Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 5
1
Behavior
Mind
and Brain
OCR for page 6
OCR for page 7
OCR for page 8
OCR for page 9
OCR for page 10
OCR for page 11
OCR for page 12
OCR for page 13
OCR for page 14
OCR for page 15
OCR for page 36
OCR for page 37
OCR for page 38
OCR for page 39
OCR for page 40
OCR for page 41
OCR for page 42
OCR for page 43
OCR for page 44
OCR for page 45
OCR for page 46
Representative terms from entire chapter:
language acquisition
~ ~~s~:~:~s~sl~Si~ ~ S~s~ss~Sss~ss~s
~ :~:~S~E~S~S
S~) ~~T ^
>:~s~sis~s~s~ ~s~s~s~s~s~s~s~s~sssl~:~:~:~:
:s:s:~:-:~:s:s:s::: :s:::s:s:s:sis:s:s:s:s:s~
I
....~$ Cost
7~.~ ~ ~~) ~~ ~~>
,,..~s~:~:~:~:~:~:~:~s~s~s~s~s~sesss~sisas~s~s~s~s .
~ -~ ~~ ~~S IS
,,,s~s~s-s s~s~:~s~:~:~:~:~:~:~s~:~s~s~s~s~sSs~s~s~:~:s his
....S rS~S~:~SS,
,~: ~~S~S~:~'i'
,.~ (S~:lS~SISISIS~:~S~S:S~S~S~S~S~S~SiSIS~SISIS~S~S~S al''';
r<~·~) )~;~$~<
s s s s s s Tsars s~s~sisi~sis~s~sisisis~s>ls~s>~ssS~ssss
.s.s.s.s.~.s.s.s.~..~.s.s S.S.S.... s s s A.
.... ~~-~si
,3.~) ~~:~S~
$~si$ Is
~~:~S
{~ ~~'
S~S~S~ ~S~S~(S~#Sl~lSl~:~Si::SISSS~S~S~SISISiS~S
As
~S~?
~S~SIS:SiS~:~:~:~:lS~S~:~S~S~S~S~SSS~:S#:lS~S
|~S~<'~
iS
S~:~S~S~:~:~:lS~S~S~S~SISIS~SSS%SSS~SS
~~:~ S ~~
S~SIS~S~S'S.~:~:~:~#Sl:~:~S!S::
{:iSi:~:~:~:~SiS~S~S~S~S~:S~S)
V~t
~~ "'
ASH ~S~:~:~SISISIS~SISISSSiSS
~~lS~
SS~S~S~SISISISISSSIS~SSS~S"''
~~i~) _
~S~SIS:S~SSS~SSS:SSSSSS
~~.
S~S~S~S~:~:lS~SISISi
\~>~#
So
~SiS~S~S~S~S~iS~:
IS:
~:~
~:~:~S
BEE ~S~:~S~S
~S~S~S~S~SS~
As
Swiss:
S~SiS~SiS~SS~S
S~S~:lSl:SS:~l
Is:
.~:~
S~S~S.~S~S~S~S
.~#S~SSSIS~:S
I':
~~.:~:~
.S..S.S.S.S.
~ ~~;~ ~!~!!!!ll!lI)~ll~i~Ii~!il~l~f~l~l~
~ S~ ~ ~~ ~~ SO:,. ~ S~S Sag ?~ ~~ ~ S~ ~ {~'~S~:~iS~>S~
~ I ~
>~S~ ~~<~<~<~ ,, ,Z,,, ,Z, :,Z, ~ ~ a, ,,....
~~ ~ ~ ~ ~ ~ ~ ~ ~ ~ 7 ~ ~ a: ~ ~~ ~~ ~~ ~~ ~ ~ ~ ~ ~~ ~~
1
B-
e Pavlov
, Mind, and Brain
From the beginnings of scientific inquiry, researchers have tried to understand
the workings of the mind and its relationship to behavior. In modern terms,
scientists seek to answer such questions as:
How does an individual manage to see coherent objects in changing patches
of multicolored light or to hear speech and music in bursts of sound varying
in loudness and pitch?
How do people remember, even imperfectly, the vast storehouse of factual
and functional information that each of us carries about?
How do infants and other nonverbal creatures think, and what are their
thoughts like?
How does an individual learn new ideas, create concepts, organize his or
her knowledge, and act upon what he or she knows?
Until about 100 years ago, these and similar questions led mainly to spec-
ulation, to sophisticated but untestable guesses about the nature of the mind
and its function in behavior. Many scientists thought that this confinement to
speculation would always be the case, on the presumption that mental life,
unlike the world of objects, is not directly observable and so is not amenable
to systematic, scientific study. But during the second half of the nineteenth
century, experimental and close observational approaches to such questions
began to appear. Some of the first work involved the study of sensory processes
7
8 / The Behavioral and Social Sciences
that underlie organic sensitivity to different physical stimuli, rigorous obser-
vations about how behavior is shaped by experience, and the relation of ob-
servable behaviors to certain brain regions and neural events.
Both theory and method have since advanced at an accelerating pace, with
extensive revisions of earlier questions to make them more tractable and to
take advantage of the growing linkages between knowledge about behavior,
the mind, and the brain. Instead of asking only how best to measure the
intelligence of people, researchers now ask about the nature of intelligent ac-
tion, whether exhibited by humans, animals, computers, or robots. In addition
to seeking laws that directly relate external stimuli to behavioral responses,
researchers now attempt to unravel the basic processing that the brain must
carry out in order to generate behavior and to simulate that processing on
computers to formalize and test theories. These are the new questions—with
their extensions into the details of visual and auditory processing, memory
formation and retrieval, language, cognition, and action that drive the re-
search investigations highlighted in this chapter.
SEEING AND HEARING
As you look around, you see a variety of objects at various distances, some
still and others moving, some transparent and some densely colored, some
partially obscured and some not. All is perfectly ordinary and seen without
effort if the light is adequate. Your cat or dog sees these things, too, although
somewhat differently from the way you do. And so does the fly that evades
your swat. And when you listen to a musical recording, you have little trouble
hearing the separate instruments. If someone speaks while you are listening to
the music, you have no difficulty in understanding the words, unless the music
is very loud or you suffer from a certain form of hearing loss that is common
in older males.
It is all so commonplace but no one yet knows exactly how the brain does
any of these things. No one yet knows enough to program a computer to pick
out a wide range of objects from a scene, to isolate a violin in an orchestral
passage, or to separate speech from noise and to partition it into words. Ena-
bling machines to do these things would surely affect the way people live as
much as the telephone, the thermostat, or the radar have done. And it is clear
these tasks can be done because the human brain does them, continuously and
apparently effortlessly. One approach to studying how these tasks are carried
out is to work with computers and sensing devices, without much regard for
how the tasks are done in the brain. Another approach is to focus directly on
unraveling nature's way of doing them.
These tasks may be very complex. Or they may be simple but involve prin-
ciples that scientists do not yet understand. For example, between one-quarter
Behavior, Mind, and Brain / 9
and two-fifths of the total cortex of the human brain is devoted to vision,
suggesting that it is one of the most complex brain functions. However, abilities
somewhat similar to human vision are exhibited by animal brains far more
modest. For example, pigeons trained to identify the occurrence of trees (in
contrast to bushes or other plants) in color slides- whether alone or in a forest,
in whole or in part, leafy or bare are then able to identify trees as accurately
as people can, in slides they have not previously seen. Pigeons can also be
trained to recognize a particular person in slides of individuals or crowds. Yet,
with similar training, pigeons cannot identify simple line drawings, such as
cartoon characters. The difficult is easy and the easy difficult for that tiny brain,
or so it seems.
Hearing similarly involves complexity and simplicity. The only physical event
on which hearing is based is the variation of air pressure at the ears. A plot of
the sound pressure generated by a person saying "science" in a quiet room is
completely different from the plot of that same word said by the same person
against a background of white noise, such as cocktail party chatter. A person
has no difficulty hearing the message in the noise, although the physical sound
patterns are quite different. The ability of the human ear and brain to recognize
the word in the noise is vastly superior to that of any machine.
Understanding such abilities is the current focus of research on perception.
This research includes innovative theoretical work and sophisticated experi-
ments with animals, and it is increasingly able to use computers to develop
theoretical models and simulate experimental tests. The research involves psy-
chologists, biologists, physiologists, physicians, graphic artists, physicists, en-
gineers, and computer scientists. The work is highly interactive: for example,
behavioral analyses of sensory and perceptual processes provide evidence for
functional distinctions that can then be sought in the workings of the brain.
Another example comes from theoretical advances in the development of re-
cursive computer models that capture the simultaneous use of multiple sources
of information. These are often called activation models because each piece of
information that plays a part in a particular mental process is viewed as an
active influence on the direction of the process. These activation models are
being applied to previously intractable problems in research on human and
. . .
an~ma . vision.
Work with various species of animals commonly used in laboratory exper-
iments is playing an important role in learning how brains accomplish the tasks
of seeing and hearing. By recording the electrical behavior of single nerve cells
and studying the wiring diagram of how these cells are arranged in the brain,
researchers are beginning to understand the architecture of perceptual systems
and how they process information. In addition to monkeys, animals such as
cats, rabbits, goldfish, and barn owls have yielded valuable information about
sensory systems. Which species is appropriate depends on the particular ques-
10 / The Behavioral and Social Sciences
Behavior, Mind, and Brain / 11
tion being asked and the similarity of that aspect of its perceptual system to
the human system.
Visual Analyzers
The retina of the eye really a bit of exposed brain is a regular mosaic of
five different layers of nerve cells whose spatial arrangement affects the nature
of vision. Aided by recent technical advances in electronics, researchers have
shown that in one early stage of visual processing, millions of small groups of
neurons in the eye and the rest of the brain work as parallel microprocessors
or analyzers. Each analyzer responds preferentially to stimulus components
within a narrow range of size and orientation; they decompose the scene into
a vast number of overlapping components. Precise quantitative models based
on such analyzers can be explored by complex computer simulations and can
now explain how the human eye and brain detect and identify low-contrast
visual patterns, such as nighttime shadows. Some current work suggests that
the brain may be able to rapidly select and regroup the analyzers, depending
on what is being perceived.
One application of such models is to evaluate, at the design stage, the relative
clarity and efficiency of a variety of visual displays: flight simulators, video
HEARING How do the senses, such as hearing, work? How does a per-
son's auditory system identify a single voice in a buzz of conversation and
pick words out of a vocal stream?
The figures on the facing page show the sound pressure generated by a
human male voice saying the word "science." They are digitized plots of the
amplitude of a sound wave over time; the total time is 0.85 seconds; the
amplitude is sampled at a rate of 10,000 observations per second.
In the top figure, the person spoke in the stillness of a soundproof cham-
ber. In the bottom figure, he spoke while his voice was masked by noise that
simulated a roomful of people talking, such as would occur at a cocktail
party.
Neither the human eye nor any known form of sound filter or computer
analysis can detect the wave form of the top figure in the bottom figure. That
is, they cannot extract the vocal wave form from the noise. But a normal
listener has no trouble hearing and understanding the word through the
noise. How the human ear and brain perform the analysis needed to rec-
ognize the word in the noise is a fascinating and complicated scientific
problem.
12 / The Behavioral and Social Sciences
terminals, warning lights, and information signs. Another application is image
augmentation. The inherent inadequacies of electrical and optical imaging sys-
tems such as blurring due to the atmosphere can be compensated for by
computers. Information that is particularly useful to human perception can be
improved at the expense of less useful information. These applications are
aimed at speeding up such tasks as reading x-ray plates and satellite recon-
naissance displays, tracing circuit diagrams, and following blueprints and maps,
which are done slowly and are error-prone without augmentation. Another
application in the future will be the possibility of more efficient transmission
of visual images over telephone lines.
Researchers have discovered that some visual tasks are so complex that
humans, some primates, and possibly many other animal species use special-
purpose brain areas to solve them efficiently. For example, the human ability
to recognize and discriminate readily among a seemingly endless variety of
faces is now known to make use of certain neural circuits in the right-posterior
cerebral hemisphere. Facial recognition may be different from the recognition
of most other kinds of objects in the world, but it may be that other forms of
expertise in pattern recognition hinge at least in part on engaging those neural
circuits. Researchers are beginning to explore in detail the extent and nature
of such special-purpose systems in the human and animal brain and their
relation to perception.
Color Constancy
An outstanding example of a very important special feature of visual per-
ception is color constancy: color relations between objects in a scene seem
relatively invariant to the eye, independent of the light playing on the scene.
A blue shirt looks blue whether seen in sunlight or in a dimly lit restaurant.
Yet this invariance cannot be replicated by color photographic media.
Recently, some investigators have proposed a theory about how the eye and
brain may separate the inherent color relations among objects from the light
impinging on them. Certain newly defined mathematical algorithms, if carried
out simultaneously over the entire visual field, presumably by millions of visual
analyzers, have been shown in theory to be able to separate the inherent colors
of objects from incident lighting. This theoretical model assumes certain math-
ematical conditions that correspond to qualities of colors, of light, and of the
high-level "programming" of the brain. Large-scale computations are now being
carried out to test and refine the model and to find ways to simulate it in real
time with new computer architectures.
If, as the theory suggests, this decomposition works efficiently and if, as is
anticipated, the electro-optical technology for recording a visual scene digitally
can be made sufficiently compact, a new form of photography could result.
Instead of the various "speeds" of film suited to different lighting conditions,
Behavior, Mind, and Brain / 13
one would need only one type of photosensitive medium that could be used
under any circumstances. The "developing" process would then decompose
what is inherent in the scene from the lighting that happened to exist at the
time of the photograph. With such a process, one could display or print the
picture with whatever lighting conditions one wanted. The developing and
printing processes, carried out on a computer, would be analogous to what the
brain does all the time. The ability to adjust lighting as if from the mind's eye
would mean unprecedented expansion of the capabilities for the scientific study
of vision and for applications in graphic arts, industrial design, and computer
vision.
Depth and Motion
That the world is spatially three-dimensional while the retina is spatially
two-dimensional creates inherent visual ambiguities. The eye cannot directly
capture motion perpendicular to it; the eye and the rest of the brain must infer
depth and motion from ambiguous clues. Resolving these ambiguities, which
is usually essential for accurate vision, is possible because of certain constraints
that occur in natural scenes concerning shading, motion, texture, and so on.
Recent results of behavioral experiments with humans and animals, which have
inspired corroborating neurophysiological experiments, reveal that higher-level
neural analyzers of motion in visual images are sensitive to the global direction
of a pattern's motion, even when parts of the pattern are moving in other
directions. These analyzers give rise to moving optical illusions, such as rigid
objects that are perceived as flexible and vice versa. Another example is the
illusion that very large objects are moving more slowly than they actually are.
Landing or taking off, a jumbo jetliner appears to be flying more slowly, by a
sizable factor, than a small, executive jet, when in fact their speeds are very
similar. This kind of size-speed illusion is believed to be the underlying cause
of many railroad-crossing accidents when motorists drastically underestimate
the speed of an approaching train. These illusions, which are difficult and in
some cases virtually impossible to study experimentally except with modern
computer-graphic technology, imply that immense distortion in visual displays
may be tolerated or even go unnoticed by both humans and animals.
Perceiving Objects
The process by which visual analysis fills in a big picture from a collection
of details means that mere fragments of objects suffice in normally cluttered
scenes to permit observers to produce coherent, conceptually appropriate per-
ceptions. Adults make use of several experimentally confirmed theoretical prin-
ciples to do this, such as assuming that objects in a scene do not share bound-
aries. People also tend to see correlated movement of disconnected parts in
natural scenes as a single moving object partially masked by another object.
This latter principle is apparently a deep-seated tendency and has even been
14 / The Behavioral and Social Sciences
demonstrated in very young infants. Increasing knowledge about visual per-
ception can be expected to improve the design of factory robots that must make
complex identifications of objects in order to carry out their function without
selecting the wrong object or inadvertently hurting someone.
Temporal Auditory Patterns
Hearing problems having to do with space (for example, distance and lo-
calization) are very important in the design of sound equipment and audito-
riums and have long been studied. Much of the current focus of research on
auditory perception, however, concerns complex patterns of sound stimuli
varying in time, such as speech and music. Auditory signals can now be de-
signed and stimulated quite precisely by a computer driving a digital-to-analog
signal coverter, and these artificial sound patterns are used to study specific
aspects of the hearing process.
One type of study measures a listener's ability to discriminate changes in the
intensity of one of several tones that are played simultaneously. Initially, the
ability to sense a change in intensity within a complex of other tones is far less
than when a single tone is played by itself. But with practice in listening to
complex tones repeated with little variation, the ability to recognize the vari-
ations when they do occur becomes very good. Studies are now under way to
see whether practice in listening to complex sound patterns is characteristic of
how an infant reams to identify the particular phonemic differences that char-
acterize its prospective language.
An important area of application of research on the perception of auditory
patterns (including speech recognition, described more fully below) is in the
design of hearing aids. They have been greatly improved in the past decade by
coupling what was known about the nature of hearing loss with modern elec-
tronics. Whenever the auditory deterioration is peripheral rather than central,
certain aspects of the loss, such as reduced loudness at certain frequencies, can
be mitigated by appropriately altering the sound waves at the eardrum. Whether
it will also prove possible to compensate for the inability to separate speech
signals from background noise may depend similarly on the nature of the loss,
which further research is likely to reveal.
Music is the focus of much scientific research. Musical keys and scales are
highly structured, and musicians, mathematicians, and philosophers have long
speculated about the underlying nature of that structure. More recently, the
rules of counterpoint and orchestration are being explained, not as arbitrary
requirements of particular musical styles, but as selective adaptations to basic
tendencies of the human auditory system to group sounds in certain ways as
a step in auditory pattern recognition. Researchers some time ago postulated
that a set of ratios corresponding to a helical geometric structure underlies tonal
perception. Recently, far more systematic studies, using trained musicians as
Behavior, Mind, and Brain / IS
respondents and multidimensional scaling methods (discussed in Chapter 5)
to analyze the results, have led to a modified understanding of the geometric
relations implicit in tonal perception; they appear to be conical rather than
helical configurations. Stimulated by the success of linguists in analyzing lan-
guage as a rule-governed system, behavioral researchers working with musi-
cians have also successfully begun to define the deeply embedded rules that
become internalized in the course of growing up in a musical culture, and they
have demonstrated how such deep rule structures strongly affect listeners'
perception and memory of what is heard.
MEMORY
Most people can readily kind their way around a once frequented neighbor-
hood, improve a sporting skill with practice, or recall an old story. These deeds
are possible because the capacity for memory storage is vast; indeed, virtually
everything discussed in this chapter depends critically on humans having very
large, readily accessible banks of organized information in the brain. Not only
is the capacity large, but also much of its use is automatic: learning and memory
often occur incidentally, with little special effort. Yet for some people memory
failure can become an acute problem. In certain diseases, such as Alzheimer's,
the ability to learn and remember is drastically impaired, and life becomes a
series of unconnected moments.
Some questions about memory are of long standing: How are memories
organized? How does the brain code, store, and retrieve them? What are the
computations and processes involved? What happens when the capacity for
memory becomes impaired? How can memories be improved? For both these
questions and newer ones, a whole new approach to the study of memory
derives from the development of computers. However, although much has
been learned from the way information is stored in computational systems, the
analogy between computer memory and human memory is imperfect in many
.~
secant ways.
In the past 20 years, an interdisciplinary revolution has occurred in under-
standing the organization of learning and memory and their biological foun-
dations. Behavioral studies in animals and humans are characterizing the cat-
egories and properties of learning and memory; research on human memory
and the brain is identifying the neuronal systems that serve different categories
of memory; memory trace circuits in the mammalian brain are being defined
and localized in animal models; researchers are beginning to understand the
neuronal, neurochemical, molecular, and biophysical substrates of memory in
both invertebrates and vertebrates; and theoretical mathematical analysis of
basic associative learning and of neuronal networks is proceeding rapidly.
Better mathematical and computational modeling of elaborate memory pro-
36 / The Behavioral and Social Sciences
theory. In particular, these data will permit the testing of various proposed
theories of the learning process, including computer simulations of language
acquisition, models evaluating the chances of learning a grammar from ob-
serving a modest sample of the language, efficient forms of artificial intelligence,
and explicit behavioral models of language performance and proficiency. Gram-
matical and word acquisition studies in children should help to resolve con-
troversies regarding the species-specificity of language, language-specific innate
constraints, and the relative contribution of the child and the environment to
the reaming process. The improvement of communication between people and
computers by developing programming modes closer to "natural language" will
also benefit from this work.
A major research task in the next decade is to explore more completely the
hypothesis of grammatical universality, through intensive investigation of lan-
guages that are just now being fully described and are historically unrelated to
the commonly studied ones. This work will involve considerable coordination
and a stronger international basis between theoretical linguists and those de-
scriptive linguists whose research is directed toward these frontier languages.
It is also vitally important to this work to add to the relatively sparse instances
of longitudinal studies that cover the same people over a period of many years
and across a variety of linguistic learning environments.
Machines That Talk and Listen
Shortly after World War II, attempts were initiated to devise computer schemes
to translate from one language to another, to recognize spoken and handwritten
language, and to convert text into natural-sounding speech. Abortive and highly
expensive early approaches to automatic machine translation and speech rec-
ognition were made in electrical engineering projects, but they did not address
the complexities of human language; participants in these efforts came ruefully
to appraise their results as "language in garbage out." After engaging linguistic
and phonological expertise to diversify the lines of research undertaken, team
efforts ultimately did make good progress, leading to the scientific and practical
successes seen to date.
At present, interdisciplinary research aims to explain the intricate relations
that hold between language, the world, and intelligent systems (whether natural
or artificial). In one line of research, formal linguistic models are being devel-
oped that make explicit provision for the varying computational requirements
of language understanding. Relating linguists' grammars to existing computer
systems enables computer scientists to provide a new generation of interpreter
programs that run much more efficiently than their predecessors. A new theory
of the semantics of programming languages promises to unite two formerly
separate analyses: the denotational meanings of terms in natural language and
the functional meanings of instructions in a computer program. Other devel-
Behavior, Mind, and Brain / 37
opments in this work include knowledge about the recursive (looping-back)
nature of computational processes, symbolic systems, and the importance of
shared knowledge and beliefs in conversational face-to-face interactions.
In recent decades, collaboration between linguists, communication engi-
neers, and computer scientists has led to dramatic increases in knowledge and
new methods for analyzing and synthesizing acoustic speech signals. The com-
puter revolution has made it possible to acquire and analyze in hours or days,
rather than months or years, the large phonetic data bases needed to study the
sound structures of language. As a result of the investigation of intonational
and other phonetic properties of speech, undertaken with the goal of con-
structing machines that talk and listen, there is a better understanding of how
the units of a linguistic system relate to acoustic signals, leading to more natural-
sounding, machine "voices" and more capable machine "ears." However, the
present ability to synthesize speech far exceeds the ability to automate speech
recognition and understanding, because the normal acoustic flow of speech
cannot be readily divided into neat, uniform segments corresponding to dis-
crete sounds, syllables, words, or even phrases, as is readily apparent by lis-
tening to unfamiliar languages. At any instant of speech production, several
articulators (larynx, tongue, velum, lips, jaw) are executing a complex, inter-
woven, rhythmical pattern of movements, blurring the boundaries between
words as well as between the phonemic segments that compose words. The
problem is further complicated because each perceived unit sound, syllable,
word, or phrase varies widely with phonetic context, stress or emphasis, and
the rate of speech, style, dialect, and gender of individual speakers. (The ad-
ditional problem of discriminating a speech signal from background noise,
including other speech, was noted above.)
What is invariant that enables people to recognize and understand speech?
The discoveries of the last several decades have contributed to speech-recog-
nition schemes, but they are still far from satisfactory; since they can deal at
most with 5,000 words, compared with the 14,000 known by a 6-year-old or
the 100,000 to 150,000 used by monolingual adults. The current solutions are
still not very subtle. It is expected that dramatic improvements will accompany
increased understanding of how the brain carries out the segmentation and
analysis provided by the linguistic system.
Reading
Reading incorporates a complex hierarchy of skills, each raising a long list
of questions in its own right. What are the fundamental perceptual processes
involved in registration of visual print? How does word recognition occur? In
the absence of aural and gestural cues, how do readers accomplish phrase
interpretation? How are processes of inferential reasoning, critical thinking,
and behavioral response catalyzed by the text? How predictable are these re-
sponses?
38 / The Behavioral and Social Sciences
There have been fairly dramatic advances recently in understanding each of
these interactive processes linking reader and text, especially as a result of
developments in the cognitive sciences. These advances have yielded new in-
sights into what the eyes and brain do that enable a reader to comprehend
written language. These insights have progressed to the point that computer
simulations can pinpoint exactly where and why a given reader (with known
skills) encounters difficulty in a text and disclose what might be done to avoid
or ameliorate those difficulties.
Knowledge from this research has led to educational strategies that result in
improved reading skills in children. For example, in some projects, children
who are poor readers have been taught particular cognitive strategies discovered
in the skilled adult reader, such as how to identify and retain the most central
information in a text. Children given this training show large and general
improvements in comprehension. Other studies are implementing new pro-
grams to teach word recognition, taking advantage in some instances of com-
puter-based technology for practice, drill, and individual tailoring of the cur-
riculum. The research has practical importance for scientific and technical
documentation, display technologies, and instructional methods, and it also
provides a major opportunity to expand the frontiers of knowledge about
interactive multilevel reaming.
Decoding and Dyslexias
The process of decoding letters and words is critical in learning to read and
has received a great deal of attention over the years, constituting the bulk of
the research relating to reading. One central issue in research and pedagogical
practice is whether a reader must use the printed words to retrieve a sound
and then use the sound to retrieve the meaning of the word. Most of the
evidence shows that a mature reader needs an intervening phonological code
only if a word is unfamiliar or if the material is particularly difficult. But
phonological decoding is critical when first learning to read. For example,
awareness of the phonological structure of language is one predictor of the
rapidity of early reading progress; S-year-old prereaders who can segment a
spoken word into its constituent phonemes (who can, for example, follow an
instruction to say "table" without the t) tend to be better at word recognition
at the end of second grade. Such results reinforce the importance of a phonics
approach at the earliest stages of reading instruction, when fluency of word
recognition is the key factor to rapid progress.
By third or fourth grade, children have generally mastered recognition skills
sufficiently so that individual differences in reading skills no longer closely
reflect differential word-decoding abilities. At these stages, knowledge of par-
ticular kinds of text structure, such as narrative, becomes increasingly impor-
tant to comprehension, and these matters are now receiving greatly increased
attention. Even very young children are aware of narrative conventions and
Behavior, Mind, and Brain / 39
use this knowledge in understanding stories that are read or told to them.
Recent analyses of grade-school materials suggest that many selections violate
conventional story structures, making it difficult for a learning reader to estab-
lish the coherence of the story. Research on reading comprehension will lead
to better selection of useful texts for teaching.
Some grammatical structures are known to be used much less in speaking
than in writing. For example, cleft constructions like "It was John who won
the trophy" are rarely spoken in English; rather, vocal stress on the name-
"]ohn won the trophy" is sufficient to convey that the identity of the winner
is the key information in the sentence. These differences between reading and
speaking seem small, but for some dialects they may result in serious interfer-
ence between the spoken language and comprehension of the written language.
Such difficulties have been noted in some speakers of black English and some
deaf children, for whom written English primers are virtually samples of a
foreign dialect. While learning to read a second language without first knowing
how to speak it is not uncommon for adults who are already literate in their
native speech, it is a very unusual and difficult hurdle for children who are
first reaming how to read.
Explaining the wide variation among early and middle readers in the rates
at which basic skills become automated and more advanced ones develop and
finding the causes behind reading disabilities (dyslexias) are major challenges.
Research on reading disability has proven to be especially difficult. There is
still no general consensus about the nature or definition of dyslexia, and indeed
there are probably several distinct kinds of dyslexia, some of which are cor-
ollaries or causes and some of which are effects of reading dysfunctions. The
field has discarded numerous theories that have not stood up to close testing,
such as the hypothesis that the disorder is visual, involving reversals between
letters such as b and d or words such as "saw" and "was"; that the disorder
involves particular difficulty in associating visual and verbal elements; or that
poor readers have difficulty in maintaining information about sequences. Other
theories that have some support, but not full assent, include the idea that there
is a deficiency in the specifically auditory-linguistic background and that the
problem results from contracted vocabulary, low verbal fluency, inappropriate
grammar or syntax, or difficulty or slowness in word retrieval. The most prom-
ising approaches to these tangles of cause and effect appear to be continued
fundamental research on normal acquisition of reading skills, more detailed
examination of disabled individuals by interdisciplinary terms psychologists,
cognitive scientists, medical researchers, and educators and more longitu-
dinal research, starting with prereading children.
Interpretation and Comprehension
A number of techniques for studying human reading skills have in recent
years achieved high levels of sophistication, enabling progress in areas that had
40 / The Behavioral and Social Sciences
Typical Colleg - Level Reader
Flywheels are one of the oldest mechanical devices known to man. Every
~ ~ ~$ ~3 ~ ~
internal-combustion engine contains a small flywheel that converts the jerky
motion of the pistons into the smooth flow of energy that powers the drive shaft.
Trained Speed Reader
Colter understood enough of what they said to realize that some of
them were proposing to set him up as a shooting target. Others were
arguing for a more lingering death by tomahawk. Colter waited.
Dyslexic Reader
In appearance the surface of Mars is more like rocky
~)~ (by)
, - -
volcanic deserts on the earth than it is like the highly cratered surface
of the moon, yet Mars, once visualized as being largely a world of gently
rolling dunes, seems to possess little sand.
Behavior, Mind, and Brain / 41
seemed quiescent for many decades. One such technological advance is eye
fixation research, which provides a detailed characterization of where in a text
and for how long a reader looks, pauses, skips, or looks back. This kind of
research was impossibly cumbersome before the availability of laboratory com-
puters and precision optical and video devices and is providing clear answers
to questions that have been on the table since early in this century. Eye fixation
READING How do a reader's eyes move across the written page? What
does the mind do as the eye moves? How are marks on paper interpreted as
words that convey ideas?
Researchers have developed methods to mechanically track a person's eye
movements while reading a text and to meticulously analyze the patterns of
gaze and subsequent accuracy of comprehension of the material. Such ex-
periments are used to test and develop theoretical models of how people
process what they see on paper. The different gaze patterns in this figure,
represented by the pause durations (in milliseconds) and retracking lines,
are typical of different kinds of readers. (There is no significance to the
different texts shown; they are standard passages used in experiments.)
Though texts differ in complexity, a given reader processes them the same
way and at relatively small variation in overall speed.
Most people sample a written text densely, fixing their gaze (for more
than 50 milliseconds) on about two-thirds of the words and rarely skimming
past more than one word at a time; in most cases, the skimmed words are
articles, prepositions, and conjunctions, "function words." Nearly all readers
interpret each word immediately, inferring its meaning even when important
clues to that meaning come later in the text. The duration of gaze on each
word depends on the word's length, familiarity, and grammatical or contex-
tual clarity. There is also a tendency to pause longer on the word at the end
of a printed line rather than at the end of a sentence- to "wrap up" any
loose ends in interpretation. Normally, readers retrace their gaze only when
a grammatical or semantic ambiguity leads them to rethink how to interpret
a phrase.
Readers trained to "speed read" fixate on fewer words and spend less time
on each fixation. Like normal readers, they retain very little information
from words they do not fixate. The major skill that speed readers learn is to
infer the meaning of a line of text based on a smaller sample of the words
in it than before and not to tarry or retrack in order to resolve local ambi-
guities in meaning. The cost of these rapid inference procedures is less
accurate comprehension, especially for nonrecurrent details.
Dyslexic readers seem to have trouble mainly in decoding written words,
that is, segmenting the letters into syllables and retrieving their sounds
correctly. These readers spend extra time on each word and frequently re-
track in order to reinterpret inaccurately perceived words. And even at their
slower reading rates, dyslexic readers generally develop a much less accurate
understanding of a written text than other readers a problem they do not
necessarily have in interpreting spoken words.
r --r ~ r
42 / The Behavioral and Social Sciences
research has revealed that the instantaneous perceptual window during reading
is only two or three words wide even for highly literate readers, and that readers
fixate on 70 to 85 percent of the content words of an unfamiliar or technical
text, but only 40 percent of the short function words, such as "an" and "of."
Also, readers attempt to understand the meaning of each word as soon as they
see it rather than postpone a choice of interpretation until more data (from the
remaining parts of the sentence) are collected. If the interpretation proves
incorrect, readers then seek the source of the problem and choose an alternative
interpretation that works better. The characteristics of each content word, such
as its familiarity and the moment-to-moment demands of linguistic and task
processing, control the duration of pausing on the word. Pause duration is thus
a gauge of processing load or difficulty at each point in a text.
Since a reader is constructing a mental representation of a text while pro-
cessing only a single phrase or sentence at a time, short-term memory limita-
tions may play an important role in reading comprehension. This is not a new
hypothesis. But the original theoretical claims that short-term memory may be
an important bottleneck for comprehension processes were contradicted by
the fact that standard short-term memory tests (for example, digit span) did
not correlate with reading comprehension skills. The more recent experimental
work does not evaluate short-term memory capacity in static contexts or at
nonreading tasks; rather, it evaluates how much excess capacity different read-
ers have available during the actual reading process: the resulting measures are
highly predictive of comprehension levels.
Detailed temporal analyses of this sort have also made it possible to write a
computer program that reads newly encountered though very carefully edited
and prepared texts about as well and as rapidly as human readers. The program
"pauses" taking longer to process a passage when there is difficulty in com-
prehension and speeds up when comprehension is easy. It can produce a
summary or answer certain kinds of questions about the content of texts about
as well as humans can.
Efficient readers adapt their expenditure of effort and ingenuity to their goal
in reading, such as "light" reading for relaxation, reading to learn, or reading
to react to an action document, such as a request for funds. Mature readers
possess a complex repertoire of reading and study strategies for enhancing
understanding and for detecting and overcoming comprehension difficulties.
Considerable advances have been made in understanding of how these learning
activities develop and how they can be enhanced by instruction. All of these
advances can now be described and measured in terms of scientific theory
rather than just intuitively, as was once the case.
No artificial intelligence reading program comes close to the seemingly ef-
fortless way in which humans seem able to call up just the right information
needed to interpret a text and ignore the mass of irrelevant data. But recent
work has begun to create substantially better insight into how people do this,
Behavior, Mind, and Brain / 43
and one interesting fact is that readers do not simply ignore irrelevant infor-
mation. Consider the sentence: "Thieves broke into the vault of a bank and
stole a million dollars." Do you even consider interpreting"bank" in the sense
of "river bank?" No? Sophisticated methods show otherwise. If a person is
{lashed the word "bank" and then, immediately, "money," the response to
"money" is about 40 milliseconds faster than when an unrelated word (say,
"bark") comes first. This is called a priming effect. In a context like the sentence
above, if"bank" is followed quickly by "river," instead of"money," the priming
effect on recognition is present for this word as well, showing that both possible
meanings had been activated. After a 300 millisecond interval, however, prim-
ing would occur only for the contextually appropriate meaning ("money"~. The
results of such laboratory experiments suggest that a rich knowledge of rela-
tionships among words may facilitate decoding and perhaps other levels of
processing, even though there is no conscious awareness by a reader of how
such knowledge is being processed in the course of reading.
A recent goal of language research is to develop a scientific basis for writing
comprehensible texts. Previously, the classical view was that texts are less
readable when they contain long sentences and use uncommon words. But
research shows that revising a difficult text by shortening its sentences and
simplifying the vocabulary does not make it substantially (or sometimes even
at all) more comprehensible. One idea being studied is that some texts may be
hard to read because the reader must repeatedly search in long-term memory
for specific information needed to interpret phrases or sentences. Other re-
search focuses on how well the syntactic devices and cues in a text focus the
reader's attention on its major themes or most important information. Related
research has shown that successful didactic writing highlights information that
the reader needs in order to act or that sets forth specific examples that the
reader will most likely have encountered or expect to encounter.
Continued work along these lines should lead to improved and simplified
models of how to make written documents more comprehensible, but this is
only a proximate goal. Making a text comprehensible for a given reader will
certainly contribute to understanding, but it will not necessarily lead to learn-
ing. People can comprehend, remember, and summarize a text and still be
unable to use the information acquired; learning requires the integration of
textual information with previous knowledge. There is much work to be done
to discover how people learn, how they use the information acquired from a
. . .
text in new situations.
OPPORTUNITIES AND NEEDS
The results of research on basic processes linking the mind, the brain, and
behavior have grown impressively in the recent past and show even more
44 / The Behavioral and Social Sciences
promise in the immediate future. Some of those results have led or soon will
lead to valuable applications, such as new types of photography; economical
telephone-line transmission of video pictures; better hearing aids; ways to im-
prove normal and impaired memory; enhanced computer abilities to synthe-
size, decode, and translate natural speech; vastly improved expert systems to
aid in such matters as medical diagnosis and car repair; and innovations in
teaching students across all areas and levels of learning. But the major contri-
bution of this research is that of sheer knowledge about human beings and
other intelligent beings a wealth of insights into the nature, possibilities, and
limits of intelligent individual action. In this section we propose increased
expenditures of approximately $61 million annually for this research: for equip-
ment, investigator-initiated grants, new data collection, research centers, pre-
doctoral and postdoctoral fellowships, postdoctoral training institutes, and
multidisciplinary collaborative activities.
Research in these areas depends heavily on instrumentation for simulating,
modeling, and controlling experiments, generating stimuli, recording and ana-
lyzing data, and developing new theoretical models, which require substantial
amounts of computational power. Some of this equipment is currently only
available, if at all, in such special settings as clinics, national laboratories, or
expensive commercial centers. Increasingly, there is a call for greater access to
powerful workstations and supercomputers. There is also a lively expectation
that massively parallel computer architectures will be especially well suited to
many behavioral and cognitive research problems. To some extent the com-
putational need is filled by machines available at major research universities,
but during the recent period of serious funding cutbacks and stringencies
imposed on the behavioral and social sciences and the increased regulatory
demands on research involving animals and humans, the overall level of in-
strumentation in laboratories has fallen seriously short of the research needs.
Many formerly up-to-date university laboratories are no longer adeauatelv
equipped to do the research that is now possible.
In light of the increasingly advanced and specialized technology required to
carry out experiments on behavior, mind, and brain, an estimated $25 million
in new funds annually are needed to acquire or provide access to new equip-
ment. We estimate that about one-half of this amount ($12 million) should be
allocated to new and upgraded laboratory equipment and service facilities,
about one-fourth ($7 million) to new computer hardware and software devel-
opment, about $4 million to the improvement of animal care for research
animals, and about $2 million for access to major neuroimaging devices.
The principal mode of studies on behavior, mind, and brain is carried out
by small groups of investigators: one or two principal investigators working on
a separately funded project (though each investigator may have more than one
project) with one or a few assistants, who may range in level of training, skills,
C, , ,
Behavior, Mind, and Brain / 45
and independence from technicians to postdoctoral scientists. This approach
has worked well, and we do not recommend major changes.
Such research by small groups is largely supported by investigator-initiated
grants. The aggregated funding level of these grants has been somewhat buff
ered from roller-coaster trends in behavioral and social sciences spending at
the federal level, due partly to the linkages and overlaps with life science and
computer science research, which have been on more monotonic funding paths.
Cognitive and behavioral sciences support nevertheless has not kept up with
the growth in scientific opportunities, such as the rapid shifts in technological
capabilities that have resulted from the microprocessor revolution. A rapid
increase in funds for investigator-initiated grants- by about $20 million an-
nually is a high priority. These funds should be used to increase research
productivity in three ways. First, they should be used to reverse the lack of
change or actual reduction in the size of grants that has occurred at a time
when real costs are increasing. Those real cost increases have resulted from
improvements in animal care, the need for auxiliary staff in experiments in-
volving infants and children, and the routine costs of supplies and services,
among other matters. Second, they should be used to increase the duration of
a typical grant to an experienced investigator from 3 to S years. Such an increase
will sharply reduce the burden of time imposed both on investigators preparing
renewal grant proposals and on the individuals asked to read and evaluate each
proposal (generally 6, but sometimes as many as IS), which is a nonnegligible
cost in research time. Third, they should be used to increase somewhat the
total number of grants available because many promising proposals are now
being turned down due to lack of funds.
A great deal of the pioneering research lies at the interface of disciplines and
often requires a great deal of highly specialized technical expertise in fields
ranging from biochemistry and physics to neuroscience, physiology, psychol-
ogy, linguistics, economics, statistics, mathematics, and computer science. This
inherent feature of much current work on behavior, the mind, and the brain
calls for a number of developments to foster and facilitate more collaboration
among people from very different backgrounds. This facilitation may take dif-
ferent forms, such as short-term visits, new centers of research, interdisciplinary
training in the predoctoral curricula of universities, or the growth of postdoc-
toral programs that broaden rather than intensify the focus of new scientists'
thesis work. We especially note the need for advanced training institutes for
younger postdoctoral-level researchers working on newly emerging methods
in highly technical specialities and recommend an additional $1 million be
spent annually for such institutes. Given the geographical dispersion of inves-
tigators, collaboration can often be sustained only if there are opportunities to
come together periodically in joint working sessions. Short-term workshops,
seminars, and conferences focused on exchanges of current ideas and methods
46 / The Behavioral and Social Sciences
are the ideal medium to encourage this, and we recommend that $1 million
be added to current levels of funding for such activities.
The picture of recruitment into graduate schools in the behavioral, cognitive,
and brain science disciplines continues to reflect a growth trend. But much of
this growth leads toward clinical and other nonresearch employment, and there
is growing concern about the number of highly motivated and talented young
people who are entering research careers. In particular, there is a relative lack
of opportunity for individuals to gain intense experience at the predoctoral or
postdoctoral levels in a variety of research methods and theoretical disciplines.
This is only in small part a problem of curriculum: it is in large part a problem
of funding, which is available more for teaching and clinical or applied work
than for research apprenticeships. We therefore recommend that an additional
$5 million annually be invested in research fellowships at the predoctoral and
postdoctoral levels, with the majority share ($3 million) directed to the post-
doctoral level.
There is in certain areas a critical need to inaugurate new data bases that can
be accessed by large numbers of investigators; in particular, longitudinal data
bases concerning cognitive and educational development. The cost of these
new data collection effort would be about $5 million annually.
Finally, there is at present momentum in certain fields toward major new
interdisciplinary research enterprises. In the recent climate of restrained budget
growth, it has not been possible to organize the sort of stimuli either in the
form of planning grants or competitive announcements for new center pro-
grams that would lead to competitive, full-fledged center proposals. We rec-
ommend that there be an initial commitment of $4 million to create two to
three new interdisciplinary centers, with a view toward increasing this by as
much as threefold if the experience of productivity warrants. We should note
that some increment in support staffing in research agencies is a necessary
element to generate, evaluate, and monitor good center grants.
