A BIBLIOGRAPHICAL GUIDE TO THE MIND
Piero Scaruffi
Draft of April 1997
Preface
This document provides a guided bibliography to literature published
(mainly, but not exclusively, over the last two decades) on the sub-
ject of the mind. The exponentially growing interest on the mind,
consciousness and life is altering the course of many disciplines and
opening new fascinating horizons for science. Subjects created by this
trend include Artificial Intelligence, Artificial Life, Neural Net-
works, Cognitive Science and Complex Systems. Physics itself is being
rewritten, in the quest for a grand theory of nature that will unify
the physical and the psycological sciences.
Books reviewed in this bibliography therefore span Philosophy,
Psychology, Biology, Computer Science, Neurophysiology, Mathematics
and even Cosmology.
This document contains an alphabetical list of books with a short
review of their contents. The reader can use it to decide which books
to buy for futher information, or just "cut and paste" the information
for her or his personal research.
In a sense, this document provides the researcher, or the merely curi-
ous, with the "tools" to work out her or his own theory of the mind.
At the same it provides everybody with an updated survey of one of the
most exciting fields of today's science.
Proceedings of conferences have been generally omitted, but collection
of historical articles are included. Books that have been made
obsolete by new editions or by new books by the same author are gen-
erally omitted.
The decision of which books had to be included was largely subjective.
Recent books have been given a higher priority, both because of their
availability and because they are likely to include information about
older texts.
In reviewing a book I have often quoted liberally from the author.
Each review is meant to deliver the main original points of the book.
It is not meant to be an exaustive review of the entire content.
The author will gladly receive information about books that should
have been included and were not. A future edition will hopefully do
justice to the ones who were forgotten this time.
Nobody has time anymore to read all the interesting books that are
written in the world. This is a humble effort to make it possible to
be at least aware of their existence.
Piero Scaruffi
scaruffi@hpp.stanford.edu
Introduction
These days something is happening that is likely to have deep reper-
cussions on the future of science. A new view of nature is emerging,
which encompasses both galaxies and neurons, gravitation and life,
molecules and emotions. As a culmination of centuries of studying
nature, mankind has been approaching the thorniest subject of all:
ourselves. We are part of nature, but science leaves us in the back-
ground, limiting our role to the one of observers.
For a long time we have enjoyed this priviliged status. But we seem no
longer capable of eluding the fundamental issue: that what we have
been studying for all these centuries is but us, albeit diguised under
theories of the universe and theories of elementary particles. And now
it is about time that we focus on the real subject. The mind appears
to us as the ultimate and most refined product of life. And life
appears to us as the ultimate and most refined product of the
universe. Life and mind must follow from a logical inference on the
essence of the universe. If we had the right theory of the universe,
we would need no effort in explaining why life happened and what the
mind is.
The fact that we don't have yet a good theory of the mind means that
probably we don't have a good theory of the universe. Therefore, in a
sense, the science of the mind is doing more than just studying the
mind: it is indirectly reformulating the whole of science.
Thanks to progress in all fields, from Mathematics to Neurobiology,
our knowledge has been immensely enriched by a wealth of empirical
data and by a wealth of theoretical tools. While differing on the
specifics, many scientists and philosophers feel that mankind is now
ready for a momentous synthesis. The main theme of such a synthesis
may be that of the spontaneous "emergence" in our universe of such
unlikely properties as life and consciousness. If we can explain how
it developed, we can explain what it is and how it works.
Ageno Mario: LE ORIGINI DELL'UNIVERSO (Boringhieri, 1992)
Mario Ageno shows that Boltzmann's proof contains two errors: 1.
Boltzmann's model of a gas represents a discrete set of molecules
as a continuum of points; 2. Boltzmann assumes that the walls
containing the closed system are perfectly reflecting. If these
arbitrary assumptions are dropped, no rigorous proof for the
irreversibility of natural processes exists.
Aggleton John: THE AMYGDALA (Wiley-Liss, 1992)
The book explores various neurobiological aspects of emotion and
memory. Emotions are key to learning and behavior as fear condi-
tioning imprints emotional memories that are quite permanent. The
relationship between emotion and memory goes beyond fear, but
fear is the emotion that has been studied more extensively. As a
matter of fact, fear seems to be a common ground for (at least)
all vertebrates. Memories about fearful experiences are created
by interactions among the amygdala, the thalamus and the cortex.
Emotional memory (stored in the amygdala) differs from declara-
tive memory (which is mediated by the hippocampus and the cor-
tex). Emotional memory is primitive, in the sense that only con-
tains simple links between cues and responses. A noise in the
middle of the night is enough to create a state of anxiety,
without necessarily bringing back to the mind full consciousness
of what the origin of that noise can be. This actually increases
the efficiency (at least the speed) of the emotional response.
Emotional and declarative memories are stored and retrieved in
parallel. Adults cannot recall childhood traumas because in
children the hippocampus has not yet matured to the point of
forming conscious memories, but the emotional memory is there.
Allen James: NATURAL LANGUAGE UNDERSTANDING (Benjamin Cummings,
1995)
The new edition of one of the best textbooks on natural language
processing, from basic parsing techniques to anaphora resolution,
discourse structure to speech acts.
Allen James: READINGS IN PLANNING (Morgan Kaufmann, 1990)
Allen's temporal logic is based on a many-sorted predicate cal-
culus with variables ranging over "properties", "time intervals",
"events", etc. Temproal relations such as "during", "before",
"overlap", "meets" and "equal" are primitive, are represented by
predicates and are controlled by the axioms of the logic. An
instant is defined as a very small interval. Properties hold at
intervals.
Amari Shun-ichi & Freeman Walter: NEURAL NETWORKS AND CHAOS
(Lawrence Erlbaum, 1994)
A collection of papers for a workshop on the subject.
Anderson James A. & Rosenfeld Edward: NEURO-COMPUTING (MIT
Press, 1988)
A comprehensive collection of historical papers on brain anatomy,
cognitive psychology, cybernetics and neural networks.
William James had a number of powerful intuitions: that the brain
is built to ensure survival in the world; that cognitive func-
tions cannot be abstracted from the environment that they deal
with; that the brain is organized as an associative network; that
associations are governed by a rule of reinforcement.
Warren McCulloch's and Walter Pitts' 1943 "A logical calculus of
the ideas immanent in the nervous system" is a seminal paper that
laid down the foundations for the computational theory of the
brain. Their binary neuron can only be in one of two possible
states, has a fixed threshold below which it never fires, can
receive inputs from either inhibitory synapses and/or excitatory
synapses, and integrates its input signals at discrete intervals
of time. If no inhibitory synapse is active and the sum of all
excitatory synapses is greater than the threshold, the neuron
fires. A network of binary neurons is fully equivalent to a
universal Turing machine (i.e., that any finite logical proposi-
tion can be realized by such a network, i.e. every computer pro-
gram can be implemented as a neural net).
Featured are the main forefathers of today's neural architec-
tures. Oliver Selfridge's 1958 "Pandemonium" employs a battery
of multiple independent units analyze the input, each specialized
in a different recognition task, so that the input can be pro-
gressively identified through a number of hierarchical layers,
each one relying on the conclusions of the lower ones.
Rosenblatt's 1958 "Perceptron", based on a non-linear model of
memory, was probably the first artificial neural network for
learning concepts.
Bernard Widrow's and Marcian Hoff's 1960 "Adaptive switching cir-
cuits" yield the ADALINE, a variation on the perceptron based on
a supervised learning rule, the "error correction rule", that
could learn in a faster and more accurate way: synaptic strenghts
are changed in porportion to the error (what the output is and
what it should have been) times the input.
Briefly mentioned are also Teuvo Kohonen's linear model for
memory and Stephen Grossberg's non-linear quantitative descrip-
tions of brain processes.
John Hopfield's 1982 "Neural networks and physical systems"
developed a model inspired by the "spin glass" material, which
resembles a one-layer neural network in which weighs are distri-
buted in a symmetrical fashion, the learning rule is hebbian,
neurons are binary and each neuron is connected to every other
neuron. As they learn, Hopfield's nets develop configurations
that are dynamically stable (or "ultrastable"). Their dynamics is
dominated by a tendency towards a very high number of locally
stable states (or "attractors"). Every memory is a local
"minimum" for an energy function similar to potential energy.
Hopfield's nets exhibit the ability to correct incomplete or
incorrect information (because deviations from local minima are
attracted towards one of those minima and therefore canceled
away). Compared with the perceptron, a Hopfield net is asynchro-
nous (which is a more plausible model of the nervous system) and
employs backward coupling. In a later paper (also included here)
Hopfield replaced the binary neuron with a more plausible neuron.
More than anything else, Hopfield proved that, despite Minsky's
critique, neural networks are feasible and can even be useful.
Fukushima's 1983 "Neocognitron" is a multi-layered network with
strong self-organizing properties, based on Hubel' and Weisel's
model of the visual system. A number of modules are triggered by
a retina of photoreceptors. Each module has a number of simple
"S-cells" and more complex "C-cells", driven by "S-cells" layers
so that they abstract the features that the "S-cells" pick up.
In Geoffrey Hinton's and Terrence Sejnowsky's 1985 "A learning
algorithm for Boltzmann machines" Hopfield's basic architecture
(binary neuron, energy function and so on) is retained, but
Hopfield's learning rule is replaced with the rule of annealing
(start off the system at very high "temperature" and then gradu-
ally drop the temperature to zero), which Kirkpatrick and others
had just proposed as a general-purpose optimization rule. The
new model, Boltzman's machine, is more stable than Hopfield's
model as it will always end in a global minimum (the lowest
energy state).
David Rumelhart's and Geoffrey Hinton's "back-propagation" algo-
rithm, originally proposed in 1986, considerably faster than the
Boltzmann machine, quickly became the most popular learning rule
for multi-layered netowrks. The generalized "delta rule" was
basically an adaptation of the Widrow-Hoff error correction rule
to the case of multi-layered networks by moving backwards from
the output layer to the input layer. This was also the definitive
answer to Minsky's critique, as it proved to be able to solve all
of the unsolved problems.
Anderson James A.: NEURO-COMPUTING 2 (MIT Press, 1990)
Another set of historical articles, including seminal papers on
Caianiello's neural equations, Wiener's cybernetics, Pribram's
holographic model, Minsky's critique of perceptrons and Fodor's
And Pylyshyn's "Connectionism and cognitive architecture" on the
feasibility of a compositional theory.
Anderson John Robert: THE ARCHITECTURE OF COGNITION (Harvard
Univ Press, 1983)
ACT, as developed in 1976, was a cognitive architecture capable
of dealing with both declarative knowledge (represented by propo-
sitional networks) and procedural knowledge (represented by pro-
duction rules). The production system worked as the interpreter
of the propositional network.
New production rules are learned as the system works. Complex
cognitive skills can develop from a simple architecture.
ACT assumes that a cognitive system has two short-term memories:
a declarative memory (that remembers experience) and a productive
memory (that remember rules learned from experience). Knowledge
is compiled into more and more complex procedural chunks through
an incremental process of transformation of declarative knowledge
in procedurale knowledge. An incremental process transforming
declarative knowledge into procedural knowledge consolidates
knowledge into ever more complex procedural chunks. Each rule is
weighed according to how often it is used, and the weight deter-
mines its priority.
Anderson John Robert: THE ADAPTIVE CHARACTER OF THOUGHT
(Lawrence Erlbaum, 1990)
The book explores the cognitive architecture known as ACT, which
broadens the principles of production systems.
Anderson has developed a probabilistic method to explain how
categories are built and how prototypes are chosen. Anderson's
model maximizes the inferential potential of categories (i.e.,
their "usefulness"): the more a category helps predict the
features of an object, the more the existence of that category
makes sense. For each new object, Anderson's model computes the
probability that the object belongs to one of the known
categories and the probability that it belongs to a new category:
if the latter is greater than the former, a new category is
created.
Anderson John Robert: RULES OF THE MIND (Lawrence Erlbaum, 1993)
In this book Anderson looks for the psychological evidence of
production systems (in particular in the area of acquisition of
cognitive skills) and refines ACT into ACT-R, which includes a
neural-network implementation of a production system. The book is
structured as a set of articles by Anderson and others, and it
includes simulation software.
Anderson James: AN INTRODUCTION TO NEURAL NETWORKS (MIT Press,
1995)
A very up-to-date 600-page survey of the mathematical foundations
of neural networks that neatly organizes linear associators, per-
ceptrons, gradient descent algorithms (ADALINE, back propaga-
tion), nearest neighbor models, Kanerva's sparse distributed
memories, energy-based models (Hopfield model, Boltzmann
machine), Kohonen's adaptive maps, the BSB model, etc. The sonar
system of the bat is also reviewed.
Anderson Norman: A FUNCTIONAL THEORY OF COGNITION (Lawrence Erl-
baum, 1996)
Anderson presents a unified theory of functional cognition, i.e.
a cognitive theory of everyday life, centered around the founda-
tion axiom of purposiveness.
Aoun Joseph: A GRAMMAR OF ANAPHORA (MIT Press, 1986)
Aoun deals with reciprocals and reflexives by proposing a gen-
eralized government-binding theory that leads to a structural
unification of the notions of pronouns, empty categories and ana-
phors.
Arbib Michael: THE HANDBOOK OF BRAIN THEORY AND NEURAL NETWORKS
(MIT Press, 1995)
This 1,000-page handbook (compiled by dozens of experts under the
direction of Michael Arbib) covers topics in Psychology, Philoso-
phy, Neurophysiology, Artificial Intelligence, self-organizing
systems, neural networks, etc.
Arbib Michael: METAPHORICAL BRAIN (Wiley, 1972)
This introduction to cybernetics begins with dividing simulation
and emulation approaches to modeling intelligent behavior, i.e.
artificial intelligence and neural networks. Then the book
focuses on brain theory, considering the brain as a particular
type of machine.
Arbib Michael: CONSTRUCTION OF REALITY (Cambridge University
Press, 1986)
The mind constructs reality through a network of schemas. A
schema is both a mental representation of the world and a process
that determines action in the world. Arbib's theory of schemas
is based on Pierce's notion of a "habit" (a set of operational
rules that, by exhibiting both stability and adaptability, lends
itselft to an evolutionary process) and Piaget's notion of a
"scheme" (the generalizable characteristics of an action that
allow the application of the same action to a different context).
Both assume that schemas are compounded as they are built to
yield successive levels of a cognitive hierarchy. Categories are
not innate, they are constructed through the individual's experi-
ence. What is innate is the process that underlies the construc-
tion of categories (this is similar to Chomsky's view of the
rules of language).
The theory of schemas is consistent with a model of the brain as
an evolving self-configuring system of interconnected units.
The construction of reality is also guided by social conventions,
as the formation of new schemas is sometimes a social process.
Language arises from such a process. Arbib argues that all
language is metaphorical and bases its theory of language on
Black's interaction theory of metaphor: to understand the meaning
of a sentence is not only to be able to identify its referent,
but also to call to mind all the schemas associated to it. Meta-
phor is a necessary ingredient of any symbolic system.
The theory is applied to a wealth of issues in psychoanalysis,
hermeneutics, epistemology and even theology.
Arbib Michael: FROM SCHEMA THEORY TO LANGUAGE (Oxford Univ
Press, 1987)
A theory of language based on Arbib's theory of schemas, with a
practical implementation.
Arbib Michael: BRAINS MACHINES AND MATHEMATICS (Springer Verlag,
1987)
An introduction to some topics of cybernetics, neural networks,
Turing machines, self-reproducing automata and to Godel's incom-
pleteness theorem.
Arbib Michael: METAPHORICAL BRAIN 2 (Wiley, 1989)
The second volume greatly expands the contents of the first
volume. Besides a little neuroanatomy, the focus is on mathemat-
ical analyses of neural phenomena from the perspective of
action-oriented perception and in the light of Arbib's own theory
of schemas. Schema theory is applied to the vision of the frog
and high-level recognition, hand control and speech understand-
ing. Along the way, mathematical models are offered to explain
locomotion and eye movement; and all the main learning models
(from perceptrons to the HEARSAY system, from Hopfield nets to
Boltzmann machines, from backpropagation to the NETTALK system)
are formally introduced.
Arbib advances a theory of consciousness: first language
developed, as a tool to communicate with other members of the
group in order to coordinate group action; then communication
evolved beyond the individual-to-individual sphere into the self
sphere.
Armstrong David Malet: BELIEF, TRUTH AND KNOWLEDGE (Camrbidge
University Press, 1973)
Beliefs are maps of the world (with the believer as central
reference) by which the believer's actions are guided. Beliefs
are states that have an internal structure: the content of the
proposition believed. Beliefs may be reduced to the deep struc-
tures of Chomsky's linguistic theory. Beliefs often come in
degrees: a partial belief is a degree of causal efficacy of the
belief state in relation to action.
Armstrong David Malet: THE NATURE OF MIND (Cornell Univ Press,
1981)
A philosophical treaty on the dualism of the mind, which also
presents Armostrong's causal theory of the mind. Mental states
and physical states are identical (just like we perceive many
natural phenomena without perceiving the corresponding micros-
copic physical processes) and a mental state is causally con-
nected with a physical state. A state of the brain causes a men-
tal state. Consciousness of a mental state is a perception of
that mental state.
Consciousness is the perception of mental states. Its special
status is purely illusory. The self is the single continuing
entity that appears from the organization of introspection. The
biological function of consciousness is to sophisticate the men-
tal processes so that they yield more interesting action.
Ashby William: AN INTRODUCTION TO CYBERNETICS (Chapman & Hall,
1956)
In this book Ashby summarized a number of influential concepts.
He placed emphasis on feedback, the process that allows for
"homeostasis". Both machines and living beings tend to change to
compensate variations in the environment, so that the combined
system is stable. For living beings this translates into "adapta-
tion" to the environment. The "functioning" of both living
beings and machines depends on feedback processes. Ashby also
emphasized the power of self-organizing systems, systems made of
a very high number of simple units which can evolve autonomously
and adapt to the environment by virtue of their structure.
In 1962 Ashby also formulated his principle of self-organization:
"in any isolated system life and intelligence inevitably
develop". In every isolated system subject to constant forces
"organisms" arise that are capable of adapting to their environ-
ment.
Ashmore Richard & Jussim Lee: SELF AND IDENTITY (Oxford Univ
Press, 1997)
A series of article on the relationship between the self and the
idea of identity.
Ashtekar Abbay: CONCEPTUAL PROBLEMS OF QUANTUM GRAVITY (Bir-
khauser, 1991)
Ashtekar is a proponent of the loop-space theory of quantum grav-
ity. To quantize gravity physicists only need to show that grav-
itational waves consist of quantum force-carrying particles, or
gravitons. The perturbation methods that have been developed to
this purpose (and which gave rise to the theory of superstrings,
infinitesimal loops of energy whose wrigglings should generate
particles and forces) have largely failed because gravitons,
unlike other force carriers, alter the very geometry of space and
time, which in turn affects their behavior; in other words,
because of gravity's inherently self-referential, non-linear
nature.
By using Amithaba Sen's variable, time and space can be split in
two distinct entities subject to quantum uncertainty just like
position and momentum. AShetkar's equations generate exact solu-
tions for quantum gravitational states that can be represented by
loops (as in knot theory). The loops are tightly knitted
together. Gravitons are embroidery knitted into the loops.
Austin John Langshaw: HOW TO DO THINGS WITH WORDS (Oxford Univ
Press, 1962)
Austin handles language as a particular case of action, "speech
action".
Austin introduced a tripartite classification of acts performed
when a person speaks. Each utterance entails three different
categories of speech acts: a locutionary act (the words employed
to deliver the utterance), an illocutionary act (the type of
action that it performs, such as warning, commanding, promising,
asking), and a perlocutionary act (the effect that the act has on
the listener, such as believing or answering).
A locutionary act is the act of producing a meaningful linguistic
sentence. An illocutionary act sheds light on why the speaker is
uttering that meaningful linguistic sentence. A perlocutionary
act is performed only if the speaker's strategy succeeds.
Austin believes that any locutionary act (phonetic act plus
phatic act plus rhetic act) is part of a discourse which bestows
an illocutionary force on it. All language is therefore an illo-
cutionary act.
Austin John Langshaw: SENSE AND SENSIBILIA (Clarendon, 1962)
We cannot directly perceive material objects, but only sense-
data.
Austin John Langshaw: PHILOSOPHICAL PAPERS (Clarendon, 1961)
A collection of all the philosophical papers of the philosopher
famous for his theory of truth as grounded in historical situa-
tions: "a statement is true when the historic state of affairs to
which it is correlated by the "demonstrative" conventions is of a
type with which the sentence used in making it is correlated by
the "descriptive" conventions. Descriptive conventions correlate
sentences with types of situation. Demonstrative conventions
correlate statements with historic situations.
Baars Bernard: A COGNITIVE THEORY OF CONSCIOUSNESS (Cambridge
Univ Press, 1993)
Conscious experience is distributed widely throughout the nervous
system. Any conscious experience emerges from cooperation and
competition between the many processing units of the brain work-
ing in parallel and occurs within a background of unconscious
contexts. The self is the dominant, enduring context of many
conscious experiences. Baars brings forth a wealth of psychologi-
cal and neurophysiological data to justify his views. The ner-
vous system can be viewed as a set of independent intelligent
agents which broadcast messages to the other agents through a
common workspace (just as if they were writing on a blackboard
visible to every other agent). That workspace is consciousness.
He starts by proving that the mind contains unconscious mental
representations, such as episodic memories and linguistic
knowledge; that the mind originates from the work of many
independent, specialized "processors", i.e. skills that have
become highly practiced, automatic and unconscious. Baars
emphasizes the striking differences between conscious and uncons-
cious processes: unconscious processes are much more effective
(e.g., we parse sentences unconsciously all the time, but cannot
consciously define how we parse them), they operate in parallel
(whereas we can only have one conscious process at the time),
they appear to have almost unlimited capacity (conscious
processes have very limited capacity).
Contexts are created by a dual process of searching for informa-
tion and adaptation to information, the former leading to more
conscious access, the latter reducing conscious access (things
become habitual and automatic). Baars emphasizes the relation-
ship between information and consciousness (perceptual systems
are more sensitive to information than energy, redundant informa-
tion fades from consciousness). Conscious experience is informa-
tive and triggers widespread adaptive processes. Conscious
experience is the product of biological adaptation.
Baars Bernard: IN THE THEATRE OF CONSCIOUSNESS (Oxford Univ
Press, 1996)
Baars conceives consciousness as a theatrical stage for emotion,
perception and thoughts.
Back Thomas, Fogel David & Michalewicz Zbigniew: HANDBOOK OF
EVOLUTIONARY COMPUTATION (Oxford Univ Press, 1997)
The ultimate handbook for professional genetic algorithm users.
Bach Emmon: UNIVERSALS IN LINGUISTIC THEORY (Holt, Rinehart &
Winston, 1968)
A collection of four essays on linguistics, the longest one being
Charles Fillmore's seminal "The case for cases".
Fillmore's grammar assumes that each sentence represents expli-
citly the relationships between concepts and action. A universal
underlying set of caselike relations play a key role in determin-
ing syntactic and semantic relations in all languages. A sen-
tence is represented by identifying its "cases", analogous to
noun cases. Sentences that deliver the same meaning with dif-
ferent words are therefore represented in the same way.
Bach Emmon: CATEGORIAL GRAMMARS (Reidel, 1988)
A collection of essays on what Yehoshua Bar-Hillel defined in
1960 as categorial grammar, that provide an excellent historical
introduction to the field.
In contrast to linguistic analyses based on phrase structure
grammars, in a categorial grammar every item of a language
belongs to one or more categories; a category can be either basic
or derived; derived categories are defined in terms of basic or
derived categories in a compositional way. Expressions belonging
to derived categories may be identified with funtions that map
expressions of one constituent category into expressions of
another constituent category.
Categorial grammars adhere to three principles: language is seen
in terms of functions and arguments rather than constituent
structure (dependency grammar rather than phrase-structure gram-
mar); a tight correspondence is imposed between syntax and seman-
tics such that every rule of syntax is also a rule of semantics
(the rule-to-rule hypothesis); monotonicity is always favored at
the expense of destructive devices which characterize transforma-
tional grammars.
Categorial grammars are based on the algebraic notions of func-
tion and argument and can therefore be represented using Church's
lambda operator. The Lambek calculus was the first major
mathematical tool for the field.
Categorial grammars involve semantic categories, in agreement
with Edmund Husserl's meaning categories and Stanislaw
Lesniewski's and Kazimierz Ajdukiewicz's logics.
Bach has improved the original model by allowing categories to
have internal structures that define the features that are
relevant to determine lexical and syntactic properties.
Categories can then be viewed as clusters of features.
Bach Emmon: SYNTACTIC THEORY (Holt, Rinehart & Winston, 1974)
An in-depth treatment of transformational grammars for linguists
that summarizes the progress made in the early Seventies and
updates Bach's earlier "Introduction to Transformational Gram-
mars". It contains a long introduction to Chomsky's "Aspects of
the Theory of Syntax".
Baddeley Alan: WORKING MEMORY (Clarendon Press, 1986)
Baddeley developed a theory of work memory based on three subsys-
tems: a central control (for residual ignorance) and two passive
storage systems, a speech system and a visual system.
Baddeley Alan: YOUR MEMORY (MacMillan, 1982)
An introduction to the functioning and structure of memory for
the broad audience. Baddeley assumes the existence of three types
of memory: long-term (both episodic and semantic), short-term and
sensory memory.
Baddeley Alan: ATTENTION (Oxford University Press, 1994)
A tribute to Donald Broadbent in the form of a collection of
essays on his contributions to various cognitive tasks.
Baddeley Alan: HUMAN MEMORY (Simon & Schuster, 1990)
An introduction to the theories of memory for the broad audience.
Ballard Dana: COMPUTER VISION (Prentice Hall, 1982)
This monumental book describes a detailed computational model of
how physical objects can be constructed from images.
Baltes Paul: LIFE-SPAN DEVELOPMENT AND BEHAVIOR (Academic Press,
1984)
Baltes' theory of dual processes assumes that intelligence as
information processing is universal and biological, whereas
intelligence as knowledge pragmatics is acquired through experi-
ence and therefore influenced by cultural factors.
Bar-Hillel Yehoshuas: LANGUAGE AND INFORMATION (Addison Wesley,
1964)
By building on Lesniewski's and Ajdukiewicz's semantic
categories, Bar-Hillel defined a variant of phrase structure
grammar that he called categorial grammar in which "every sen-
tence is the result of the operation of one continous part of it
upon the remainder, these two parts being the immediate consti-
tuents of the sentence, such that these constituents are again
the product of the operation of some continous part upone the
remainder, etc".
Barkow Jerome, Cosmites Leda, Tooby John: THE ADAPTED MIND
(Oxford Univ Press, 1992)
A collection of articles on evolutionary psychology (i.e., the
evolution of the mind).
The mind consists of specialized modules designed by natural
selection to solve problems in the environment that have to do
with survival and reproduction.
Social darwinism is the evolution of Wilson's sociobiology
Barr Avron & Feigenbaum Ed: HANDBOOK OF ARTIFICIAL INTELLIGENCE
(William Kaufmann, 1982)
A monumental catalog of models and techniques for A.I. profes-
sionals and researchers.
Barsalou Lawrence: COGNITIVE PSYCHOLOGY (Lawrence Erlbaum, 1992)
An introduction to the field.
Bartlett Frederic Charles: REMEMBERING (Cambridge Univ Press,
1967)
In 1932 Bartlett developed one of the earliest cognitive models
of memory. Bartlett noted how memory cannot remember all the
details, but can "reconstruct" the essence of a scene. Events
cannot be stored faithfully, but must have been summarized into a
different form, a "schema". Individuals do not passively record
stories verbatim, but rather actively code them in terms of sche-
mas, and then can recount the stories by retranslating the sche-
mas into words.
Each new memory is categorized in a schema which depends on the
already existing schemas. In practice, only what is strictly
necessary is added. When a memory must be retrieved, the
corresponding schema provides with instructions to reconstruct
it. Bartlett notes how much easier it is to recognize an object
in its typical environment.
That is why recognizing an object is much easier in its typical
context than in an unusual context.
Barwise John & Perry John: SITUATIONS AND ATTITUDES (MIT Press,
1983)
Inspired by Gibson's ecological realism, Barwise proceeds to undo
Frege's theory of meaning (that meaning is located in the world
of sense). The world is full of meaning and information that
living organisms can use.
Meaning is not an exclusive of language, it is pervasive in
nature ("smoke means fire"). Meaning involves the informational
content of situations and arises from regularities in the world.
Reality is made of situations. Sentences stand for situations.
The semantic value of a sentence is a set of abstract situations.
Meaning arises out of recurring relations between situations.
Barwise's formalism employs Kleene's partial functions (which
deal with finite amounts of information).
Reality comes in situations. Situations are made of objects and
spatio-temporal locations; objects have properties and stand in
relations. Therefore, a situation is described by a set of rela-
tions between objects.
A situation-type is a partial relation from n-ary relations and n
individuals to the values true and false. A course of events is
a partial function from locations to situation-types. Therefore
a course of events at a location on which it is defined yields a
situation-type. A state of affairs is a course of events which
is defined on just one location.
A living organism (a part of reality capable of perception and
action) must be able to cope with the ever new situations of its
course of events and to anticipate the future course of events.
It must be able to pick up information about one situation from
another situation. This can be realized by identifying similari-
ties between situations and relations between such similarities.
Each organism performs this process of breaking down reality in a
different way, as each organism "sees" reality in a different
way, based on its ecological needs.
The type of a situation is determined by the regularities that
the situation exhibits. Regularities are invariants differen-
tiated by organism, acquired by adaptation to the environment,
that define its behavior in the environment. These similarities
between various situations make it possible for an organism to
make sense of the world. At the same time they are understood by
all members of the same specie, by a whole "linguistic commun-
ity".
Formally, one situation can contain information about another
situation only if there is a relation that holds between situa-
tions sharing similarities with the former situation and situa-
tions sharing similarities with the latter situation. In that
case the first situation "means" the second. A meaning is a
relation between different types of situations. In situational
semantics the meaning of a declarative sentence is a relation
between utterances and described situations.
Therefore, constraints between types of situations are actual and
yield meaning. Meaning is defined as relations that allow one
situation to contain information about another situation.
Situational semantics solves the semantic problems of granular-
ity, context and factorization by expressing properties and rela-
tions as primitive entities. By assuming that sentences stand for
situations, it avoids all the pitfalls of the logical tradition,
for which sentences stand for truth values.
Situations are more flexible than possible worlds because they
don't need to be coherent and don't need to be maximal. Just like
mental states.
Indexicals are held to represent not only a few isolated words
such as "I" and "now" but the way the speaker exploits the
discourse context. They play a key role in the way language con-
veys information.
Propositional attitudes report relations to situations.
The book also contains a theory of knowledge and beliefs that is
similar to Dretske's. An agent knows that p if the agent has a
belief that p and that belief carries the information that p.
Barwise Jon: THE SITUATION IN LOGIC (Cambridge Univ Press, 1988)
A collection of a few historical papers by Barwise on situation
theory and situation semantics, including philosophical discus-
sions, replies to criticism and introduction to the mathematical
rudiments. Barwise also extendes and refines a few of his origi-
nal concepts.
Logic should be studied from the perspective of information,
information processing and information communication. Barwise
emphasizes the relational nature of information (e.g., perception
is a relation between perceiver and perceived) and the cir-
cumstantial nature of information (information is information
about the world).
Situation semantics emphasizes two related phenomena: efficiency
of language and partiality of information. Situation semantics
offers a relation theory of meaning: the meaning of a sentence
provides a constraint between the utterance and the described
situation.
Bechtel William: PHILOSOPHY OF MIND (Lawrence Erlbaum, 1988)
A broad and accessible survey of various schools of philosophy of
mind. The book is organized around three topics: language,
intentionality and mind-body problem. As far as language goes it
covers referential analysis of meaning (Frege, Russell), speech
act theory (Austin, Searle, Grice), holistic analysis of meaning
(Quine, Davidson), Kripke's possible world semantics and Putnam's
causal theory of reference. The chapters on intentionality deal
with the computational theory of mind, cybernetics, Dennett's
intentional stance. The mind-body problem is summarized from
Descartes' dualism to behaviorism, identity theories, eliminative
materialism and functionalism.
A survey of ancient and modern theories of the mind.
Bechtel William: PHILOSOPHY OF SCIENCE (Lawrence Erlbaum, 1988)
An introduction to logical positivism and most recent theories
(Kuhn, Feyerabend, Lakatos).
A survey of modern theories of science.
Bechtel William & Adele Abrahamsen: CONNECTIONISM AND THE MIND
(MIT Press, 1991)
Drawing from James McClelland, David Rumelhart and Geoffrey Hin-
ton, the book provides a primer to connectionist networks, with
examples on connectionist simulations of language and reasoning.
The book includes a lengthy defense of connectionism against cri-
ticism and a survey of the impact of connectionism on other dis-
ciplines.
Behe Michael: DARWIN'S BLACK BOX (Free Press, 1996)
Behe is skeptical about Darwin's theory of evolution because
cells are too complex to have evolved spontaneously. Most cellu-
lar systems are "irreducibly complex", i.e. they could not work
without some of their parts. If one of the parts is not there,
the system does not operate, and therefore cannot reproduce and
evolve. Such systems cannot be built by gradual evolution: too
many of their parts must be there in order for them to be able to
start evolving. Their structure cannot be due to evolution
because their function cannot be built incrementally. For exam-
ple, a mousetrap is not a mousetrap until it has a spring: a
mousetrap with the spring cannot evolve from a mousetrap without
a spring because the latter would have no function, therefore
would simply not reproduce. Organisms are even more complex than
mousetraps: they require sophisticated mechanisms for storing and
transporting enzymes and proteins, among other things. The cell
is too complicated, and it needs to be that complicated in order
to be a living cell, and therefore it cannot have evolved from
something that was less complicated. Behe concludes that life
must have been designed by an intelligent agent.
Berwick Robert: PRINCIPLE-BASED PARSING (Kluwer Academic, 1991)
A collection of articles in principle-based parsing a small set
of fundamental principles is used to derive sentence types (such
as passive). The principles interactive deductively to construct
sentence types. Parsers are highly specialized inference pro-
cedures.
Bickerton Derek: LANGUAGE AND SPECIES (Chicago Univ Press, 1992)
Bickerton thinks that language was invented to represent the
world and guesses what long series of evolutionary events helped
develop that faculty.
Language is sufficient to account for the rationality and intel-
ligence of humans. Language created the human species and the
world that humans see. Language is a biological feature that
arises from the genetic code. Language was created during the
evolutionary path by a change in neural organization
Syntax is the fulcrum of language.
Bischof Horst: PYRAMIDAL NEURAL NETWORKS (Lawrence Erlbaum,
1995)
Bischof thinks that the complex task of vision is performed
effortlessly by the brain because of a massive use of hierarchi-
cal structures.
Black Max: MODELS AND METAPHORS (Cornell Univ Press, 1962)
Black's interaction theory of metaphor views the metaphor as a
means to reorganize the properties of the destination. A meta-
phor is not an isolated term, but a sentence. The metaphorical
sentence, or "frame", contains the words that are used metaphori-
cally, or the "focus". A metaphor involves two subjects, and one
of them, the secondary subject, comes with a system of associated
stereotyped information which can be used as a filter on the
principal subject. Therefore, there is a tension between the two
subjects of a metaphor, each subject is a system and the metaphor
consists in a transaction between the two systems.
A metaphor does not express similarities: it creates similarity.
Metaphors are based on similitude, not analogy. Metaphors act on
the organization of the lexicon and the model the world. Meta-
phorizing is related to categorizing (the choice of a category in
which to place an object is a choice of perspective), but is dis-
tinguished from it by an incongruity which causes a reordering
and a new perspective.
Language is dynamic: what is literal may become metaphoric and
viceversa.
Block Ned: READINGS IN PHILOSOPHY OF PSYCHOLOGY (Harvard Univ
Press, 1980)
A collection of articles on behaviorism (Putnam, Skinner, Chom-
sky), physicalism (Davidson, Fodor, Putnam, Kripke, Nagel), func-
tionalism (Armstrong, Nagel, Lewis, Putnam, Kim), mental
representations (Fodor, Dennett), imagery (Dennett, Fodor,
Kosslyn, Pylyshyn), linguistics (Stich, Chomsky, Fodor, Katz)
Block offers his own critique of functionalism and his own theory
of the mind.
The psychological state of a person can be identified with the
physical process that is taking place in the brain rather than
the state in which the brain is. The psychological state can be
represented as the operation performed on a machine, i.e. with
the computational state of the machine. The psychological state
does not depend on the physical state of the machine and can be
the same for different machines that are in different physical
states.
Qualias (sensations that are associated to the fact of being in a
given psychological state) are not easily explained in a func-
tionalist view. An organism whose functional states are identi-
cal to ours, but in which pain causes the sensation that we asso-
ciate to pleasure (inverted qualia), and an organism whose func-
tional states are identical to ours, but in which pain causes no
sensation (absent qualia). Functionalism cannot account for
either case.
Functionalism does not prescribe how we can limit the universe of
organisms who have mental states. A functionalist might think
that Bolivia's economy, as expertly manipulated by a financier,
has mental states. Class identity requires also identical inter-
nal processes, but this way it excludes beings that we might be
tempted to consider having mental states, such as an extraterres-
trial being who behaves like us but is made of different
material.
Block Ned: IMAGERY (MIT Press, 1981)
A collection of articles on mental imagery, including results of
psychological experiments and philosophical theories. The start-
ing point for the debate is that scientists have found no pic-
tures or images in the brain, no internal eye to view pictures
stored in memory and no means to manipulate them. Either (Fodor,
Kosslyn) the brain has mental pictures that somehow represent the
real-world images, or (Dennett, Pylyshyn) the brain represents
images through a non-imagistic system, namely language, i.e. all
mental representations are descriptional.
Bobrow Daniel: QUALITATIVE REASONING ABOUT PHYSICAL SYSTEMS (MIT
Press, 1985)
This is the first volume (a special issue of the Artificial
Intelligence Journal) that brought together the main names in the
then still young discipline of qualitative reasoning. They all
share the aim of explaining a physical system's behavior through
something closer to common sense than Physics' dynamic equations.
They conceive a physical system as made of parts that contribute
to the overall behavior through local interactions. They all
employ some variation of Hayes' measure space (a discrete
representation of a continous space that only deals with the sig-
nificant values that determine boundaries of behavior).
The main difference in the way they model a system is in their
ontologies: Kuipers adopts qualitative constraints among state
variables; DeKleer focuses on the devices (pipes, valves and
springs) connected in a network of constraintsl; Forbus deals
with processes by extending Hayes' notion of history. The system
behavior is almost always described by constraint propagation.
Johan DeKleer describes a phenomenon in a discrete measure space
through "qualitative differential equations", or "confluences".
An envisionment is the set of all possible future behaviors.
Ken Forbus defines a quantity space as a partially ordered set of
numbers. Common sense is interested in knowing that quantities
"increase" and "decrease" rather than on formulas yielding their
values in time.
Benjamin Kuipers formalizes the problem as a sequence of formal
descriptions; from the structural description derive the
behavioral description ("envisionment") and from this derive the
functional description. In his quantity space, besides the signs
of the derivatives, what matters most are critical or "landmarks"
values, such as the temperature at which water undergoes a phase
transition.
The other papers mainly cover practical applications.
Bobrow Daniel: ARTIFICIAL INTELLIGENCE IN PERSPECTIVE (MIT
Press, 1994)
An excellent selection of the articles (originally published by
the Journal of Artificial Intelligence) that made Artificial
Intelligence, from McCarthy's circumpscription to Moore's autoep-
istemic logic, from Newell's knowledge levels to Pearl's belief
networks, from DeKleers', Forbus' and Kuipers' qualitative
reasoning to Hayes-Roth's blackboard systems. With a chapter-
tribute to Newell.
McCarthy's circumpscription starts from the "close-world assump-
tion", that all relevant information is known (or, all informa-
tion that is not known can be considered false).
Boden Margaret: PHILOSOPHY OF ARTIFICIAL INTELLIGENCE (Oxford,
1990)
A collection of historical papers, starting with Warren
McCulloch's and Walter Pitts' "A logical calculus of the ideas
immanent in the nervous system" (1943).
In "Computing machinery and intelligence" (1950) Alan Turing pro-
posed his famous "Turing test" to prove whether a machine is
intelligent or not (a computer can be said to be intelligent if
its answers are indistinghishable from the answers of a human
being).
John Searle's "Mind, Brains and programs" (1980) summarizes his
view that computers are purely syntactic and therefore cannot be
said to be thinking. His famous thought experiment of the
"Chinese room" (a man who does not know how to speak chinese but
is provided by formal rules on how to build perfectly sensible
chinese answers would pass the Turing test, even if he will never
know what those questions and those answers were about) opened
the floodgates to the arguments that computation per se will
never lead to intelligence.
In the introduction Boden surveys the arguments pro and against
Turing's test and the possibility of thinking machines.
Drew McDermott's "A critique of pure reason" (1987) is a critique
specifically of Pat Hayes' attempt at formalizing common-sense
knowledge. Most of reasoning is not deductive and therefore can-
not be reduced to first-order predicate logic. McDermott proves
that all logistic approaches, in particular non-monotonic logics
as the one advocated by McCarthy (circumscription), yield very
weak solutions to the problem of representing knowledge in a
tractable way: one cannot write axioms independent of a program
for manipulating them if the inferences to be performed from them
are not deductions.
In "Motives, mechanisms and emotions" Aaron Sloman analyzes emo-
tions as states in which powerful motives respond to relevant
beliefs by triggering mechanisms required by resource-limited
systems. An autonomous system having many motives and finite
resources is prone to internal conflicts whose resolution
requires emotion-based mechanisms. Emotion is not a separate
subsystem of the mind, but a pervasive feature of it. Sloman
even proposes a generative grammar for emotions.
Boden Margaret: THE CREATIVE MIND (Basic, 1992)
An analysis of human creativity.
Bogdan Radu: GROUNDS FOR COGNITION (Lawrence Erlbaum, 1994)
Bogdan' teleo-evolutionary theory claims that cognitive systems
are guided by the environment in their goal-driven behavior.
Cognitive systems actually are the product of the evolutionary
pressure of guiding behaviors towards goals. Organisms are sys-
tems that are genetically programmed to maintain and replicate
themselves, therefore they must guide themselves to their goals,
therefore they need to obtain relevant information about their
environment, therefore they need to be cognitive. It makes evo-
lutionary sense that cognition should appear. Central to his
thinking is the concept of "goal-directedness", the result of
prebiological evolution which is constantly reshaped by natural
selection. Natural selection presupposes goal-directedness.
Goal-directedness arises from the genes themselves, which operate
goal-directedly.
Organisms manage to survive and multiply in a hostile world by
organizing themselves to achieve specific, limited goals in an
ecological niche. To pursue their goals, organisms evolve ways
to identify and track those goals. Such ways determine which
knowledge is necessary. To obtain such knowledge, organisms learn
to exploit pervasive and recurrent patterns of information in the
world. The information tasks necessary to manipulate such infor-
mation "select" the appropriate type of cognitive faculties that
the organism must be capable of.
Bond Alan & Gasser Leslie: READINGS IN DISTRIBUTED ARTIFICIAL
INTELLIGENCE (Morgan Kaufman, 1988)
A collection of articles, and a subject-indexed bibliography.
Distributed information processing systems, i.e. collection of
"intelligent" agents, embody a variety of strategies of decompo-
sition and coordination. Research in distributed A.I. focuses on
such methods, and on the forms of interaction that make such
methods effective.
Mike Georgeff discusses multi-agent planning. Barbara Hayes-
Roth's "A blackboard architecture for control" is included.
Frederick Hayes-Roth discusses ABE. Also articles by Victor
Lesser, Carl Hewitt, etc.
Gasser thinks, with Mead, that intelligent behavior is essen-
tially a social behavior and emphasizes the social aspects of the
interaction among intelligent agents.
Brachman Ronald: READINGS IN KNOWLEDGE REPRESENTATION (Morgan
Kaufman, 1985)
A collection of milestone essays on the topic of knowledge
representation from a semantic perspective and of knowledge
representation frameworks (mainly semantic networks and frames).
It includes Pat Hayes' "The logic of frames" (1979) and William
Woods' "What's in a link" (1975).
Hayes proves that the language of frames (with the exclusion of
stereotipical reasoning) can be reduced to a notational variant
of predicate logic. A frame is a micro-theory which allows very
rapid inferences. On the other hand, stereotipical reasoning of
default values goes against the monotonicity of classical logic.
Woods highlights that a semantic network confuses two types of
representation: assertions and definitions (a taxonomic relation
between concepts). A concept is equivalent to a first-order
predicate. As first-order predicate logic cannot handle the
intension of a concept, a semantic network must exhibit the same
limitation. Woods proposes to define a concept as a set of suffi-
cient and necessary conditions.
Ross Quillian's "Word concepts" (1967) originated the idea of a
semantic network of nodes interconnected with associative links.
Marvin Minsky's "A framework for representing knowledge" (1975)
presented the fundamental idea of a frame, a knowledge represen-
tation formalism based on prototypes, defaults, multiple perspec-
tives, analogies and partial matching.
James Allen's "Maintaining knowledge about temporal intervals"
(1983) claims that common sense's time is subject to a number of
principles, such as relativity (a date is usually specified rela-
tive to another date) and decomponibily (any event can be
described as a sequence of component events that take place in
the same interval). These principles state the preminence of the
"interval" of time (time as partial ordering of intervals) over
the "instant" of time (time as total ordering of instants).
Brady Michael & Berwick Robert: COMPUTATIONAL MODELS OF
DISCOURSE (MIT Press, 1983)
Bonnie Webber is looking for a formal language to represent
utterances. In addition, Candace Sidner also tries to track
discourse entities (especially, the focus) over the entire dura-
tion of discourse; that involves an understanding of how (defin-
ite) anaphoras work.
James Allen thinks that minds are connected to objects via the
causal connection between actions and objects, i.e. via beliefs
and desires. Allen is trying to marry Austin's and Searle's
theory of speech acts with Artificial Intelligence's theory of
planning by assuming that speech acts are just particular cases
of actions that, like all actions, must be planned. The speaker
that asks a question must have a plan of speech acts in mind and,
in order to answer appropriately, the other speaker must first
unravel that plan. Understanding the purpose of a question helps
understand indirect speech acts.
Brandon Robert: GENES ORGANISMS POPULATION (MIT Press, 1984)
A collection of seminal papers on the subject of the level at
which natural selection operates.
Evolutionary theory is based upon the idea that species evolve
and their evolution is driven by natural selection, but what
exactly evolves and what natural selection acts on is still not
clear. Nature is organized in a hierarchy: genes are located on
chromosomes, chromosomes are located in cells, cells make up
organs which make up organisms which make up species which make
up populations which make up ecosystems: at what level does
selection act?
Darwin's theory implies that what evolves is a population and
what selection acts on are the competing organisms of a genera-
tion within the population.
Alfred Russel Wallace thinks that selection acts on populations
as well as individuals. Wynne-Edwards (1963) thinks that selec-
tion acts on groups of organisms. Ernst Mayr (1975) thinks that
genes cannot be treated as separate, individual units, that their
interaction is not negligible. The units of evolution and natural
selection are not individual genes but groups of genes tied into
balance adapative systems. Natural selection favors phenotypes,
not genes or genotypes.
Lewontin thinks that all entities that exhibit heritable variance
in fitness (from prebiotic molecules to whole populations) are
units of selection.
William Wimsatt thinks that the notion of selection must be
grounded around the notion of "additive variance". This quantity
determines the rate of evolution. Variance in fitness is totally
additive when the fitness increase in a genotype is a linear
function of the number of genes of a given type present in it.
Additivity can be proven to be a special case of context-
independence. If variance in fitness at a given level is totally
additive, then this is the highest level at which selection
operates (the entities at that level are composed of units of
selection, and there are no higher-level units of selections).
Robert Brandon distinguishes levels of selection from units of
selection.
David Hull distinguishes replicators (units that reproduce their
structure directly, such as genes) from interactors (entities
that interact directly with their environment, such as organ-
isms). Differences in the interactions of interactors with their
environment result in differential reproduction of reproductors.
Hamilton (1975)'s kin-selection theory, and more general group-
selection theories are also introduced.
Brandon Robert: ADAPTATION AND ENVIRONMENT (Princeton Univ
Press, 1990)
Natural selection is defined as the process of differential
reproduction due to differential fitness to a common selective
environment. The "selective" environment (measured in terms of
the relative fitnesses of different genotypes across time or
space) is distinguished from the "external" environment and the
"ecological" environment (measured using the organism itself as
the measuring instrument so that only that part of the external
environment that affects the organism's contribution to popula-
tion growth is taken into account). The selective environment is
the one that is responsible for natural selection.
Following David Hull, Brandon generalizes phenotype and genotype
to "interactor" (Dawkins' "vehicle") and "replicator" and posits
that selection occurs among interactors. The biosphere is
hierarchically arranged and, in agreement with Lewontin, natural
selection applies to any level of the hierarchy. Selection
applies at different levels of the hierarchy of interactors.
Interactors can be lengths of RNA or species, or even replicators
(but even they behave as interactors when "naturally selected").
Brandon thinks that adaptation defines the function of a property
of the organism. The only process one needs to study to under-
stand the properties of a living organism are those that contri-
bute to adaptation.
Bresnan Joan: MENTAL REPRESENTATIONS OF GRAMMATICAL RELATIONS
(MIT Press, 1982)
A monumental work on grammars as mental representations, that led
to the definition of a lexical functional grammar. Half of the
chapters are by Bresnan in person.
Bresnan's lexical functional grammar posits the existence of an
intermediary functional level between syntactic structures and
semantic structures. Two levels of syntactic structures are pos-
tulated: constituent (a standard context-free surface parse of a
sentence) and functional (generated by equations associated with
the context-free rules). Transformations are avoided in favor of
a richer lexicon and links between nodes in the constituent and
functional structures.
Brillouin Leon: SCIENCE AND INFORMATION THEORY (Academic Press,
1962)
A seminal book on information theory, which employed the theory
of thermodynamics to formulate the "negentropy principle of
information".
A basic point is that information does not reside within the sys-
tem and is thus phenomenological.
Entropy (a measure of randomness in the state of the system)
measures the lack of information.
Information is defined as the amount of uncertainty which existed
before a choice was made. Information is thus the difference
between the entropy of the observed state of the system and its
maximum possible entropy.
Brillouin proved that the minimum entropy cost for obtaining one
bit of information is 10 to the -23 joules per degree K.
Broadbent Donald: PERCEPTION AND COMMUNICATION (Pergamon, 1958)
Broadbent is one of the psychologists who identified two types of
memory, a "short-term memory", limited to few pieces of informa-
tion, capable of retrieving them very quickly and decaying also
very quickly, and a "long-term memory", capable of large storage
and much slower in both retrieving and decaying. Broadbent thinks
that short-term memory is a set of pointers to blocks of informa-
tion located in the long-term memory.
Broadbent enunciated the principle of "limited capacity" to
explain how the brain can focus on one specific object out of the
thousands perceived by the retina. The selective character of
attention is due to the limited capacity of processing by the
brain, which can only be conscious of so many events at the same
time. Attention originates from a multitude of attentional func-
tions in different subsystems of the brain.
Broadbent's 1958 model of memory reflected well-known features of
memory: information about stimuli is temporarily retained but it
will fade unless attention is turned quickly to it. The unat-
tended information is "filtered out" without being analyzed. He
draws a distinction between a sensory store of virtually unlim-
ited capacity and a categorical short-term store of limited capa-
city. This is the way that a limited-capacity system such as
human memory can cope with the overwhelming amount of information
available in the world.
Broadbent proposes a block diagram which was similar to those
used by computer science, thereby approaching the first computa-
tional model of memory.
Broadbent Donald: DECISION AND STRESS (Academic Press, 1971)
In 1971 Broadbent modified his original information-flow model of
1958 by taking into account new physiological and psychological
findings. Foremost among the changes is that stimuli may be
selected by the attentional filter on the basis of semantic pro-
perties, besides their physical properties.
In 1984 Broadbent will also propose his "maltese cross" model
consisting of four stores (sensory, short-term, long-term and
motor output) with a central processing unit that controls the
flow of information among them.
Brooks Daniel & Wiley E.O.: EVOLUTION AS ENTROPY (Univ of Chi-
cago Press, 1986)
The goal of this unified theory of evolution is to integrate
Dollo's law (the irreversibility of biological evolution) with
natural selection. Natural selection per se only states an
environmental constraint, but no directionality in time. Dollo's
law is considered as a biological manifestation of the second law
of thermodynamics.
Unlike Prigogine, Wiley and Brooks believe that biological sys-
tems are inherently different from dissipative structures. Bio-
logical systems owe their order and organization to their genetic
information, which is inherent and inheritable. Both during
growth and during evolution entropy of biological information
constantly increases. Evolution is a particular case of the
second law of thermodynamics and biological order is a direct
consequence of it.
The creation of new species is made necessary by the second law
and is a "sudden" phenomenon similar to phase changes in Physics.
Phylogenetic branching is an inevitable increase in informational
entropy. The interaction between species and the environment is
not as important in molding evolution: natural selection mainly
acts as a pruning factor. Species are systems in a state of
non-equilibrium and new species are created according to the
second law.
Biological systems differ from physical dissipative systems in
that their order is based on properties that are inherent and
heritable. Their relevant phase space is genetic. The total
phylogeny is characterized by an ever increasing genetic phase
space. Dissipation in biological systems is not limited to
energy but also involves information. Information is transmitted
to subsequent generations.
Unlike most theories of information, that use information to
denote the degree to which external forces create structure
within a system, Brooks-Wiley's information resides within the
system and is material, it has a physical interpretation. Such
information resides in molecular structure as potential for
specifying homeostatic and ontogenetic processes. As the organism
absorbs energy from the environment, this potential is actualized
and is "converted" into structure. Over short time intervals bio-
logical systems behave like dissipative structures. Over longer
time intervals they behave like expanding phase space systems.
In concluding, by studying entropy in biological systems, Wiley
and Brooks propose a nonequilibrium approach to evolution.
Reproduction, ontogeny and phylogeny are examples of biological
organization that exhibit irreversible behavior. Biological sys-
tems are nonequilibrium systems.
Brooks Rodney & Luc Steels: THE ARTIFICIAL LIFE ROUTE TO ARTIFI-
CIAL INTELLIGENCE (Lawrence Erlbaum, 1995)
A collection of papers on the paradigm of situated cognition.
Brooks' 1991 papers, "Intelligence without representation" and
"Intelligence without reason", were instrumental in creating a
new, "situated" approach to cognition by emphasizing the interac-
tion between an agent and its environment.
Situated agents have no knwoledge. Their memory is not a locus of
representation but simply the place where behavior is generated.
In Brooks' subsumption architecture behavior is determined by the
structure of the environment. The cognitive system has no need to
represent the world, but only how to operate in the world. There
is no centralized function that coordinates the entire cognitive
system, but a number of distributed decisional centers that
operate in parallel, each of them performing a different task.
The system does not have the explicit representation of what it
is doing. It does have parallel processes that represent only
their very limited goal.
The system decomposes in layers of goal-driven behavior, each
layer being a network of finite-state automata, and incrementally
composes its behavior through the interaction with the world.
Brooks can therefore account for the response times required in
the real world. In the real world there is no clearcut difference
between perception, reasoning and action.
Brooks' tolemaic revolution in cognitive science turns the mind
into one of many agents that live in the environment. The
environment is the center of the action, not the mind.
The environment is action, continous action, continously chang-
ing. Only a system of separate, autonomous control systems could
possibly react and adapt to such a context.
The world contains all the information that the organism needs.
Therefore there is no need to represent it in the mind. The
environment acts like a memory external to the organism, from
which the organism can retrieve any kind of information through
perception.
"Intelligent" behavior can be partitioned into a set of asynchro-
nous tasks (eating, walking, etc), each endowed with a mechanism
of perception and action. An organism can be built incrementally
by gradually adding new tasks.
In other words, every intelligent being has a body!
Cognition is rational cinematics.
Brown Frank: THE FRAME PROBLEM (Morgan Kaufmann, 1987)
Proceedings of a workshop on the frame problem. Yoav Shoham
identifies a qualification problem and an extended prediction
problem that subsume the frame problem. Frank Brown presents a
modal logic approach. Matthew Ginsberg's "Reasoning About Action"
offers a solution based on the search for the nearest possible
world to the current one.
Bruner Jerome: A STUDY OF THINKING (Wiley, 1956)
A book that helped launch the cognitive revolution in Psychology
and Philosophy. Bruner concentrates on how human beings categor-
ize objects. All cognitive activity depends upon the process of
categorizing events. A category is a set of events that can be
treated as if they were equivalent. Bruner employs techniques of
game theory and communication theory to explain how the environ-
ment is partitioned into equivalence classes. Concept forma-
tion, or "attainment", is achieved via a number of selection
(choice of instances) and reception (revision of hypothesis)
strategies. In general, though, subjects categorize with proba-
bilistic cues.
Bruner Jerome: ACTS OF MEANING (Harvard University Press, 1994)
A manifesto of methodology from the man who set up the first
Center for Cognitive Studies (in Cambridge, MA, in the Sixties)
proposes a "cultural psychology" that is centered on meaning, not
information, and on the construction of meaning by the mind, not
on the processing of information by the mind. To understand
humans one must understand how their experiences are shaped by
their intentional states. The form of these intentional states
depend upon the symbolic systems of their culture. Biological
inheritance merely imposes constraints on action. Culture enables
humans to transcend those biological limits. Folk psychology is
but a device for people to organize their views of themselves,
the others and the world they share with them. Folk psychology
is not grounded on a logical system, but on narratives. Narra-
tive skills arise somehow from a biological need to narrate. Even
selves must be viewed in the context of culture and society: a
self is distributed interpersonally.
Buchler Justus: METAPHYSICS OF NATURAL COMPLEXES (Columbia
University Press, 1966)
A general discussion of complexity from a philosophical point of
view. The world is unlimitedly complex and complexity if the
result of multiple relatedness among processes. Buchler adopts
an ontology of processes instead of things.
Buck Ross: THE COMMUNICATION OF EMOTION (Guilford Press, 1984)
HUman behavior is a function of several systems of organization:
innate special-purpose processing systems (reflexes, instincts,
etc) concerned with bodily adaptation and the maintenance of
homeostasis and that employ a holistic, syncretic type of cogni-
tion (knowledge by acquaitance); and acquired general-purpose
processing systems, concerned with making sense of the environ-
ment and that employ sequential, analytic cognition (knoweldge by
description). The former (associated with the right emisphere)
carry out spontaneous communication involving emotional expres-
sion, the latter (associated with the left emisphere) carry out
symbolic communication involving propositions. The former is
primitive, the latter also requires the former, and may be based
upon it both phylogenetically and ontogenetically.
Buck grounds his model of communication of emotions on Shannon-
Weaver's theory of communication and assumes that such
communication occurs via two parallel streams, one spontaneous
(emotions) and one symbolic (propositions).
Communication occurs when the behavior of an individual influ-
ences the behavior of another individual. Communication of emo-
tions, in particular, is a biologically shared signal system that
has been created through an evolutionary process.
Emotion is defined as a readout of motivational systems. Buck
identifies three functions of emotions: bodily adaptation to the
environment, social communication with other aware beings and
subjective experience. All originate from motives that must be
satisfied. The emotion is a measure of how far they have been
satisfied.
Buck provides both a general cognitive model of emotions and a a
detailed physical model of their neural processes.
Buck thinks that behavior is governed by biological, or innate,
epistemic, or acquired, and rational, or processed, factors.
Bundy Alan: THE COMPUTER MODELLING OF MATHEMATICAL REASONING
(Academic Press, 1983)
Bundy introduces the notation of propositional logic and predi-
cate logic, higher-order logics and lambda calculus. Then
explains how a computer can perform automatic theorem proving by
using resolution, along the way defining Horn clauses, Kowalski
form and Skolem normal form. The book also touches on Douglas
Lenat's concept formation and Daniel Bobrow's theory formation.
Bunge Mario: TREATISE ON BASIC PHILOSOPHY (Reidel, 1974-83)
A monumental seven-volume synthesis of modern philosophical
themes.
Volume one deals with sense and reference. Reference is non
equated to extension. Intension is an irreducible semantic
object. The sense of a construct is relative to the theory in
which it occurs (sense depends on the context).
Volume two deals with interpretation and truth. Meaning is sense
together with reference. Meaning is not verifiability, truth con-
ditions, information, etc. Bunge develops a calculus of meaning.
A truth measure function (a continous function) allows for the
expression of partial truth, or degrees of truth.
Volume three and four deal with ontology (substance, properties,
change, spacetime). Reality is the aggregation of things holding
spatiotemporal relations: spacetime can be understood only in
terms of changing things. Spacetime must be anchored to things,
not the other way around. A system is identified by three com-
ponents: its composition, environment and structure. The universe
is a system composed of subsystems. Everything is a system or a
system component.
Organisms are particular systems with emergent properties. The
unit of biological study is the organism-in-the-environment
together with its subsystems (from cells to organs) and its
supersystems (from population to biosphere). The mind is a col-
lection of processes of neural systems. Society is a system made
of people linked by social relations.
Volume five and six deal with epistemology. Every cognitive
activity is a neural process. Language is for transmitting
knowledge and influencing behavior. Perception yields a subjec-
tive type of knowledge. Conceptualizing yields objective
knowledge. Perception is like copying reality to the brain. Con-
ceptualizing goes beyond mere copying: it can form new proposi-
tions out of nonpropositional knowledge (percepts) or it can form
new propositions out of old propositions (inferring). Inference
yields new propositions, not new concepts.
Bunt Harry: MASS-TERMS AND MODEL-THEORETIC SEMANTICS (Cambridge
Univ Press, 1985)
The book deals with the semantic problems related to mass nouns
(such as "water", "music", "luggage", etc), as opposed to count
nouns. The semantics for mass terms is built on ensemble theory
(an extension of mereology built around the concept "part of").
Buss David: THE EVOLUTION OF DESIRE (Basic, 1994)
Research on sexual behavior reveals a distinct gender gap.
Natural selection has molded the brains of men and women in very
different ways as a result of their different reproductive goals.
Cairns-Smith A. G.: GENETIC TAKEOVER (Cambridge University
Press, 1982)
Cairns-Smith argues that the first living beings were not carbon
composts but metallic crystals, i.e. minerals. Life's ancestors
were self-replicating patterns of defects in metallic crystals.
One day those patterns started replicating in a different sub-
stance, carbon molecules.
Cairns-Smith A. G.: EVOLVING THE MIND (Cambridge University
Press, 1995)
The author reviews theories of consciousness and is skeptical
about the possibility of deriving consciousness from matter.
Calvin Melvin: CHEMICAL EVOLUTION (Clarendon, 1969)
Calvin explores different autocatalytic scenarios for the origin
of life which assume life spontaneously bootstrapped itself from
simple molecules and don't require any unlikely event to produce
very complex molecules.
Calvin William: THE ASCENT OF MIND (Bantam, 1991)
Calvin looks for the causes of the evolution of the human brain
in ice-age climates.
The brain got bigger and bigger through a three-part cycle of
evolutionary alterations in body proportions which involves a set
of genes that regulate fetal and childhood growth.
Calvin William: THE CEREBRAL CODE (MIT Press, 1996)
Campbell John: PAST, SPACE AND SELF (MIT Press, 1994)
Campbell examines how human thinking about space and time differs
from animals' thinking about space and time (in particular the
ability to think about the past). Campbell then examines the
consequences on self-consciousness.
Carbonell Jaime: MACHINE LEARNING (MIT Press, 1989)
Contains nine articles from prominent researchers in the area of
machine learning. Carbonell's introduction compares the tradi-
tional inductive paradigm (constructing the symbolic description
of a concept from a set of positive and negative instances) with
the new analytic (i.e., deductive) paradigms. The latter utilize
past problem solving experience to formulate the search strategy
in the space of potential solutions. Deductive learning systems
include: Jerry DeJong's "explanation-based learning", Allen
Newell's "chunking", and Carbonell's own "derivational analogy".
Pat Langley and others cover concept formation, reviewing histor-
ical systems such as Langley's own BACON, Doug Lenat's AM, Ed
Feigenbaum's EPAM, Michael Lebowitz's UNIMEM, Doug Fisher's
COBWEB, Jan Zytkow's FAHRENHEIT.
An explanation-based learning system is given a high-level
description of the target concept, a single positive instance of
the concept, a description of what a concept definition is and
domain knowledge. The system generates a proof that the positive
instance satisfies the target concept and then generalizes the
proof. Richard Fikes' STRIPS is recognized as a forerunner of
explanation-based learning.
Derivational analogy solves a problem by tweaking a plan
(represented as a hierarchical goal structure) used to solve a
previous problem. Jack Mostow surveys a few applications.
John Holland and Geoffrey Hinton touch briefly on two alternative
and extreme paradigm, respectively genetic algorithms and connec-
tionism.
Holland's "Classifier systems and genetic algorithms" provides
his definitive version of classifier systems.
Classifier systems are defined as "massively parallel, message-
passing, rule-based systems that learn through credit assignment
(the bucket brigade algorithm) and rule discovery (the genetic
algorithm)". When a message from the environment matches the
antecedent of a rule, the message specified in the consequent of
the rule is produced. Some messages produced by the rules cycle
back into the classifier system, some generate action on the
environment. A message is a string of characters from a specified
alphabet. The rules are not written in the first-order predicate
logic of expert systems, but in a language that lacks descriptive
power and is limited to simple conjunctive expressions.
Credit assignment is the process whereby the system evaluates the
effectiveness of its rules. The bucket brigade algorithm assigns
a strength (a maesure of its past usefulness) to each rule. Each
rule then makes a bid (proportional to its strength and to its
relevance to the current situation) and only the highest bidding
rules are allowed to pass their messages on. The strengths of the
rules are modified according to an economic analogy: every time a
rule bids, its strength is reduced of the value of the bid while
the strength of its "suppliers" (the rules that sent the mes-
sages matched by this bidder) are increased. The bidder strength
will in turn increase if its consumers (the rules that receive
its message) will become bidders. This leads to a chain of
suppliers/consumers whose success ultimately depends on the suc-
cess of the rules that act directly on the environment.
Then the system replaces the least useful (weak) rules with newly
generated rules that are based on the system's accumulated
experience, i.e. by combining selected "building blocks"
("strong" rules) according to some genetic algorithms.
Hinton focuses on gradient-descent learning procedures of connec-
tionist systems. Each connection computes the derivative, with
respect with its strength, of a global measure of error in the
performance of the network, and then adjusts its strength in the
direction that decreases the error. Hinton is interested in
learning procedures that lead to internal representations of the
environment. His survey starts with associative memories without
hidden units (linear and nonlinear associators) and supervised
networks without hidden units (least squares and perceptron con-
vergence algorithms) and proves the deficiencies of such
approaches. Backpropagation (a multi-layer least squares algo-
rithm) can instead lead to the discovery of semantic features,
but it too exhibits limitations, specifically computational
intractability and biological implausibility.
Hinton also surveys Boltzmann machines (where units update their
state based on astochastic decision rule), Hebbian learning
(where weight modification depends on both presynaptic and post-
synaptic activity), competitive learning (where units in a hidden
layer compete to become active) and reinforcement learning (where
credit is assigned to a local decision by measuring how it corre-
lates with the global reinforcement signal.
John Anderson's "A theory of origins of human knowledge" general-
izes the results of his systems, in particular ACT and his latest
PULPS. They organize knowledge in three levels: knowledge level
(information acquired from the environment and innate principles
of inference), algorithm level (internal deductions, inductions
and compilations) and implementation level (setting strengths for
the encoding of specific pieces of information).
Carvalo Marc: NATURE, COGNITION AND SYSTEM (Kluwer Academic,
1988)
A collection of articles on cybernetics applied to the nature of
living systems, autopoiesis and self-organization. One of the
main themes is that of the "two arrows of time": the second law
of thermodynamics pointing towards entropy increase and therefore
disorder increase, and evolution pointing the other way by build-
ing increasingly complex structures of order.
Castaneda Hector-Neri: THINKING, LANGUAGE, EXPERIENCE (Univer-
sity of Minnesota Press, 1989)
The book advances a general semantics of thinking that accounts
for the unity of experience: "guise theory". According to its
ontological scheme, properties are the building blocks of the
world.
Singular reference (reference to individuals insofar as they are
thought of individuals) is achieved through a combination of one
of four linguistic mechanisms: indexical reference (required for
a person to have experience), quasi-indexical reference (required
to conceive of other subjects with experience), descriptive
reference and reference by proper names. We refer to ourselves
and to objects indexically.
Believing and intending partition the class of mental states in
two categories, corresponding to contemplative thinking ("propo-
sitions") and practical thinking ("practitions").
Proper names are not individuating devices (they are not genuine
singular terms, they are free variables of quantification).
Proper names have an epistemic role (they are useful to organize
beliefs) and a causal role (they allow the retrieval of informa-
tion).
The individuality of an individual consists in the set of that
individual's differences from everything else, the set of dif-
ferentiating properties. The units of individuation are
"guises".
Castaneda emphasized the fundamental indexality of practical
thinking (exercized in acts of willing, commanding, advising,
etc). Indexical reference is the backbone of perceptual refer-
ence. Indexical reference is experiential reference. Therefore, a
theory of indexical reference (and a semantics of indicators)
depends on a theory of perception.
In order to deal with indexicals and demonstratives, one must
appreciate the difference between sense and meaning: the word "I"
has the same meaning, no matter who utters it, but different
senses, and different references.
Guise theory is a theory of predication. Properties are the ulti-
mate components of the world. Concrete objects (or "guises") are
bundles of properties. A concret object is made of the members of
a set of properties plus an operator: the operator (sort of the
inverse of the abstraction operator) is what turns the properties
into a concret object. For each distinct set of properties there
is a distinct concrete object that results from the application
of the operator on that set. Therefore, "the thing that doesn't
exist" is a concrete object, because it is made of a bundle of
properties.
When assertions of ordinary discourse are made explicit, proper-
ties turn out to be predicated of the guises which constitute the
domain. They are predicated either internally (if the property
belongs to the core of a guise which is the subject of predica-
tion) or externally. In other words, the disguised predications
of ordinary discourse are, when made explicit, either internally
or externally "guised" depending upon the form of reference to
the subject of predication.
An object can stand in a number of relationships to a property:
constitution (the property is a member of the core of the
object), identity, consubstantiation, consociation, conflation.
Chalmers David: THE CONSCIOUS MIND (Oxford University Press,
1996)
Chalmers argues that consciousness cannot be explained with a
reductionist approach, because it does not belong to the realm of
matter. Chalmers proposes to expand science in a fashion that is
still compatible with today's science (in the areas where it is
successful) and that allows for a dualist approach.
Chalmers distinguishes between a phenomenal concept of mind (the
way it feels) and a psychological concept of mind (what it does).
Every mental property is either a phenomenal property, a psycho-
logical one or a combination of the two. The mind-body problem
is therefore made of two parts, one that deals with the mental
faculties and one that deals with how/why those mental faculties
also give rise to awareness of them (Jackendoff's "mind-mind
problem"). The same distinction applies to consciousness, with
psychological consciousness being commonly referred to as "aware-
ness" (phenomenal consciousness always comes with psychological
consciousness). Awareness is having access to information that
may affect behavior.
Chalmers shuns the problem of identity and prefers to focus on
the notion of supervenience. Consciousness supervenes on the phy-
sical, just like biological properties supervene on physical pro-
perties (any two situations that are physically identical are
also biologically identical). Chalmers defines logical superveni-
ence (to be interpreted loosely as "possibility", and as logi-
cally possible worlds that supervene on the physical world) and
natural supervenience (to be interpreted as a real empirical pos-
sibility, when two sets of properties are systematically and pre-
cisely correlated in the natural world). Logically possible
situations are not necessarily also naturally possible situations
(e.g., any situation that violates the laws of nature). Logical
supervenience implies natural supervenience, but not viceversa.
A natural phenomenon can be reduced to a set of lower-level pro-
perties when it is logically supervenient on those properties,
i.e. it can be reduced to the physical when it is logically
supervenient on the physical. From his analysis it turns out
that "almost everything" is logically supervenient on the physi-
cal.
Using arguments about zombies, inverted spectrum, epistemic asym-
metry with respect to consciousness and Jackson's thought experi-
ment of the blind neurologist, Chalmers then proves that cons-
ciousness is naturally, but not logically, supervenient on physi-
cal properties. That means that it cannot be reduced to the phy-
sical. Chalmers therefore criticizes cognitive architectures
(such as Dennett's), neurobiological theories (such as Edelman's)
and hypotheses based on quantum mechanics (such as Penrose's).
Chalmers' "naturalistic monism" admits both physical and non-
physical features in the world. His dualism is different from
Descartes' in that it claims that "consciousness is a feature of
the world" that is somehow related to its physical properties. A
new, fundamental, irreducible feature (a set of "protophenomenal"
properties) must be added to space-time, mass, charge, spin, etc,
and a set of "psychophysical" laws (explaining how phenomenal
properties depend on physical properties) must be added to the
laws of nature.
Consciousness is viewed as "organizationally invariant", i.e.
every system organized in the appropriate way will experience the
same conscious states, regardless of what substance it is made
of. In this sense, a computer can be intelligent and conscious.
In order to build a scientific theory of consciousness, Chalmers
outlines a few candidate psychophysical laws, such as the princi-
ple of coherence between consciousness and cognition and the
principle of organizational invariance.
Still looking for fundamental laws of consciousness, Chalmer
offers an interpretation of his theory based on the dualism
between information and pattern: information is what pattern is
from the inside. Consciousness is information about the pattern
of the self. Information becomes therefore the link between the
physical and the conscious.
Chalmers also offers his own interpretation of quantum theory:
Changeux JeanPierre: NEURONAL MAN (Pantheon, 1985)
Changeux is one of the brain scientists who maintain that the
mental and the neural are simply two aspects of the same physical
state.
From neuroanatomy Changeux derives a view of the complexity of
the brain: the evidence for specific localization of particular
functions always comes with evidence for diffuse control and
interaction of parts.
The human brain is priviliged by the (relatively recent) develop-
ment of the neocortex. The human brain contains representations
of the world in its cortex, is capable of building new mental
representations and is capable of using them for computations. A
mental object corresponds to the activity of a population of neu-
rons.
Changeaus notes that at the level of communication between nerves
nothing distinguishes the brain from the peripheral nervous sys-
tem, or, for that matter, from any other animal.
Changeux proposes a "neo-darwinian" theory for the development of
the set of nerve connections that underlie memories and percep-
tions. The nervous system makes very large numbers of random mul-
tiple connections. External stimuli cause differential elimina-
tion of some connections. Phenotypic variability is the result of
experience.
His theory of "epigenesis by selective stabilization of synapses"
stems from a number of observations: the main organizational
features of the nervous system are determined by a set of genes;
phenotypic variability increases in organisms with the increase
in brain complexity; during development connections are created
and destroyed in large numbers; neurons communicate even at very
early stages of development.
The theory explains the nonlinearity between the complexity of
the genome and that of brain complexity. The evolutionary advan-
tage of the human species stems from the individual, epigenetic
variability in the organization of neurons, which resulted in
greater plasticity in adapting to the environment.
Changeux JeanPierre: ORIGINS OF THE HUMAN BRAIN (Oxford Univer-
sity Press, 1995)
A collection of essays from neurobiologists, anthropologists and
psycholigists, covering the anatomy of the brain, genetics, and
consciousness/mind.
Charniak Eugene: ARTIFICIAL INTELLIGENCE PROGRAMMING (Lawrence
Erlbaum, 1987)
The second edition of a classic textbook of practical Artificial
Inteligence techniques (very LISP-oriented).
Chauvin Yves & Rumelhart David: BACKPROPAGATION (Lawrence Erl-
baum, 1995)
Theory and practice of the most popular training algorithm for
neural networks.
Chierchia Gennaro: DYNAMICS OF MEANING (Univ of Chicago Press,
1995)
A few linguistic phenomena constitute evidence in favor of a view
of meaning as "context change", as opposed to the traditional
view of meaning as content. Context updating would be an
integral part of the compositional system of meaning.
Chierchia proposes a "dynamic binding" theory (based on
Montague's intensional logic) as an alternative to classical
"discourse representation theory".
Chierchia Gennaro: MEANING AND GRAMMAR (MIT, 1990)
A seminal textbook on semantics.
The empirical domain of semantics is defined according to the
linguistic phenomena that a semantic theory is required to
account for: entailment (an implication both in terms of truth
and information that is conveyed), presupposition (an implication
which does not depend on the truth of the premise because the
truth of the conclusion is implied in the wording itself of the
premise), anaphora (expressions that are connected to previous
expressions), ambiguity (lexical, syntactic and scope ambiguity),
synonymy (mutual entailment of two expressions), contradiction (a
sentence that can never be true because of incompatible entail-
ments), anomaly (a sentence that can never be true because of
incompatible presuppositions), appropriateness (in the context).
Theories of meaning include referential or denotational theories
(meaning lies in the relations of symbols to what they stand
for), psychologistic or mentalistic theories (meaning lies in
their mental representation), social or pragmatic theories (mean-
ing lies in the social interaction of agents), but all aspects
should contribute to a complete theory of meaning.
Problems with denotation (especially Frege's take on reference
and sense) and truth (Tarski's correspondence theory) are intro-
duced. Kripke's and Putnam's causal theory of reference (which
assumes a causal link between a word and what it stands for) is
sketched.
Chapters are devoted to: how to derive truth conditions of sen-
tences containing quantified expressions; the relation between
the meaning of an expression and the meaning of the speaker (as
in Grice); speech acts (as in Austin and Searle); intensionality
(as in Montague); discoursse analysis (indexicals, contexts,
filters, ...); Lambda abstraction; lexical semantics (including
thematic roles).
Child William: CAUSALITY, INTERPRETATION AND THE MIND (Oxford
University Press, 1994)
The nature of intentional phenomena, such as belief and desire,
in a causal theory of the mind.
Chomsky Noam: SYNTACTIC STRUCTURES (Mouton, 1957)
With this book Chomsky striked a fatal blow at the behaviorist
tradition of Skinner and others that research should be focused
solely on external, measurable stimuli and responses, and not to
abstract mental entities. At the same time Chomsky reacted to
structural linguistics that was content with describing and clas-
sifying languages. Chomsky extended the idea of formal systems
to linguistics by using the logical formalism to express the
grammar of a language.
Chomsky's idea was to concentrate on the study of grammar, and
specifically syntax, i.e. on the rules that account for all valid
sentences of a language. The idea was that language is based on
a system of rules determining the interpretation of its infin-
itely many sentences.
Chomsky argued for the independence of syntax from semantics, as
the notion of a well-formed sentence in the language is distinct
from the notion of a meaningful sentence.
The phrase structure model, based on immediate constituent
analysis, is a more powerful tool for the purpose of grammar than
other existing tools, but not adequate enough. A grammar needs
to have a tripartite structure: a sequence of rules to generate
phrase structure, a sequence of morphophonemic rules to convert
strings of morphemes into strings of phonemes, and a sequence of
transformational rules that transform strings with phrase struc-
ture into new strings to which the morphophonemic rules can
apply.
Chomsky proposed a hierarchy that categorizes languages according
to the complexity of the grammars that generate them. The sim-
plest languages are regular languages, or type-3; type-2
languages are context free; type-1 are context-sensitive; and
type-0 are recursively enumerable languages. The definitions are
based on the type of rules needed to generate all the sentences
of the language.
Chomsky posited the existence of two levels of language: an
underlying deep structure, which accounts for the fundamental
syntactic relationships among language components, and a surface
structure, which accounts for the sentences that are actually
uttered, and which is generated by transformations of elements in
the deep structure.
A generative grammar is a rules system that generates the gram-
matical sentences of the language that it describes and assigns
to each sentence a grammatical analysis. The simplest type of
generative grammar is the finite-state grammar, but no natural
language is finite. In a phrase structure grammar the elements of
the sentences are identified by constituents (noun phrase, verb
phrase, etc). In a transformational generative grammar the
phrase structure (which produces the "deep structure" of a
sentence) is supplemented by a transformational component and a
morphophonemic component (which transform the deep structure into
the surface structure of the sentence, e.g. active or passive
form).
Chomsky's computational approach had its flaws. Each Chomsky
grammar is equivalent to a Turing machine. From Godel's theorem,
the processing of a Turing machine may never come to an end.
Therefore a grammar may never find the meaning of a valid sen-
tence, although we have no evidence that our brain may never find
the meaning of a valid sentence in our language. Later, Gold
proved that no amount of correct examples of sentences are enough
to learn a language.
The book was one of the milestones of cognitive science.
Chomsky's formal method was influenced by mathematical logic
(particularly formal systems) and the computer model (the
information-processing paradigm).
Chomsky Noam: ASPECTS OF THE THEORY OF SYNTAX (MIT Press, 1965)
In order to explain the difference between "performance" (all
sentences that an individual will ever use) and "competence" (all
sentences that an individual can utter, but will not necessarily
utter), Chomsky posits the existence of some innate knowledge.
Chomsky proved that the grammar of a natural language cannot be
reduced to a finite-state automaton. Later, Gold proved that no
amount of correct examples of sentences are enough to learn a
language.
Chomsky argues for the existence of two levels of language: an
underlying deep structure, which accounts for the fundamental
syntactic relationships among language components, and a surface
structure, which accounts for the sentences that are actually
uttered, and which is generated by transformations of elements in
the deep structure. Transformational analysis does overcome the
limitations of phrase structure.
Chomsky's "standard theory" defines a grammar as made of a syn-
tactic component (phrase structure rules, lexicon and transforma-
tional component), a semantic component and a phonological com-
ponent. The lexicon is modeled after Katz's lexicon. Context-
sensitive rules determine the legal positions in the sentence of
lexical items. The semantic component is also inspired by Katz,
as it uses projection rules and semantic markers.
The deep structure of a sentence is a tree (the phrase marker)
that contains all the words that will appear in its surface
structure.
Chomsky starts coupling syntax and semantics when including an
account of the relation between sound and meaning in the con-
struction of a grammar. The "standard theory" syntax provides
the mechanisms for transforming a meaning (a deep structure) into
a phonetic representation (a surface structure).
Chomsky decomposes a user's knowledge of language into two com-
ponents: a universal compenent (universal grammar), which is the
knowledge of language possessed by every human, and a set of
parameter values and a lexicon, which together constitute the
knowledge of a particular language.
Chomsky Noam & Halle Morris: THE SOUND PATTERN OF ENGLISH
(Harper & Row, 1968)
A classical textbook on generative phonology. Besides detailing
the formal structure of a phonological theory, the book tried to
define a way in the formal expressions of these processes that
would predict which phonological processes were likely and which
were not. An evaluation metric ranks rules according to how
likely they are to occur (inversely proportional to the number of
features needed to express it).
Chomsky Noam: REFLECTIONS ON LANGUAGE (Pantheon, 1975)
Chomsky's standard theory assumed that each sentence exhibits a
surface and a deep structure. Many sentences may exhibit the same
deep structure (e.g., active and passive forms of the same
action). Understanding language consists in transforming surface
structures into deep structures.
These transformations can be seen as corresponding to mental
processes, performed by mental modules, each independent of the
others and each guided by elementary principles.
Fundamental to his theory is the belief that there exist "innate
structures", that the ability to understand and utter language is
due to a "universal grammar" that is common to all humans and is
somewhow encoded in the human genome. Then experience "selects"
the specific grammar that the individual will learn. Grammar is
learned not in stages, as Piaget thought, but simply by gradually
fulfilling a blueprint that is already in the mind.
Children do not learn, as they do not make any effort. Language
"happens" to a child. The child is almost unaware of the language
acquisition process. Learning to speak is not different from
growing, maturing and all the other biological processes that
occur in a child. A child is genetically programmed to learn a
language, and experience will simply determine which one. The
way a child is programmed is such that all children will learn
language the same way.
Chomsky also notes how the language spoken by a linguistic com-
munity is so identical to the smallest detail, even if no indivi-
dual of the community has been exposed to all the smallest
details.
The universal grammar is the linguistic genotype. Its principles
are invariant for all languages. The values of some parameters
can be "selected" by the environment out of all valid values.
This process is similar to what happens with other growth
processes (e.g., with the immune system).
Chomsky Noam: THE LOGICAL STRUCTURE OF LINGUISTIC THEORY
(University of Chicago Press, 1975)
A detailed, technical exposition of Chomsky's early linguistic
theory.
Chomsky Noam: RULES AND REPRESENTATIONS (Columbia Univ Press,
1980)
Chomsky defends (on philosophical and psychological grounds) his
position that grammars are internally represented in the mind and
that an initial state of knowledge is shared by all individuals
and then developed by social and cultural interactions.
Chomsky Noam: THEORY OF GOVERNMENT AND BINDING (MIT Press, 1982)
Chomsky Noam: LECTURES ON GOVERNMENT AND BINDING (Foris, 1981)
Chomsky's hypothesis is that sound and meaning are mediated by
syntactic representations. A universal grammar, an innate pro-
perty of the human mind, defines what is a possible grammar, and
therefore a possible language.
The government-binding theory puts constraints on which features
can occur in the same rule, so that grammatical information is
modularized and localized (e.g., the "projection principle"
states that lexical properties must be satisfied in the same way
at all levels of syntactic representation). This process of
constraining minimizes the effort required to learn a grammar (it
limits possible rule applications).
Universal principles of grammar limit language-specific options
to a (small) set of "parameters".
The lexicon is the repository of lexical information that cannot
be predicted from the universal principlesor from choices of
parameters.
The final level of syntactic derivation, that of "logical form",
must meet the "theta criterion" (every theta role must be
uniquely assigned).
Every sentence has a quadruple structure: the D-structure is gen-
erated by phrase-structure rules, the S-structure is obtained
from the D-structure by applying transformational rules, the P-
structure (a phonetical structure) and a logical form (a semantic
component, a first-order logical representation of the semantics
of a natural-language sentence).
An anaphor is bound in its local domain. A pronominal is free in
its local domain. An r-expression (non-anaphoric and non-
pronominal) is free.
His 1970 X-bar theory eliminated the distinction between features
and categories, and reduced every expression to a set of
features. This way verbs and nouns (e.g., "destroy" and "destruc-
tion") are related in virtue of the features they share. The X-
bar theory was made possible by the separation of the lexicon
from the phrase structure rules (i.e., from the computation).
The projection principle, the theta theory and the X-bar theory
compose the structure-building tools of the theory of government
and binding.
A universal grammar should include a number of interacting sub-
systems to deal with specific problems, such as the relations of
anaphors to their precedents (theory of binding) and the rela-
tions between the head of a construction and categories dependent
on it (theory of government). Other subsystems involve determin-
ing thematic roles, assigning abstract cases, posing locality