Download ling411-13 - Rice University

Ling 411 – 13
Words in the Brain:
Six Hypotheses
WORDS IN THE BRAIN
Hypothesis I: Functional Webs
•
•
A word is represented as a functional web
Spread over a wide area of cortex
 Includes perceptual information
• Relating to the meaning
 As well as specifically conceptual
information
• For nominal concepts, mainly in
» Angular gyrus
» (?) For some, middle temporal gyrus
» (?) For some, supramarginal gyrus
 As well as phonological information
• Temporal, parietal, frontal
Part of the functional web for fork
A Set of Hypotheses
C
T
M
PP
P
PA
V
All of these
lines
represent
bidirectional
connections
(hence, reverberation)
Compare Pulvermüller’s
version
Phonological representation:
a distributed representation
in the perisylvian area
Meaning
of a verb
Meaning
of a visual
object
Friedemann Pulvermüller, The Neuroscience of Language, 2002
WORDS IN THE BRAIN
Hypothesis II:
Nodes as Cortical Columns
 Information is represented in the cortex in the
form of functional webs (Hypothesis I)
• A functional web is a network within the
cortical network as a whole
 consisting of nodes and their
interconnections
• connections represented in graphs as lines
 Nodes are implemented as cortical columns
 The interconnections are implemented as intercolumnar neural connections and synapses
•
•
Axonal fibers
Dendritic fibers
The node as a cortical column
 The properties of the cortical column are
approximately those described by Vernon
Mountcastle
• Mountcastle, Perceptual Neuroscience, 1998
 Additional properties of columns and
functional webs can be derived from
Mountcastle’s treatment together with
neurolinguistic findings
•
Method: “connecting the dots”
 Hypothesis IV: (Coming Soon!)
Quote from Mountcastle
“[T]he effective unit of operation…is not
the single neuron and its axon, but
bundles or groups of cells and their
axons with similar functional properties
and anatomical connections.”
Vernon Mountcastle,
Perceptual Neuroscience
(1998), p. 192
Findings relating to columns
(Mountcastle, Perceptual Neuroscience, 1998)
 The column is the fundamental module of
perceptual systems
• probably also of motor systems
 This columnar structure is found in all
mammals that have been investigated
 The theory is confirmed by detailed
studies of visual, auditory, and
somatosensory perception in living cat
and monkey brains
A Memory Experiment
(Pulverműller 2002: 26-27)
 Performed with macaque monkeys
 Delayed matching – monkey must remember
• Monkey must keep in mind the shape or color of
an object and perform a matching response
after delay of several seconds
 Neural activity detected in frontal and temporal
lobes
 Temporary lesion of frontal or temporal area leads
to impaired stimulus specificity in other area
 Supports the hypothesis of a functional web
including sites in frontal and temporal areas
•
Why?
Pulvermüller’s line of reasoning
1.
2.
3.
“If neurons in the functional web are strongly
linked, they should show similar response properties
in neurophysiological experiments.
“If the neurons of the functional web are necessary
for the optimal processing of the represented
entity, lesion of a significant portion of the network
neurons must impair the processing of this entity.
This should be largely independent of where in the
network the lesion occurs.
“Therefore, if the functional web is distributed
over distant cortical areas, for instance, certain
frontal and temporal areas, neurons in both areas
should (i) share specific response features and (ii)
show these response features only if the respective
other area is intact.”
(2002: 26, see also 27)
Pulvermüller’s line of reasoning
1.
2.
3.
“If neurons in the functional web are strongly
linked, they should show similar response properties
in neurophysiological experiments.
“If the neurons of the functional web are necessary
for the optimal processing of the represented
entity, lesion of a significant portion of the network
neurons must impair the processing of this entity.
This should be largely independent of where in the
network the lesion occurs.
“Therefore, if the functional web is distributed
over distant cortical areas, for instance, certain
frontal and temporal areas, neurons in both areas
should (i) share specific response features and (ii)
show these response features only if the respective
other area is intact.”
(2002: 26, see also 27)
A Memory Experiment (2002: 26-27)
 Performed with macaque monkeys
 Delayed matching – monkey must remember
• Monkey must keep in mind the shape or color of
an object and perform a matching response
after delay of several seconds
 Neural activity detected in frontal and temporal
lobes
 Temporary lesion of frontal or temporal area leads
to impaired stimulus specificity in other area
Reasoning from memory experiment
 Temporary lesion of frontal or temporal area leads
to impaired stimulus specificity in other area
 “Together, these data provide evidence that
neurons in both temporal and frontal areas (a)
showed the same specific response features and
(b) showed these response features if and only if
the respective other area was intact…”
 Compares language impairment vis-à-vis
Wernicke’s and Broca’s areas (2002: 28)
Not so fast!
… the same specific response features?
Elsewhere he writes “similar”
“If neurons in the functional web are strongly
linked, they should show similar response
properties in neurophysiological experiments.”
(2002:26)
N.B.: similar – not same!
 Similar:
•
•
Sharing some features
There may be differences with respect to
other features
Pulvermüller’s reasoning (cont’d)
“These results obtained in memory experiments
with macaque monkeys are reminiscent of wellknown facts from … investigation into acquired
language disorders … . These … studies … showed
that prefrontal and temporal areas are most
crucial for language processing. They also showed
that lesions in either area can lead to aphasia,
which in the majority of cases include deficits in
both language production … and perception … .”
(2002: 28)
Pulvermüller on Wernicke’s aphasia
“… patients with Wernicke’s aphasia have
difficulty speaking…. These deficits are
typical…and cannot be easily explained by
assuming a selective lesion to a center
devoted to language comprehension.”
(2002: 36-37)
Pulvermüller’s hypothesis on
phonological word forms
“The functional webs realizing phonological word forms
may be distributed over the
perisylvian area of the
dominant left hemishpere.
Circles represent local
neuron clusters and lines
represent reciprocal
connections between them.”
Friedemann Pulvermüller, The Neuroscience of Language, 2002: 52
Basic and complex functions
 Phonological recognition is a basic function
• Located in Wernicke’s area
 Speaking is a complex function
• A cooperative effort of several areas,
•
including Broca’s area and Wernicke’s area
Phonological recognition is a necessary
component of speaking
Wernicke:
“Primary functions alone can be referred to specific areas…. All
processes which exceed these primary functions…are dependent
on the fiber bundles, that is, association.”
Aphasia Symptom Complex (1874)
Wernicke’s Area and Speaking
 Phonological images guide speech production
 Phonological recognition monitors production
 Compare..
• Painting without visual perception
• Playing a piano without auditory perception
 Conclusion: Of course phonological
recognition (i.e. Wernicke’s area) plays a
role in speech production
Paraphrasing Pulvermüller
…patients with Wernicke’s aphasia have difficulty
speaking…. These deficits are typical…and cannot be
easily explained by assuming a selective lesion to a
center devoted to language comprehension.
The Neuroscience of Language (2002)
Altered quote:
…patients with damage to the occipital lobe have
difficulty drawing pictures…. These deficits are
typical…and cannot be easily explained by assuming
a selective lesion to a center devoted to visual
perception.
Re-examining the monkey memory experiment
 Compare short-term verbal memory
•
•
Hypothesis: reverberating activation between
Broca’s area and Wernicke’s area
If one of those areas is impaired, the
reverberating activity is disrupted, leading to
diminished activity in the other area
 Same principle could apply in memory test
in macaque monkey
•
Reverberation between temporal lobe
(recognition zone) and frontal lobe (action zone)
 Does not require that the two areas share
“same specific response features”
Conclusion:
The components of a functional web are diverse
 The phonological representation of a word
may be seen as a functional web in the
perisylvian area
 But each component of the web has its own
specific local function within that
representation
• For example, phonological recognition in
Wernicke’s area
 If they are all the same, why have many of
them, spread out over different areas?
Compare Hypothesis III: Nodal specificity (below)
Elsewhere, Pulvermüller gets it right
“…activation of the web, so to speak, completes
itself as a result of the strong web-internal links.
If the web of neurons is considered a memory
representation of an object and each neuron to
represent one particular feature of this object
memory, the full ignition would be the neuronal
correlate of the activation of the stored object
representation. Such full activation of the object
memory could occur if only a fraction of the
features of the object are present in the actual
input.” (2002: 29)
Why do the nodes in a web appear to
have similar response features?
 Not because each node has – on its own –
response features similar to those of
other nodes in the web
 Simply because all the nodes are “tied
together” in the web
• Therefore, all respond when the whole web
is ignited
 Actually they have, individually, very
different response features
• E.g. in Wernicke’s area and Broca’s area
Reverberation in functional webs
 Reverberation among connections in an
established web strengthens activation
 Experimental verification:
• Compare words and pseudo-words
•
•
 Pseudo-words: phonologically OK but no meaning
Real words show greater activation
“About one-half second after the onset of spoken
one-syllable words, high-frequency brain responses
were significantly stronger compared to the same
interval following pseudo-words.” (Pulverműller: 53)
Another word : pseudo-word experiment
(Pulverműller 2002: 54-56)
 Finnish
• pakko ‘compulsion’
• takko : a pseudo word
• Same 2nd syllable
 Measurements used: MMN and MMNm
• (MMN : mismatch negativity)
• Larger for –ko of real word
 Strongest difference at 200 ms
 Subjects were watching a silent movie
•
I.e., not paying attention
Finnish ‘pakko’ experiment: discussion
 [-ko] produces activation in either context,
since it is a syllable occurring in Finnish
 Stronger activation in pakko
• pakko is an established word in Finnish
• That means it has established connections
• Established connections provide stronger
activation
WORDS IN THE BRAIN
Hypothesis III: Nodal Specificity
in functional webs
 Every node in a functional web has a
specific function
 The nodes in each area of a functional web
• Constitute a subweb
• Their function fits the portion of cortex in
•
which they are located
 For example,
• Phonological recognition in Wernicke’s area
• Visual subweb in occipital and lower temporal lobe
• Tactile subweb in parietal lobe
Each node of a subweb also has a specific
function within that of the subweb
Support for Nodal Specificity:
the paw area of a cat’s cortex
Column (node) represents
specific location on paw
Support for Nodal Specificity:
Columns for orientation of lines (visual cortex)
Microelectrode
penetrations
K. Obermayer & G.G. Blasdell, 1993
Support for Nodal Specificity:
Map of auditory areas in a cat’s cortex
A1
AAF – Anterior auditory field
A1 – Primary auditory field
PAF – Posterior auditory field
VPAF – Ventral posterior
auditory field
WORDS IN THE BRAIN
Hypothesis III(a): Adjacency
 Nodes of related function are in adjacent
locations
• More closely related function, more closely
adjacent
 Examples:
• Adjacent locations on cat’s paw represented by
•
adjacent cortical locations
Similar line orientations represented by
adjacent cortical locations
Support for Nodal adjacency:
the paw area of a cat’s cortex
Adjacent column in cortex
for adjacent location on paw
Hypothesis III(b)
 The nodes in each area of a functional web
• Constitute a subweb
• Each node of a subweb has a specific function
within that of the subweb (Hypothesis III)
 Functional specificity of subwebs:
• Each subweb has specific function within the web
 Fits its location in the cortex
 For example,
• Visual subweb in occipital and lower temporal lobe
• Tactile subweb in parietal lobe
A phonological subweb: /bil/
bil
Cardinal node for bill
Subweb for bill
bi-
-il
An activated functional web
showing two subwebs
T
C
PP
PR
PA
M
Control of articulation
V
Visual features
WORDS IN THE BRAIN
Hypothesis IV: Extrapolation to Humans
 Hypothesis: The findings about cortical
structure and function from experiments on
cats, monkeys, and rats can be extrapolated
to human cortical structure and function
 In fact, this hypothesis is simply assumed
to be valid by neuroscientists
 Why? We know from neuroanatomy that,
locally,
• Cortical structure is relatively uniform across
•
mammals
Cortical function is relatively uniform across
mammals
WORDS IN THE BRAIN
Hypothesis IV(a):
Linguistic and conceptual structure
 Hypothesis IV(a): The extrapolation can be
extended to linguistic and conceptual
structures and functions
 Why?
• Local uniformity of cortical structure and
function across all human cortical areas except
for primary areas
 Primary visual and primary auditory are
known to have specialized structures, across
mammals
 Higher level areas are – locally – highly
uniform
Objection
 Cats and monkeys don’t have language
 Therefore language must have unique
properties of its structural representation
in the cortex
 Answer: Yes, language is different, but
• The differences are a consequence not of
•
different (local) structure but differences of
connectivity
The network does not have different kinds of
structure for different kinds of information
 Rather, different connectivities
WORDS IN THE BRAIN
Hypothesis V:
Hierarchy in functional webs
 A functional web is hierarchically organized
• Bottom levels in primary areas
• Lower levels closer to primary areas
• Higher (more abstract) levels in
 Associative areas – e.g., angular gyrus
 Executive areas – prefrontal
 These higher areas are much larger in
humans than in other mammals
 Hypothesis V(a): Each subweb is likewise
hierarchically organized
Hierarchy in a visual subweb
V
FORK
A network of
visual features
Etc. etc.
(many layers)
Properties of Hierachy
 Relates to general hierarchy in the cortex
 Each level has fewer nodes than lower
levels, more than higher levels
• Compare the organization of management of a
corporation
 Top level has just one node
• Compare the “CEO”
• The “C” node of a word web
 Cardinal node
• Hypothesis VI
WORDS IN THE BRAIN
Hypothesis VI:
Cardinal nodes
 Every functional web has a cardinal node
• At the top of the entire functional web
• Unique to that concept
• For example, /cat/ at “top” of the web for CAT
C
 Hypothesis VI(a):
• Each subweb likewise has a cardinal node
 At the top level of the subweb
 Unique to that subweb
 For example, V/cat/
• At the top of the visual subweb
Cardinal nodes of a functional web
Some of the cortical structure relating to fork
Each
node
shown
here is
the
cardinal
node of
a
subweb
Cardinal
node of the
whole web
T
M
PP
C
P
PA
V
Cardinal
node of
the visual
subweb
(Part of) the functional web
for the concept CAT
The cardinal node for the
entire functional web
T
C
P
A
M
V
Cardinal nodes
of subwebs
The argument against cardinal nodes
 Pulvermüller: “It is not necessary to
assume a cardinal node” (p. 24)
 Arguments by others (directed against
“grandmother nodes):
•
•
Not enough flexibility
Not enough availability
•
The cardinal node of a hierarchical web is not
a “grandmother node” as usually understood
It is supported by the hierarchy principle
Compare
 CEO of a corporation
 President of the U.S.
 Response to the arguments:
•
•
The “Grandmother Node”
(Cardinal node for your grandmother)
 Grandmother node
• A node that represents “grandmother”
 An untenable hypothesis, according to the
usual conception
 But two separate conceptions to be
distinguished
• A node that represents “grandmother” all by
•
itself
A node whose receptive field is “grandmother”
The untenable grandmother node
 A node that would recognize grandmother all
by itself
• Such a node would have to be extraordinarily
•
complex
 How could one node recognize grandmother
• In different positions/postures
• In different clothing
• At different ages
Criticisms of such a conception are well-founded
 Such a hypothesis involves local
representation without distributed
representation
A sophisticated grandmother node
has a distributed representation
 It also has a cardinal node (local representation)
 GRANDMOTHER
• A ‘grandmother node’ in the sophisticated sense
• It represents a specific value: GRANDMOTHER
• Its receptive field is “grandmother”
 It works because it is the cardinal node of an
entire functional web
• Other nodes in the web handle
 The details
 A range of diverse perceptual properties
Arguments against ‘grandmother nodes’
 They usually assume that the local
representation is representing a concept
(like ‘grandmother’) all by itself
• i.e., Local representation without distributed
representation
 i.e., without a supporting web
Arguments against local representation
1. Recognizing new things and producing
motor responses to new things are
problematic on the local-coding theory
2. The patterns recognized visually by a
human in a lifetime vastly outstrip the
number of sensory processing neurons in
the entire human nervous system
Churchland & Sejnowski
The Computational Brain
MIT Press, 1992, p. 163
Arguments against local representation
1. Recognizing new things and producing
motor responses to new things are
problematic on the local-coding theory
2. The patterns recognized visually by a
human in a lifetime vastly outstrip the
number of sensory processing neurons in
the entire human nervous system
These arguments are directed against the
naïve conception of the grandmother node.
Arguments against local representation
1. Recognizing new things and producing
motor responses to new things are
problematic on the local-coding theory
This argument assumes that such a node recognizing
grandmother all by itself. But it is the whole
functional web that recognizes grandmother. Each
part of this web naturally responds to a wide range
of values, including novel values.
Arguments against local representation
2. The patterns recognized visually by a
human in a lifetime vastly outstrip the
number of sensory processing neurons in
the entire human nervous system
Churchland & Sejnowski 1992:163
On the contrary, the web can accommodate recognition
of multiple new exemplars without the need for
recruiting additional nodes. Not a problem after all.
New nodes are needed only for new learning.
Support for the cardinal node hypothesis
1.
It follows from the hypotheses of
nodal specificity and hierarchy
•
•
A hierarchy must have a highest level
The node at this level must have a
specific function
2. It is needed for ignition of the whole
web from activation of part of it
•
For example, to activate the phonological
representation from the visual
3. It is automatically recruited in
learning anyway, according to the
Hebbian learning hypothesis
More support for cardinal nodes
 The distributed network as a whole

represents the concept (e.g. FORK)
The whole can evidently be activated
by any part of the network
• From seeing a fork
• From eating with a fork
• Etc.
 The cardinal node provides the
coordinated organization that makes
such reactivation possible
Reactivating the functional web
 When the cardinal node (the integrating node)
is activated, it can activate the whole
(distributed) functional web
• Without it, how would that be possible?
• E.g., activating conceptual and perceptual
properties of cat upon hearing the word cat
• From phonological recognition to concepts
• From visual image to phonological
representation
Cardinal nodes and the linguistic sign
 Connection of conceptual to phonological

representation
Consider two possibilities
1. A cardinal node for the concept connected
to a cardinal node for the phonological
image
2. No cardinal nodes: multiple connections
between concept representation and
phonological image
•
supported by Pulvermüller (2002)
Pulvermüller’s hypothesis:
No cardinal nodes
Phonological representation:
a distributed representation
in the perisylvian area
Meaning
of a verb
Meaning
of a visual
object
Friedemann Pulvermüller, The Neuroscience of Language, 2002
Functional Webs acc. to Pulvermüller
1. Distributed representation of form and
of meaning
•
This part is correct
•
•
Runs counter to the linguistic evidence
Implication: parts of the phonological
representation connect to parts of the
meaning
Example: walk - WALK
 [w-] or [-k] for action with legs?
2. Multiple connections between form and
meaning
•
Implications of possibility 2



No cardinal nodes: multiple connections
between concept representation and
phonological image
I.e., different parts of meaning connected
to different parts of phonological image
Consider fork
•
•
•

Maybe /f-/ connects to the shape?
Maybe /-or-/ connects to the feeling of
holding a fork in the hand?
Maybe /-k/ connects to the knowledge that
fork is related to knife?
Conclusion: Possibility 2 must be rejected
Support for the cardinal node hypothesis – 3
 It is automatically recruited in learning
according to the Hebbian learning principle
 Even if it weren’t there it would soon be
recruited as a result of co-activation of its
linked properties
 This is the operating principle for building a
functional web from bottom up
• At each level, co-occurring properties will
•
activate a node at next higher level
 That newly activated node represents the
combination of those properties
This process continues up to top of hierarchy
end