cognitive neuroscience a virtually completely reductionist...
TRANSCRIPT
• SYSTEMS+ NEUROSCIENCES
• Empirical methods developed out of standard biomedical sciences over the last 140 years and were in a reasonably modern form by about 1970 (except 4 below)
• COGNITIVE SCIENCES
• Empirical methods developed out of conceiving of the ‘human operator’ as a machine, all founded between 1945 and 1970
Cognitive Neurosciences• NEUROSCIENCES• 1. Effects of lesions in
animals• 2. Single (multiple-) cell
recording • 3.Human neuropsychology• 4. Functional Imaging: PET,
fMRI, EEG, MEG
• COGNITIVE SCIENCES• 1.Information-processing
psychology• 2. Neural network modelling
(connectionism)• 3. Symbolic artificial
intelligence• 4. Linguistics((All derived historically from
the computer revolution of World War II)
COGNITIVE NEUROSCIENCE• Theoretical questions basically arise from the
Cognitive sciences but the empirical methods they have available (in human experimental psychology and linguistics) often cannot differentiate between alternative theoretical attempts to answer the Qs. Neither set of methods alone are sufficiently powerful. Hence C.N. imports empirical measures/subjects from the neuroscience disciplines with tasks and models derived from the cognitive science disciplines.
A virtually completely reductionistapproach to cognitive
neuroscience?• Can it work, say by single-cell recording for a
series of tasks and modelling of the responses, as post-Dubner & Zeki it could be argued to be the case for a structure like MT/V5 ?
• The case of the hippocampus• The case of dorsolateral prefrontal cortex
Hippocampus and its cortical connections
Approach 1 Hippocampus as the seat of cognitive maps (O’Keefe &
Nadel 78 for navigation )• ‘Place cells’- cells which
respond when the animals enter a particular part of the environment.
• Can be modelled – predicts changes to responses when environment changed (Burgess & O’Keefe, 1996) with CA3 and CA1 as locus of the place cells); fits head-direction cells in subiculum 4 fields of place cells
for an animal in a square box
Hypothetical function derived fromsingle-cell recording
Virtual Town Navigation (BurgessN& O’Keefe) Human Hippocampus and Navigation
• A- Activation produced by controls when navigating in a virtual town
• B – Volumetric brain Mapping (VBM) (Taxi Drivers vs Controls)
• C – VBM Correlation with amount of time spent taxi-driving
A – Maguire et al Science 98; B,C – Maguire et al PNAS 2000
2.Hippocampus as the seat of episodic memory
• Tulving’s 72 concept for basis of autobiographical memory characterised in contrast to ‘semantic memory’ (knowledge)
• First applied to amnesia by Kinsbourne & Wood 75• Pure amnesia associated with hippocampal lesions
particularly in patient HM by Milner 1962 (see egSquire Psych Rev 92; Cipolotti et al Neuropsychologia 01; Spiers et al Neurocase 01)
Cipolotti et al (Neuropsychologia2001)
Selective damage to hippocampus (5SDs below normal)with all other parts of temporal cortex being well within normal limits except for Left parahippocmpal gyrus(2SDs below normal –presumed to arise from loss of white matter fibres from hippocampus – by eye)
VC (Cipolotti et al 2001)
• Grossly impaired on many quantitative tests of episodic memory function - both verbal and non-verbal, both recall and recognition
• For example knowledge of events famous for brief periods of time but then little referred to e.g. incident in British nuclear power facility in 1957 leading to major release of radioactive material
Treves & Rolls model
• Detailed specification presented for CA1, CA3, pathway from entorhinal to CA1 which bypasses the dentate gyrus etc
Hippocampus• Need to decide which or both of two approaches are
on the right lines, and if so how they are related • This has not occurred despite 30 years of detailed
investigation of anatomy and neurophysiology of the hippocampus
• Assumptions on function permeate all models of the structure. Issue must therefore be addressed on functional level as well as standard neuroscience levels
Functions of Dorsolateral Prefrontal Cortex
Paradigms activating DorsolateralPrefrontal Cortex
DR = Delayed response; DA = delayed alternation: DMS = Delayed matching to Sample
Funahashi et al J Neuroscience 1993
Left sulcus principalislesions in oculomotordelayed response tasks
Funahashi et al
• ‘Disruption of a transient on-line memory process that holds the memory of a given spatial coordinate ‘online’ to direct the response at the end of the delay’
D’Esposito & Postle 2002
Dominant perspective – dorsolateral prefrontal cortex is the seat of working memory; Goldman-Rakic identified with the executive component of Baddeley-Hitch information-processing model of working memory
Baddeley model inappropriate• Baddeley & Hitch 1974
produced a model of working memory with one central executive system and two slave systems
• In Baddeley 1986 identified the ‘central executive’ with Norman & ShalliceSupervisory System which is not essentially a storage system
Rowe et al Science 2000• This experiment separated maintenance of items
- exact positions of three dots - from selection. After a variable interval a line appeared on the screen thru’ one of the dots.
• Ss had to move the cursor to the position of THAT dot.
• GREEN = maintenance• RED = selection
Maintenance
Selection Green – maintenance; Red - selection
Baddeley model inappropriate
• In Baddeley 1986 identified the ‘central executive’ with Norman & Shallice Supervisory System which is not essentially a storage system
Supervisory System Model• FIGURE 4
Baddeley(1986) identified ‘executive store’ in his working memory model with the S.A.S component of Norman- Shallice model which is not a store
Marr’s (1982) tri-level approach• 1. Domain of abstract problem analysis• 2. Computational (software and functional
architecture) domain• 3. Hard-ware (computer/brain-based)
• NB ‘Level’ – a misnomer as three domains of questions can be posed at many different grains
Two dominant functional architectures – brain-derived
• Modules/ isolable subsystems
• Computational – need to separate different computations
• Neuroscientifically plausible because of the anatomical separatons in system level of organisation of the nervous system BUT as part of more complex systems
• Neural Networks (can be ‘artificial’)
• Plausible due to local network characterisics of nervous system BUT to be tractable often are grossly simplified and with non-brain based elements: a necessary part of modelling
• To be discussed in lecture 2
Neuropsychology & Functional Imaging
• Three levels of theoretical inferences• 1. From dissociations (N), main activation cross-over activation
contracts & conjunction analysis (FI) -> functional architecture• 2. Boot-strapping based on plausible inferences from model to
prediction (examples from N)• 3. From more complex aspects of data (e.g. error patterns (N),
change in activation correlations over time(FI)) -> more specific aspects of operation of components
• For N because of the rarity of particular specific lesions one often needs the methodology of single case studies (lectures 1 & 2) –require replication across patients - as well as group studies (lecture 3)
Single Dissociation :Amnesia
• Verbal span – can be normal ie 6 or 7 (Drachman& Arbit, 1966)
Recognition Memory for 50 word lists - grossly impaired ( <36)
(see Shallice FNMS 88 chap 2)
• Does this imply that the two tasks are carried out by different subsystems only one of which is damaged
• NO -Span may just be an easier
Span – number of random digits that can be repeated at 1/sec presentation rate eg4,8,1,5,3,9,6,7,
Recognition Memory - 50 words (eg join) presented at 1 per 3 sec rate. Tested by 2 –alternative forced choice (eg mind or join?)
Conceptual framework derives from work on attention –particularly dual tasks. If S devotes more attention to one task devotes less to the other. Each task has its own performance-resource curve
If curves monotonic upward AND if the same system is the critically involved one THEN iff for two patients A,B :
Perf (A) > Perf (B) for task I THEN:
Perf (A) > Perf (B) for task II
Therefore crossover interaction shows assumption false – 2 systems involved
Performance – Resource Curves
Derives from attention literature –Norman & Bobrow1975
Amnesics Vs STM patients (initially Warrington & Shallice Brain 1969)
• Amnesics : Verbal span –can be normal ie 6 or 7 (Drachman & Arbit, 1966)
Recognition Memory for 50 word lists - grossly impaired ( <36)
(see Shallice 88 chap 2)
• STM patients : Verbal span - grossly impaired (<4)
(Shallice & Vallar, 1990)
Recognition memory for 50 word lists – normal – 45 (KF, JB) (Shallice, 88 chap 3)
Classical Double Dissociation – Good task at normal level; other grossly impaired
But what is the deficit in STM patients
• Normal ability to read and comprehend single words.
• In some (see Shallice & Butterworth Neuropsychlogia 1977) speech production essentially normal.
• Damage to input phonlogical buffer in Baddeley-Hitch model
Free Recall Serial Position Curves
• Amnesics – Baddeley & Warrington JounalVerbal Learning and Verbal Behavior 1970, 9, 176.
• STM Patients – Shallice& Warrington QJEP 1970; Vallar & Papagno Brain & Cognition, 1986, 5, 428
N word lists presented at a standard eg 1 per 2 sec rate – followed by recall in any order. Plot serial position in the N words where words recalled
Free recall serial position curves in amnesic patients – Baddeley & Warrington (J. Verbal Learning Verbal Behaviour 1970)
Vallar & Papagno (Brain & Cognition 1986)
Short-term memory patient PV
Controls
STM patient does not showstandard recency effect
Functional specialisation of cortex – eg input phonological buffer
• Dissociations observed in neuropsychology
• Local areas of activation from functional imaging
Shallice & Warrington Quart J Exp Psychology 1972 – Peterson paradigm – with short-term memory patient KF – visual v auditory input Paulesu et al Nat. 1993, 362, 342
• STM task• Sets of 6 letters sequentially
visually presented followed by a recognition probe
• 1.Vis. letter processing• 2. Articul. Recoding• 3. STM stor. + retriev.Control –Vis equiv (Korean
letters)
• Rhyming Task• Series of letters
presented visually – if any rhyme with B detect
• 1. Vis letter processing.• 2. Articul. Recoding• 3. Rhyme processingControl – Vis analogue
(Korean letters)
A= ISP +STS
B = ISP
ISP = Inner Speech Processing
Paulesu et al Nature 1993 Warrington Logue and Pratt Neuropsychologia 1971
• Two early cases of short-term memory syndrome
• Case 1 _ Warrington & Shallice Brain 1969
• Case 2 – Shallice & Butterworth Neuropsychologia 1977
Vallar (Cortex 2004) review –Baddeley & Hitch phonological
buffer slave system
Required for span –impaired in short-term memory patients
Quite distinct from hippocampal and medial-temporal lobe regions involved in episodic long-term memory
Connectionist models of Phonological STS eg Burgess & Hitch (Psychological Review 1992) etc.
Input Phonological Buffer: Isolable subsystem plus Anatomy
• Patients: STM syndrome (Warrington & Shallice, Brain 1969)
• Left BA 39, 22, 40? (Warrington et al Neuropsychologia, 1971)
• Functional Imaging: Paulesu et al, 1993
• Left BA 39, 40• Fits with many other
studies
Dissociations only suggest the existence of isolable subsystems –possibilities include
Inference to modularity – stronger with classical dissociations
Functional Imaging
• 1.Change in cognitive demand• 2.Change in neural activity• 3.Change in vascular system properties • 4a.Blood flow thru’ radioactive tracers (PET• 4b.Blood oxygenation level dependent(BOLD)
signal (fMRI)
Statistical Parametric Mapping (Friston et al HBM 1995)
• Realignment to reduce movement artefacts• Normalise to standard brain (Talairach space*)• Spatially smooth – increases signal-to-noise
ratio and improves cross-subject averaging (iereduces effects small diffs in anatomy)
• Use general linear model – statistics (related to ANOVA)
• *In early SPM
Glass Brain Representations
Activation levels across conditions
Design of Functional Imaging Experiments re Seperability of Resources
• Great problem in FI for cognitive inferences –need to have a complete specification* of subsystems involved in the task AND OFTEN of the resource demands made by each condition
• Initial approach eg Peterson et al 88 – cognitive subtraction
• More appropriate I – conjunction designs• More appropriate II - factorial designs
*Only required for systems closely interacting with damaged systems for neurosychology
Functional Imaging : methodological problems (Price &
Friston, Neuroimage, 1997)Methodology of ‘pure insertion’ (adding a
component X which will be subtracted to produce the activation corresponding to process X) leads to 2 difficulties:
1. Other components may be affected by the addition
2. It is often difficult to create satisfactory control tasks
Hence -> conjunction analysis
Design of Functional Imaging Experiments re Seperability of Resources
• Great problem in FI for cognitive inferences – need to hace a complete specification of subsystems involved in the task AND OFTEN of the reesourcedemands made by each condition their Initial egPeterson et al 88 – cognitive subtraction
• More appropriate I – conjunction designs• More appropriate II - factorial designs given
activation in critical regions is monotonically related to resource demands made on region (see ShalliceNeuroImage 2003 for more detailed specification)
Advantages of Conjunction Design
• If the control task for an experimental condition does not have quite the same set of other components it is much less serious as will only be a problem if all control tasks are similarly flawed
Designs
• Top – Cognitive Subtraction
• Bottom - Cognitive conjunction
Price & Friston NeuroImage 1997
• Hypothetical stages in various naming tasks
Aguirre & D’Esposito (in Moonen& Bandettini Functional MRI,
1999)
• Criticise conjunction analysis as it presupposes that a task requires a component an equal amount in each task– in fact they may lie on a continuum e.g. possible effects of change in attention to object processing in above tasks.
• In the Price & Friston study the subject just said Yes to all stimuli in the control conditions.
• Thus attentional demand of experimental tasks higher • Analogous problem to resource variation in patient studies
(Shallice, 1988)
Aguirre & D’Esposito• Argue that parametric designs are the answer ie
where a variable takes several values on a continuum.
• In my view does not deal with the problem since the different values of the variable could require different levels of a resource.
• Best bet for separable systems – double dissociation. Examples in next lecture
• However conjunction analysis also good if non-critical components are made really unrelated
Burgess et al (Neuropsychologia2001) fMRI study on multi-tasking
• ON-LINE Task• 3 different types
(Unrelated to each other)
• Operates continuously through scan
• SECOND TASK• If event X occurs do Y
• 3 Conditions• 1.On-Line Task only• 2. S expects second task but it
does not occur (ie holds intention to do it)
EXPECTATION CONDITION• 3. S expects 2nd task and it
occurs
COLOUR BAR
COLOUR BAR
hammer
3 9
Burgess, Quayle and Frith (2001) Neuropsychologia 39, 545-555
Burgess, Quayle and Frith (2001) Neuropsychologia 39, 545-555.
Expectation – Baseline (Conjunction Analysis)
Shallice et al (Nature 1994)
Left Dorsolateral PFC sig more active in (Mem – PL) x (Easy - Diff)
Anterior Cingulate sig more active in (Mem – PL) x (Diff – Easy)
Supervisory System Model• FIGURE 4
Baddeley(1986) identified ‘executive store’ in his working memory model with the S.A.S component of Norman- Shallice model which is not a store
Rumiati et al NeuroImage 2004• Factorial Design• A. Two types of stimuli : 1. pantomime of meaningful
actions (on video) eg cutting as with scissors, 2. objects with characteristic actions eg scissors
• B. Two types of response : 1. action (all single restricted hand movement), 2. name
• Critical condition for assumption of trigger stimuli –interaction with A2 x B1 more strongly activated than the rest.
• Note visual and motor processes controlled
Action vs Object Stimuli
Object vs action stimuliDLPFC
-48, +8, +44; T=5.46
AAC-4, +30, +34; T=5.55
VLPFC-44, +46, +6; T=6.46
d IPL-52, -44, +46; T=5.19
v IPL-58, -32, +30; T=5.30
Rumiati et al.
Sig. Interaction Term
IA = Imitate Action
IO = Act given Object
NA = Name Action
NO = Name Object
Prefrontal Cortex: Three Positions
1. (Goldman-Rakic) Stores contents of working memory (i.e. memory required for cognitive operations)
2. (Duncan) Equipotential structure – the substratum of operations requiring General Intelligence (g)
3. (Shallice) A set of different but connected abstract Supervisory operations
Data Inference - Functional Specialisation only (ie not
modules or isolable subsystems)• Patients:• Many other types of
system theoretically compatible with observed dissociations
• Functional Imaging:• Critical comparison
subtractions between activation in two conditions
• Hence cannot infer absolute levels of activation in particular regions
Functional specialisation of cortex
• Neither method (neuro. or func.imag activation diffs.) allows one to infer the presence of modules in Fodor’s 1983 sense) or isolable processing systems. Other possible architectures are compatible with the evidence.
• However isolable subsystems which are only partially informationally encapsulated and which are part of larger systems are the most plausible account of the type of evidence just presented
Double Dissociations
• To avoid resource artefacts technically require:• Task A patient I > patient II• Task B patient II > patient I• NOT Patient I task A> task B• Patient II task B > task A
Group Studies
• Objection made by Caramazza (Cog Neuro1986) that average of a group need not correspond to possible values of ANY possible member of group
• Not a problem for dissociations per se as could occur only in highly implausible situations –‘monsters’ (Lakatos – Brit J Phil Science 1963)
Double Dissociations in FuncImag. (Fletcher et al Brain 1998)
• Retrieval of 2 types of material• A - Organised word list (16 related words, eg 16
foods)• B – Paired associates(eg wine – Burgundy) • Right Dorsolateral Prefron. A > B > Con.• Right Venterolateral Prefron. B > A > Con
Functional Architecture (Subsystems)
• Relation between task characteristics and underlying subsystems can be far from transparent (e.g. abstract, concrete)
Functional Architecture (Networks)
• Supports• 1. Subsystem contains ‘attractor’ characteristics• 2. Concrete nouns and abstract nouns
differentiated by different type of features AND different quantitative aspects (ie simple no of features)