene ration effect: Delineation of a phenomenon. Journal of Experimental and Psychol ogy: Human Learning and Memory, 4 592604. Smith, E. E., Shoben, E. J., & Rips, L. J. (1974). Structure and process in semantic memory: A featural model for semantic decisions. Psychological Review, Vol. 81 (3), 214-241.

PAGE 147

133 Smith, S., Faust, M., Beeman, M., Kennedy, L., & Perry, D. (1995). A property level analysis of lexical semantic repr esentation in Alzheimer's disease. Brain and Language, 49 (3), 263-279. Smith, S. R., Murdoch, B. E., & Chenery, H. J. (1989). Semantic abilit ies in dementia of the Alzheimer type. 1. Lexical semantics. Brain and Language, 36 (2), 314-324. Snodgrass, J. G., & McCullough, B. (1986). The role of visual similarity in picture categorization. Journal of Experimental Ps ychology: Learning, Memory, and Cognition, 12 (1), 147-154. Stanczak, L., Waters, G., & Capla n, D. (2005, May 31-June 4, 2005). Typicality-based differences in treatment of naming deficits. Paper presented at the Clinical Aphasiology Conference, Sanibel Island, FL. Stanczak, L., Waters, G., & Caplan, D. (2006). Typicality-based learning and generalisation in aphasia: Two cas e studies of anomia treatment. Aphasiology, 20 (2-4), 374-383. Stokes, T. F., & Baer, D. M. (1977). An implicit technology of generalization. Journal of Applied Behavioral Analysis, 10 349-367. Thal, L. J. (1999). Clinical trials in Alzheime r disease. In R. D. Terry, R. Katzman, K. L. Bick & S. S. Sangram (Eds.), Alzheimer Disease (2nd ed.). Philadelphia: Lippincott Williams & Wilkins. Thompson, C. K. (1989). Generalization issu es in the treatment of communication disorders. In L. V. McReynolds & J. E. Spradlin (Eds.), Generalization in the treatment of aphasia (pp. 82-115). Toronto; Philadelphia: Decker. Thompson, C. K., Ballard, K. J., & Shapir o, L. P. (1998). The role of syntactic complexity in training wh-movement struct ures in agrammatic aphasia: Optimal order for promoting generalization. Journal of the International Neuropsychological Society, 4 (6), 661-674. Thompson, C. K., Shapiro, L. P., Ballard, K. J., Jacobs, B. J., Schneider, S. S., & Tait, M. E. (1997). Training and generalized production of whand NP-movement structures in agrammatic aphasia. Journal of Speech, Language, and Hearing Research, 40 (2), 228-244. Thompson, C. K., Shapiro, L. P., Kiran, S., & S obecks, 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 (3), 591-607. Thompson-Schill, S. L., Gabrieli, J. D., & Flei schman, D. A. (1999). Effects of structural similarity and name frequency on pict ure naming in Alzheimer's disease. Journal of the International Neur opsychological Society, 5 (7), 659-667.

PAGE 148

134 Tippett, L. J., Grossman, M., & Farah, M. J. (1996). The semantic memory impairment of Alzheimer's disease: Category-specific? Cortex, 32 (1), 143-153. Tulving, E. (1983). Elements of episodic memory Oxford: Oxford University Press. Uyeda, K. M., & Mandler, F. (1980). Protot ypicality norms for 28 semantic categories. Behavior Research Methods and Instrumentation, 12 (6), 587-595. Warrington, E. K. (1975). The selective impairment of semantic memory. Quarterly Journal of Experime ntal Psychology, 27 (4), 635-657. Warrington, E. K., & Shallice, T. (1984). Category specific semantic impairments. Brain, 107 829-854. Watson, M. E., Welsh-Bohmer, K. A., Hoffma n, J. M., Lowe, V., & Rubin, D. C. (1999). The neural basis of naming impairments in Alzheimer's disease revealed through positron emission tomography. Archives of Clinical Neuropsychology, 14 (4), 347357. Wiegel-Crump, C., & Koenigsknecht, R. A. (1973). Tapping the lexica l store of the adult aphasic: Analysis of the improvement made in word retrieval skills. Cortex, 9 (4), 411-418. Wilkinson, G. S. (1993). Wide Range Achi evement Test-Third Edition (WRAT-3): Jastak Associates, A Division of Wide Range, Inc. Williams, B. W., Mack, W., & Henderson, V. W. (1989). Boston Naming Test in Alzheimer's disease. Neuropsychologia, 27 (8), 1073-1079. Williamson, D. J., Adair, J. C., Raymer, A. M., & Heilman, K. M. (1998). Object and action naming in Alzheimer's disease. Cortex, 34 (4), 601-610. Wilson, M. (1987). MRC Psycholinguistic Database : Machine Usable Dictionary Retrieved May, 2006, 2006, from http://www.psy.uwa.edu.au/mrcdatabase/uwa_mrc.htm Wittgenstein, L. (1953). Philosophical investigations New York: Macmillan.

PAGE 149

135 BIOGRAPHICAL SKETCH Claudia A. Morelli received her Bachelor of Arts degree from Emmanuel College, Boston, MA. She also has a Master of Scie nce degree from the University of Rhode Island, Kingston, Rhode Island, and is a certifie d speech-language pathologist. Claudias interest in providing treatment for adults with neurogenic communication disorders motivated her desire to complete the doctora l program at the University of Florida. During that time, Claudia also had the opport unity to teach underg raduates for seven consecutive semesters, and from these expe riences, developed a great appreciation for teaching. She has been trained in suppor tive and collegial academic and research environments driven by a core set of valu es focusing on the betterment of students, clients and their families, and ultimately th e profession. In Fall, 2006, Dr. Morelli will be an assistant faculty member in the Depa rtment of Speech-Language Pathology/Audiology at Loyola College in Maryland.