Transcript No Slide Title

Michael Arbib: CS564 - Brain Theory and Artificial Intelligence
University of Southern California, Fall 2001
Lecture 17.
From the FARS model to the Evolution of Language
Reading Assignment:
Arbib,
M.A., 2001, The Mirror System, Imitation, and the Evolution of
Language, in Imitation in Animals and Artifacts, (Chrystopher
Nehaniv and Kerstin Dautenhahn, Editors), The MIT Press, to
appear.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
1
Two stages in the Evolution of Human Language
 Biological Evolution: Yielding a language-ready brain:
Language-readiness: the brain capacity needed to acquire and use
language.

 Cultural Evolution: From hominids with a language-ready
brain and rudimentary manual-vocal communication to humans
with full language capability

Social organization of the developing brain.
 Stressing the rich historical processes whereby groups of language
arose and “cross-pollinated”.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
2
5 million years of hominid evolution




Adapted from Clive Gamble:
Timewalkers Figure 4.6
What were the biological changes supporting language-readiness?
What were the cultural changes extending the utility of language as a socially
transmitted vehicle for communication and representation?
How did biological and cultural change interact “in a spiral” prior to the
emergence of Homo sapiens?
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
3
Deep Time
The divergence of the Romance languages took about one thousand
years.
The divergence of the Indo-European languages with their immense
diversity

Hindi, German, Italian, English, ...
took about 6,000 years.
How can we imagine what has changed since the emergence of Homo
sapiens some 200,000 years ago?
Or in 5,000,000 years of prior hominid evolution?
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
4
Broca’s and Wernicke’s Aphasias
Warning: Localization of Aphasias is HIGHLY Variable
Wernicke’s original drawing
(wrong hemisphere!)
“Perception”
Wernicke:
“Perception”
“Production”
Broca:
“Production”
MRI-scans from
Keith A. Johnson, M.D. and J.
Alex Becker
The Whole Brain Atlas
Broca’s Area (Negative Image)
http://www.med.harvard.edu./
AANLIB/home.html
Slice viewed from below:
So “right” is left
Wernicke’s Area
An Observation/Execution Matching System in
Humans
Rizzolatti, Fadiga, Matelli, Bettinardi, Perani, and Fazio:
Broca's region is activated by observation of hand gestures: a PET study.
PET study of human brain with 3 experimental
conditions:
Object observation (control condition)
A key language area!!!
 Grasping observation
 Object prehension.
The most striking result was highly significant activation in the
rostral part of Broca's area.
Another PET data, by Petrides et al., showed that during
execution of a sequences of self-ordered hand movements there
was a highly significant activation of Broca's area.

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
6
Neural Substrate of Vocalization
The neural substrate for primate calls is in a region of cingulate cortex
distinct from F5, the monkey homologue of human Broca's area
For most humans language is heavily intertwined with speech
Why is F5, rather than the cingulate area already involved in monkey
vocalization, homologous to the Broca's area's substrate for language?
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
7
A New Approach to the Evolution of Human Language
Rizzolatti, G., Fadiga L., Gallese, V., and Fogassi, L., 1996,
Premotor cortex and the recognition of motor actions.
Cogn Brain Res., 3: 131-141.
Rizzolatti, G, and Arbib, M.A., 1998, Language Within Our Grasp,
Trends in Neuroscience, 21(5):188-194:
The Mirror System Hypothesis: Human Broca’s area
contains a mirror system for grasping which is homologous
to the F5 mirror system of monkey, and this provides the
evolutionary basis for language parity - i.e., an utterance
means roughly the same for both speaker and hearer.
This adds a neural “missing link” to the tradition that roots speech
in a prior system for communication based on manual gesture.
[See most recently: William C. Stokoe (2001) Language in
Hand: Why Sign Came Before Speech.]
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
8
Monkey
and
Human
My goal: A fully articulated model of the monkey mirror system
(grounded in neurophysiology of macaque [and other?] monkeys; a
cooperative computation model of interacting brain regions for
human neurolinguistics as well as human mirror systems; and a coherent
evolutionary framework which links them, both by synthetic brain imaging
and by brain imaging across monkeys, chimps, and other primates.
Not AIP Homologue:
Let’s discuss this!
F5 Homologue
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
9
A Gene for Universal Language?
Lai, C.SL., Fisher, S.E., Hurst, J.A., Vargha-Khadem, F., &
Monaco, A.P., 2001,
A forkhead-domain gene is mutated in a severe speech and
language disorder, Nature 413:519-523
We have studied a unique three-generation pedigree, KE, in which a severe
speech and language disorder is transmitted as an autosomal-dominant
monogenic trait. Our previous work mapped the locus responsible,
SPCH1, to a 5.6-cM interval of region 7q31 on chromosome 7. We also
identified an unrelated individual, CS, in whom speech and language
impairment is associated with a chromosomal translocation involving the
SPCH1 interval. Here we show that the gene FOXP2, which encodes a
putative transcription factor containing a polyglutamine tract and a
forkhead DNA-binding domain, is directly disrupted by the translocation
breakpoint in CS. We suggest that FOXP2 is involved in the
developmental process that culminates in speech and language.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
10
Probably Not!
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
11
Criteria for Language-Readiness
A hypothesis on which human brain mechanisms underlie language
Properties Supporting Prelanguage Communication:
Symbolization: The ability to associate an arbitrary symbol with
a class of episodes, objects or actions.
(At first, these symbols may not have been words in the modern sense.
Nor need they have been vocalized.)
Intentionality. Extension of communication to be intended by the utterer to
have a particular effect on the recipient.
Parity: What counts for the speaker must count for the listener
(Mirror Property)
More General Properties:
Hierarchical Structuring: Perception and action involving components with
sub-parts (Action-oriented perception)
Temporal Ordering: Coding hierarchical structures “of the mind”
Beyond the Here-and-Now: The ability to recall past events or imagine
future ones.
Paedomorphy and Sociality: Conditions for complex social learning
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
12
Criteria for Language
What cultural evolution and learning add to the brain’s capabilities
On the basis of the Parity, Hierarchical Structuring, and
Temporal Ordering of Language-Readiness:
Symbolization: The symbols become words in the modern sense,
interchangeable and composable in the expression of meaning.
Recursivity of Syntax and Semantics: The matching of syntactic to
semantic structures co-evolves with the fractionation of utterances
Beyond the Here-and-Now: Verb tenses or other circumlocutions
express the ability to recall past events or imagine future ones.
Learnability: To qualify as a human language, it must contain a
significant subset of symbolic structures learnable by most human
children. [It is not true that children master a language by 5 or 7 years of age.]
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
13
From Grasp to Language:
Seven hypothesized stages of evolution
grasping
 a mirror system for grasping (i.e., a system that matches
observation and execution) [Shared with common ancestor
of human and monkey]
 a simple imitation system for grasping [Shared with common
ancestor of human and chimpanzee]
 Pre-Hominid
 Hominid Evolution
 a complex imitation system for grasping,
 a manual-based communication system, breaking through the
fixed repertoire of primate vocalizations to yield an open repertoire
 proto-speech resting on the "invasion" of the vocal apparatus by
collaterals from the communication system based on F5/Broca's area
 Cultural Evolution in Homo Sapiens
 language: the change from action-object frames to verb-argument
structures to syntax and semantics: Co-evolution of cognitive and
linguistic complexity
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
14
Stage 3: Simple Imitation
Masako Myowa-Yamakoshi:
the form of “imitation” employed by chimpanzees is a long and laborious
process compared to the rapidity with which humans can acquire novel
sequences;
 the focus is on moving objects to objects rather than on the structure of
movements per se.

Monkeys less so and chimpanzees more so (and, presumably, the
common ancestor of human and chimpanzees) have
Simple imitation: imitating simple novel behaviors but only through
repeated exposure.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
15
Stage 4: Complex Imitation
Humans have complex imitation: they can acquire (longer)
novel sequences in a single trial if the sequences are not too
long and the components are relatively familiar.
The very structure of these sequences can serve as the basis for
immediate imitation or for the immediate construction of an
appropriate response, as well as contributing to the longer-term
enrichment of experience
Extension of the mirror system from single actions to compound
actions adequate to support complex imitation was an evolutionary
change of key relevance to language-readiness
Hypothesis: This emerged on the hominid line after the divergence
from the common ancestor of humans and chimpanzees.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
16
Two Roles for Imitation in the Evolution of Manual-Based
Communication
1. Extending imitation to pantomime to provide ad hoc gestures that may
convey a situation to the observer
2. And then extending the mirror system from the grasping repertoire to
mediate imitation of gestures to support the transition from ad hoc gestures
to conventional signs which can reduce ambiguity and extend the semantic
range.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
17
“Beyond” the Mirror System
F5 alone is not the “full” mirror system

We want not only the “unit actions” but also
sequences and more general patterns
The FARS model sketched how to generate a sequence positing roles
for SMA and BG.
Our proposed mirror model must match this with a model of how


the units of a sequence (cf. current MNS model) and
their order/interweaving (extending the MNS model)
can be recognized and imitated.
This new model requires recognition of a complex behavior on multiple
occasions with increasing success in recognizing component actions
and in linking them together.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
18
The Ancestral Communication System
For want of better data, we will assume that our common
human-monkey ancestors shared with monkeys the following:
Primate Call System
a limited set of species-specific calls
Oro-Facial Gesture System
a limited set of gestures expressive of
emotion and related social indicators
Note the linkage between the two systems: communication is inherently multi-modal.
Note the role of body posture as well.
Combinatorial properties for the openness of communication are virtually
absent in basic primate calls and oro-facial communication though individual
calls may be graded.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
19
From Praxis to Communication
Stage 5: Gestural Communication Emerges
Our hypothetical sequence for manual gesture:
pragmatic action directed towards a goal object
pantomime in which similar actions are produced away from the goal
object
Imitation is the generic attempt to reproduce movements performed by
another, whether to master a skill or simply as part of a social interaction.
By contrast, pantomime is performed with the intention of getting the
observer to think of a specific action or event. It is essentially
communicative in its nature. The imitator observes; the panto-mimic intends
to be observed

abstract gestures divorced from their pragmatic origins (if such
existed) and available as elements for the formation of compounds
which can be paired with meanings in more or less arbitrary fashion.
A distinct manuo-brachial communication system evolved to
complement the primate calls/oro-facial communication system
On this view, the "speech" area of early hominids
 i.e., the area somewhat homologous to monkey F5 and human
Broca’s is not yet even a proto-speech area!
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
20
Noun/Verb pairs differentiated by movement
A change in the extent of movement will change the meaning
of a sign
Stokoe
Language in Hand
Figure 1
Figure 3
Figure 6
A change in the speed of movement will change the
meaning of a sign
Here the noun is characterized by short, repeated movements,
while the verb is characterized by a single, prolonged
movement
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
21
Facial Expressions as Components of Sign
Stokoe
Language in Hand
Figure 7
Here facial expression changes a statement to a question.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
22
Stage 6: From Manual Gesture to Proto-Speech
The "generativity" which some see as the hallmark of language
is present in manual behavior. Combinatorial properties are
inherent in the manuo-brachial system. This provided the
evolutionary opportunity for:
Stage 6. The manual-orofacial symbolic system then “recruited”
vocalization.
Association of vocalization with manual gestures allowed them to
assume a more open referential character.
This explains why F5, rather than the primate call area provide the
evolutionary substrate for speech
This yields our explanation for the
evolutionary prevalence of the lateral motor system over the medial
(emotion-related) primate call system
in becoming the main communication channel in humans.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
23
Gesture Remains




McNeill has used videotape analysis to show the crucial use that
people make of gestures synchronized with speech
Even blind people use manual gestures when speaking
Sign languages are full human languages rich in lexicon, syntax,
and semantics.
Moreover: not only deaf people use sign language, so do some
aboriginal Australian tribes, and some native populations in
North America
All this suggests that we locate phonology in a
speech-manual-orofacial gesture complex.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
24
Not three separate systems but a single system
operating in multiple motor and sensory modalities
Primate Call System
a limited set of species-specific calls
Larynx and
Vocal Cords
Genuine Cooperation
Oro-Facial Gesture System
a limited set of gestures expressive of
emotion and related social indicators
Facial Muscles
Manual Gesture System
an open set of communicative gestures
Arm and
Hand
Speech System
an open set of communicative gestures
Caution: One system but many brain regions, each with its own evolutionary story.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
25
Linking the “F5-Broca” and Vocalization Systems
Rizzolatti & Arbib (1988) thus showed why speech did not
evolve “simply” by extending the classic
primate vocalization system.
We now note the co-evolution of the two systems:
 Lesions centered in the anterior cingulate cortex and supplementary
motor areas of the brain can cause mutism in humans, similar to the
effects produced in muting monkey vocalizations
I hypothesize cooperative computation between cingulate cortex and
Broca’s area,
with cingulate cortex involved in breath groups and
emotional shading (and imprecations!), and
 Broca’s area providing the motor control for rapid
production and interweaving of elements of an utterance.

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
26
Language acquisition
Locating phonology in a
speech-manual-orofacial gesture complex
we see that
language acquisition takes various forms:
 a hearing person shifts the major information load of
language -- but by no means all of it -- into the speech
domain, whereas
 for a deaf person the major information load is removed
from speech and taken over by hand and orofacial
gestures
 and note that blind children accompany speech with
hand movements
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
27
From Action-Object Frame to
Verb-Argument Structure
to Syntax and Semantics
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
28
The Action-Object Frame
The action-object frame is non-linguistic:
the representation of an action involving
one or more objects and agents.
(Composing them yields “schema
assemblages”)
Verb-argument structure is an overt
linguistic representation;
in modern human languages, generally
the action is named by a verb and the
objects are named by nouns (or noun
phrases). (Composing them yields
semantic structures.)
A grammar for a language is then a
specific mechanism (whether explicit or
implicit) for converting semantic
structures into strings of words, and vice
versa.
Cognitive Structures
(Schema Assemblages)
P
r
o
d
u
c
t
i
o
n
Semantic Structures
(Hierarchical Constituents
expressing objects,
actions and relationships)
P
e
r
c
e
p
t
i
o
n
“Phonological” Structures
(Ordered Expressive
Gestures)
Cautionary Note: In the brain there is
probably no single grammar, but rather a
“direct model/grammar” for production
“inverse model/grammar” for perception
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
29
The Biological Basis of Language-Readiness 1
 “Knowing
there are things and events”: The ability for
perception of Action-Object Frames in which an actor, an action,
and related role players can be perceived in relationship –
was well established in the primate line
 Recognizing action-object frames
 Extending the mirror system beyond single actions to a repertoire
of action-object frames which is unbounded a priori.
 Naming action-object frames (the “names” can be manual/oro-facial)
creation of a “symbol toolkit” of meaningless
[less so for sign; very much so {phonemes} for speech; cf./cx.
Chinese]
elements from which an open ended class of symbols can be
generated
 abstract symbols are grounded in action-oriented perception

Note that such naming does not imply separate names for the actions and
objects or their attributes; i.e., it does not entail that utterances of
prelanguage were compounded from words akin to those we see in, e.g.,
the Indo-European languages.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
30
The Biological Basis of Language-Readiness 2
Hypothesis: The ability to communicate a fair
number of action-object frames was established
prior to Homo sapiens.
The Transition to Language
Fractionation of symbols to yield symbols for actions and
objects, yielding verb-argument structures linked to actionobject frames
The ability to compound those structures in diverse ways.
Recognition of hierarchical structure rather than mere sequencing
could provide the bridge to constituent analysis in language –
 Relating particular subactions (themselves further decomposable)
to achievement of certain subgoals in a complex manipulation.
 Abstraction and compounding of more generic verb-argument
structure
 Syntax and semantics: compounding utterances, “going recursive”

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
31
Claim: Homo sapiens had a language-ready brain
but did not have language
Grounding Hypothesis: Many ways of expressing
relationships were the discovery of Homo sapiens:
adjectives, conjunctions such as but, and, or or and that,
unless, or because, etc., might well have been “post-biological” in
their origin.
The one word ripe halves the number of fruit names to
be learned
 Separating verbs from nouns lets one learn only m+n
words to be able to form m*n*m of the most basic
utterances.

The result: A spiraling co-evolution of communication and
representation, extending the repertoire of achievable, recognizable
and describable actions.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
32
The spatial basis for “prepositions”
Consideration of the spatial basis for “prepositions”
may help show how visuomotor coordination underlies some aspects of language
and makes clear the “naturalness” of sign.
Stokoe
Language in Hand
Figure 10
The addition of movement
transforms IN to INTO and
exemplifies the differences in
meaning between the two signs
However, the basic semantic-syntactic correspondences have been overlaid by a multitude of
later innovations and borrowings.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
33
The Mirror Neuron System (MNS) Model
If the monkey needs so many brain regions for the mirror system for grasping, just
think how many more brain regions we will need for an account of language-readiness
that goes beyond the mirror to develop a full neurolinguistic model that extends the
linkages far beyond the F5  Broca’s area homology
Object features
cIPS
Object
affordance
extraction
7b: PF/PG Object affordance
-hand state
association
Hand
shape
recognition
Hand
motion
detection
STS
AIP
7a
Motor
program
(Grasp)
Integrate
temporal
association
Action
Mirror
Feedback recognition
Hand-Object
spatial relation
analysis
F5canonical
(Mirror
Neurons)
F5mirror
Object
location
Motor
program
(Reach)
Motor
execution
M1
F4
work with
Erhan Oztop
"What" versus "How"
DF: Jeannerod et al.
Inability to Preshape (except for
objects with size “in the semantics”
reach programming
Parietal
Cortex
How (dorsal)
grasp programming
Visual
Cortex
Inferotemporal
Cortex
What (ventral)
AT: Goodale and Milner
Inability to verbalize or pantomime size or orientation
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
35
Goodale and Milner
Our evolutionary theory suggests a progression from action to
pantomime to (proto)language
 object  AIP  F5canonical: pragmatics
 action  PF  F5mirror: action understanding
 scene  Wernicke’s  Broca’s: utterance
The "zero order” model of AT and DF data is:
 Parietal “affordances”  preshape
 IT “perception of object”  pantomime or verbally describe size
Inference: one cannot pantomime or verbalize an affordance; one needs
a "unified view of the object" (IT) to express attributes.
The problem with this is that the “language” path as shown in  is
completely independent of the parietal  F5 system, and so the data
seem to contradict our view in .
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
36
Recall:
FARS (Fagg-Arbib-Rizzolatti-Sakata) Model Overview
•AIP extracts a set of
affordances but
•IT and PFC are
crucial to F5’s
selection of the
affordance to execute
AIP
AIP
Dorsal
Stream:
dorsal/ventral
Affordances
streams
Ways to grab
this “thing”
TaskConstra
Constraints
Task
ints ( F6)
(F6)
Working
Memory
W
orking Me
mory (46)
(46?)
Instruction
Stimuli
Instruction
Stim
uli (F2)
(F2)
PFC
Ventral
Stream:
Recognition
F5
“It’s a mug”
IT
An Early Pass on the AT/DF Challenge
Visual Input
AIP
F5canonical
Choosing an Action
PF
F5mirror
Recognizing an Action
Wernicke’s
Area
Broca’s
Area
Describing an Object or
an Action
STS
IT
Recognizing an
Object or an
Action
Prefrontal
Memory
Do these link the right boxes? What is the relationship?
Is PF a homologue of Wernicke’s area? How does the role of PFC in the
FARS model relate to its roles in the mirror system of monkey and in
language?
To be continued ...
Many Challenges Lie Ahead at the Interface between Computer Science and
Cognitive Neuroscience
Analyze the brain regions involved in tasks involving varied combinations
of action, vision and language, to probe the overlapping or distinctive roles
of specific regions.
 Develop a neurally plausible model which explains how, given a video, the
viewer's attention may be drawn to a specific object or action, and then expands that
attention to determine the minimal subscene containing that focus of attention.
 Develop a neurally plausible model for how, given a minimal subscene, the viewer
generates sentences to describe it, analyzing the extent to which the initial focus of
attention biases the type of sentence structure used for the description.
 Develop a neurally plausible model for how a question about a visual scene provides
a top-down influence on mechanisms of attention as the viewer examines the scene in
preparation to answer the question.
 Explore how the expansion of attention in detail, space, time and factuality grounds a
cognitive-based functional expansion of syntax and semantics.
 Use comparative and historical studies to tease out the universals of language and
test them against functional explanations versus Universal Grammar.
 Explore the relation between the child’s exploration of its world and the influence of
mother and community in the acquisition of language.

Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
39
An Invitation
All these topics will be explored in CS 664
taught by Michael Arbib and Laurent Itti
in the Spring of 2002.
Michael Arbib CS564 - Brain Theory and Artificial Intelligence, USC, Fall 2001. Lecture 17. FARS to Language
40