Generation of language in normal males


Material Information

Generation of language in normal males a regional cerebral blood flow study
Alternate title:
Regional cerebral blood flow study
Physical Description:
vii, 163 leaves : ill. ; 29 cm.
Williamson, David James Graves, 1966-
Publication Date:


Subjects / Keywords:
Language   ( mesh )
Regional Blood Flow   ( mesh )
Cerebrovascular Circulation   ( mesh )
Temporal Lobe   ( mesh )
Tomography, Emission-Computed, Single-Photon   ( mesh )
Magnetic Resonance Imaging   ( mesh )
Neuroanatomy   ( mesh )
Men   ( mesh )
Department of Clinical and Health Psychology thesis Ph.D   ( mesh )
Dissertations, Academic -- College of Health Related Professions -- Department of Clinical and Health Psychology -- UF   ( mesh )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1992.
Includes bibliographical references (leaves 154-162).
Additional Physical Form:
Also available online.
General Note:
General Note:
Statement of Responsibility:
by David James Graves Williamson.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
oclc - 51586282
System ID:

Table of Contents
    Title Page
        Page i
        Page ii
        Page iii
    Table of Contents
        Page iv
        Page v
        Page vi
        Page vii
    Chapter 1. Introduction
        Page 1
        Page 2
        Page 3
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
    Chapter 2. Review of the literature
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
        Page 24
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
        Page 41
        Page 42
        Page 43
        Page 44
        Page 45
        Page 46
        Page 47
        Page 48
        Page 49
        Page 50
        Page 51
        Page 52
        Page 53
        Page 54
        Page 55
        Page 56
        Page 57
        Page 58
        Page 59
        Page 60
        Page 61
        Page 62
        Page 63
        Page 64
        Page 65
        Page 66
        Page 67
        Page 68
        Page 69
        Page 70
    Chapter 3. Materials and methods
        Page 71
        Page 72
        Page 73
        Page 74
        Page 75
        Page 76
        Page 77
        Page 78
        Page 79
        Page 80
        Page 81
        Page 82
        Page 83
        Page 84
        Page 85
        Page 86
        Page 87
        Page 88
        Page 89
        Page 90
        Page 91
        Page 92
    Chapter 4. Results
        Page 93
        Page 94
        Page 95
        Page 96
        Page 97
        Page 98
        Page 99
        Page 100
        Page 101
        Page 102
        Page 103
        Page 104
        Page 105
        Page 106
    Chapter 5. Discussion and conclusions
        Page 107
        Page 108
        Page 109
        Page 110
        Page 111
        Page 112
        Page 113
        Page 114
        Page 115
        Page 116
        Page 117
        Page 118
    Appendix A. Directions to subjects in pilot #2
        Page 119
        Page 120
    Appendix B. Ratings of the imageability and concreteness of 246 verbs
        Page 121
        Page 122
        Page 123
    Appendix C. Instructions for pilot study #3
        Page 124
        Page 125
        Page 126
    Appendix D. Instructions to experimenter and subject for main study cognitive stimulation
        Page 127
        Page 128
        Page 129
        Page 130
        Page 131
        Page 132
        Page 133
        Page 134
        Page 135
        Page 136
        Page 137
        Page 138
        Page 139
    Appendix E. Experimental stimuli
        Page 140
        Page 141
        Page 142
        Page 143
    Appendix F. Correlations between adjusted regional counts and effort indices (reaction time to novel stimuli and subjective rating of difficulty) within and across conditions
        Page 144
    Appendix G. Scatterplots of regions of interest by side and task
        Page 145
        Page 146
        Page 147
        Page 148
        Page 149
    Appendix H. Partial correlations (controlling for global counts) between regions
        Page 150
        Page 151
        Page 152
        Page 153
        Page 154
        Page 155
        Page 156
        Page 157
        Page 158
        Page 159
        Page 160
        Page 161
        Page 162
    Biographical sketch
        Page 163
        Page 164
        Page 165
        Page 166
Full Text







Copyright 1992


David James Graves Williamson


I wish to express my sincerest thanks to Dr. Bruce Crosson and Dr. Stephen Nadeau for their

invaluable support and guidance throughout this project. I also wish to express my thanks to Dr.

Kenneth Heilman, Dr. Leslie Gonzalez Rothi, Dr. Eileen Fennell, and Dr. Russell Baner for invaluable

help with the design of the study, and to Dr. Shailendra Shukla, Dr. Clyde Williams, and Dr. Michael

Fagen for their technical support of this project. Special thanks go to Morris Thompson whose

patience and flexibility was crucial to the completion of this study. Thanks also goes to Randi

Lincoln for her help with subject screening, and I am indebted to Janice Honeyman for her

programming skills. I am appreciative to Dr. James Scott for his support and constructive criticism as

well. I also wish to express my appreciation to my wife Linda, who provides support for all my

endeavors as well as the best sounding board for conceptual and statistical dilemmas that I could hope

for. Finally, I would like to acknowledge my parents, James and Barbara Williamson, for the support

and encouragement they have continued to provide me over the years.


ACKNOWLEDGEMNENTS ........................... ..................................................................................................... iii

ABSTRACT ...................................................................................................................................................................


I INTRODUCHON ......................................................................................... ................................... 1

Definition of Terms ................................................................................................................................ 2
Methodological Considerations............................................................................................... 4

II REVIEW OF THE LITERATURE ................................................................................................. 17

III M ATERIALS AND METHODS .................................................................................................... 71

P ilo t S tu d y # 1 ........................................................................................................................................... 7 1
Pilot Study #2 .......................................................................................................................................... 74
Pilot Study #3 .................................................................................................................................. 76
Primary Study ........................................................................................................................................... 82

IV RESULTS ............................................................................................... .............................................. 93

Daa Analyic Stra gy ........................................................................................................................... 93
Investigation of Cognitive Effort .................................... 95
Tests of Hypotheses ........................................................................................................ 95

V DISCUSSION AND CONCLUSIONS ........................................................................................... 107


A DIRECTIONS TO SUBJECTS IN PILOT #2 .............................................................................. 119


C INSTRUCTIONS FOR PILOT STUDY #3 ............................................................................ 124

COGNITIVE STIM ULATION .......................................................................................... 127

E EXPERIM ENTAL STIM ULI ......................................................................................................... 140



BE E R EG IO N S ............................................................................................................. 150

R E FE R HCAE S .................................................................................................................................................................. 154
B IO G R A PH IC AL SK E TC H ....................................................................................................................................... 163

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



David James Graves Williamson

December, 1992

Chairman: Bruce Crosson, Ph.D.
Major Department: Clinical and Health Psychology

This investigation examined regional cerebral blood flow during the generation of language.

Twelve neurologically normal right-handed males participated in this study. Subjects performed

three cognitive activation tasks, after which distribution of regional cerebral blood flow was

determined via single photon emission computerized tomography (SPECT). The three tasks were

designed to preferentially stress the phonological, semantic, and grammatical components of

language in a stepwise, hierarchical fashion. Image analysis was performed using a neuroanatomical

atlas and a magnetic resonance scan of each subject's brain in conjunction with the data given by the

SPECT scans.

ntertask analyses suggest an important role for the left posterior superior temporal region

(i.e., Wernicke's area) in tasks stressing the generation of lexical-semantic information. In addition,

intratask analyses suggest that while there are no differences in whole-brain counts across tasks, each

task does engender a distinct pattern of asymmetries between area homologues. In particular, the

findings support the role of Wernicke's area in lexical-semantic processing and provide further

support for the previously reported findings of consistent left frontal opercular involvement in

generative language and left Brodmann's area 37 in tasks requiring visual analysis of word forms. In

addition, results are consistent with the view that practice of a given activation task may change the

magnitude and pattern of rCBF observed during the subsequent performance of that task. The results

are discussed in light of current findings in the functional neuroimaging and lesion literature, and

directions for future research are discussed. This investigation reinforces the merits of studying the

neural instantiation of language in the human brain with functional neuroimaging.



The study of language has a rich history, from the seminal clinico-anatomical correlation of

Paul Broca in 1861 to the functional neuroimaging studies of today (Wise, Hadar, Howard, &

Patterson, 1991). The most fruitful of these efforts have been built upon careful consideration of the

observed language phenomena and systematic investigation of the "building blocks" upon which these

phenomena are based. For almost one hundred years, these efforts were limited to observations of

patients with brain pathology that could not be quantified until autopsy. Despite these limitations,

theorists such as Wernicke and Lichteim were able to generate a general framework for the

instantiation of language in the brain which has held up well to the test of time (Liehteim, 1885;

Wernicke, 1874). Evolving technologies and the rediscovery and elaboration of these theories by

Geschwind in the late 1960's ushered in a new age of language investigation (Geschwind, 1972). The

rich complexity of both the neuroanatomical and psycholinguistic components of language has sparked

the interest of a variety of disciplines, and the increasing inter-reliance of cognitive, psycholinguistic,

neuroanatomical, and neurophysiological investigations upon each other has the potential to provide a

more comprehensive understanding of language than has been available to previous generations.

Investigations built upon the theoretical foundations laid down by previous generations which exploit

the capabilities provided by technological advances will be an important component to the further

investigation of the "building blocks" of language in a normal population. The focus of the current

study is an examination of specific components of language comprehension and generation utilizing

functional neuroimaging.

Definition of Terms

A careful examination of language requires an explicit definition of the phenomena to be

investigated. For the purposes of this study, language will be broken down into three components:

phonology, lexical-semantics, and grammar. For the purposes of this study, phonological processes

will be defined as those that deal with the sequencing of individual phonemes into morphemes, which

are then used to form words. Other phonological processes (e.g. linguistic prosody, accent) are not

explicitly addressed in this study. Lexical-semantic processes are those that deal with meaningful

concepts and the symbolic representation of those concepts (i.e. words). Specifically, lexical processes

deal with words, while semantic processes deal with conceptual knowledge. While these processes

have been shown to dissociate (Hillis & Caramazza, 1991), the current study makes no attempt to

investigate this dissociation. Grammatical processes deal with the word- and sentence-level aspects of

language which are used to conceptually elaborate upon single words and to denote relationships

between words, respectively. At the single word level, changes in the inflectional and/or derivational

morphology of the word are used to specify its properties. In contrast, the sentential-level properties

denote the relationships between words; these properties may be referred to as the syntactic properties

of a sentence. Although the terms grammar and syntax have been used interchangeably in the

literature, Nadeau (1988) points out that syntactic and morphological processes are distinct both

conceptually and anatomically. Further investigation of the dissociability of these components is

beyond the scope of this study, and the term grammar will be used to incorporate both syntactic and

morphological processes.

Functional neuroiinaging studies have varied a great deal in the extent to which they have

investigated these components of language. In general, the contributions of the three components to

the experimental tasks which have been used to investigate language in normals have been hopelessly

confounded with each other. While these investigations have been successful in showing that most

language tasks do engender different patterns of regional physiological activity (blood flow or glucose

metabolism) than those that are seen at rest, they have done relatively little to disambiguate the ways in

which disparate psycholinguistics processes are instantiated in the brain. Perhaps the most consistent

finding to date is that measurable changes in regional measures of activity will occur during tasks that

require subjects to generate language (Friston et al., 1991a; Frith et al., 1991; Ingvar & Schwartz,

1974; Petersen et al., 1988; Warkentin et al., 1991; Wise et al., 1991). Given these techniques'

consistent sensitivity to gross changes involved in these tasks, the experimental manipulation of

generative language tasks may be a powerful method of exploring the instantiation of different

language components in the brain.

"Generation of language" is a broad construct which may be broken down conceptually. The

amount of "generative capacity" demanded for a given language task may be seen to lie on a

conceptual continuum. The defining characteristic of this continuum is the degree to which novel

formulations of language must be produced. On one end of the continuum are tasks in which the

material to be generated is largely predefined. A prime example of this sort of task is repetition.

While repetition undoubtedly accesses phonological, lexical-semantic, and grammatical processes,

only phonological processes (parsing of auditory stimuli into individual morphemes, sequencing these

same morphemes for output, motor programming of output) need be intact to successfully perform the

task across the entire range of linguistic stimuli. In addition to the predefinition of generated material

by external sources, material may be predefired internally as well. An example of this sort of

generative task is the production of "automatic language." Language is considered automatic to the

extent to which it can be produced without cognitive effort on the part of the speaker. Examples

include counting, reciting the alphabet, or reciting the months of the year. In each of these examples,

the material to be generated is overlearned and exquisitely predefined.

Tasks which require novel formulations of language vary in their demands as well. A task

frequently used to assess one's ability to produce language is alternately termed "verbal fluency,"

"controlled oral word association," or the Thurstone task (Thurstone & Thurstone, 1943). In variants

of this task, output is strongly mediated by lexical or semantic constraints set up by the examiner, but

the subject is responsible for the generation of single words which fit into these constraints. External

mediation of language generation may be decreased by asking patients to describe pictures or generate

stories around a particular theme, in which case the subject must utilize phonological, lexical-semantic,

and grammatical processes to successfully complete the task. Finally, the least externally-mediated

language generation is the subject's own spontaneous speech, which the subject is solely responsible

for conceptual, lexical-semantic, grammatical, and phonological formulation.

While spontaneous language generation is perhaps the best measure of the subject's overall

language competence, its inherent variability between subjects and within subjects across situations

makes it the most difficult to study in the laboratory. Especially during the use of functional

neurounaging, when differences in quantity of input or output may confound results meant to examine

qualitative aspects of comprehension or production, the highest degree of experimental control possible

while maintaining the integrity of the phenomenon under study is extremely desirable. Thus, the focus

of this language investigation will be the strongly-mediated generation of novel responses in

hierarchically structured phonological, lexical-semantic, and grammatical tasks in a normal population.

As visually-displayed words will be used as the mediating stimuli in this study, the following

review will focus on functional neuroimaging investigations of the instantiation of single-word reading

and strongly-mediated language generation in the normal brain. Consideration will be given to the

ways in which these abilities have been investigated and the ways that these findings may be

interpreted in the light of findings from classical aphasiology. The goal of this review is to integrate

the data provided by these investigations into a coherent set of hypotheses which may be

systematically examined in the context of a functional neuroimaging study.

Methodological Considerations

An important step in interpreting the available data is understanding the ways in which the

data have been collected. While the assumptions made in interpretation of data from aphasic

populations have been described extensively (Kertesz, 1983), the assumptions inherent in functional

neuroimaging studies have received relatively little attention in the sources typically read by behavioral

scientists. As this review makes substantial mention of both of these bodies of literature, the

assumptions and limitations inherent to each will be discussed; however, given the imbalance of

attention that each has received in the literature to date, relatively more attention will be given to

assumptions important to the design and interpretation of functional neuroimaging results.

Aphasia/Brain Lesion Studies

Clinico-anatomical correlation is the most time-honored method of brain-behavior

investigation, dating back to Broca's seminal effort in 1861. This approach assumes that the

importance of a given region of brain tissue to a given behavior is manifested in the degree to which

that behavior is disrupted when the region is damaged. The majority of progress in understanding

brain-behavior relationships has been built upon theories and investigations based upon this principle.

These patients have long provided the most natural "laboratory" to investigate hypotheses built upon

earlier work, and a great deal of effort has been expended to define parameters which mediate the

validity of these observations (such as recovery curves, stability of lesions, etc.).

Nevertheless, this approach has limitations. First, the definition of "damage" has varied

considerably from study to study and has included such disparate processes as infarction, laceration,

necrosis secondary to radiation, excision secondary to neoplastic process, and electrophysiological

disruption (either due to a seizure focus or due to electrical stimulation). The differences between

these types of lesions in terms of etiology and effects on neighboring structures may well lead to

different behavioral observations when examining the effects of a given cerebral lesion, While this

limitation needs to be kept in mind when evaluating data, it may be mitigated to a large degree by

careful selection of the data upon which conclusions are based. A related limitation is the uncertainty

of lesion extent. The best neuromaging technologies available today are not 100% accurate in

identifying sites of cerebral pathology, and recent functional neuroimaging studies suggest that

structural imaging may significantly underestimate the extent of "functional pathology" as defined by a

decrease in regional glucose metabolism (Metter et a., 1986). The most severe limitation in

interpretation of data from these studies, however, applies to even those studies with extensive

definition of lesion location and size: the uncertainty as to whether one is observing the functioning of

a damaged functional system or an alternate compensatory system. In the absence of subsequent

lesions which serve to disambiguate these issues (Heilman, Rothi, Campanella, & Wolfson, 1979), this

question is unanswerable in the context of a single lesion study.

Functional Neuroimaine Studies

Functional neuroimaging studies, on the other hand, have the potential to disambiguate these

issues to a greater degree due to their depiction of which areas of the brain are most physiologically

active (or undergo the most change in activity) during a given cognitive task. This potential has

combined with the ever-improving resolution of functional neuroimaging techniques such as positron

emission tomography (PET), single photon emission tomography (SPECT), and nuclear magnetic

resonance spectroscopy (NMR spectroscopy) to spur increasingly widespread use of these techniques

in the investigation of brain-behavior relationships. As a result of this rapidly growing interest,

however, behavioral scientists have been increasingly thrust into the role of assimilating and evaluating

data which they have very little formal training in interpreting. While a great deal of excitement has

accompanied the use of these "windows into the living brain," there are a number of fundamental

assumptions made by many researchers and consumers alike that vary widely in terms of the extent to

which they have been supported by empirical data. These assumptions are made at the levels of (1)

global assumptions across techniques of functional neuroimaging, and (2) the technique-specific

assumptions of image acquisition and data analysis. As a full understanding of these assumptions is

necessary to the informed evaluation of these studies, they will be examined in some detail.

Global Assumptions

1. Local metabolism is indicative of local neuronal activity. The most fundamental

assumption common to all functional neuroiniaging studies is that the data being collected are a valid

indication of neural activity. Given the critical importance of this assumption to the interpretation of

functional neuroinsaging studies, it has received a great deal of attention. As a result, this assumption

has received the most empirical support of all of the global assumptions. If "neuronal activity" may be

defined as neuronal energy production, then one of the most widely accepted indices of neuronal

activity is oxygen consumption. The vast majority of the metabolic demands of the brain are met by

oxidative metabolism of glucose. Although there are rare instances in which oxygen utilization may be

an inaccurate indicator of ATP production (Frackowiak & Lammertsma, 1985), these conditions are

uncommon enough that 02 metabolism is considered a reliable index of neuronal activity.

Importantly, both regional cerebral metabolic rate for glucose (rCMR) and regional cerebral blood flow

(rCBF) have been shown to be very highly correlated to oxidative metabolism measures given by

autoradiographic techniques (Kurschinsky, 1987). These relationships hold particularly strongly under

normal conditions (Raichle, Gmbb, Gado, Eichling, & Ter-Pogossian, 1976). However, the

relationships between these variables become less consistent under certain clinical conditions

(Frackowiak & Lammertsma, 1985; Harper, 1989), and some evidence suggests that they may

dissociate in normals under conditions of abnormally high stimulation, although this remains a matter

of some debate (Chadwick & Whelan, 1991; Collins, 1991; Fox & Raichle, 1986). When rCBF and

rCMR do dissociate, rCMR seems to be the more predictable of the two indices (Chadwick & Whelan,


2. The changes seen in functional images of the brain represent changes in neuronal activity,

Recent efforts have examined the way in which the phenomena being viewed in functional

neuroinsaging scans are related to neuronal activity at the microscopic level (Collins, 1991). The

results of these efforts have thrown an interesting wrinkle into the traditional interpretation of

functional neuroimaging scans, particularly in studies of normal subjects. Since the inception of

functional neuroimaging techniques in the 1960's, the majority of investigators have described

significant regions of change in their data as signaling the populations of neurons that are most active

(in terms of increasing or decreasing activity) in a given comparison. In fact, it has now been relatively

firmly established that the changes seen in these images occur almost exclusively at the synaptic level,

not at the granular level. Thus, it is not change in activity in cell bodies that is reflected in these scans,

but rather change in the level of synaptic activity in a region (Chadwick & Whelan, 1991; Collins,

1991). This dynamic has been supported in a number of investigations. Examples include

autoradiographic investigations of the effects of eye-patching on glucose utilization in striate cortex in

rhesus monkeys (Kennedy et al., 1976), the effects of salt-loaded diets on hypothalamic-hypophysial

glucose utilization in rats (Schwartz et al., 1979), and the effects of sciatic nerve stimulation on glucose

utilization in the dorsal root ganglia and lumber spinal cord in anesthetized rats (Kadekaro et al., 1985).

This latter study showed a quantitative relationship between the frequency of stimulation and the rate

of glucose utilization in the dorsal horn of the lumbar spinal cord, while no significant changes in

glucose metabolism occurred in the dorsal root ganglia.

Such a dynamic is also supported by recent findings in the microarchitecture of the brain

(Collins, 1991). Regions of high metabolic activity in the brain are marked by increased

concentrations of the mitochondrial enzyme cytochrome oxidase (Wong-Riley, 1989).

Correspondingly, there is a clear relationship between cytochrome oxidase and capillary density in

laminated structures in the rat brain, and capillary density has shown to vary closely with variations in

the magnitude of brain glucose utilization (Borowski & Collins, 1989a). The relationship between

cytochrome oxidase and capillary density appears to be stronger in the neuropil than in cell bodies.

Regions that show the highest degree of cytochrome oxidase activity are typically dendrite tips with

spines and glomeruli (Collins, 1991). Conversely, proximal dendritic shafts show lower concentrations

of cytocrome oxidase and higher concentrations of glycolytic enzymes, such as lactate

dehydrogenase. There is an inverse relationship between capillary density and lactate dehydrogenase

activity (Borowski & Collins, 1989a).

These findings have profound implications for interpretation of functional neuroimaging

studies. For instance, use of the term "activation" in describing the neuronal events occurring during

stimulation may well be a misnomer. If one is examining a region in which the majority of the activity

in the dendritic fields is inhibitory in nature, then an "increase in activity" may in fact signal a decrease

in the activity level of the cell bodies in the region. Thus, the only way to determine the actual change

in activity level of the region is to look "downstream" at the region's efferent sites and see if they are

affected as would be expected given an increase in the region's activity. Given the multiplicity of

afferents to most regions in the brain, this is extremely difficult to do using in vivo studies in which

areas within human brains are activated through behavioral methods. Studies using animal models

(e.g., Mitchell, Jackson, Sambrook, & Crossman, 1989) have shown greater success in tracing

"downstream activity," although such studies are not without problems.

An important related issue which has yet to be sufficiently addressed in the literature is the

relative contributions of intra- and inter-regional inputs into a given region. If the data are correct in

suggesting that functional images are indicative of synaptic rather than cell body activity, then the

relative contributions of these sources play a critical role in the interpretation of "activation." If,

indeed, the majority of the synaptic activity in a region of interest is the result of local interneuronal

activity, then the traditional interpretation of functional images may be closer to the truth than would

be the case in a region whose synaptic activity was primarily the result of afferent pathways from other

regions in the brain. Intuitively, it is likely that the ratios of these inputs vary across different regions of

the brain, and it is conceivable that the ratios may even vary within regions according to the demands

placed upon a given region at a given time.

3. The deumre of activity in a region is indicative of the importance of that region to the

phenomena being studied, This assumption appears to stem indirectly from the clinico-anatomical

correlational approach described earlier. While the technological limitations of the times limited the

data available to the original European investigators of aphasia (Broca, Wernicke, Lichteim, etc.) to

post-mortem structural information, later developments enabled investigators to extend this model to

more accurately-defined anatomical lesions as well as to metabolic abnormalities secondary to these

lesions (Alexander, Naeser, & Palumbo, 1990; Benson, 1967; Metter et al., 1989; Mohr et al., 1978).

As the reasoning in this research suggested that those regions which were hypodense or hypoperfused

were regions important in the etiology of the language disturbance, it was a relatively short step to

assume that those areas that "light up" in functional neuroimaging studies of the normal brain must be

those that are most important to the behavior at hand.

The recent anatonucal findings mentioned earlier complicate this leap in reasoning, however.

In patients with brain lesions, interpretation is simplified by the identifiable lesion. Hypoperfusion in

the area of the lesion is accounted for by death of the afferent terminals and/or cessation of

interneuronal activity, while remote hypometabolic effects are explained by a lack of activity in the

damaged region's efferent terminals as well as silent ischemia and neuronal dropout (Metter et aL.,

1986; Nadeau & Crosson, 1992). In normals, one is seeing analogous input and output functions, but

there is no area of cell death to easily demarcate the activity level of a given region. Thus, in

functional neuroimaging studies, one is limited to a modification of assumption #3, which might be

stated as 'The degree of change in synaptic activity in a given region is indicative of the importance of

that region to the phenomena being studied." While the original assumption of neuronal activity in a

region being important to the behavior being studied is certainly not discarded, it is much more

difficult to investigate given that these techniques do not show change in activity at the level of the cell


4. The scale by which activity is quantified in functional neuroimaging studies is nEorortional

to a phvsioloicallv "meaningful" scale in some sense. and this scale has the same meaning across all

regions of the brain. This assumption is made by those investigators who compare changes in activity

levels across regions or who use global change as an index to which regional changes are referenced.

Thus, if one observes a 30% increase in rCBF in the right dorsolateral frontal lobe and a 15% increase

in the left dorsolateral frontal lobe, then one assumes that the right dorsolateral frontal lobe is "more

important" to the phenomena at hand. Furthermore, it is presumed that the 15% increase in activity in

one region is functionally equivalent to a 15% increase in the another region (in terms of the size and

increments of the scale on which rCBF is measured). Thus, a consistent "potential change in blood

flow" range is posited over the entire brain. Given the wide variability in arterial supply across regions

of the brain as well as the varying neuron densities across regions, this may well be a tenuous

assumption. Given the strong relationship between capillary density and regional glucose metabolism

(Borowski & Collins, 1989a), it may be that regions with greater capillary density have a greater range

of potential metabolic values. If this is the case, then percentage increases in regional

metabolism/blood flow may have dramatically different implications across different regions, as the

differences in potential range of values serve to put the different regions on different scales of


A concrete example might help to illustrate the point. Investigations comparing the reactivity

of different areas of cortex suggest that behavioral activation may induce changes in primary motor

and sensory cortices of up to 40%; in contrast, the changes induced in association cortices may be as

low as 2 to 5% (Raichle, 1987). This potential confound must be kept in mind for any algorithm that

references change to other regions. Indeed, depending upon the size of the investigator's regions of

interest, this confound may have a substantial impact upon data from within regions. As different areas

of cortex may have different scales of potential activity levels (and gray and white matter are certainly

different), one must temper conclusions about changes in activity with the knowledge of exactly which

regions are being measured.

5. The activity measured is relatively consistent across the time of measurement. The issue

of temporal resolution is one that is receiving increasing attention in the literature (Chadwick &

Whelan, 1991). Despite years of electrophysiological data detailing the change in neural areas in

increasingly short intervals, few functional neuroimaging studies discuss this issue. In fact, the time

taken to acquire data varies tremendously between techniques, with a current low of approximately 40

seconds for PET using 150-labeled water to approximately 40 minutes for PET using 18F-2-fluoro-2-

deoxy-D-glucose. The methods that have been used to investigate language in normals acquire the

majority of their data within six minutes of tracer administration. To date, no group has systematically

evaluated the impact that differences in data acquisition time have upon the pattern of regional activity

that is observed, although some work has been done showing that relatively small differences in

stimuli presentation rate can engender different patterns of activation during performance of the same

cognitive task (Raichle, 1991).

Techniue-Secific Issues

In addition to the assumptions made across functional neuroimaging techniques, each

technique must come to grips with a common set of issues. These issues involve (1) the manner in

which data are acquired, (2) the construction of the images to be analyzed, and (3) the correspondence

of these images to the physiological phenomena they are meant to reflect (both in terms of numbers to

actual metabolism/blood flow and in terms of localizing a given structure). Finally, the nature of the

experimental tasks themselves is crucial to any interpretation of data. An exhaustive review of the

techniques that have been used to deal with these issues is beyond the scope of this paper. Instead,

attention will be paid primarily to those methodologies used in the work most applicable to the study of

single-word reading and language generation in normals.

Single Photon Emission Tomouranhv (SPET) The earliest functional neuroimaging studies of

language generation in normals (Ingvar & Schwartz, 1974; Larsen, Skinhpj, & Lassen, 1978) were

done using single photon emission tomography (SPET). In the early versions of this technique, the

radiopharmaceutical is administered via intracarotid injection and blood samples are taken from the

jugular vein to provide indices which are later used in data correction. Localization is performed by

superimposing markers placed in certain detector fields onto markers placed in the same location and

then scanned by X-ray or CT.

Obviously, the method of data acquisition employed by this technique has major limitations.

It is extremely invasive, such that only subjects undergoing carotid angiography for suspected arterial

aneurysm or A-V malformation are eligible for participation (Lassen, 1985). This casts immediate

doubts on the "normality" of the population when exploring the neural instantiation of "normal"

cognitive function. In addition, the subjects are in an extremely unusual setting, with collimators

placed directly over their heads and needles in their necks, and the impact that these far-from-normal

circumstances has upon "normal" cognitive activity is not clear. The construction of the images is a

relatively straightforward matter, but the early investigations were limited to studying one hemisphere

at a time. This obviously limits the questions that may be asked of the data. The extent to which the

counts are reflective of neural activity in a given region is also difficult to interpret in this

methodology, due to the manner in which the data are acquired. Since the detectors are placed over

fixed locations for a given period of time, they are integrating all activity in their "field of view" over

that period of time. Given the orientation of the detectors, this means that cortical, subeortical, and

possibly even contralateral areas may be contributing to the activity index at any given detector

(Risberg, 1980).

Later SPET research (Risberg, 1980; Warkentin et al., 1991) utilizes 133Xe inhalation rather

than intracarotid radiopharmaceutical injection. In addition, this technique allows simultaneous

scanning of both hemispheres. The primary advantage of this technique over its predecessor is its

noninvasiveness. In addition, given the nature of the radiation emitted by 133Xe, a number of repeat

scans may be performed on the same subject at 30 minute intervals without danger of excessive

radiation exposure. Another advantage of the later SPET research is the use of numerical algorithms to

reduce the contribution of counts from regions other than the cortex of interest to the acquired data

(Risberg, 1980). A related limitation of this technique is the steps required in the analysis of the data.

Due to radiation scatter (primarily due to airway artifacts), the data that "count" towards the indices of

activity are not collected until 1.5 to 2 minutes after the administration of the 133Xe, and data

collection is then continued for five to ten minutes afterwards (depending upon the specific technique).

These aspects of the temporal resolution of the technique must be kept in mind when interpreting


Single Photon Emission Comuterized Tomosranhv (SPECTI SPECT utilizes some of the

same principles pioneered in the SPET research and applies them to three dimensional computerized

tomography. The method of data acquisition in this technique depends upon the tracer that is used. In

those studies using 133Xe-inhalation, the data acquisition is virtually identical to the 133Xe-inhalation

SPET studies. The primary advantage of this technique over the SPET studies is the improved image

resolution and the capability to examine the data in three dimensions. In addition, newer tracers used

with SPECT have compared very favorably to PET in terms of image resolution and replicability

(Gemell et al., 1990; Inugami et al., 1988).

Positron Emission Tomoraphv (PET). The results which have received the most attention in

the functional neuroimaging of normal brain function are those produced by laboratories utilizing PET.

In PET, data are acquired by means of detectors which encircle the head of a subject who has received

an injection of radiopharmaceutical. The length of time which elapses between administration of the

tracer and data acquisition varies according to the tracer. The tracers most commonly administered in

investigations of normal language are IV injections of 150-labeled water and inhalation of C1502.

150 has a very short half-life (2.05 minutes), and may thus be used for a number of repeat scans of the

same subject with relatively little radiation exposure. In addition, a complete scan may be taken as

quickly as 40 seconds following administration of the tracer. This short interval is not without costs in

terms of image sensitivity, however (lida et al., 1991). Due to the radioactive properties of the tracers

used in PET, spatial resolution is improved relative to SPECT. The construction of the images to be

analyzed varies from lab to lab (Fox et al., 1988; Friston et al., 1990), and these will be discussed later

in context of the studies that use them.

Many aspects of the data acquisition in PET are superior to those of other methodologies

(Raichle, 1983). The nature of the radionuclides (particularly 150, 13N, and llC) are such that they

decay with a relatively high degree of energy, thus enhancing their detectability and the resolution with

which images may be reconstructed. As previously mentioned, their short half-life may be

advantageous in some situations. They are also readily incorporated into substances that will be

included in most metabolic processes, thus increasing their flexibility. The disadvantages of PET are

primarily procedural in nature. Due to the nature of 150 and the current models of analysis, one is

limited to no more than 40 seconds of an experimental manipulation before analysis of regional

cerebral blood flow, on peril of severely underestimating true rCBF (lida et al., 1991; Raichle, 1985).

While this facet has the advantages enumerated above, it also limits the phenomena one is capable of

studying. The highly mathematical transformation of raw count data to the three dimensional

topographic image is also not without error, although it is assumed that the error due to estimation of

rCBF using 150 is less than 10% (Herscovitch, Markham, & Raichle, 1983; Jones, Greenberg, &

Reivich, 1982).

In terms of localization of activity, the three most frequently cited methodologies (those of

Raichle and colleagues at Washington University in St. Louis, Friston, Frackowiak, and colleagues at

Hammersmith Hospital in London, and Evans and colleagues at Montreal Neurological Institute) share

the common feature of image standardization. These laboratories convert each subject's PET data to a

standardized three-dimensional Cartesian format (that of Talairach and Tournoux, 1988). Although

each lab accomplishes this in slightly different ways (Fox, Perlmutter, & Raichle, 1985; Friston et al.,

1989), the end result is that each subject's brain is altered so as to fit the standard Cartesian coordinate

space to enable comparisons between subjects. While this is an admirable goal in terms of maximizing

statistical power and generalizability of results, it has some conceptual problems. As Steinmetz &

Seitz (1991) point out, the structural variability between individual brains is so great as to severely

compromise the interpretation of scans that have been "averaged" across subjects. Likewise, recent

work by Black et al. (1990) and Dietrich et al. (1982), as cited by Collins (1991), has suggested that the

synaptic, enzymatic, and microvascular structure of the rat brain is subject to change based upon the

animal's experience. Assuming that this dynamic may be generalized to the human brain, then the

interindividual variability discussed by Steinmetz and Seitz is compounded even further. In addition,

work by Phillips and his colleagues (1990) suggests that measurements of regional cerebral glucose

may err by up to 20% as a result of very minor misalignments of multiple scans of a single subject.

When one considers the variability already inherent in interindividual comparisons, this error rate

suggests that standardization of images will underestimate areas of regional change at best and will

falsely identify areas of regional activation at worst. Thus, conclusions based upon standardized scans

must be tempered with a great deal of caution, especially since these groups tend to choose individual

pixels (local maxima) as their regions of interest.

Another trend in the analysis of functional images is the normalization of the physiological

values across scans to some index, usually either mean whole brain or hemispheric blood flow. This is

done to allow comparison of various regions while controlling for fluctuations in whole brain blood

flow that may be unrelated to the task at hand, such as alterations in pCO2 or differences in tracer

uptake; thus, observed changes are not merely reflections of global phenomena affecting the whole

brain. However, this technique makes a number of assumptions. First, one assumes that the changes

in the regions of interest do not play a significant role in influencing the index to which the values are

being normalized. In comparing cognitive activation studies, this assumption has not yet been

disconfirmed. To date, significant changes in whole-brain blood flow have been seen only when

comparing cognitive activation studies to either a "rest" task or a motor task.

The second assumption that such normalization makes, however, has recently come under

attack. This assumption is that a change in the reference region (i.e. whole brain) will have the same

proportional effect on all regions of the brain. Recent findings by Friston et al. (1990) suggest that this

is not the case, but rather that areas of high flow are disproportionately increased relative to low flow

areas given a fixed increase in global brain blood flow (gCBF). Such a relationship would tend to

magnify the changes in high flow areas relative to low flow areas after normalization. Friston and his

colleagues suggest that this difficulty may be circumvented by removing the variance in the ROI due to

changes in gCBF by conducting an analysis of covariance (ANCOVA) with ROI activity as the

dependent variable, gCBF as the covariate (or "nuisance variable"), and task as the categorical

independent variable to determine the effects of cognitive activation on the ROI with the effects of

gCBF "partialled out." The manner in which this is done will be discussed later in context of the

Hammersmith studies. The issue to be emphasized here is that the assumption of proportional effects

across regions of whole-brain normlization may be flawed, to the detriment of the ability to detect

change in low-flow regions. In important related question which has yet to be fully addressed is the

extent to which this discrepancy between high-flow and low-flow regions is attributable to differences

between gray and white matter or to differences between regions of gray matter.

A vitally important issue to the interpretation of functional neuroimaging studies is the

manner in which change in activation is assessed. Perhaps the most widely-cited algorithm for

assessing change between images is the subtraction methodology devised primarily by the St. Louis

group (Fox, 1991). As pointed out by Raichle (1991), the subtraction paradigm is based on thinking

that can first be seen in the 19th-century work of Donders, who used increments in reaction time to

dissect out the components of mental operations (Donders, 1969). As applied to subtraction

methodologies in functional neuroimaging, this thinking suggests that if one subtracts the pattern of

brain activation seen in task A from the pattern of brain activation seen in task B, one discovers which

populations of neurons are recruited to handle the increased demands of task B. Furthermore, if one

averages the changes engendered by a given task across subjects, one can minimize the "noise"

inherent in individual subtractions and obtain a more accurate picture of the common regions which are

most important to the tasks.

There are a number of assumptions made by this sort of analysis which may be problematic.

First, this approach assumes that the "noise" that is washed out by the combination of images does not

yield valuable information in and of itself. In addition, the subtraction methodology employed by

Petersen and colleagues implicitly assumes that the brain handles increasingly complex language

activation by simply recruiting more neurons, while, in fact, some research suggests that the brain may

handle different linguistic tasks by the activation of different systems rather than by addition of

systems (Bemdt, 1988; McCarthy & Warrington, 1984).

There are a number of more technical details of functional neuroimaging which have

remained unaddressed by this brief review (attenuation, smoothing, and reconstruction algorithms,

among others) that are not without effect on the data reported by various authors. However, for the

sake of brevity, these issues will not be discussed in the current paper, and the reader is referred back

to the cited sources for more information.



The topics of single word reading and strongly mediated language generation have received a

great deal of attention from a number of labs employing widely different investigative techniques. Due

to the plethora of data available on these topics, selected studies will be presented which are felt to

reflect the current understanding of the instantiation of these functions within the brain. Particular

emphasis will be placed upon interpretation of the data provided by a relatively exhaustive review of

the normal-population functional neuroiiaging studies in light of data provided by investigations of

aphasic populations.

In order to focus the manner in which the review will be conducted, a conceptual framework

is useful to define terms and illuminate specific areas of investigation. Given the array of disciplines

with a vested interest in the investigation of language and the fine detail required to do careful studies

of psycholinguistic phenomena, a number of conceptual frameworks have been proposed. These

frameworks are most useful as heuristic devices which define the investigators' thinking rather than

guidelines by which the brain must behave in processing language. It is in this capacity that the

present study will use a framework based on that proposed by Ellis and Young (1988: see Figure 2-1).

It is readily acknowledged that other models have been presented to (1) describe similar phenomena,

(2) break down aspects of phenomena considered relatively unitary in this model, and (3) explain

language phenomena not included in this model. A discussion of these alternative models is beyond

the scope of the current review, as is a complete description of the derivation of Ellis and Young's

model. Rather, the model is presented and modified to provide a focus on the language phenomena to

be examined in this study; namely, the phonological decoding of nonwords, the comprehension of

single words and the generation of single conceptual units (nonwords, real words, or sentences). The

two processes of interest to this study that are not explicitly addressed by this model are the drive to

generate language and the grammatical formulation of output, both of which will be discussed in the

review of the available data.

Alexic Deficits in Sinele Word Readin,

Like all "omnibus" linguistic functions, reading single words is a complex act made up of a

number of dissociable components. A fundamental component of reading is, of course, the visual

analysis of the written word, A detailed discussion of the basic perceptual mechanisms involved in this

process is beyond the scope of this review; rather, the focus will remain on psycholinguistic analysis of

the visual stimulus. An extensive body of literature examining the neural instantiation of single word

reading has been compiled in investigations of aphasic and alexic patients. These investigations have

enabled the delineation of certain dissociations in the language system in a broad sense and within the

reading system specifically.

The process by which one reads has been a topic of ongoing investigation. Specifically, the

question arises as to whether all words are read in a similar manner, or if the reading process varies

according to some systematic criteria. Investigations of these questions have consistently shown that

reading in fact is not a unitary phenomenon, and that reading is susceptible to lexical, semantic,

grammatical, and contextual effects (Ellis & Young, 1988). For the purposes of this review, only those

dissociations dealing directly with lexical and semantic influences on reading performance will be


Consistent differences between reading real words and reading nonwords have been shown by

a number of investigators. For instance, in normal readers, real words are read more quickly than

nonwords, even when the two are equated for length and phonological complexity (Monsell, Graham,

Hughes, Patterson, & Milroy, 1992). These differences become even more apparent in phonological

and surface alexias. Patients with phonological alexia are able to read real words without difficulty,

but they are not able to read nonwords without a disproportionate amount of effort (if at all).

Conversely, patients with surface alexia are quite capable of reading nonwords, but their reading of

irregular real words is impaired (Beauvois & Derousne, 1979; Marshall & Newcombe, 1973; Rapcsak,

Figure 2-1. Organizational Heuristic based on Ellis & Young (1988).

Gonzalez Rothi, & Heilman, 1987). This double dissociation of reading abilities has been interpreted

as suggesting the existence of two routes of reading, one relying upon lexical identification and one

relying upon letter-by-letter analysis (or grapheme-to-phoneme conversion). Data suggest that normal

readers rely primarily upon lexical identification in order to maximize reading efficiency (Colitheart.


Data describing the lesions responsible for these deficits are somewhat inconsistent. The

angular gyros has been identified as important to both reading and writing, as lesions to this region

have been shown to produce alexia with agraphia (Benson & Geschwind, 1967; Dejerine, 1891).

Alexia without agrapbia, on the other hand, is typically associated with lesions to left calcarine cortex

and the splenium of the corpus callosum (Damasio & Damasio, 1983). These lesions are felt to

disconnect the angular gyms from visual input, thus preventing its performing its function of

processing of word images (Dejerine, 1891, 1892; Geschwind, 1965). Recent data have broken down

this scheme even further. Based on the phonological alexia of a patient with a left inferior temporo-

occipital lesion (Brodmann's areas 21, 37, and underlying white matter), Rapcsak, Gonzalez Rothi, &

Heilman (1987) suggested that the lexical reading route may be mediated by an intact dorsal pathway

from the inferior visual association cortex to Wernicke's area via the angular gyms, while the

nonlexical phonological reading route may be mediated by a ventral pathway from inferior occipital

association cortex to Wernicke's area via the posterior-inferior portion of the left temporal lobe.

Anderson, Damasio, and Damasio (1990) have also reported an interesting case in which a small left

premotor lesion (in the area classically defined as Exner's area) appears to have severely impaired

lexical reading and totally impaired nonlexical phonological reading for words but not numbers. The

patient was also agraphic. The alexia, combined with the authors' report that three of four others

patients with similarly located lesions were alexic acutely, suggests that this region may play some role

in conjunction with the classically defined posterior cortices in the reading process. The nature of this

role is uncertain, however.

To date, only two groups of investigators have reported results pertaining to basic processes

involved in normal reading. The most completely reported work is that of Petersen and his colleagues

(1989, 1990). The first study by this group used PET to study changes in rCBF engendered by a

variety of cognitive tasks. One of the analyses that was performed compared the pattern of activity

seen during the passive observation of high frequency English nouns to the pattern of activity seen

during visual fixation on a crosshairs. The nouns were presented at a rate of 1 Hz. The authors report

that this comparison identified striate cortex, left basal ganglia (possibly putamen), bilateral temporo-

occipital cortex (junction of Brodmanns areas 37 and 19), and an area in right temporo-occipital cortex

(area 37) slightly inferior to the bilateral foci as becoming significantly activated. The bilateral

activation temporo-occipital activation was asymmetrical, with the left side increasing approximately

30% more than the right. In their 1990 study, this group further explored the response of the visual

system to word-like stimuli. In this study, the investigators compared the patterns of activity seen

during observation of (1) a blank screen (control condition), (2) false fonts, (3) consonant strings (e.g.,

JVJFC), (4) nonwords that followed spelling rules of English (pseudowords), and (5) regular common

nouns. Three control scans were taken, and the one (or the combination) used in comparisons with

stimulation tasks was not identified. In addition to the asymmetrical activation at the junction of areas

37 and 19 described in the earlier study, the investigators also noted a significant increase in left medial

extrastriate cortex counts (perhaps medial area 37) during the observation of real words and

pseudowords. This activation was not present in the comparison of the control scan(s) to

orthographically irregular letter strings or false fonts. In addition, subtraction of the pseudoword image

from the real word image suggested an area of activation in the left inferior-lateral prefrontal region (in

Brodmann's area 45 or 47, anterior to Broca's area).

The other group that has explored the response of the visual system to real words is based at

the Montreal Neurological Institute. Unfortunately, none of the work with visual stimuli done by this

group has been published in complete form, and references are limited to presentations and abstracts.

As such, informed evaluation of these studies is impossible. The results that have been reported will

rUUBW i lCUnlfl [ l l W % u z

be discussed with these cautions in mind. Similar to Petersen and colleagues (1989), Marrelt and

coworkers (1992) contrasted the rCBF pattern seen while passively reading words to that seen when

fixating on a crosshairs. They reported strong bilateral activation in striate and extrastriate cortices as

well as unilateral activation in the left temporo-occipital and anterior temporal cortices (exact location

unspecified). Chertkow (1990, as cited by Posner & Carr, 1992) reported similar results in the same


Obviously, there are processes other than basic visual analysis occurring when one reads

single words. Both lexical and semantic processes are involved, as are processes involved in the

comprehension of inflectional and derivational morphology. Although the dissociation of these

components has received a great deal of attention in the cognitive neuropsychological literature (Hillis

& Caramazza, 1991), the neural instantiation of these components is probably beyond the capacity of

the current functional neuroimaging techniques. Thus, investigations have focused on paradigms

which lump the disparate processes together and seek to differentiate their neural instantiation from

that of the processes involved in the translation of visual patterns to phonological information.

The investigations examining the "higher cognitive processing" (i.e. combination of lexical,

semantic, and grammatical functions) of single words explicitly require the subject to take an active

role in processing the stimuli in order to examine processes beyond simple visual analysis of the

stimuli. In one of these studies, Petersen and his colleagues (1989) instructed subjects to keep track of

the proportion of dangerous animals that were named in the presented words. The two versions of this

task set the ratios at 1/40 and 20/40. The rCBF pattern was then compared to that engendered by

maintenance of visual fixation on a crosshairs that was placed above nouns. In both tasks, words were

presented at a rate of 1 Hz. Subtraction of the crosshairs-observation scan from the animal-tracking

scan revealed a significant rCBF increase in the anterior cingulate gyros, This response was

impossible to lateralize due to its medial position and technological constraints on the spatial resolution

of the technique. The change in rCBF in this region was proportional to the ratio of dangerous

animals. The authors also noted that rCBF in the left inferior lateral prefrontal cortex (anterior to

Broca's area) increased as well, but not to a statistically significant extent. In another study (also

described in Petersen et al., 1989), subjects were instructed to press a key whenever a visually-

presented pair of words rhymed. Word pairs included non-rhyming, visually dissimilar pairs (dog,

cat), nonrhyming, visually similar pairs (have, wave), rhyming visually dissimilar words (weigh, they),

and visually similar rhymes (dog, bog). These scans were compared to those acquired while the

subject maintained visual fixation on a crosshairs between pairs of words (one each above and below

the crosshairs). Subtraction of the crosshairs-observation scan from the rhyme-monitoring scan

suggested activation in left temporoparietal cortex "in a location near that found for auditory word

input [in a previous study]" (p. 155). Interestingly, this is the only report by Petersen and his

coworkers suggesting peri-sylvian activation during a language task in which the stimuli were

presented visually. The presence of other areas of activation was not discussed.

An understanding of the algorithm by which these results were obtained is essential to their

interpretation. First, scans were standardized to the coordinate system of Talairach, Szikla, &

Tournoux (1967) as described by Fox et al. (1985). The data were then linearly normalized by

dividing each pixel value by the global blood flow value. The direction and extent to which global

blood flow varied between subjects or across tasks within subjects is not reported. Paired scans were

then subtracted from each other on an intrasubject basis in order to isolate "the regional blood flow

changes associated with the operations of each cognitive level." These "subtraction images" were then

averaged across subjects in order to "increase the signal to noise ratio" of each comparison of

conditions. Each average image in the 1989 study was based upon 5 12 intrasubject subtractions,

while 8 subtractions were averaged in the 1990 study. Given the assumption that averaging the

subtraction images would result in greater disparity between consistently activated areas and areas not

consistently involved in the tasks, statistical significance of activated regions within each distribution

(averaged subtraction image) was determined by a two step process. First, the gamma-2 statistic was

used to determine if there were any significant outiers in the averaged subtraction image. If this test

was significant, the magnitudes of outliers are expressed in terms of a Z score relative to the "noise

level" of the distribution (response magnitude/standard deviation of the averaged subtraction image

[change scores over the standard deviation of the change scores]). Pixels with Z-scores greater than

2.17 (p>.03) are then interpreted as significant.

The methodology used in Montreal is similar in many ways (Zatorre, Evans, Meyer, &

Gjedde, 1992). The principle of the subtractive hierarchy is used in the design of their studies, and

analyses are based on averaged subtraction images. Unlike the St. Louis methodology, however, the

Montreal group tries to link their PET data to each subject's MR data as tightly as possible. It is then

parameters derived from the MR data which are used to standardize the PET data (Evans, Marrett,

Torrescorzo, Ku, & Collins, 1991; Marrett et al., 1992). Although this method may be able to

capitalize on MR's better spatial resolution to provide more accurate localization than algorithms based

on PET data alone, it is still subject to the previously-discussed limitations inherent in any

standardization of brain coordinate systems.

There are some methodological aspects of these designs to be commended. As only some of

the Petersen et al. studies have been presented fully, only these may be evaluated. First, they are

designed to fractionate complex cognitive acts (reading single words, in this case) and examine

changes in patterns of blood flow engendered by discrete components of these acts. As such, the

cognitive breakdown of tasks is generally more complete than is seen in the majority of the functional

neuroiuaging literature. Likewise, the investigators made a point to make the input and output

demands of each task as similar as possible, thus controlling for basic processes unrelated to the

cognitive phenomena of interest. In addition, the short duration of the stimulation task (not more than

90 seconds) minimizes the impact of fatigue or wavering attention.

Despite these advantages, there are a number of methodological limitations which cloud

interpretation of the results. Foremost are the previously-described limitations in the interpretation of

stereotactically-normalized, averaged subtraction images pertaining to the anatomical and

physiological variability across brains. In addition to the limitations inherent in the analysis algorithm,

however, there are also limitations in the design and report of the studies. In the 1989 study, the

number of subjects upon which the each subtraction image was based is not specified. Given the

authors' report that scans were based upon 5-12 intraindividual subtraction images, the number

averaged per scan used in analyses varied across conditions. Furthermore, the description of the total

number of scans in the study suggests that at least a subset of subjects performed one of the stimulation

tasks more than once, and the scan (or combination or scans) that was used in deriving the averaged

subtraction images is not specified.

There are conceptual limitations in the design of the studies as well. These limitations have to

do with the simplistic interpretation of the cognitive (and therefore neural) events occurring during

each task. Petersen and his colleagues appear to take the stance that each task reflects only those

processes which are necessary for performance of the stimulation task at hand. Thus, during visual

fixation, only those areas of the brain which are necessarily involved in fixating on a crosshairs

displayed in the middle of a cathode-ray tube are activated. Likewise, when "passively observing"

single words, only those regions of the brain responsible for processing the visual attributes of visually

presented single words are active. Clearly, this reasoning is inadequate. In fact, when one uses a

resting state in which the subject is given no instruction other than to visually fixate on a given spot,

one has no idea what cognitive state is being used as a "control" condition. And numerous

investigations in semantic priming have shown that mere exposure to a word results in processing

above and beyond the simple "visual pattern identification" level (Neely, 1991).

A summary of these findings may be found in Table 2-1. Keeping the above limitations in

mind, there are a number of consistencies between the two groups of studies. First, as expected,

significant increases in counts are seen in primary visual cortex regardless of the nature of the visual

stimulus. Likewise, a consistent asymmetrical bilateral response is seen in temporo-occipital

association cortices (perhaps Brodmanns convexity area 37) across different types of word-like

stimuli, with the left side increasing more than the eight. Finally, a particular region in medial

extrastriate cortex (perhaps medial area 37) is consistently identified as becoming activated during the

observation of orthographically-regular word forms. These results are somewhat consistent with the

lesion data in alexic patients, in that damage to left medial occipital cortex has been shown to lead to

an alexic syndrome that typically recovers (Greenblatt, 1983). The recovery curve is slower than could

be explained by simple edemous effects on the surrounding white matter tracts, however, suggesting

Table 2-1

Summary ot urictional Neuroumagme ginumoun visual LalLin adnrLvaut nre. a n I W e.

if en] Dr- flonend

StUdy NSrvlati n 'lk U control lad Siate Medial te p ore oteier titeror 0l er
Exrslriate Ocipital Suprior Lar
Tenipeal Prefrontal
Pteenet al. (1949) Passivelyrregle words ViSlalitxnon B Rea Ligtfletl
crolmiats (LIR 30%)

wytnle orinlg vi t on on
croesbain between words

earket croasllro above weed (tred)
eter et n0ervng nsi- s rinig s visual te n o
ircervillg 0velevgraphcally...efsa nS xttiec,
ieregolar srgs erorohars

...g .. visal f io en on
pned.r chair

nve redg Is ia o one.

U ing reeglar herring e no, g
evrdrowordn __

Macnet rim). (1992) tasierreadinig vlsalivettoe B L AnMte lemioral
Clinhweln ain PaItie reading Visal fiction B
(1991) a cited in
Peter & Cain
Note. B denotes bilateral change in activity, L denotes change in left hemisphere, R denotes change in right hemisphere, M denotes change in a midline region.
Empty cells denote those that are not discussed, while "-" denotes areas that are specifically identified as not changing.

that the "normal" reading system may be preferentially wired to left medial occipital regions than to the

right side. However, the good recovery typically seen in these patients suggests that this asymmetrical

preference may not be necessary to reading.

The finding of significant change in blood flow in temporo-occipital association cortex

(posterior area 37, anterior area 19) is a remarkably consistent finding. The fact that rCBF in this

region increased significantly every time a comparison was made between tasks involving word-like

stimuli and tasks not involving these stimuli suggests that it is involved in the early in the cognitive

process of reading, perhaps in direct conjunction with secondary visual cortices. This is supported by

the lack of change in this region when comparing two tasks that both involve these stimuli. Recent

neuroanatomical findings also support this hypothesis, as it has been shown that occipital association

cortex (V4) has extensive projections to the posterior two-thirds of inferior temporal cortex in the

macaque via U fibers in the white matter underlying the cortex (Tusa & Ungerleider, 1985). Rapcsak

et al. (1987) postulated that this connection is important to the process of grapheme-to-phoneme

conversion based on the phonological alexia exhibited by a patient with a small lobar hemorrhage in

the temporo-occipital junction.

The data interpreted as pertaining to "higher" processes are less consistent between groups.

Petersen and his colleagues noted an rCBF increase in the putamen when comparing passive reading to

visual fixation, while Marrett and colleagues discussed an tCBF increase in left anterior temporal lobe

(not further specified). Neither of these regions was replicated across labs. To date, only Petersen and

colleagues have reported findings from investigations using visual word-like stimuli other than single

real words as a control for visual processing demands, so these results must be interpreted with caution.

Again, it should be noted that full reports of the investigations of Marrett et al. and Chertkow et al., are

not available, so any mention of consistency between studies is severely limited.

Three of the findings and interpretations mentioned by Petersen and colleagues have spurred a

great deal of further discussion, First is the finding that rCBF response in the anterior cingulate is

related to the number of targets in an array of stimuli. This finding has been expanded upon in

investigations of attention which are beyond the scope of this review (Posner & Petersen, 1990). The

second finding is the increase in rCBF in left posterior superior temporal gyrus during the rhyme

monitoring task. Since this was the only subtraction using visual stimuli that showed a response in this

area, this was interpreted as an area important to the phonological monitoring of single, visually

presented words. This interpretation is consistent with the performance of patient H.R. (Friedman &

Kohn, 1990), who evidenced significant difficulties with a number of rhyming tasks. CT on H.R.

suggested a patchy lesion involving the temporal isthmus, the posterior half of Wernicke's area, and the

posterior supramarginal and angular gyri.

The finding which has produced the greatest controversy in the studies of language

comprehension is the increase in rCBF in the inferior lateral prefrontal region during the reading of real

words. At first glance, the 1989 and 1990 conclusions of Petersen and colleagues about this issue seem

contradictory. In the 1989 study, the investigators showed no activation of the left inferior lateral

prefrontal region (Brodmann's areas 45, 47) when subtracting the rCBF pattern engendered by fixating

on a crosshirs from the pattern engendered by silently reading real words. In contrast, this region was

shown to increase significantly when subtracting the rCBF pattern engendered by fixating on false font

strings from that engendered by reading silently. A plausible explanation offered by Petersen (personal

contact) to explain this apparent contradiction is that there was a great deal of variability in the

subtraction image generated when comparing the reading and rest conditions due to the extreme

increase in activity in the occipital and temporo-parieto-occipital regions. As such, the relatively

modest increase seen in left inferior lateral prefrontal cortex was "washed out." When the activity

related to visual processing was "controlled" by comparing reading silently to observing strings of false

font characters, the same response in the inferior prefrontal region was comparatively strong enough to

reach statistical significance. This response is taken by the authors to indicate that the left inferior

prefrontal region is involved in semantic processing that occurs during reading. This interpretation

will be discussed further after presenting evidence related to the auditory comprehension of single


Aphasic Deficits in Spoken Word Comnrehension

One way of increasing one's confidence about those regions which may be involved in single

word comprehension is to consider the findings of studies using aural presentation of single words.

Comprehension of the spoken word has received a great deal of attention in the aphasia literature and,

in fact, has been shown to fractionate in a manner similar to that seen in single word reading (Ellis &

Young, 1988). The comprehension of single spoken words may be broken down into (1) the

perception of the sound pattern, (2) the decoding of this pattern into identifiable phonemes, (3)

identification of the sequence of phonemes as a lexical entry, and (4) accessing the semantic

information associated with the identified lexical entry. Again, examination of the available data

suggests that current functional neuroimaging technology is probably not capable of making the fine

discriminations necessary in dissociating lexical identification and semantic access in a normal

population, so this review will knowingly commit the error of referring to these as a unitary construct

to be referred to as "lexical-semantic processes."

The phonological and lexical-semantic components of spoken word comprehension have been

shown to dissociate in pathological populations. The syndrome of "pure word deafness" reflects an

impairment m the phonemic decoding of the auditory language stream. Although the heating of

patients with pure word deafness is within normal limits, these patients have extreme difficulty

understanding spoken language, despite intact comprehension of written language and normal

expressive language. This syndrome has been broken down into two relatively distinct syndromes

(Auerbach et al., 1982: Kertesz, 1983). The first type, commonly associated with bilateral superior

temporal lobe lesions, is marked by deficits in the general ability to properly sequence perceived sound

patterns (Auerbach et al., 1982). The second subtype, conversely, is typically seen after unilateral

lesions in the dominant periauditory cortex and is more specifically linked to difficulties in linguistic

auditory discrimination (Saffran et al., 1976).

In contrast to pure word deafness, patients with transcortical sensory aphasia (TCSA) have no

difficulties decoding the stream of language, as evidenced by their normal repetition. However, these

patients have profound difficulties with language comprehension, and expressive language is marked

by fluent but semantically empty speech. This behavioral syndrome may be the result of a number of

different etiologies, from dementia of the Alzheimer's type to infarction of the left temporo-occipital

junction (Alexander, Hiltbrunner, & Fischer, 1989). In fact, some data suggests that there are distinct

subtypes of TCSA which dissociate according to the functional system(s) left intact to handle repetition

(Coslett, Roeltgen, Rothi, & Heilman, 1987). The behavioral and pathophysiological variability

manifested by patients with this syndrome complicate the process of identifying the neuroanatomic

correlates responsible for the observed deficits. However, analysis of the available data from a number

of sources suggest that it may be the function of the dominant inferolateral temporo-occipital cortices

and the regions that integrate this region with other functional subsystems (i.e. pathways in the

posterior periventricular white matter adjacent to the posterior temporal isthmus) that are critical in the

manifestation of TCSA (Alexander et al., 1989). The individual behavioral manifestation of each case

of TCS A may be a function of the parameters of the pathology within these regions (Coslett et al.,


The most pervasive disturbance of language comprehension is seen in Wemicke's aphasia, a

syndrome in which comprehension, repetition, and expressive speech are all severely impaired.

Expressive speech is fluent but marked by frequent semantic and phonemic paraphasias, often

deteriorating to jargon. This syndrome is typically seen following damage to the temporoparietal

junction, with the posterior third of the posterior superior temporal gyrs (Wemicke's area) typically

involved. The increased frequency of receptive and expressive phonological difficulties in Wernicke's

aphasia relative to TCSA suggests that the temporoparietal junction plays an important role in both the

morphological decoding of the incoming language stream and the conceptual morphological encoding

necessary to expressive language, The frequency with which one sees semantic paraphasias after

damage to this region also suggests some role in lexical-semantic processes. A number of studies have

been done in an attempt to disambiguate the phonological-morphological functions of the region from

the lexical-semantic processes (Cappa, Cavallotti, & Vignolo, 1981; Kertesz, 1983). The results of

these studies are not crystal clear, however. On the one hand, some data has been presented which

suggests that phonological-morphological processes are impacted in proportion to the involvement of

the supramarginal gyrs in addition to the posterior superior temporal region, while lexical-semantic

deficits are more heavily affected when the lesion is slightly more ventral and posterior (Cappa,

Cavallotti, & Vignolo, 1981; Kertesz, 1983). These positions seem to be supported by the findings that

conduction aphasics, in whom phonological processing is clearly impaired, typically have lesions in

the region of the supramarginal gyrs, while transcortical sensory aphasics have the semantic

difficulties and inferolateral temporo-occipital lesion locus described earlier. On the other hand, recent

findings suggest that deficits in semantic comprehension of single words is most strongly tied to

pathology of the dominant posterior superior temporal and inferior parietal cortices (Wernicke's area

and supermarginal gyms) in patients with single left-hemisphere CVA's (iart & Gordon, 1990).

Thus, the posterior third of the dominant superior temporal gyms appears to play a primary role in

lexical processes underlying the comprehension of language. As will be shown, this consistency has

received mixed support from the functional neuroimaging literature examining language

comprehension in normal individuals.

Functional Neuroimagin, of Snoken Word Comorehension

A number of functional neuroimaging studies have examined the comprehension of single

spoken words. Bartlett, Brown, Wolf, and Brodie (1987) measured regional cerebral glucose

metabolic rates (rCMR) in two groups of right-handed normal males in an investigation of regional

intercorrelation during single word monitoring. The language stimulation group was notably older

than the control group (mean ages 40.8 and 28.4, respectively). The 17 control subjects were scanned

lying quietly on the scanner table with the eyes open and their ears plugged. The 12 language

stimulation subjects were presented with monosyllabic English words at a rate one every 3.7 seconds

and instructed to make a foot movement every time they heard the phoneme Ib or //. Seven of the

subjects were instructed to make this movement on the left, while five made it on the right. The eyes

of the language stimulation group were closed throughout the stimulation.

Thirty regions of interest (ROt's) were identified using a combination of a neuroanatomical

atlas, superimposition of images, and experimenter judgment in order to get ROl's with the highest

metabolic rates and smallest regional standard deviations. ROI's were then averaged into six peri-

sylvian (left and right inferior frontal, superior temporal, and lower temporal), six control (left and

right anterior superior frontal, posterior temporal, occipital), two motor response (left and right mesial

sensorimotor areas associated with foot and leg movement), and two subcortical regions (left and right

thalamus). Given the thickness of the slices examined and the lack of specific description of

localization, finer identification of the regions examined is not possible.

Interregional relationships were explored using partial correlations (i.e. correlations

controlling for mean whole-slice value). In the unstimulated subjects, significant relationships were

found between left and right anterior superior frontal regions (roughly areas 9 and 46: r = .84), left and

right Broca's area (r = .70), left and right occipital cortex (r = .72), and left thalamus with right

Wernicke's area (r = -.66). Significance was determined utilizing Bonferroni's procedure due to the

high number of correlations. In the stimulated subjects, a number of areas were found to be

significantly correlated with left Broca's area, including left Wernicke's area (r = .88), right Wernicke's

area (r = .87), left Heschls gyrus (r = .75), left anterior superior frontal (r = .87), right anterior superior

frontal (r = .83), and left parieto-temporal cortex (r = .78). In addition, significant relationships were

found between left and right Wernicke's areas (r = .86), left and right anterior superior frontal (r = .87),

left frontal and right Wernicke's (r = .79), left and right occipital cortex (r = .84), and left and right

sensorimotor cortex (r = .86 roughly face region). Mean metabolic values were consistently lower in

stimulated subjects, while standard deviations of regional means were higher. Unfortunately,

interpretation of these findings is complicated by the finding that stimulated subjects were significantly

older than the unstimulated subjects, and regional metabolism has been shown to decrease with age

(Martin, Friston, Colebatch, & Frackowiak, 1991). Although a comparison of simple correlations and

partial correlations with age as the independent variable suggested that age played little role in

influencing the relationships between regions, it is unclear how age interacted (if at all) with

stimulation in contributing to the lower rates of metabolism in the stimulation task.

Petersen and his colleagues performed a number of studies analogous to those using visual

stimuli (Petersen et al., 1989). Only one of these studies was aimed at elucidating which areas are

involved in the "passive auditory perception" of single words (nouns). In this study, subjects were

required to visually fixate on a crosshairs and either remain in a resting state or listen to nouns

presented at 1 Hz. Results of the averaged-image subtraction suggested bilateral increases in primary

auditory cortex and unilateral increases in left anterior superior temporal cortex oust below the

Rolandic fissure), left temporoparietal cortices, and inferior anterior cingulate cortex. The authors also

reported activation in the right superior temporal region slightly posterior to primary auditory cortex.

Of these regions, the greatest increases were seen in right primary auditory cortex, right posterior

superior temporal cortex, and left temporoparietal cortex.

A more detailed breakdown of auditory comprehension was performed in Montreal by

Zatorre, Evans, Meyer, & Gjedde (1992). Using a subtraction methodology, these investigators had

subjects perform a number of tasks: (1) recline silently, (2) press a key to alternate pairs of noise

bursts (noise condition), (3) press a key to alternate pairs of consonant-vowel-consonant syllables

(passive speech condition), (4) press a key when the syllables ended with the same phoneme (phonetic

condition), (5) press a key when the second syllable had a higher pitch than the first (pitch condition).

The same syllables were used in tasks 3, 4, and 5. Subtraction of the resting baseline from the noise

condition showed bilateral increases in activity in primary auditory cortex, unilateral increases in left

sensory-motor hand area, left anterior superior temporal gyrus, and right lateral cerebellum, and

midline increases in supplementary motor cortex and anterior cingulate cortex. Subtraction of the

noise condition from the passive speech condition showed bilateral increases in anterior superior

temporal gyros and unilateral increases in right anterior superior temporal gyms (more anterior than the

bilateral focus) and left inferior prefrontal cortex (anterior to Broca's area). Subtraction of the passive

speech condition from the phonetic condition showed unilateral increases in left Broca's area, "near the

superior aspect of the supermarginal gyms," left inferior temporal gyrus, and the right occipital pole in

addition to midline increases in anterior and posterior cingulate cortex. Finally, subtraction of passive

speech from the pitch condition showed unilateral increases in the right inferior and middle frontal gyri

and left SMA in addition to increases in midline occipital cortex.

The group of investigators based at Hammersmith Hospital in London has also been active in

the investigation of single-word processing. Wise and his colleagues (Wise et al., 1991) had subjects

perform four tasks: (1) rest, during which subjects were instructed to "empty their minds," (2) listen to

nonwords conforming to typical English phonological structure, (3) decide whether two aurally-

presented nouns were correctly categorized and signal this decision by opposing the thumb and the

forefinger of the left hand, and (4) decide whether a noun and a verb were correctly categorized, as in

task 3. In order to control for the decision to signal and its action as behavioral variables between

conditions, subjects were instructed to oppose their left thumb and forefinger every few seconds in the

rest and nonwords tasks. For the first three subjects, nonwords were presented at a rate of 40 per

minute (wpm), while 13-15 noun-noun or verb-noun pairs were presented per minute. For the last

three subjects, the rates were 60 wpm and 25-27 pairs per minute, respectively. In the categorization

tasks, all subjects achieved 95% or greater accuracy, although the authors do not report if this figure

varied by stimulus presentation rate. Results showed virtually identical patterns of activity when

comparing the nonword, noun-noun categorization, and verb-noun categorization tasks to the resting

baseline. This pattern involved bilateral tCBF increases in Heschl's gyms, posterior superior temporal

cortex, and middle superior temporal cortex. Direct comparisons of (1) the noun/noun and noun/verb

categorization conditions with the nonword condition, and (2) the categorization conditions to each

other yielded no significant differences. In addition to the comparisons of the different tasks to each

other, the authors examined the correlations between the percentage increase in rCBF in the superior

temporal regions and the frequency of stimuli presentation. Significant positive correlations with

presentation rate were found bilaterally with rCBF in Heschls gyrus and middle superior temporal

gyrs and unilaterally the right posterior superior temporal region (all p<.01). Conversely, the

correlation between presentation rate and rCBF in the left posterior superior temporal region did not

reach statistical significance (r=.36, p<.08).

In a chapter reviewing PET investigations of language in normal subjects (Wise et a, 1991),

Wise and his colleagues anecdotally report a number of other investigations that have been carried out

in their lab as well. One of these has particular relevance to the comprehension of single words. The

pattern of activation seen while listening to reversed English words was compared to that seen during

the previously-described resting state. This comparison reportedly led to a pattern of change virtually

identical to that seen in the comparisons performed in the previous study (Wise et al., 1991 a): that is,

bilateral activation of Hesch's gyms, middle superior temporal gyms, and posterior superior temporal


In evaluating the value of these studies, it is important to keep in mind the advantages and

limitations of the techniques. These same considerations described earlier apply to the St. Louis and

Montreal studies. In addition, Zatorre and coworkers failed to provide information about the rate at

which subjects performed the different tasks. Since this appeared to be dependent upon subject

reaction times, the impact of this factor on the amount of processing required of each subject across

tasks is unknown. It is also important to note that the Hammersmith group uses a slightly different

imaging algorithm than that employed at St. Louis and Montreal, so this will be briefly reviewed. In

concordance with the St. Louis and Montreal studies, the Hammersmith images were standardized for

size and shape to the coordinate system of Talairach & Toumoux (1988). The methodology used by

this group is detailed by Friston et al. (1989). However, instead of subtraction of averaged scans from

each other, data were analyzed via analysis of covariance (ANCOVA) with global blood flow as the

covariate. This was strategy was based on the finding of Friston and his colleagues (1990) that the

relationship of global blood flow to rCBF may vary across regions. In this strategy, each pixel in the

standardized coordinate system is analyzed via ANCOVA, and the resulting adjusted mean scores are

compared with pre-planned t-tests. The total number of pixel-by-pixel comparisons required to

examine the entire brain is approximately 147,000 (Friston, Frith, Liddle, & Frackowiak, 1991). The

resulting t-statistics are then presented as a statistical paranetric map (or SPM). The number of

significant t-statistics is then compared to the number expected by chance with Bartlett's X2. If the

number of significant t-tests is significantly greater than would be expected by chance, individual

pixels are then interpreted (Friston et al., 1991).

There were a number of good points about the methodology used by Wise and colleagues.

First, they are the only group to date to attempt to systematically investigate the effect of presentation

rate on changes in rCBF in a task using aural stimulation. Furthermore, the study employed a direct

comparison of conditions rather than a subtraction methodology, this avoiding the interpretive

complications inherent in averaged subtraction images.

Although the use of ANCOVA may be an advance to account for the variability across

regions, the technique has a number of limitations. In deriving their SPM's, the authors smooth their

data in order "to account for interindividual differences in gyral anatomy." Smoothing is a function

whereby each pixel x comes to represent the average of the pixels in a predefined radius from the pixel

x. In the Friston et al. methodology, the degree to which this smoothing "blurs" the data varies

according to the statistical properties of the distribution of the images being examined. In addition, in

order to ascertain which areas were significantly activated, the authors compared "average images" for

each condition, rather than obtaining values for each individual brain. Given that the authors had

"standardized" their individual images and that their design anticipated analyses of certain regions a

priori, it is unclear why rCBF values were not obtained from individual scans using the same algorithm

used to locate pixels on the average images and then analyzed. Given the variability which will remain

in exact localization even after standardization (Steinmetz & Seitz, 1992), obtaining values from

average images may impact upon the findings in any number of ways. Since no data were given about

the variability in the average scans, it is difficult to estimate what this effect may have been. There

were other limitations to the study as well. First, it was reported that the first author was one of the

subjects. Thus, one wonders how familiar the subjects were with the hypotheses and stimuli of the

study, and what effect this may have had on their performance of the tasks. Another issue clouding

interpretation of the data is the fact that, although the issue of disentangling stimuli presentation rate

from type of stimuli in affecting rCBF is an important one, the division of the six subjects into two

groups of three varying on stimulus presentation rate raises some concern. In particular, even though

the authors showed that increased presentation rate led to higher flow in 5 of 6 superior temporal areas,

these two groups of three were combined in drawing between-task conclusions. Also, different tasks

called for substantial differences in output, as the motor response would increase by approximately

33% in the 60 stimulilminute condition. Given that the authors only reported the correlations between

the rate of presentation and changes in percentage increase for superior temporal regions, it is unknown

if and to what extent varying presentation (and thus output) rates affected any other areas. Thus, the

effect that this combination of images may have had on the pattern of results is uncertain. Finally, the

tasks were administered to each subject in the same order, thus making an order effect on the results

impossible to rule out.

A summary of the results regarding the comprehension of aurally presented words is found in

[able 2-2. There are a number of interesting consistencies across studies. First, for the most part,

primary and secondary auditory cortices area are affected in predictable ways by auditory stimuli. The

exceptions are (1) Zatorre's finding of increase limited to the left side when comparing alternating

noise bursts to rest, and (2) the consistent lack of change in primary or secondary auditory cortices

when comparing vocal tasks to rest. It seems as if the cortical response to one's own speech does not

activate the language comprehension system in the same way that exterually-generated language does.

Another relatively stable finding across laboratories is an increase in rCBF in the left posterior superior

temporalItemporo-parietal region whenever language is perceived at the phonemic level. Finally, the

functional breakdown performed by Zatorre and colleagues suggests a role for both Broca's area and

the superior portion of the supramarginal gyrus in tasks which require careful monitoring of phonemic


Most of the above results are consistent with what would be expected from the lesion

literature. The early activation of primary and secondary auditory cortices may be reflective of some

of the processes impaired in pure word deafness, while the consistent involvement of the posterior

superior temporal regions/temporoparietal junction at the morphemic level is consistent with the

analyses that are known to break down in patients with lesions in this region. Furthermore, Zatorre and

colleagues' report of activity in Broca's area and superior supramarginal gyms during phonological

monitoring is not inconsistent with the phonological deficits seen in conduction aphasia. This

syndrome has been associated with lesions in the connections between the temporoparietal junction

and the inferior premotor cortices (Wericke, 1874). A lesion in the region of the supramarginal gyrus

has been hypothesized to disrupt the connection between these cortices. In addition, recent data suggest

that while the classically defined arcuate fasciculus is a primary pathway between these two regions,

Table 2-2
Summa of Functional Neuroimaing Findings during Auditory Language Stimulation Tasks without Verbal Response Demand

Study Stmulation flask Control lak Primary Postenor remproo- nteor Anterior Atenor tr
Aoditory Sopeior Parietl Lateral Telapornl Ctgulale
Cortex Teseteral Prefrontl

eiesteneat (1959) 5sssve listeig Ios gle vt Is'Moteoon L L M
amrds cossairo lmioso)

atrre ear (19921) Pres a cry t. a lerante Reclne silently L K L a L Sensory-MotOr
noise busts Ho.a
R Lateral

syllables o.o boia (2 ci, at Temporal
aren 21/22
and 38)
Press a key when last Mes a key to alterOne M LiBroca'
phonemes idemical syllables 1a1so a aros.
posterior L Soperior
cingulair Parietal;
loca) L Inferior
R Occipital Pole
HRN s key whto sen PFeakeyoaienate R Middle Fodnt
pisth higher syllables Gyr s;
M Occipital
Wise eaol (1991) LiWten an anWdt, oppose Recline sently, oppose let It B 5 Mid-superior
lef thumb and fostoge tb nd rfitger every Tempora
sle t seconds fe- seconds .
Opys lIba ned Reclnewalety oppose ert B B B Mtd-Sue0r
frefcge if.t sn are thumb ad fortfiger every Temporal
semantically related wr ads
O ipose let thumb ad Recline sdsndy,oppose let B B B Mid-Superiso
forefner if a non ada thumb ad forfinger every Temporal
verb ase wtiCally few waonds

forefinger if tw aots are left th adofisger
semaosnlny related et fre seconds
IWos thumb and Listea to nonords, oppose
foeiger if a noos ad o left thumb and fcefesger
verb ae sesmmtcally esesy se seonds

tlet faron o refiger ons ado
sematically related verb one seiantcally
Note. B denotes bilateral change in activity, L denotes change in left hemisphere, R denotes change in right hemisphere, M denotes change in a midline
region. Empty cells denote those that are not discussed, while "-" denotes areas that are specifically identified as not changing.

there are at least one other pathway running beneath the insular cortex in the extreme capsule (Damasio

& Damasio, 1983). Thus, it is conceivable that the projections into both parietal cortex and inferior

premotor cortices may become activated during the normal process of attending to phonemic attributes

of syllables. An important difference between the findings of Zatorre et al. and those in the lesion

literature is the locus of activity in the supramarginal gyrus. Zatorre et al. reported activation in the

superior portion of this region, while the lesion literature suggests a more inferior focus is important to

the phonological processing of language. As Zatorre et al published no figures showing the region of

activation, it is difficult to know it (a) there was overlap with inferior supramarginal cortex, (b) there

was close enough proximity to inferior supramarginal cortex to attribute the discrepancy to localization

limitations inherent in standardized images, or (c) the finding of Zatorre et al was simply contradictory

to what would be expected from the lesion literature.

As mentioned earlier, the potential role of the inferior lateral frontal prefrontal cortex (areas

45, 47) during language comprehension has received a great deal of recent interest (Carr, 1992;

Compton, Grossenbacher, Posner, & Tucker, 1991; Damasio & Damasio, 1992; Posner & Carr, 1992).

As the reader will recall, Petersen and his colleagues noted an increase in activity in this region during

the reading of real words in their paradigm using visual stimuli once the response attributed to primary

sensory processes had been "controlled." Although Petersen and colleagues have yet to control the

primary sensory processes in a paradigm using auditory stimulation, Zatorre and colleagues (1992)

performed an analogous comparison when subtracting alternatingg noise bursts" from "alternating

syllables." This comparison also engendered an increase in rCBF in the left inferior lateral prefrontal

region anterior to Broca's area. Both of these groups have related this response to semantic processing.

Wise and colleagues, on the other hand, cite their data to dispute this claim. This group found no

significant changes in the inferior lateral prefrontal cortex in any of their comparisons. They maintain

that their data support the role of left posterior superior temporal gyros (Wernicke's area) in the

comprehension of language. They explain the activation of left posterior superior temporal gyrus

during the monitoring of nonwords as being due to the conceptual dynamics proposed in parallel

distributed processing models of language comprehension (Rumelhart & McClelland, 1986), in which

it is posited that comprehension is arrived at though the emergent realization of the auditory pattern of

the word. Theoretically, the cortices (or neural networks) involved would automatically engage in

many of these processes regardless of the nature of the linguistic stimuli perceived; in the case of real

words, the network concerned with lexical identification would settle into a coherent pattern of

activity, while the attempted comprehension of a nonword would activate the same network (concerned

with lexical identification) but would not settle into a coherent pattern of activity.

Examination of the lesion literature supports neither position strongly. As noted above,

Wericke's area proper does not appear to be strongly implicated in the semantic processing of visual

or auditory input strings beyond the phonemic or possibly even lexical level. Likewise, there is very

little lesion literature to support a role for Brodmann's areas 45 and/or 47 in semantic processing. The

pair of studies from outside the functional neuroimaging literature that are frequently cited to support

this role are those by Milberg, Blumstein, & Dworetzky, which explore the way in which semantic

primes affect the response times and accuracy of Broca's and Wemicke's aphasics in lexical decision

tasks (Milberg & Blumstein, 1981; Milberg, Blumstein, & Dworetzky, 1987). The results of these

studies suggest that, while Broca's aphasics are more accurate in their lexical decisions than are

Wernicke's aphasics, their response times do not show the expected effects of semantic cuing.

Wemicke's aphasics, on the other hand, benefit from semantic primes in a pattern similar to that shown

by normnals. An obvious limitation of using these studies as support for the findings of Petersen et al.

and Zatorre et al. is that Broca's aphasia does not necessarily imply a lesion in the inferior lateral

prefrontal cortex, and, in fact, examination of the CT data cited by Milberg and colleagues suggests

that less than half of the Broca's aphasics had structural lesions in this area. Specifies as to when these

scans were acquired were not presented, however, and structural imaging techniques such as CT are

limited in their ability to delineate "functional" lesions as defined by hypometabolism (Metter et al.,

1986). Thus, such a lesion cannot be ruled out in most of Milberg et al.'s patients; however, the

relationship between the anatomical focus defined by Petersen and colleagues (1990) and the sites of

lesions explored by aphasia investigations to date is far from clear. Another limitation in the

interpretation of these findings as suggesting a role for Brodmann's areas 45 and/or 47 in semantic

processing emerges from the work of Goldman-Rakic and colleagues (1992). Briefly, the work of

these authors indicates that the prefrontal cortex is important to working memory, and the methods

used by Milberg, Blumstein, and colleagues do not allow one to rule out the possibility that one is

observing the consequences of an impairment in working memory (i.e., difficulty establishing or

maintaining an internal representation to guide further responses) rather than an impairment in

semantic processing.

Thus, an integration of the functional neuroimaging and lesion data concerning the

comprehension of the written or spoken single word leaves many questions unanswered. Foremost

among these questions is the role of the inferior lateral prefrontal region in single word comprehension.

To date, the data seem to support different interpretations depending on the modality of the stimulus,

and neither of the available interpretations are particularly well supported by the lesion literature.

Another interesting issue arising from comparing the two literatures is the striking lack of perisylvian

activation during language tasks mediated by visual stimuli. Although posterior area 37/anterior area

19 seems to be consistently involved, the response in this appear to be more related to visual

processing than to language per se, although the apparent asymmetry favoring the left suggests some

degree of specialization. In addition, anecdotal reports by Raichle (1991) suggest that a change in

presentation rate dramatically increases the response in posterior superior temporal cortex, suggesting

that methodological factors may well play a role in the data that is acquired. Examination of the

available data, then, suggests that the lack of consistent perisylvian response may be due largely to (1)

lack of control of primary sensory responses, which have consistently been shown to be much greater

than responses related to the "higher functions" involved in the comprehension of language, and (2)

time-linked variables, such as the rate of stimulus presentation.

Aphasic Deficits in Lanauaae Production

The generation of language may be conceptualized as involving (1) the drive to generate

language, (2) the formulation of the concept to be communicated, (3) the identification of the open-

class lexical items appropriate for communicating the concept, (4) the manipulation of the morphology

of these open-class lexical items and the relationships between them in order to precisely communicate

the concept, (5) the programming of the phonological sequences necessary for communication of the

formulated message, and (6) motor execution of the phonological program. These steps are not seen as

a strictly linear sequence but rather as psycholinguistic subcomponents which must be accomplished in

some manner during the production of language. The dissociation of the first three components is

remarkably difficult to do in a normal population, and is probably far beyond the current capabilities of

functional neuroimaging due to its apparent difficulties in discriminating between tasks with relatively

subtle differences in cognitive demands. Thus, although all of these components are important in the

production of language, the lexical-semantic, grammatical, and phonological components of this

process are those with which this study will directly address. For the purposes of this study, the

lexical-semantic component may be seen as a combination of steps 2 and 3, the grammatical

component as step 4, and the phonological as step 5. These will be reviewed after briefly discussing

the general issue of generativity.

As discussed earlier, the production of language may be viewed on a continuum defined by

the amount of novel language formulation required. Fundamental to this formulation is the drive to

generate language. A few studies in the lesion literature have addressed this issue, primarily with

regards to dynamic and transeortical motor aphasic syndromes. Dynamic aphasia, a term coined by

Luria, may be defined as a selective impairment of this drive in the relative absence of other language

impairment (Luria, 1970). In its most selective form, this impairment seems to be most strongly tied to

lesions in superior medial frontal regions (SMA) and their connections to dorsolateral frontal cortices

(Alexander, Benson, & Stuss, 1989). It is lesions in these same regions, anterior and superior to

Broca's area, that have been tied most strongly to transeortical motor aphasia. In this syndrome,

fluency of spontaneous speech is also impaired; however, the clinical picture is usually more complex,

in that greater impairment is typically seen in syntactic function and occasional paraphasias are noted.

Notably, decreases in "verbal fluency," or the ability to generate words beginning with a given letter,

have been repeatedly shown to be associated with lesions in left dorsolateral frontal cortex (Benton,

1968; Miceli et al, 1981; Pendieton et al., 1982), although findings also suggest this deficit is also seen

in patients with nonfrontal lesions (Pendleton et al., 1982). An important aspect to both of these

clinical syndromes is the preservation of repetition and the relative sparing of the semantic integrity of

the language formulation.

Lexical-semantic processes in the production of language are tied more closely to lesions in

the posterior cortices. Typically, selective impairments in these functions are seen only in TCSA,

while striking impairment is seen in conjunction with greater difficulty in phonological processes in

Wemicke's aphasia. The previous discussion of the psycholinguistic and pathophysiological

mechanisms underlying each of these syndromes discussed their impact on the generation of language.

Impairment in the grammatical formulation of language has received increasing attention

throughout the past 20 years. Initially, this impairment was felt to be one of the hallmarks of Broca's

aphasia (Goodglass & Menn, 1985). However, traditional conceptualizations of Broca's aphasia have

undergone intense scrutiny during the past two decades. While the classic Wemicke-Lichteim model

of language hypothesized the inferior foot of the third frontal gyrus (Broca's area) to play a critical role

in the phonological and grammatical formulation of language, recent findings have suggested that the

traditional syndrome of Broca's aphasia (sparse, effortful, dysfluent agrammatic speech with relatively

spared comprehension) is composed of a number of mutually dissociable components whose neural

instantiation may have little if anything to do with Broca's area (Alexander, Naeser, & Palumbo, 1990;

Nadeau, 1988). With regards to grammatical formulation in particular, impairments in expressive

grammar have been seen in aphasic patients with both anterior and posterior lesions (Blumstein, 1988).

However, on closer inspection, these grammatical impairments appear to differentiate into impairments

at the sentential level (difficulties with syntax) and impairments at the inflectional morphological level

(difficulties with concatenation) (Nadeau, 1988). Some recently presented data suggest that syntactic

difficulties appear to be preferentially bound to damage in the anterior cortices, while concantenation

difficulties appear to be more strongly bound to damage in the perisylvian regions (Kolk. van

Grunsven, & Keyser, 1985; Nadean, 1988).

Similarly, as mentioned in the previous discussion of Wernicke's aphasia, current evidence

does not support the notion that Broca's area is solely responsible for the phonological coding of

spoken language. This notion is critically flawed in two respects: (1) lesions in the posterior superior

temporal gyrus disrupt phonological aspects of output despite spared anterior cortices, and (2) there

have been a number of case reports of patients with lesions including Broca's area that have no

demonstrable deficit in expressive language (Alexander et al., 1990; Mohr et al., 1978; Nadeau, 1988).

Relatedly, the majority of patients diagnosed as Broca's aphasics have lesions incorporating insular,

anterior parietal, and subcortical regions in addition to their underlying white matter pathways (Mohr

et al., 1978). In effect, given the extent of the damage typically seen in Broca's aphasia, and the fact

that many lasting Broca's aphasias are global aphasias which have resolved into this syndrome, the

traditional syndrome of Broca's aphasia may be in fact no more than the representation of the right

hemisphere's relative ineptitude at handling phonological and syntactic function (Baynes, 1990).

Rather, the behavioral manifestations of the transcortical aphasias make a strong argument that it is the

perisylvian cortices, not the anterior cortices, that support the processes fundamental to phonological

decoding and encoding of language, as preservation of the perisylvian regions typically preserves

repetition despite potentially pervasive pathology across other aspects of language functioning.

Investigation of the strongly-mediated generation of language in normals with functional

neuroimaging has typically been conducted by having subjects perform tasks that are devised require

differing amounts of novel formulation of language. Thus, tasks conceptualized as "low-level" include

repetition and automatic language. This may be problematic, as both repetition and automatic

language have been shown to dissociate from other aspects of language. In transcortical mixed

aphasia, these may be the only two sorts of clear verbalizations available to the patient, while even

global aphasics generally show relatively good retention of at least some aspects of automatic

language. Thus, it may well be that these sorts of productions disproportionately rely on neural

substrates that are relatively peripheral to normal language generation. If this is the case, the studies

employing this sort of hierarchy may not be looking at the hypothesized additive hierarchy but rather at

a hierarchy that incorporates qualitatively (and neurally) different processes at different levels.

Nevertheless, as some aspects of language are certainly required for these productions (such as

phonological programming), this review will incorporate the studies which have examined these

functions in order of increasing demand for novel language formulation.

In 1974, Ingvar and Schwartz's SPET investigation of changes in rCBF in the left hemisphere

during automatic speech marked the beginning of functional neuroimaging studies of normal language.

One of the comparisons made by these authors was between a resting control state and a task in which

subjects were instructed to recite either the months of the year or the days of the week. The authors

noted no significant changes in global blood flow between conditions, but regional increases were

observed throughout most of the frontal cortex and in the superior parietal regions. Likewise, increases

in rCBF were observed in the middle and anterior portions of the temporal lobe. In contrast, rCBF

decreased in the occipito-parietal-temporal region. Larsen, Skinhoj, and Lassen followed this study in

1978 with a more detailed look at automatic speech. These investigators compared rCBF patterns

during (1) rest, and (2) counting from I to 20 or reciting the days of the week at a rate of one word per

second. Localization was performed by using three external markers and the proportional system of

Talairach et al. (1967). These investigators found a significant increase in right hemisphere flow

(mean average increase 10%) and no difference in left hemisphere flow (mean average increase 3%).

Bilateral regional increases were seen in SMA, the face sensorimotor region, and the posterior superior

temporal region. On the left side, increase in SMA flow was more pronounced, and face sensorimotor

and temporal regions of activation were more distinct from each other. Bilateral regional decreases

were seen in parietal regions. An analysis of repeated measurements on a subset of patients revealed

the same pattern of activation with a decrease in the magnitude of change from rest to speech. The

poor resolution of the images acquired in these studies precludes further anatomical specification of the

regions of interest.

As the SPET technique used in these studies is substantially different from the previously

reviewed PET techniques, it is important to keep in mind a different set of limitations to the

interpretability of these studies. The most obvious of these is the nature of the experimental samples.

Ingvar and Schwartz's sample consisted of ten or eleven psychiatric patients "with normal neurological

findings." The data show ten subjects, but eleven are described. Five were chronic schizophrenics,

two had recent manic episodes, and one each had "endogenous depression," alcohol abuse (dry at the

time of the experiment -- it is unclear for what period of time), "chronic lymphadenopathy of uncertain

etiology," and depressive symptoms following a mild head injury. The medication status of the

subjects was not discussed. Likewise, the sample that Larsen and colleagues used had clinical

symptoms sufficient to warrant clinical angiography, and an unspecified number of subjects received

"a small dose of diazepam."

As discussed earlier, the SPET technique has its limitations as well. Both studies were

extremely invasive, as all of the subjects received their isotope (3-5 mCi of 133Xe) into their

cannulated internal carotid artery (side not documented), and seven of Ingvar and Schwartz's subjects

also had the jugular vein cannulated. Although Larsen et al.'s study had greater localizing power due

to a greater number of cameras, it is important to keep in mind that the detectors used in these studies

are "looking straight through" the brain. Although these detectors measure the activity in the outer

third three times as effectively as the inner third (Risberg, 1980), the extent to which subcortical

structures contribute to the observed levels of activity is uncertain. A number of mathematical models

have been derived to minimize the impact of this problem (Risberg, 1980); however, the most effective

of these were not employed in these studies. Finally, the measurement of only one hemisphere in each

subject makes it difficult to disentangle the contributions of regional and global changes to the

asymmetry observed in the sample of Larsen and colleagues.

A summary of the findings of these two studies may be found in Table 2-3. Technological

limitations make interpretation of these studies beyond a gross level difficult. The consistent data from

the two studies suggest that the generation of automatic language is marked by an increase in rCBF in

SMA (particularly on the left) and a decrease in rCBF in the temporo-parieto-occipital association

cortices. The SMA increase is consistent with what would be expected from the lesion literature, while

the lesion literature does not speak directly to the temporo-parieto-occipital decrease. Given the

similarities of the stimulation and control tasks used in the two studies, it is difficult to determine what

sorts of differences in cognitive processes would account for the different conclusions about changes in

other regions. Examination of the reports of the two studies suggests that these differences may be

Table 2-3
Summary of Functional Neuroimaeine Findinss durine Automatic Lanaiage Stimulation Tasks

5stry Stimulation losk Control task Frontal ioesnlaterl lnoerton Latetal Medil Postenor Fewpo- "hter
Operoluin Prefrotal Prefrotal Supeetor Srpetior Prieto-
Frontal (SMA) Temporal Occiptal
00g,0 & St walMz (1914) Recte tays 01 01h weel no Keclise Silently with eyes L L L L L"Wie 0114
(ONLY LEFT mont of the yew cdosd (deoee) Anterior

L 0. 51.0j, & .10 rote to o Reclroe siently wi eyes B H U seno-r
(1978) recite daysofth e eek ti ed d -erease) F re;
R Heiphere
Note. B denotes bilateral change in activity, L denotes change in left hemisphere, R denotes change in right hemisphere, M denotes change in a midline
region. Empty cells denote those that are not discussed, while "-" denotes areas that are specifically identified as not changing.

more a function of technological and data-analytical techniques rather than the patterns of change in


Changes engendered by reading have also been used to investigate language production. In

the functional neuroimaging literature, reading aloud is typically conceptualized as the visually-

mediated analog of repetition. In their SPET study, lngvar and Schwartz (1974) had their subjects

"read aloud from simple texts from an ordinary weekend magazine" and then compared the resulting

rCBF pattern to a blindfolded resting control state. Again, there were no significant changes in mean

hemispheric flow between the two conditions. However, regional increases were noted in premotor,

sensorimotor, superior sylvian, middle temporal, and temporo-parieto-occipital regions. The increase

in the temporo-parieto-occipital activity was greater than the increase seen when comparing the

automatic speech task to the resting control state. Petersen and coworkers (1989) included a vocal

single word reading task in their paradigm, and they compared the resulting rCBF pattern to the pattern

seen during silent reading of single words. Results suggested that vocalization increased activity

bilaterally in the primary sensorimotor mouth region, SMA, superior anterior cerebellum, and superior

colliculus. Unilateral increases were seen in the left inferior premotor cortex, left rolandic/sylvian

junction, and right mid-sylvian superior temporal region. Of these increases, the sensorimotor mouth,

cerebellar, and collicular increases were greatest.

Two functional neuroimaging studies have been performed to examine the pattern of rCBF

engendered by repetition of aurally presented words. Petersen and colleagues (1989) required subjects

to repeat aurally presented nouns. When subtracting the pattern of blood flow during passive listening

to single words from the pattern seen during repetition, the authors reported bilateral increases in the

sensorimotor mouth region and SMA and unilateral increases at the left rolandic/sylvian junction, in

left inferior premotor cortex, and in right mid-sylvian superior temporal cortex. Of these increases, the

largest changes were the bilateral sensorimotor and right superior temporal changes. In contrast, Wise

and colleagues (1991) bad subjects perform a repetition task aimed specifically at localizing regions

involved in the processing of novel lexical-semantic information. In the stimulation task, subjects were

required to repeat single aurally-presented concrete and abstract nouns. In the control task, subjects

were required to say the same word when aurally presented with reversed words. Comparison of these

two patterns of activation suggested that there was significantly greater activity in the left posterior

superior temporal region during the repetition of concrete and abstract nouns than during the control


Petersen and colleagues (1989) note that previous work from their lab suggests that a number

of the increases seen during reading aloud and repetition may have more to do with motor

programming in general than with single word processing per se. For instance, previous studies have

shown very similar activations in left inferior premotor and left inferior sensorimotor cortex in simple

tongue movement and hand movement (Fox, Pardo, Petersen, & Raichle, 1987). Similarly, rCBF in

SMA has been shown to increase during simple tongue, eye, and hand movement (Fox et al., 1987;

Fox, Fox, Raichle, & Burde, 1985). In addition, Fox, Raichle, and Thach (1985) reported that superior,

anterior cerebellum and colliculus have been shown to increase in terms of rCBF during hand and eye

movements. The authors account for the lack of activation of the latter structures in the repetition of

aurally presented words by reporting that these structures (at least the cerebellar areas) did, in fact,

increase in flow, but not to a statistically significant extent.

A summary of the findings related to vocal reading and repetition is contained in Table 2-4.

When comparing these findings to those noted in the investigations of automatic language as well as to

the motor findings of the St. Louis lab, it is apparent that the majority of changes during these tasks

appear to be related to the initiation and execution of speech rather than language programming per se.

Exceptions are seen when comparing reading and repetition to control tasks in which there was little or

no language stimulation (Ingvar & Schwartz, 1974; Wise et al., 1991). Both of these groups noted

increases in left posterior superior temporal cortex. It is impossible to differentiate receptive from

expressive demands in these two comparisons, however. Another exception which appears to be

language-related is the consistent activation found in Brocas area.

The task which has been used the most frequently in exploring language production with

functional neuroimaging is single word generation. This has been done using a number of different

formats. The format requiring the least generativity to a single stimulus is that used by Petersen and

Table 2-4
Summary of Functional Neuroimazing Findinas during Reading Aloud and Repetition

Study Stamlahmon Task Cont.ol task Frontal ltsoOlateral Interior Leateral Medial Pasteitor Ieto Uher
Opeanutumt Itefrontai Prefrontal Superior Superior Peto
Frontal (SMA) Teoral Occipital
lgyar & Schwartz (1914) Read atood fno tagezine Reoline sientl wita ee L L L Mddle emtporal
(ONLY LEFT article aoned
Poteraen et al. (1989) Red single words fload latively read stogie woLd B Seasontotor
atat stituli) Faotr
B Superior Anterior
B Colltlus
L Rolandid
Sylian junction
R Mid-enpetor
Te ral

(.Uauhstmli) wood, Ea-
L Rolandid
Sylvian juaction
R Mid-superior
Ta ral

Wise et al. (1991) Repeat onrete a Say the same -od eachL t ie revesedtw ord io
(only reversed wood,

Note. B denotes bilateral change in activity, L denotes change in left hemisphere, R denotes change in right hemisphere, M denotes change in a midline
region. Empty cells denote those that are not discussed, while "-" denotes areas that are specifically identified as not changing.

colleagues (1989). The basic format is the same as described previously, in that subjects are presented

with high frequency English nouns at the rate of 1 Hz. In the verb generation task, subjects are

instructed to generate a verb associated with each noun. This task was presented with both visual and

aural stimuli. When the pattern of rCBF observed during vocal reading was subtracted from the

pattern engendered by verb generation (visual stimuli), increased activity was noted in anterior

cingulate cortex (two foci, one inferior to the other), left prefrontal cortex (inferior, lateral, and

dorsolateral; roughly areas 44-47), and throughout the cerebellumn (bilateral posterior, right inferior

lateral, and anterior cerebellum/colliculus). Anecdotal report also suggests that a bilateral decrease

was seen in "sylvian-insular cortex" (Raichle, 1991). When the pattern of rCBF observed during

repetition was subtracted from the pattern engendered by verb generation (aural stimuli), activation

was noted in the anterior cingulate regions, left inferior frontal cortex, right inferior lateral cerebellum,

and anterior cerebellum/coUiculus. It is unclear whether the presence of the aforementioned bilateral

decrease in sylvian-insular activity occurs in both versions of the verb generation task.

In addition to the previously discussed limitations which apply to the studies of Petersen and

colleagues, another complication arises in comparing the verb generation condition to the vocal

reading/repetition tasks. Although the structure of the verb generation condition minimizes the amount

of generativity associated with one stimulus and maintains consistency in presentation rate between it

and the other Petersen et al. tasks, behavioral data suggest that the task is much more demanding than

the others. The authors anecdotally report that the subjects reported this task to be the most difficult

(Petersen et al., 1989), and Fox (as cited in Raichle, 1991) claimed that verbs were successfully

generated to only 40% of the nouns on average. Thus, both rate of language output and relative effort

may be having some systematic impact unaccounted for by the design of the tasks. Studies examining

the effects of task difficulty upon rCBF suggest that such variation may lead to systematic differences

in prefrontal flow indices, with more difficult tasks leading to greater rCBF increases in this region

than easier tasks (Gur et al., 1988). Indeed, recent anecdotally-reported findings suggest that when the

presentation rate is changed from 1 Hz to 1.5 Hz, one observes a marked difference in the resulting

subtraction image (Raichle, 1991). For instance, one observes a dramatic increase in the activity of the

left posterior superior temporal region. Raichle notes that this response was present in the 1989 study,

but its magnitude did not reach statistical significance. Unfortunately, the effect of this manipulation

on the activity observed in other regions is not discussed.

Other studies carried out in the St. Louis lab described by Raichle (1991) have examined the

effects of practice on the observed rCBF responses during the tasks. He reported that, with as few as

six practice trials with the same word list, one sees a significant reduction in the magnitude of the

rCBF response in left prefrontal cortex, anterior cingulate cortex, left posterior temporal cortex, and the

right cerebellar hemisphere. Further neuroanatomical specification was not provided. Accompanying

these reductions was a significant increase in sylvian-insular activation bilaterally. It may be more

accurate to view this latter increase as a recovery from the initial decrease seen in the early trials,

however, as the activity level in the sylvian-insular area after practice is back to levels observed during

"normal speech" rather than above its "baseline language production" level. Raichle likens the

distribution of rCBF seen in practiced verb generation task to that observed during the reading of nouns

aloud. It should be emphasized that the findings concerning presentation rate and practice are

impossible to evaluate fully since the results and methods employed in arriving at the conclusions have

never been formally reported. Thus, one is left to accept the findings "on faith," and any further

hypotheses based upon this work must be made with the utmost caution.

Another format of word generation has been used by Wise and his colleagues (1991). These

investigators had subjects silently generate as many verbs associated with an aurally-presented noun as

they could within four seconds. Subjects were instructed to signal their arrival at the first verb by

opposing thumb and forefinger of the left hand, and they were asked to limit themselves to thinking of

verbs without forming sentences. Fifteen nouns were presented per minute, and subjects

retrospectively reported an average of two to four generated verbs per noun. When comparing this

scan to the one acquired during the resting control state, the authors noted increased activity in the left

posterior superior temporal gyros, left posterior middle frontal gyrs, and midline SMA. Notably, the

activity level reported for posterior superior temporal cortex during the verbal generation task did not

vary from that reported during listening to nonwords of typical English phonological structure or

deciding if two words are correctly categorized semantically.

Single word generation has been investigated with a different methodology by another group

of investigators at Hammersmith. Friston, Frith, Liddle, & Frackowiak (1991a) had four subjects (1)

count out loud, (2) decide if aurally-presented morphemes were real words ("correct" or "incorrect"),

(3) generate animal names, (4) generate words that start with the letter "a." Two resting control tasks

during which the subjects were instructed to recline quietly with eyes closed were not used in

comparisons. As pilot work showed that the average rate of output in tasks 5 and 6 was approximately

one word every 2.5 seconds during the first 90 seconds of the task, this rate was chosen for the

counting and lexical decision tasks. In the lexical decision task, high frequency concrete words

beginning with the letter "L" were used, with half of the stimuli being orthographically regular

nonwords in which the first letter of a word was replaced by "L."

For the purposes of their analysis, the authors combined the images from the word generation

conditions into an "average word generation" image. No explanation for this combination was offered,

and no direct comparison of the two generation tasks was reported. The SPM's derived by comparing

the generation image to the counting image and the lexical decision image to the counting image,

respectively, were not displayed in the report. The authors noted that the only region to decrease

significantly in the generation tasks and increase significantly in the lexical decision tasks was left

posterior superior temporal cortex. Comparison of the pattern of rCBF in the average word generation

image to that seen in the lexical decision task suggested greater rCBF bilaterally in the anterior

cingulate and unilaterally in left dorsolateral prefrontal cortex (Brdmann's areas 9, 46), left

parahippocampal gyros (area 28), and left parietal cortex (supramarginal gyms) during the generation

conditions. During the lexical decision task, on the other hand, greater activity was seen bilaterally

along the superior temporal gyros and unilaterally in the right frontal operculum.

In addition to the changes in level of activity, Friston and coworkers also explored the

correlation between the average left dorsolateral prefrontal activity and all other regions across

scanning conditions. They noted significant negative correlations between left dorsolateral prefrontal

and superior temporal regions. Although the authors do not discriminate between superior temporal

regions when describing this negative correlation, examination of the figures suggests a substantial

asymmetry. On the right side, the negative correlation appears to encompass the entire superior

temporal region, with extension into inferior parietal and occipito-temporal cortices. In the left

hemisphere, however, the negative correlation appears to be limited to more anterior superior temporal

regions (perhaps primary auditory cortices), with much less involvement of posterior superior temporal

cortex or nearby association areas. Unfortunately, significant positive correlations were not addressed

in the report.

This same group of investigators reportedly explored single word generation again with a

methodology virtually identical to their other study (Frith, Friston, Liddle, & Frackowiak, 1991). The

authors reported that in this investigation, the word generation task was to generate as many jobs as

one can think of rather than animal names. The primary differences between the two studies reportedly

lie in the manner in which the data were analyzed. In this study, Frith and coworkers converted

radioactive counts to rCBF equivalents following the algorithm described by Mintum, Fox, and Raichle

(1989). Analyses suggested that global blood flow decreased in all activation tasks relative to the

resting control state. In contrast, there was no difference in global blood flow between cognitive

activation tasks.

Statistical parametric maps (SPM's) were also computed to determine regional changes. In

order to identify the neural substrates important to the generation of single words, the investigators

compared the mean of the images acquired during both word generation conditions to the mean of the

images acquired during rest, counting, and lexical decision tasks. Interestingly, the SPM's displayed

for this comparison appear identical to the ones displayed by Friston and colleagues (1991a) in their

comparison on the average word generation image to the lexical decision image, thus suggesting

increases and decreases in rCBF in exactly the same locations, In order to identify the neural

substrates important to lexical decision, the pattern of rCBF engendered by the lexical decision task

was compared to the mean pattern of the other tasks. Bilateral increases were noted in "periauditory"

and superior temporal regions, while unilateral increases were seen right inferior prefrontal cortex and

in the left frontal pole. Bilateral decreases were seen in the posterior cinguate gyrus. In order to

investigate the covariance of activation between regions, correlations were computed between the pixel

in the left dorsolateral region with the most significant t-statistic in the generation vs. others

comparison and all other pixels. The displayed SPM was again identical to the one displayed by

Friston and colleagues exploring their analogous correlation, suggesting negative correlations in the

same locations. Positive correlations were not reported.

The investigations conducted by Frith, Friston, and their colleagues have much in common.

Methodologically, they both use the previously-described image construction and data analysis

algorithms defined by Friston and colleagues (1989, 1990, 1991b). They also both relate their

hypotheses to an explicitly-defined, mathematically-derived model. In addition, the authors

investigated not only absolute differences in activity, but also correlations between regions in an

attempt to understand the covariance of activity between regions during specific tasks. However, a

number of factors cloud the interpretation of the studies. At the level of data analysis, the combination

of images acquired during different stimulation tasks into a mean image designed to reflect a specific

cognitive process is overly simplistic and confounds the interpretation of any comparison with this

mean image. Likewise, the use of a single pixel as the "representative" of the activity of a given region

of cortex makes one extremely susceptible to outliers, even with the "smoothing" of the images

discussed earlier. Conceptually, a major confound in drawing conclusions about the role of the

superior temporal gyrus is the fact that the amount of information aurally presented was not held

constant across conditions. Given that the greatest increase in activity in the superior temporal region

was seen in the lexical decision task, which provided the most aural stimulation, it is difficult to

separate this factor out from the contribution of cognitive processes. In addition, the rates of output,

although relatively consistent over the first 90 seconds of each stimulation task, were different in

nature. The lexical decision and counting tasks each involved a constant output demand, while the

word generation tasks involved a much less linear output function. Thus, one wonders if one is

looking at as homogeneous a cognitive process over the scanning interval in the word generation tasks

as in the lexical decision task. Also, the lack of a comparison between generativity tasks is not

discussed. This is puzzling, as these tasks reflect somewhat different processes. In addition to subtly

different cognitive requirements, these two tasks have been shown to differ in terms of level of

performance (Lezak, 1983). Some sort of ratings of the relative difficulty of each task also would have

been helpful, as this may play an important role in determining extent and/or magnitude of activation,

especially in prefrontal regions (Gur et al., 1988). Finally, the exact duplication of results on a very

small sample suggests that these results are separate reports of the same investigation rather than a

replication, although it should be noted that Frith and colleagues do not present their data as


Warkentin, Risberg, Nilsson, Karlson, and Graae (1991) investigated single word generation

in a more traditional format. These authors used 133Xe inhalation SPET in 39 subjects (19 men, 20

women) to examine rCBF patterns during a resting control state and while generating as many words

as possible that begin with a given letter for sixty seconds. Regional analyses within tasks showed

greater rCBF in frontal than in post-central or temporal regions, a finding consistent with earlier studies

(Ingvar & Schwartz, 1974). Comparison of the two rCBF patterns suggested that there were no

differences in mean hemispheric flow within or across conditions. Regionally, however, word

generation was noted to engender a significant increase in flow in the left anterior frontal region, while

flow was decreased in left superior sensorimotor and left anterior parietal/sensory regions. In temporal

and frontotemporal regions, right-sided values were greater in absolute terms and increased more than

did left-sided values during single word generation. When the investigators combined the values for

the three most anterior frontal probes into a single mean, they found that, during word generation, the

flow values from the left prefrontal region were significantly higher than the values of their

contralateral homologues. In addition, the left prefrontal mean was higher during the word generation

task than at rest. In contrast, there was no difference in the right prefrontal mean across tasks.

Although females' mean hemispheric flow was slightly higher than that of their male counterparts,

there was no difference in prefrontal rCBF change according to gender. Likewise, the prefrontal flow

values observed in sixteen "highly productive" subjects did not differ from those of the rest of the


The investigators went on to discuss some of the individual variability in their data. They

reported that six subjects (15% of their sample) showed neither left nor right increases in prefrontal

flow values during the word generation task. As analyses showed that this group's resting prefrontal

values were significantly higher than those of the other subjects, the authors hypothesized that the

change was indeed present but masked by the already high baseline level of activity. The authors do

not speculate upon how this dynamic would work, however. This hypothesis assumes one of two

possibilities. First, it may be that there is a maximal value for rCBF in the left prefrontal cortex, thus

preventing the high baseline group from showing enough of a difference to reach statistical

significance. On the other hand, it may be that there are areas in prefrontal cortex whose flow values

are not affected by word generation, and these areas were already activated to such an extent in the

resting condition that the increase in flow secondary to word generation did not constitute a statistically

significant increase in flow to the region.

There are a number of strengths to this study. Foremost is the nature of the sample. Sample

sizes under 15 are the norm for this literature, and samples are commonly made up of either inpatients

or groups of normals whose members may be familiar with the hypotheses or stimuli used in the study.

Neither of these wealmesses are present in this study, as all 39 subjects were recruited through local

advertisement. While this undoubtedly did not result in a sample representative of the population at

large, it is certainly intuitively more comfortable to make generalizations based on a large sample of

experimentally-naive subjects. In addition, these subjects were screened in order to rule out mental

illness, substance abuse, hypertension, and neurological disorders. Finally, a brief neuropsychological

battery was administered to insure that subjects were within normal limits in terms of verbal ability,

verbal memory, spatial ability, reaction time, and visual constructive ability (although details and

scores of tests were not reported).

Clear interpretation of the data is clouded, however, by a lack of control conditions. Since the

cognitive resting state as defined by Warkentin et a, is inherently uncontrolled, it is impossible to

attribute the observed rCBF changes to a specific cognitive act; indeed, the uncontrolled nature of the

resting condition may explain in part some of the variability in the observed changes. Again, as with

other SPET studies, the poor localization abilities of the technique must be kept in mind when

interpreting the results. This limitation tempers the power of the left-prefrontal activation finding in

particular, as the authors had to combine three relatively poorly-localized regions (with an unknown

subcortical component) before their findings became significant. As the resulting comparisons involve

large regions of both lateral and medial cortex, it is difficult to know exactly what the results mean.

A summary of the findings related to single word generation may be found in Table 2-5. As

with the investigations of other components of language, there is substantial variability between labs.

There are a number of consistencies, however, in examining those regions which appear to become

increasingly activated by the generation of single words. The most consistent increases in rCBF are

found in the anterior cingulate and in the left dorsolateral prefrontal cortex. Given the extensive

interconnections between these regions as well as their theorized role in systems which mediate

intention (Heilman, Watson, & Valenstein, 1985), these activations are not surprising. Most other

findings are consistent within but not across laboratories. As most experimental and data-analytic

parameters vary considerably from laboratory to laboratory, this variability is also not surprising.

Substantially different findings are even found within groups (Friston et al., 1991a; Frith et al., 1991;

Wise et al., 1991). In comparing studies, it is again evident that the factors of rate of stimuli

presentation and generation are extremely important to determining one's results and subsequent

interpretation of the neural instantiation of networks required to generate single words.

A concrete example of this point is provided by the comparison of the findings reported by the

Hammersmith group. The results reported by Friston, Frith, and colleagues' word/animal/job

generation tasks appear to be at odds with those of Wise and colleagues (1991). Wise and colleagues

find that both left dorsolateral prefrontal cortex and left posterior superior temporal cortex become

significantly activated in a verb generation task, while Friston, Frith, and colleagues report a decrease

in left superior temporal activity and an increase in left dorsolateral prefrontal activity during word

generation. There were subtle differences in the tasks which may account for the differences. Friston

et al./Frith et al. (1991) had their subjects generating animal names or words that begin with the letter

Table 2-5
Summer of Functional Neuroimatinn Findines during Word Generation

Ntdy Stimulat n Tat I C nh Oi ktla troatl iotn aterna teno, Lateral Medial Po enotr taripomr Oither
Operttlim Prefoittal Pteferatal Superior SUpeior Partii-
tertal (SMA( Terttua Ooiptd

tM eten et al ( 1" 9) deerate a so iated venl tired single no unsr lotd L I Cin)t la twO r
vitall stimuli) Cigolo)r(t
M Arter r
B Potetior
R Ialeetr Larerl
B drit r in

Geset assocIe vrb. repeat single trura M Atteolor
unnalsmuli)Cinguil'"e mto
(aural ttitulil ci)
M Ainterir

R Interior LaIral
(B dre- in
Insulr" regiohf
w taret al. (1991K RiUMer severe teil ntilalty with eyes L M
MIUttaor a ml (199111 Averae r[genaon to Letoal dstei.- in K (. B Aiteamt
g lad g t of (dt- (d-r Cingulate
a at geart~ot L Peatohp-
umnil p--WP-Gyrm;
B doteaeSuperit

1tIh et at. (1991) AIveaget gaeRahtF to Average at reI.lout a L n Antera
"."and gerfatini of jb sitieay with eyes closed, decreasee) (deW-)e Citgatden
tattet itaihgaloud. sod leiasl L Prah
decision peampl On
B- Si

Lexial decide. Average of realingt IR BB-p Spen
dleddly itt aesd ate. T"ttnii = tt
ooiathvg atood. generation IFioa oe
t "a and genterationof job B dere,

Weakentin at M. 0991) Grtare wOMr Woa single Relin itelently Wit etyesL L derise St~er-n
levee. farW d ti aol osed stuennatoor
.Peoe de(-e aSnPerMt

Note. B denotes bilateral change in activity, L denotes change in left hemisphere, R denotes change in right hemisphere, M denotes change in a midline region Empty
cells denote those that are not discussed, while ""denotes areas that are specifically identified as not changing

"a" or jobs for a minimum of 90 seconds, while Wise et al. (1991) had their subjects generating

"appropriate verbs" to concrete nouns which were presented at a rate of one every four seconds. In

addition, Wise et al.'s subjects retrospectively recalled producing 2-4 verbs every four seconds, while

three out of four of Friston et al's subjects generated words at a rate of approximately 2 words every

five seconds (the fourth subject was slower). Thus, in comparison to verb generation, the generation

task of Friston, Frith, and colleagues was a) less restrained semantically, b) less demanding in terms of

output rate, and c) more exhaustive in terms of the required amount of "searching" (lexical or semantic,


In addition, it is important to keep in mind the choice of control tasks made by each group of

investigators. Friston et al. compared his "average word generation" images to images acquired during

"automatic language" (i.e. counting aloud), while Wise et al. utilized an unstimulated rest condition as

his control. Thus, the former comparison theoretically represents those areas that become activated

primarily as a novel word generation resulting from lexical and semantic searches, whereas the latter

would encompass all language functions involved in silently generating verbs to nouns. One might

thereby conclude that the activation of posterior superior temporal cortex is important in the generation

of words at a cognitive level "lower" than the "search" function per se. If this is the case, then one

must presume that Friston's automatic speech task activated left posterior superior temporal cortex to

such an extent that word generation did not add significantly to this level of activation. Such a

presumption is consistent with the findings of Ingvar & Schwartz (1974) and Larsen et al. (1978), both

of whom showed left posterior superior temporal cortex activation (although not in as well-defined a

manner as did the PET studies) during "automatic speech" tasks as compared to rest control conditions.

Finally, a difference discussed solely in the context of interpreting Friston et al's findings may

be critically important: that is, the fact that Wise et al.'s subjects were being aurally presented with

new information every four seconds, while Friston et. al's subjects were being presented with none. If

left posterior superior temporal cortex is involved in the "decoding" of incoming linguistic information,

then this discrepancy alone would account for both a) Friston et al's differences between lexical

decision and word generation (between which there was also a great discrepancy in the rate of stimuli

presentation), and b) the differences between Friston et all's word generation findings and Wise et al's

findings. One might point to the lack of posterior superior temporal cortex response in the Petersen et

al. (1989) verb generation task, in which subjects were presented with new information (either visually

or aurally) at a rate of 1 Hz, as confusing in light of this interpretation. However, the "activation"

engendered by this task was reported in the context of a (verb generation) (repetition) subtraction

image, thus theoretically "removing" the activation engendered by the rate of presentation of stimuli

per se.

The final component which has been examined in functional neurointaging studies is the

construction of discourse. Although this represents an extremely complex combination of cognitive

components, it is the only data to date which are able to even indirectly examine the instantiation of

grammatical formulation. Lechevalier, Petit, Eustache, Lambert, Chapon, and Viader (1989) used

133Xe inhalation SPET to examine language formulation in seven men and three women on their

Neurology unit. Images were acquired for a resting control state, during which subjects were

instructed to neither move nor speak for ten minutes, and during a word definition task, during which

subjects were instructed to "say as much as they could" about a list of 17 nouns and 8 transitive verbs

of high frequency by either providing a definition of the word or using it in an appropriate context. A

new word was presented every 30 seconds. When comparing the patterns of rCBF from these two

conditions, the authors found significant bilateral increases in hemispheric flow (approximately 16%

on the right and 15% on the left). Regionally, increases were noted in virtually every part of cortex,

with the greatest increases noted bilaterally in the inferior parietal cortex, right superior parietal cortex,

right dorsolateral frontal cortex, left posterior superior frontal, and left inferior frontal cortices

(approximately Broca's area). The location of the left posterior superior frontal activation suggests

involvement of SMA, but this was not investigated further. In addition, the study was subject to the

previously-discussed limitations of the 133Xe inhalation SPET technique.

Finally, Wallesch, Henriksen, Kornhuber, and Paulson (1985) had six staff members of the

Department of Neurology and the Language Pathology Unit of Ulm University perform a number of

tasks: (a) rest, (b) oral movements without phonation, (c) repeated counting from one to twelve, (d)

production of "random" consonant-vowel and consonant-vowel-consonant syllables, (e) retelling of a

magazine story read ten minutes before starting, and (f) silent retelling of a story as in (e) without any

detectable coarticulation. Subjects were instructed to speak (silently in condition f) at a rate of 1

syllable per second. Analyses were all performed using a 2 cm thick transverse image containing

Broca's and Wernicke's areas, head of the caudate, lenticular nucleus, and the thalamus. The slices

used in the comparisons were constructed by creating normalized flow distribution images of the slice

(rCBF over gCBF for each pixel) for each subject, averaging these normalized images across subjects

within each condition after "careful positioning to minimize the effect of different size and shape of the

heads," and comparing the averaged task images to each other with a subtraction algorithm. The

average amount of change across all pixels between the tasks being compared was then calculated, and

"activated regions" were those pixels in which change was one or two standard deviations greater than


When subtracting the rest condition from the vocal storytelling condition, the authors

identified bilateral increases in posterior temporal/inferior parietal/anterior occipital regions (the

uncertain angle of the cut makes it difficult to determine precisely the location, and it is referred to by

the authors as "retrorolandic") and unilateral increases in left frontal operculum, right lenticular nuclei,

left head of the caudate nucleus, medial anterior frontal region (possibly anterior cingulate), left

anterior temporal regions, and in the left anterior thalamic/pallidal region. Subtraction of the rest

image from the silent storytelling image revealed bilateral increases in the same "retrorolandic"

regions, in the head of the caudate nucleus, and throughout the frontotemporal cortex, while unilateral

increases were seen in left posterior thalamnic regions. When subtracting the iCBF pattern engendered

by syllable generation from that engendered by vocal storytelling, the authors noted bilateral increases

in similar "retrorolandic" regions and unilateral changes in left Broca's area, left head of caudate, left

anterior thalamic region, and anterior cingulate. A weaker increase was seen in a more widespread

fashion over the left frontal lobe (both laterally and medially) and in the right head of the caudate.

Subtraction of the counting image from the vocal storytelling image revealed bilateral increases in the

same "retrorolandic" regions and the head of the caudate, while unilateral increases were seen in left

frontal operculum, left anterior medial frontal cortex (anterior cingulate), left anterior frontotemporal

regions, right anterior medial frontal regions (anterior cingulate), and left anterior thalamus.

There were a number of technological and analysis issues which cloud the interpretation of

this data. Foremost is the manner in which regional flow distributions were compared between

conditions. In constructing "average flow distribution" scans, the authors necessarily limit the

accuracy of their localization of brain areas due to the superimposition of different brains upon each

other with no attempt at standardization. This, combined with the relatively thick SPECT and CT

slices and poor resolution of the technique available at that time restrict interpretation to "general area"

statements at best. This difficulty in interpretation is compounded by a lack of presentation of the

rCBF data in numerical form, thus forcing the reader to rely upon the investigators drawings of the

regions of activation as the only measure of extent of activation of the various regions. This makes

direct comparison between regions very difficult. Furthermore, in comparing slices, decreases in flow

were not mentioned: it is unclear whether this is due to their absence or the investigators' emphasis on


The investigators' method of using the standard deviation of change in the ratio of rCBF to

mean CBF as a measure of significance also has ramifications which should be kept in mind, especially

in light of their lack of mention of decreases in activity. If most regions of the brain respond to a task

in a relatively consistent manner, then the standard deviation of change will be small, and it will take

less deviation from the mean of change to be considered significant. In contrast, if there is more

variability (in both positive and negative directions) in the degree of change across regions of the brain,

then a substantially greater change may be required to reach statistical significance.

Conceptually, this study had some limitations as well. Output in each condition was limited

to one syllable per second, thereby necessitating each subject's monitoring of rate of output. Aside

from the fact that this sort of monitoring is not usually associated with the language phenomena to

which the results were to be generalized, it is uncertain whether the effort needed to maintain this rate

of output would be consistent across the varying tasks, given their disparate nature (single word or

syllable output vs. narrative language). In addition, only those comparisons between high levels of the

hierarchy (story retelling) and lower levels of the hierarchy (rest, syllables, counting) were discussed.

One is left wondering whether the comparisons between more adjacent steps simply were not

performed, or if statistically significant differences were observable only between more distant steps.

The results of the investigations of discourse are summarized in Table 2-6. The most obvious

characteristic of these data is their variability. As the demands of these tasks are the most complex and

the designs of the comparisons to be made were among the most uncontrolled, this is not surprising.

Perhaps the most accurate statement that can be based on these data is that the production of discourse

engenders wide-spread activation both cortically and subcortically when compared to relatively basic

linguistic tasks. The involvement of left frontal operculum, anterior cingulate cortex, and left head of

the caudate nucleus appear to be the most consistent findings across comparisons, but delineation of

which specific linguistic functions these changes may be involved in is simply impossible based on the

available discourse data.

As with single word comprehension, an integration of the functional neuroimaging and lesion

data concerning the production of language yields a number of consistencies as well as unanswered

questions. It is clear that the anterior cingulate, left frontal dorsolateral/opercular (Brodmann's areas

44-46) are somehow involved in the normal production of language, as these regions are consistently

shown to increase in rCBF when comparing a language production task to rest. The anterior cingulate

activation appears to be most strongly related to the drive to generate language, while the dorsolateral

frontal cortex and frontal operculum appear to be influenced by issues of intention as well as language

formulation. It is unclear at this point exactly how the frontal opercular activation seen in normals is to

be reconciled with the finding that lesions confined to this area usually have minimal impact on

language function (Mohr et al., 1978; Nadeau, 1988); nevertheless, the consistency of the finding

suggests that this region is involved in the normal production of language in some fashion. Finally, it

is apparent by examining the available data that no functional neurounaging studies have been done

which adequately address the grammatical formulation during the production of language.

Table 2-6
Summary of Functional Neuroimagin Findings during Discourse Production

slwy stiniolation task Loorol Task Vroltl DOsoteraw inletnio Lateral Media i'seoior Teapo 6cr
Opeaulum Prefrontal Prefrontal Superior Sperior Paroelo-
Fena SMA) Temral n
LecIevaltev e l, (1939) Define or K a given wolLL L Most e-loa other
for 30 n daond th inferior
ternoral ard
supeio pan-
L It.-=
L tinisepl 1re
R Ueisiphere
ncrease 16%
iWslench r01a (1960) Vocally rete nagaine eLt m B B M Aft-
aioey read 10 mnts Cingulaie;
earlier L Anterior
L Head of Cadate.;
L Anterior
R Lettiould Nuclei
Siently ,ll ganne e-- B B a. a. .a..d a .c.ait..
story tead 10 irinotes L tasterior
earlier Thalai-
v Ty reo maz.e Gena oe m m n e L B M Ator
story ted 10 1titnu syllables Cinglate;
earlier L Head of Cadate
L Aatarior

voooly Metc .aiaagoio LatigRpaEd-lyft.oI L ai Rtead ofiScad
storyeread lominute 1to2 L Anterior
earlier Ciaglae
B Aaterior
L Asior Frato-
Note. Uwfbi~erTleaipyaralrep1looa
Note. Ideotes hilateent oheita aotltty. d teraa o ittkcemiakee.Keota e ia =fit .Mitt in a lie ion. Easty neltt cneetoettaireareot dihiteil&vee
"0denotes areas that wre especially idetifiedt nt crangiao,

It appears that a phenomenon analogous to that seen in the investigations of single word

comprehension occurs when investigating language production: that is, that one must control for the

"primary responses" before being able to elucidate the region involved in the "higher" processes of

language formulation. In the case of language production, these "primary" responses appear to be the

basic initiation and motor execution of a language response. It may well be the case that the anterior

cingulate, SMA, and dorsolateral frontal responses are strongly related to these processes, and that

these processes are so fundamental to the production of language that they in turn are represented by

the largest rCBF response (analogous to the primary sensory responses and comprehension). Thus, in

order to investigate the processes of message formulation, labeling of the message via the output

lexicon, further defining the message through concantenation and syntactic manipulation, and

phonological programming of the message, one must do so in the context of a paradigm in which both

primary sensory and output processes are controlled as much as possible.

In summary, then, there are a number of questions which remain unanswered in the

comprehension of single words and the generation of language. Although lesion studies have been

extremely informative about potential sites of specific linguistic functions, one is always limited by the

inability to ascertain whether one is observing the function of an impaired system, an intact

compensatory system, or some combination of the two. The study of normals using functional

neuroimaging is meant to address these ambiguities, but to date these investigations have been marked

by experimental paradigms that make it extremely difficult to delineate the occurrence of specific

cognitive processes and how these processes might relate to the patterns of functional activity that are

reported. Likewise, it seems that no two laboratories share the same equipment or data analysis

algorithms, and the extent to which each reported set of data actually reflects the biological data that it

is meant to reflect may vary widely. As such, the replicability of results becomes of the utmost


There are precious few results that have been replicated across a number of investigations, and

most of these are consistent with what one would predict based on the lesion literature. The responses

of primary and secondary sensory cortices are consistently shown according to expectations. Likewise,

posterior area 37/anterior area 19 appear to be consistently involved in the early stages of the

comprehension of the written word, while the left posterior superior temporal region/temporo-parietal

junction (Wernicke's area, inferior supramarginal gyros, area 37) is consistently identified in the

comprehension of the spoken word. During the production of language, one almost always sees

increased activity in the anterior cingulate, supplementary motor area, left dorsolateral frontal cortices,

and left frontal operculum.

There are a number of contradictory findings and questions which have yet to be carefully

investigated, however. There are no interpretable data available on the functional neuroimaging of

grammatical formulation in language. Likewise, the roles of the different regions represented in the

temporoparietal junction have yet to carefully investigated in tasks that were matched according to

input and output demands as well as difficulty level. In addition, the apparently- contradictory findings

about the roles of the inferior lateral prefrontal cortex (area 45 and/or 47) and posterior superior

temporal region in the semantic comprehension of language remain confounded with stimulus modality

and poor choice of disambiguating control conditions. Furthermore, the impact of task difficulty on

reports of changes in prefrontal activity have never been systematically examined in this literature.

Since these regions are certainly important to the production of single words and quite possibly to

syntactic function as well, this sort of statistical control is extremely important.

The purpose of the current study, then, is to address a number of these issues in the context of

a regional cerebral blood flow study of word generation. Tasks were designed to preferentially assess

the phonological, semantic, and syntactic aspects of language formulation. In order to control for the

"primary" responses involved in comprehension, subjects were presented with a constant number of

stimuli at a constant rate across tasks. Similarly, "primary" output responses were controlled by

requiring the subject to formulate a response in each task which was preferentially geared towards the

targeted linguistic function (i.e. phonological formulation, semantic formulation, syntactic

formulation). Also, language formulation occurred in the absence of vocalization in an attempt to

control activation related to motor execution of the response. Finally, tasks were relatively equated on

the dimensions of input demand, output demand, and difficulty.

Acaoss-task Predictions:

(a) There will be no systematic effect of type of generative task (phonological, semantic, or

syntactic) on global blood flow. This hypothesis is based on the fact that, to date, when

comparing between "linguistically activated" states, differences in global blood flow have

never been shown.

(b) Regions of interest on the right side of the brain will not change across tasks. Although

research exists implicating simple semantic functioning in the right hemisphere, this

processing does not appear to be relatively localized as it is on the left (Baynes, 1990).

(c) The activity seen in the left posterior superior temporal region (Wernicke's area) will be

significantly greater in the semantic and syntactic tasks than it will be during the phonology

task. This hypothesis is based on the apparent specialization of this region for lexico-semantic

processing indicated in both classical aphasiology and an integration of the available

functional neuroimaging data.

(d) The activity seen in left dorsolateral prefrontal cortex (Brodmann's areas 9, 45-46) will be

significantly greater in the syntactic task than in the other two tasks. This hypothesis is based

on the work on agrammatism as well as other frontal lobe lesions which indicate that syntactic

processing (particularly at the sentence level) is handled preferentially by left prefrontal

regions (Alexander, Benson, & Stuss, 1989; Goodglass & Menn, 1985; Nadeau, 1988). As

the demands for novel output will be held relatively constant across tasks, greater activity in

this region in the syntax task should be indicative of syntactic processing rather than verbal

generation per se.

(e) The activity of regions theoretically implicated in language processing will be significantly

correlated across tasks, reflecting the change in activity of an integrated network.


Within-task Predictions

(a) Each condition will be marked by an asymmetry of flow in dorsolateral prefrontal cortex. This

hypothesis is based on the consistency with which these results are found in the imaging

literature as well as the impediment in verbal fluency which is seen in lesions of left

prefrontal, but not right prefrontal, dorsolateral prefrontal regions. As this consistency holds

in the literature regardless of vocal or nonvocal language generation, it is likely to be present

in all three of the experimental conditions.

(b) The semantic and syntactic tasks will be marked by significant asymmetry in Wernicke's area

due to the lexico-semantic processing demanded by these tasks. This hypothesis is based on

the same foundations as across-task hypotheses (b) and (c) above.




The purpose of Pilot Study #1 was to determine the relative difficulty levels of the three tasks

to be used in the primary study. These tasks are as follows:

Nonword rhyming the subject is presented five nonwords conforming to the spelling rules

of the English language and is asked to produce a rhyming nonword for each stimulus;

Semantic association the subject is presented with five words and is asked to generate a

semantic associate for each one. Word sets consist of either five nouns or five verbs in an effort to

make implicit syntactic generation less likely;

Syntactic generation the subject is given five words specifying a) sentence structure (either

passive or cleft-object), b) statement or question format, c) subject, d) verb, and e) object, and is asked

to generate an appropriate sentence. The sentence structure and format is always presented in the top

and second-from-top positions, respectively, while the order of the subject, verb, and object is

randomized across stimuli sets.

Specific issues to be examined in pilot study #1 were the patterns of error and time taken to

respond to each stimulus. In an attempt to gain a measure of the effort demanded by each task,

response time for each stimulus was recorded. Results of the pilot study were used to modify stimuli

and instruction sets so as to make the tasks as consistent as possible in terms of effort and input/output



Subjects were fifteen normal volunteers (7 males, 8 females) recruited from the Department of

Clinical & Health Psychology. Subjects were screened by questionnaire to rule out neurological

dysfunction, learning disability, and alcohol or drug abuse.

All stimuli were presented on a Macintosh lIx (Apple Computer, Cupertino, CA) using the

Psychlab software package (version 0.85; Gum, 1991). All stimuli were presented in a triple-spaced

vertical fashion in 24 point Palatino font. The order of the five stimuli on each presentation slide was

randomly determined, except in the syntactic generation condition as noted above. Likewise, the order

in which the tasks were presented was randomly determined for each subject.

Prior to engaging in each task, subjects were trained according to predefined performance

criteria. Criterion consisted of appropriate response to four consecutive sets of stimuli. After training,

subjects were presented with 25 stimuli sets from each task. Response times for each set of stimuli

were obtained by having the subject press the space bar upon completion of the required output. Sets

of stimuli were presented sequentially upon completion of each previous set. There was a minimum of

five minutes between each stimulation task during which training for the next task occurred.

In pilot study #1, the words used in the semantic association and syntactic generation tasks

were chosen solely on the basis of frequency. Three frequency levels were arbitrarily created based

upon the frequency ratings given by Francis and Kucera (1982), and the proportion of words from each

frequency level is roughly equivalent between tasks. The stimuli used in the nonword condition were

rearrangements of the letters in the words presented in the semantic association condition that

conformed to English spelling rules. Efforts were made to ensure that words presented together in the

semantic association condition have no obvious semantic link. Tasks were designed to be roughly

equivalent in terms of input and required output, and constant visual presentation of the stimuli was

used to minimize the memory requirements of the tasks.

The nonword rhyming task is meant to access the phonological component with a minimum

of semantic or syntactic involvement. Psycholinguistically, the steps required to complete this task are

visual analysis, grapheme-phoneme conversion, conceptual generation of phonemically similar outpuL

motor programming of the generated output (phoneme level), and motor execution of the output.

The semantic association task is meant to access both the phonological and semantic

components with a minimum of syntactic involvement. The psycholinguistic functions required by this

task are visual analysis, identification of the stimulus by the visual input lexicon, access to the semantic

system, generation of a semantically similar concept, identification of that concept in the speech output

lexicon, motor programming of the generated output (phoneme level), and motor execution of the

output. "Generation" in this task is subtly different from that required by the nonword rhyming task.

In the nonword rhyming task, some phonemic qualities of the output are specified, but the subject is

essentially asked to generate a novel combination of phonemes (within morphemic limits of the

English language). In the semantic association task, however, the subject's task is more of an

identification task, in that the subject identifies a pre-defined, semantically similar concept in his

semantic network and produces the lexeme associated with that concept.

Finally, the syntactic generation task is meant to access phonological, semantic, and syntactic

processing. Psycholinguistically, correct performance on this task demanded visual analysis,

identification of the stimuli by the visual input lexicon, access to the semantic system, recognition of

the semantic relationships between stimuli, recall of the specified syntactic frame, mapping of the

syntactic frame onto the semantic relationship between stimuli, inflectional morphological alteration of

individual items in the sentence as required by grammatical rules, motor programming of the

generated output (phoneme level), and motor execution of the output. This task requires more working

memory than the other tasks due to (1) the required memorization of the four sentence structures, and

(2) the required perception/identification/semantic decoding of all five stimuli per set before successful

performance of the task could begin.

The time taken to complete each set of stimuli (hereafter referred to as response time, or RT)

was recorded in milliseconds across tasks. Due to the extreme susceptibility of measures such as

response time to outliers, the median RT for each task was computed for each subject. These data are

presented in Figure 3-1. Repeated-measures analysis of variance (ANOVA) suggests there were

significant differences between tasks (p<.001), with post-hoc analyses showing

phonology>semantic>syntax (all at p<.01). Since effort was one of the primary factors to be controlled

in the main study, and RT was felt to be an indirect index of effort, instructions and stimuli needed to

be modified in order to make the tasks more congruent in terms of RT.


Due to the finding of greater differences between tasks than were anticipated, it was decided

that steps should be taken to insure greater comparability between the tasks in terms of performance

time. As such, it was decided that imageability and concreteness would be added to frequency as

criteria in choosing the stimuli to use in the semantic and syntactic tasks. This decision was based on

findings which show that associative difficulty is negatively related to imageability, and, in turn,

imagery and concreteness are positively associated with each other (Brown & Ure, 1969). Thus, it was

posited that reaction time would be inversely correlated with the imagery and concreteness ratings of

the stimuli. It was hoped that by reducing the time needed to generate semantic associates, the

semantic association task and the syntactic generation task could be made more comparable. The

motivation behind Pilot Study #2 was to generate stimuli utilizing these criteria. No measures were

taken to influence the time taken to perform the nonword rhyming task. Frequency ratings were

available in Francis & Kucera (1982), and ratings of imageability and concreteness for nouns were

available from Toglia & Battig (1978). Unfortunately, ratings of the concreteness and imageability of

verbs were not available. Thus, the purpose of the second pilot study was to generate ratings of

imageability and concreteness for a wide variety of verbs.

Median Reaction Time for Each Task
in Pilot #1 (n=15)

............................................................. ............................................

.. .~* ....11....... ................
..... ... .........................................................................

................ .......................................................... ............................

.................................... I .

Phonology Semantic
.-n m 19.87 n 14.02
.d = 7.7 .d 4.84

.d = 1.95

Figure 3-1. Scatterplot of Median Response Times By Task for Pilot Study #1.


Subjects were fourteen normal volunteers (4 males, 10 females) from the Departments of

Clinical & Health Psychology and Speech and Language Pathology at the University of Florida in

Gainesville, Florida. Subjects were screened by questionnaire to rule out neurological dysfunction and

alcohol or drug abuse.


Subjects were given a list of 246 verbs to rate in terms of both imageability and concreteness.

The order of the verbs on each rating sheet was randomly determined in an effort to control for the

effects of boredom. Subjects rated each attribute on a seven point scale. Imageability and

concreteness instructions were those published by Paivio, Yuille, & Madigan (1968), modified slightly

to apply to verbs rather than nouns. The imageability ratings generated by using these instructions are

inversely related to associative difficulty (r = 0.73: Brown & Ure, 1969). Verbs were selected for

inclusion on the list based on pseudorandom selection from Francis & Kucera (1982). The method of

selection was to take every fifth word that was used primarily as a verb. Directions may be found in

Appendix A.


The average ratings of imagery and concreteness given by the raters are presented in Appendix B.

All words chosen for use in subsequent studies had imagery and concreteness ratings of five or more

(nouns as rated by Toglia & Battig, 1978). As detailed in the methods section, words were also

roughly balanced in terms of frequency ratings given by Francis & Kucera (1982).


In order to minimize the contribution of brain activity involved in the motor act of

vocalization to the data, it was decided that the optimal approach to take in the primary study was to

have the subjects perform the tasks without vocalization. A primary purpose of Pilot #3 was to assess

the extent to which the difference between response times to previously-seen sets of stimuli and novel

sets of stimuli could be used to gauge subject engagement in the task. Another goal was to use the data

gathered from Pilot #2 to make the median RT more consistent across tasks by modifying the stimuli

and instruction set.

Subjects were ten normal volunteers (4 males, 6 females) from the Department of Clinical &

Health Psychology (without overlap with the subjects used in Pilot #2). Screening for neurological

dysfunction, learning disability, and substance abuse was done as previously mentioned.


For each of the three stimulation tasks described above, subjects performed three separate

trials. In trial #1, subjects were presented 20 sets of stimuli, and they were instructed to perform the

task as described in Pilot #2 without vocalization. In trial #2, subjects were instructed to merely watch

the 20 sets of stimuli on the screen flash by without engaging in the task. In trial #3, subjects were

presented with 30 sets of stimuli, broken down as follows: 10 sets which had been presented during

trial #1, 10 sets which had been presented in trial #2, and 10 novel sets. Trials #1 and #2 were

counterbalanced in terms of order across subjects.

During trials #1 and #2, each set of stimuli was shown to the subject for 8.439 seconds, with

the screen being blank for 0.200 seconds between sets. This interval represents the shortest median

response time of the three tasks from Pilot #1. The shortest median response time was chosen in an

attmpt to ensure that subjects had to actively process the given task for the entire duration of the trial.

During task #3, response time was monitored as in Pilot #1 so that comparisons could be made

between types of stimuli. The types of sets of stimuli (engaged, not engaged, novel) were randomly

arranged in trial #3.

Imageability and concreteness were taken into account when choosing the stimuli for the

semantic and syntactic conditions. With frequency controlled as in pilot #1, only words with imagery

and concreteness ratings of five or more (on a scale of seven) were used. In addition, slight

modifications were made to the instructions based on feedback and performances of earlier pilot

subjects. Instructions may be found in Appendix C.


Response times for each set of stimuli during the post-test period were recorded in

milliseconds. As in Pilot #1, the median RT for each task was computed for each subject. The data are

represented numerically in Table 3-1 and graphically in Figure 3-2.

In comparing the results of pilot studies #1 and #3, it is evident that the first goal of the study

(i.e. to make the tasks more congruent in terms of RT) met with some success. Although differences in

RT remained (repeated-measures ANOVA F(2, 18)=21.14, p<.001), visual inspection of the data

shows that the differences in absolute RT between the tasks was reduced. In addition, informal

debriefing of the subjects suggested that the phonology and syntax tasks were relatively equal in terms

of subject estimation of difficulty, while the semantic task tended to be somewhat easier. Although the

better comparability between the semantic association and syntactic generation tasks may be

hypothesized to have resulted from the inclusion of imagery and concreteness ratings as selection

criteria, it is unclear why the nonword rhyming task also became more comparable to the others in

terms of RT.

The data in Table 3-1 and Figure 3-2 also depict the differences in RT within task according

to the manipulation of the experimental stimulation. These analyses were carried out to see if these

performances would be valid indicators of the subjects' "cognitive involvement" in the stimulation task.

If this had proven to be the case, these indices could have been used to support the subjects attention to

the stimulation task in the main study, since aural monitoring of the subjects' performance was


In the phonology task, repeated measures ANOVA suggested significant differences between

RT to novel nonwords, nonwords which were seen but to which rhyming nonwords were not


intentionally generated, and nonwords to which rhyming nonwords were generated (p<.01). However,

post-hoc analysis suggests that this difference was between novel nonwords and both of the other sets

Table 3-1
Average Median Reaction Times with Standard Deviations in Pilot Study #3


Passively Observed

Actively Petformed

Nonword Rhvming

g = 14.51 (2.71)

K = 12.83 (2.89)

3 = 12.18 (2.71)

Semantic Association Syntacic Generation

x = 11.54(3.25) g =9.26 (2.43)

x= 10.98 (3.31) R = 7.64(2.13)

X= 11.17 (3.53) R = 6.97 (1.13)

Median Reaction Times
for Each Task in Pilot #3
45 yt
Phonology Semantic Syntax
40 ..................... ..................... i ........................... ............ ..............................................

35 .............................. ........................................ ........................................... ..

30 .... n .. .. ....- ........ 1 .................... . ..... ........................... .............

25 ........ ................

8 03

Figure 3-2. Scatterplot of Median Response Times By Task for Pilot Study #3.

Note: Novel denotes stimuli that were not seen until the post-test. Passive denotes stimuli which
subjects were instructed to passively observe during the stimulation condition. Active denotes stimuli
which subjects were instructed to actively engage according to the task instructions (see Appendix C).
Means and standard deviations of these data are listed in Table 2-1.

of nonwords; nonwords to which the subject had been exposed did not differ in terms of RT regardless

of the generation of intentional generation of nonwords.

In the semantic task, repeated-measures ANOVA suggested that there were no significant

differences in the average median time it took to generate semantic associates to real words, regardless

of previous exposure to the stimuli.

In the syntax task, repeated-measures ANOVA suggested that there were significant

differences in RT across exposure conditions (p<.001). As with the phonology task, however, post-hoc

analyses suggest that these differences were only significant when comparing novel stimuli to those

which had been seen previously. There was no difference in RT when comparing those stimuli which

were passively observed to those in which the subject actively constructed sentences and questions.

The results of Pilot #3 clearly indicate that the difference in RT between "passively observed"

stimuli and "actively-engaged" stimuli would not be a valid indicator of silent engagement in the

activation task. Thus, it was decided that we would attempt to increase motivation to attend to the

stimuli during the stimulation task by offering a reward for good performance on a post-test (see




Subjects were twelve normal male, right-handed volunteers (see Table 3-2). Subjects were

recruited by an ad in the student newspaper of the University of Florida in Gainesville, Florida.

Screening for neurological dysfunction, learning disability, and substance abuse was done as described

in Pilot #1. Handedness was assessed using items derived by Raczkowski, Kalat, & Nebes (1973).

Informed consent was obtained. Before the initial task performance, subjects were administered the

Vocabulary subtest from the Wechsler Adult Intelligence Scale Revised (WAIS-R). This scale is the

most reliable subtest on the WAIS-R, is the best measure of g (i.e. "general intelligence"), and is

correlated highly with Verbal IQ (r = 0.85) and Full-Scale IQ (r = 0.81) (Satler, 1988).

Subjects were each paid $25 for the completion of each of three SPECT scans and an MR

scan as well as an additional $25 for completion of the study, for a total of $125. Subjects were mailed

their payment upon completion of their participation in the study. In addition, each subject was paid

$10.00 per SPECT scan as a motivational tool as will be explained below.

A full description of the scanning apparatus and materials has been given elsewhere (Shukla,

Honeyman, Crosson, Williams, & Nadeau, in press). Briefly, MR and SPECT measurements were

performed at the Veteran's Administration Hospital in Gainesville, Florida. MR measurements were

performed with a Siemans Magnetom 1.5 Tesla (Iselin, NJ). All SPECT measurements were

performed with a three-beaded SPECT scanner (TRIAD-88, Trionix Research Laboratory, Twinsburg,

OH). This system has an in-plane (transverse) reconstruction resolution of 9.3 mn full width-half

maximum (FWHM) and a between plane (axial) resolution of 9.3 mn FWHM at the center of the field

of view. 20 mCi injections of teclnetium-99m-dI-hexamethylpropyleneamine oxime ([99mTc-d,l-

HM-PAO) were administered prior to each SPECT scan. Studies of the pharmacokinetics of [99mTc]-

d,/-HM-PAO, a lipophilic tracer, indicate that once inside the brain, it is rapidly converted to a

hydrophilic form that is maintained for many hours. However, the tracer is not trapped instantly once

inside the brain, thus resulting in a significant amount of backdiffusion into the bloodstream

(Andersen, Friberg, Schmidt, & Hasselbach, 1988). It has been shown that, following intracarotid

injection, the total hemispheric counts reach a peak within 40 60 seconds and then decline

exponentially to a steady-state level of approximately 40-50% of peak activity after about 10 minutes.

After this time, activity decreases at a rate of approximately 0.4% per hour (Andersen et al., 1988;

Lassen, Andersen, Friberg, & Panlson, 1988).

In order to aid in orientation of and localization within the scans, plastic molding material

(JKR Laboratories, Inc. XL-100 Impression System, Wichita, KS) was used to make custom-fitted

casings for copper sulfate solution markers for MR and cobalt-57 markers for SPECT. These casings

were placed within each ear and on the glabella. As the SPECT apparatus is capable of rotating


Table 3-2
Primary Study Sample Characterics

N = 12 subjects

Age 21.00 2.30 18 -27

Education 14.58 1.50 12- 18

WAIS-R Vocab 14.33 2.15 11 -18

reconstructions three-dimensionally, the matching of marker configuration to that seen in the MR

enables accurate SPECT-slice-to-MR-slice matching, thus insuring that localization is not confounded

by rotational factors.

The stimuli for the training and test conditions (to be described below) were presented using a

Macintosh SE (Apple Computer, Cupertino, CA) with the Psychlab software package in a small, quiet

room in the department of Nuclear Medicine at the Veteran's Administration Medical Center in

Gainesville, Florida. The stimuli for the activation condition were presented running the same

software on a Macintosh lIx in the same room as the SPECT scanner. This room was chosen for its

constant level of background noise (from the fans that cool the scanner and associated computer).

Subjects' responses during the test condition were recorded on Sony metal audiocassette tapes using a

Sony TC-D5M audiocassette recorder and a Sony ECM101-SM stereo microphone.

Image Anaiys

MR slices were 7 mm thick with a 7 mm center to center distance. Pixel size on the MR is

0.9014 nun2. SPECT slices were reconstructed so as to be 7.12 nun thick, with a center to center

distance of 7.12 mm. Pixel size in the acquired scans on the SPECT is 3.56 x 3.56 mn. Acquired

SPECT images were converted from 64 x 64 pixel images to 128 x 128 pixel images, thereby reducing

interpolated pixel size to 1.78 x 1.78 mm. Analyses are based on MR and SPECT slices acquired from

the bottom marker level to the top of the brain. SPECT and MR images were rotated such that all three

markers were in the same plane of view. This plane of view corresponds approximately to the A4

angle described by Damasio & Damasio (1989).

Due to the nonlinear relationship between rCBF and HM-PAO distribution obtained in an

uncorrected SPECT scan utilizing [99mTc]-d,-lHM-PAO as a tracer, images were corrected using the

method described by Lassen et a. (1988), using whole-brain as the reference region and ot = 1.2.

Whole brain counts were determined by adding the counts in four equally spaced slices beginning just

above gyrus rectus and continuing to just above the lateral ventricles. This sum was then divided by

the total number of brain pixels in those four slices to yield an index of the average counts per pixel.

Brain pixels were defined in the following manner-. first, an outline was drawn around the identified

MR slice of interest by the experimenter, next, the matching SPECT slice (previously corrected for

rotational factors using the markers) was superimposed upon this outline; and finally, those pixels

falling within the outline were considered to be brain pixels. Pilot data have shown this index to be an

extremely consistent and accurate estimation of the whole-brain average obtained by using all brain


Slices in which the regions of interest (ROIs) were located were identified by referencing the

patient's MR to a neuroanatomical atlas (Damasio & Damasio, 1989). ROI locations are noted in

Figure 3-3. Once the region was identified on the MR, the number of slices between this slice and the

slice most clearly showing the markers was determined. This number was then used to locate the

analogous SPECT image.

In order to accurately localize the regions of interest (ROI's) within the SPECT image, two

methods were used. For cortical ROIs, an algorithm was used in which two of the three coordinates of

the ROI were referenced to the subject's MR scan, and the third was located in the center of cortex as

indicated by activity level. This algorithm is designed so as to maximize the strengths of the respective

types of scans: MR for accurate localization, and SPECT for assessment of activity level in a given

cortical area. Upon location of the ROI in the atlas, the anterior and posterior points of the region of

interest were measured and the proportional distances of the anterior and posterior points of the ROI to

the distance between the frontal and occipital poles was computed. In translating the MR coordinates

to the SPECT scan, the formulae found in figure 3-4 were used to determine the proportional distances

of the anterior and posterior points of the ROT to the distance between the frontal and occipital poles.

These coordinates were located on the patient's MR scan so that minor adjustments could be

made if necessary. After the anterior and posterior SPECT coordinates were determined, two 8.9mm x

8.9mm nonoverlapping ROT's were placed in the y-axis center of the ROT. The x-coordinate for the

center of the ROI was determined by inspecting the x-values along the predefined y-coordinate and

choosing the center of the five x-values within the confines of the ROI with the highest total value. It

is felt that this represents the center of the gray matter of interest. This algorithm simultaneously


ILF Inferior Lateral Frontal
FO = Frontal Operculum
TO = Temporo-Occipital Association Cortex
DLF Dorsolateral Prefrontal
WA Wernicke's Area
SMG = Supramarginal Gyros
AG = Angular Gyms
MS = Motor Strip
SS Sensory Strip

Figure 3-3. Location of Regions of Interest.

Da= (Ay Fy) / (Oy Fy) Dp = (Py Fy) / (Oy Fy)

where Da and Dp = distance of the anterior and posterior points,

respectively, of the ROI from the frontal

pole, expressed as a proportion of the frontal pole-

occipital pole distance,

Ay = y-coordinate of the anterior point of the


Py = y-coordinate of the posterior point of the


Fy = y-coordinate of the frontal pole,

Oy = y-coordinate of the occipital pole.

Figure 3-4. Localization Formulae.

maximizes structural imaging's the advantages in anatomical localization and functional

neuroimaging's advantages in locating the areas of greatest change in rCBF. This method has a

number of advantages: (1) it minimizes variability inherent in outlining ROI's in the cortical regions,

and (2) it enhances flexibility in matching to nonsquare structures, as RO's may be placed in any

configuration as long as they do not overlap or exceed the boundaries of the region as defined by

MR,(3) it is systematic and reproducible, (4) it eliminates the variability inherent in combining data

across subjects, and (5) it addresses the concern of intrasubject image misalignment (with subsequent

errors in estimation of change in rCBF) discussed by Phillips et al. (1990).

Since subortical structures are not defined in the neuroanatonical atlas chosen for this study,

subcortical ROI's were identified using a procedure analogous to the one used to arrive a whole-brain

counts. First, the experimenter outlined the ROI on the subject's MR. The matching SPECT slice was

then superimposed on the MR slice, and counts were taken from the outlined area.


After responding by phone to an advertisement placed in the school newspaper at the

University of Florida, subjects were contacted by phone and administered a brief screening

questionnaire(as used in the Pilot #3) to insure that the subjects were right-handed and rule out

neurological dysfunction, learning disability, and substance abuse. Those subjects who passed

screening were then scheduled to come in for an informed consent meeting in which the study was

explained and questions were answered by the principal investigator. Informed consent was obtained

from those subjects still interested in participation.

Due to the half-life of the tracer used, SPECT scans were scheduled to be no closer than four

days apart in order to assure total washout of activity from one scan to the next. The majority of scans

were one week apart, with one subject having two scans four days apart and three subjects having

scans more than one week apart. The order of scans was counterbalanced across subjects, and each

subject completed all three tasks. One subject was excluded from the data analysis due to an old lesion

in the right head of the caudate nucleus discovered on the MR scan which was completed for the

purposes of this study.

For each experimental session, subjects were trained to criterion before engaging in the

stimulation task. This was done in an attempt to familiarize the subjects with the demands of each task

as well as to minimize the potential effects of task novelty (Damasio, 1985). Criteria consisted of four

consecutive sets of stimuli performed without error. Adequate performance is defined in the different

tasks as (1) successfully generating a rhyming nonword for each stimulus, (2) successfully generating a

semantic associate for each stimulus, and (3) successfully generating a sentence of the specified

structure with the given stimuli. Although exact times were not obtained, it is estimated that training

for the phonology task took approximately 5 minutes, training for the semantic task approximately 3

minutes, and training for the syntactic task approximately 15 minutes on average. During the

stimulation task, subjects were presented with 64 stimulus sets (each on-screen for 6.966 seconds),

with 0.200 seconds of blank screen between each set. Following the stimulation task, subjects were

presented with a post-test of 20 sets of stimuli (10 novel, 10 presented during the stimulation task)

during which reaction time was recorded for each set of stimuli. Finally, subjects were asked to rate

the difficulty of the task performed that day on a 100 mm visual analog scale. SPECT scanning was

performed after rating the difficulty of the task. On one of the subject's three visits, MR scanning was

performed before SPECT scanning. Instructions for each task may be found in Appendix D, while

stimuli may be found in Appendix E..

Placement of the IV line was done prior to initiation of training by a technician in Nuclear

Medicine. The apparatus was arranged such that the heparin lock into which the radionuclide would be

injected was behind the subject, and tubing was arranged so as to minimize movement of the IV upon

injection. This was done in an effort to minimize distraction of the patient from the experimental task

at the time of injection, as subject reports during previous pilot research suggested that the movement

of the heparin lock is more distracting than is the injection of the isotope. Informal questioning of the

subjects after completion of the study indicated that this was largely successful, with most of the

subjects reporting that they did not notice the injection occurring. Those that did notice stated that the

injection did not present a major distraction to their performance of the task. The injection was

performed 90 seconds after the stimulation task was begun. The preinjection stimuli presentations are

given to insure patient understanding of the task and ongoing cognitive activity at the onset of tracer

uptake by the brain. Stimuli were continued for approximately six minutes after injection. Recent

findings suggest that a five-minute envelope should be sufficient time for the tracer pattern within the

brain to be established (Woods et al., 1990).

As mentioned earlier, subjects performed the activation task silently in an attempt to minimize

the contribution of regions involved in the motor execution of language. In order to promote subject

engagement in the task for the entire duration of the stimulation period, a number of measures were

taken. First, the shortest average median response time in pilot #3 (Syntax-Engaged: 6.966 seconds)

was chosen as the time interval for which the stimuli would remain on the screen. The median time

was chosen in order to ensure 1) that there would be very little (if any) time between stimulus sets in

which a processing demand was not placed upon the subject, and 2) that the same interval could be

used across tasks, so that subjects would be presented with an equal number of stimuli in all


The other measure which was taken in order to promote subject compliance was a deception

aimed at increasing subject motivation. Subjects were told that each session would be broken down

into three components: 1) introduction to the task, 2) silent practice of the task, and 3) test of speed of

task performance. Subjects were instructed that, although the investigators were interested what was

going on in the brain during silent practice of the task, the primary thrust of the experiment was to

determine the extent to which silent practice could improve speed of response in the three language

tasks. Subjects were told that previous research had shown that silent practice improves the speed of

response in the tasks, but only if the practice was done continuously for approximately eight minutes.

Subjects were told that if their average response times to the sets that they would see during the silent

practice were faster than their response times to the novel sets that they would also see in the post-test

(as should be expected with continuous silent practice of the task), then they would receive an extra ten


dollars at the end of the experimental session that day. Reaction times were examined while subjects

were being scanned, but subjects were awarded the $10.00 regardless of their performance.


Following each stimulation condition, subjects were asked if they had any difficulties

concentrating on the task at hand. Following the final stimulation condition, each subject was given an

opportunity to ask any questions that he might have had concerning the procedures or experimental




Data Analytic Strategv

All statistical analyses were performed with SYSTAT 5.2 for the Macintosh (Wilkinson,

1992). Following Friston et al. (1990), functional neuroimaging data were analyzed via analysis of

covariance (ANCOVA), with whole-brain counts as the covariate. As detailed earlier, this is done in

an effort to control for systematic factors affecting whole-brain counts which are unrelated to

differences engendered by the cognitive activation. There are a number of differences between this

analysis and those of Friston et al., however. At a very basic level, the data are acquired differently, in

that this study uses an a priori region-of-interest approach, as opposed to Friston et al.'s pixel-by-pixel

analysis. Thus, the data in this study are referenced to individual MR scans on which ROI's are

located, as opposed to a standardized, per-pixel analytical approach in which anatomical variability is

presumed to be controlled smoothing and image standardization.

From a statistical perspective, this analysis was performed in a repeated-measures framework,

since within-subject changes in ROI activities were the primary variables of interest. This analysis was

done in the context of a multiple regression analysis as detailed by Cohen and Cohen (1983). The

primary advantage of multiple regression format is that it allows for the determination of effect sizes of

both covariates and independent variables. Given the great variability between methodologies in the

imaging literature and the low statistical power inherent in all of them, it was felt that it would be

helpful to obtain some measure of the magnitude of effect that these sorts of activation paradigms have

on regional counts. The effect size of an independent variable, or that proportion of the variance in the

dependent variable accounted for by the independent variable while controlling for the contributions of

other variables, tends to be a much more stable index of effect strength than does statistical

significance, primarily due to the heavy reliance of traditional F-ratios on the number of subjects