Summaries of Cognitive Neuroscience Papers



Godijn, Theeuwes JEP:HPP 29:5:882 2003

Parallel allocation of attention prior to the execution of saccade sequences [Behaviour, Attention, Saccade]

Had to execute sequence of 2 saccades to 2 peripherally or centrally pre-cued locations (in any order) as fast as possible. In 50% of trials, secondary delayed task to identify letters presented for 47ms at 35-130ms after initial sequence-cue by 2AFC at end of trial. Findings: saccade errors dual-task 22%, single-task 15%. Letter accuracy 85% at both targets, 60% elsewhere.

Separate experiment using determined order of saccades (central precue with two arrows long and short). Letter accuracy at first saccade location 80%, second location 70%.

Separate experiment with ordered saccades, and 2 letters presented: either both simultaneous, or at 1st saccade location then 2nd saccade location, or 2nd then 1st. Both locations show benefit at the same time though 2nd saccade has less; order of letter presentation makes no difference → both saccadic target locations get attentional benefits around the same time. But does this happen on different trials (averaging)?

Separate experiment with same-different decision on letters at two locations. Letters at 25% non-target, 50% one-target, 25% both-target locations. Interaction shows that letter at both target locations has much greater effect on accuracy than at 2*(one location) → attentional benefit to both saccade targets in the same trial. However: if sequential presentation, then no interaction – i.e. attention is no longer present at both locations.


**Wolf, Deubel, Hauske 1984: psychological refractory period for combinations of manual and saccade responses.

** Shiffrin & Gardner 1972: timecourse of attention

Rizzolatti – premotor theory of attention




Barton,...Edelman, Manoach EBR 168:1-2:76 2005

Switching, plasticity and prediction in a saccadic task-switch paradigm [Saccade, Executive]

Centrally task-cued antisaccade or prosaccade to peripheral cue. Random order or predictable sequences. CTI 200 or 2000ms. Monetary reward. Constant 1.7s intertrial gap. Findings: cost for antisaccades = 50ms, 6%, cost for switching = 10ms, 5% - switching costly for prosaccades, and antisaccades if the CTI is short. Long CTI paradoxically speeds antisaccades after a task switch. Effect remains when subjects informed of predictability.

Explained by combination of inhibitory effects from previous trial, plus advance reconfiguration cost independent of predictability. Antisaccade in previous trial generally suppresses preparatory pre-target activity in FEF/SC – i.e. instead of task-set inertia, the non-dominant task-set’s inhibition applies to the whole response-system.

task-switch cost for pro-saccades, but benefit for antisaccades.

Switching costs can’t be eliminated by a predictable sequence.



Fielding, Bradshaw, Millist... Cog Brain Res 25:251 2005

No sequence-dependent modulation of the Simon effect in PD [Disease, Conflict]

Saccadic Simon effect: peripherally presented cue symbolically instructing L/R saccade. Spatially incongruent trials ↑RT and ↑error.

PD patients have short latencies (550 vs 700ms), large Simon effect (90ms), irrespective of previous incompatible trial (controls 100ms → 0ms).


**Ansorge & Wuhr, ‘A response-discrimination account of the Simon effect,’ JEP HPP 30 (2004) 365– 377: Modulation by instructions, context and set-effects – contingent on WM representation of S-R mapping. Particularly, previous trial compatible causes ↑RT difference between compatible and incompatible; Simon effect vanishes after incompatible trial.



Everling 1999 – suppression causing longer latency and lower errors




Chica, Lupiez, Bartoloemo Cogn Neuropsychol 23:7:1015 2006

Dissociating inhibition of return from endogenous orienting of spatial attention: evidence from detection and discrimination tasks [Attention]

Spatially congruent or incongruent precue, either expected (75%) or unexpected, either discrimination or detection task, at varying SOA. Found that exogenous and endogenous are independent; that IOR occurs ~400ms even if attention does not need to be disengaged (when expectation is at cued location, i.e. endogenous).



Van der Lubbe, Jaskowski Psychological Res 69:179 2005

Mechanisms underlying spatial coding in a multiple-item Simon task [EEG, Attention]

Six boxes, three on left, three on right, each containing ‘R’ or ‘L’. Central arrow determines which box governs the manual response. Manipulation of order of presentation of letters or arrow (500ms before or after) does not affect effect magnitude; therefore it is only when the target is actually there at the cued (incongruent) that the ‘spatial code’ begins. Unable to distinguish between referential coding vs attentional shift model. ERP also. Note: symmetrical arrow cue; Magic square design.






Rogers & Monsell JEP: General 124:2:207 1995

Costs of a predictable switch between simple cognitive tasks [Executive]

Predictable task switch every 2 trials. ‘T9’ → ‘consonant’ vs. ‘T9’ → ‘odd’.

Task-set theory predicts reconfiguration cost vanishes if enough time between trials, but there is residual cost. Explained by endogenous vs. exogenous task-set selection.

Endogenous: Controlled processing (Atkinson & Shiffrin 1968), Central executive (Baddeley 1986),

(Norman & Shallice 1986) allow involuntary task-set competition driven by available environmental stimuli, which is overridden by SAS when needed. (explains capture, utilisation, stroop).

Allport 1994 (Attention and Performance XV) find switching is costly if recent switch to a task involving same stimulus set → ‘task-set inertia’.



** Laming 1979, Acta Psychologica 43:3:199 Choice reaction performance following an error

Examples of parametric mapping of task sets –

  • Shaffer 1965, JEP 70:284 – Choice reaction with variable mapping – precue determines which response for single stimulus.
  • Dixon & Just 1986, Memory & Cognition 14:488 – a chronometric analysis of strategy preparation in choice reactions. ↓RT for valid precueing of required task operation / mapping; this benefit increases over time between cue and task. But: also reflects time to perceptually process the cue as well as reconfigure task-set.
  • Duncan 1979, JEP:HPP 5:216, - Divided attention: the whole is more than the sum of its parts





McClelland Psychological Rev 86:4:287 1979

On the time relations of mental processes: an examination of systems of processes in cascade




Tanaka J Neurosci 27:44:12109 2007

Cognitive signals in the primate motor thalamus predict saccade timing [Neuron, Saccade]

VL thalamus neurones during ‘memory saccades’ – brief cue then saccade after fixation offset (1s or 1.5s), ‘self-timed saccades’ – brief cue and saccade after without any go signal that had to occur at 800-1600ms in order to get reward, and simple visually triggered saccades. Preparatory ramping predicts time of self-timed saccades. Very similar firing pattern for memory and internal saccades.


Malapani, Rakitin, Meck... JCN 10:316 1998

Coupled temporal memories in Parkinson’s disease: a dopamine-related dysfunction [Disease, Time]

Feedback learning and then reproducing time intervals 6s,17s. Off medication, migration of shorter durations to longer, and longer durations to shorter.



Koch, Costa, Brusa... Neuropsychologia 2007

Impaired reproduction of second but not millisecond time intervals in Parkinson’s disease



Merchant H, Zarco W, Prado L J Neurophysiol 99:2:939-949 2008

Do we have a common mechanism for measuring time in the hundreds of millisecond range? Evidence from multiple-interval timing task [Behaviour, Timing]

Perception vs production, auditory vs visual, 1 or 4 intervals.

Variance increases linearly with interval duration in each task. Greater in perceptual than motor tasks. Greater in visual than auditory.

Conclude: partially overlapping distributed mechanism underlying quantifying time in different contexts



Clark, Bechara, Damasio, Aitken, Sahakian, Robbins Brain 131:1311 2008

Differential effects of insular and ventromedial prefrontal cortex lesions on risky decision making [Lesion, Reward]

Comparison of normals, lesion controls (DL / VL), vmPFC, and insular patients on Cambridge Gambling task. Gamble on red or blue boxes, where odds range from 1:9 to 5:5. Patients select gamble colour then percentage of their money; percentage presented in ascending or descending order to examine impulsivity. vmPFC bet more all the time; insula bet more when odds are not so good – i.e. can’t adjust their risk, and go bankrupt quicker. No difference in impulsivity nor in choosing most probable colour. Insular lesions slower all the time, vmPFC slow particularly when greater uncertainty.

Upward bet=downward bet → no impulsivity or delay aversion. vmPFC have a criterion shift in risk-taking behaviour. Insula: less sensitive to odds of winning.


Cycyk & Wright Aphasiology 22:4:422 2008

FTV vs PNFA vs Semantic dementia. Clinical features


Schenkluhn, Ruff et al. J Neurosci 28:44:11106 2008

Parietal stimulation decouples spatial and feature-based attention [TMS, Attention]

TMS to supramarginal gyrus vs anterior IPS vs posterior IPS. Measure performance on singleton detection task with either endogenous location precue or feature (colour) precue. SMG impairs only spatial cueing advantage (d’), ant IPS impairs both cue types, post IPS impairs neither.




Berdyyeva TK, Olson CR J Neurosci 29:3:591 2009

Monkey SEF neurons signal the ordinal position of both actions and objects [Neuron, Memory]

Single neurone recording from SEF during a) three identical pictures, the identity of which cues which of 6 orders to visit them, refixating on the centre each time, b) three different pictures, the order of which is randomised on each display, and the saccade must be in a fixed order of objects, giving 27 different saccade orders, c) reward modulation and d) prolonged waiting period. Conditions c & d eliminate confounding factors for ‘ordinal position’.

Findings: 1) interaction in 50% between rank selectivity (in 90%) and direction selectivity (in 75%), 2) neurones’ rank selectivity is correlated between tasks a and b.


Dorris MC & Munoz DP J Neurosci 18(17):7015-26 1998

Saccadic probability influences motor preparation signals and time to saccadic initiation [Saccade, Behaviour]

200ms gap paradigm, target either in or opposite cell’s RF. Blocks of either ‘always in RF’, ‘always opposite’, or ‘50% chance in RF’. Anticipatory saccades and RT correlated with likelihood. Superior colliculus buildup neurones correlate with these.


Ikeda T, Hikosaka O Neuron 39:693-700 2003

Reward-dependent gain and bias of visual responses in primate superior colliculus [Neuron, Saccade, Reward]

Asymmetrically rewarded version of memory-guided saccade task. Visual response of SC neurons often increased when a cue stimulus indicated an upcoming reward. Interaction with whether saccade was directed into the neurone’s RF.



Serences JT Neuron 60:1169-1181 2008

Value-based modulations in human visual cortex [fMRI, Reward]

Choice between two coloured discs plus three-level choice of ‘value’. Very complex reward scheme with pseudorandom reward probability, ‘baiting’ to prevent choosing only the most likely colour, change-over-delay to discourage ‘win-stay lose-switch’ strategies. Sometimes both stimuli were rewarded.

Shows reward-related modulation of V1.


Sugrue LP, Corrado GS, Newsome WT Science 304:1782 2004

Matching behaviour and the representation of value in the parietal cortex [Neuron, Reward]

Macaque trained to saccade to one of two coloured targets for juice. ‘Foraging’: Poisson assignment of reward to targets over time. Monkey follows Herrenstein’s matching law but quickly adapts to probability changes. Therefore fitted with leaky integrator model, t ~= 12 trials which is in fact optimal.

LIP cells chosen by RF on simple memory-guided saccade task, showed appropriate correlation with local fractional income.


Itti L, Koch C NRN 2:3:194 2001

Computational modelling of visual attention [Theory, Attention]

Saliency map: salience depends on surrounding context & object recognition. Implements IOR and saccadic coordinates. Input feeds forward from other feature maps, with centre-surround inhibition, then back-propagates to early visual areas. Cf Desimone & Duncan – no global map, just enhancements of feature maps. Compare Koch & Ullman 1985 saliency map model with Zetzsche 1998 model of maximum feature density, Wolfe’s guided search model (1994) implicating top-down feature selection, Mozer & Sitton 1996 (MORSEL) in which object knowledge causes features to be enhanced. Then discusses several systems of object recognition that involve sequential direction of attention.



Greenberg DS et al Nature Neuroscience 11:749 2008

Population imaging of ongoing neuronal activity in the visual cortex of awake rats

Two-photon population calcium imaging using fluorescence in awake rats, validated with local field potential. Found correlation between neighbours in awake state could not be predicted from correlation in anaesthetised state.


Horowitz TS & Wolfe JM Nature 394:575 1998

Visual search has no memory [Attention, Behaviour]

Conjunction search set-size 8/12/16 items with present/absent decision. Stimuli moved randomly to one of 4 random positions every 111 ms. Search time was longer but slope (RT vs. set-size) was the same as for static stimuli. Analysed ‘target-present’ responses only, then repeated experiment with target identification of one of two target letters.


Mohanty A, Gitelman DR, Small DM, Mesulam MM Cerebral Cortex 18:11:2604 2008

The spatial attention network interacts with limbic and monoaminergic systems to modulate motivation-induced attention shifts [fMRI, reward]

Directional central precue (60% valid, 15% invalid or 25% uninformative) followed by peripheral images as targets (or null trials). SOA 200-800. Separate food and tool blocks: respond Left for donut or hex-nut 90%, Right for pastry or screw 10%. Performed when starved (with a donut incentive afterwards) or satiated on donuts.

Findings: small RT benefit for food. Parietal regions (IPS TPJ and PC) more correlated with speed of attentional shifts to food targets when hungry than full; opposite trend for tools.


Uchida Y, Lu X, Ohmae S, Takahashi T, Kitazawa S J Neurosci 27:50:13750 2007

Neuronal activity related to reward size and rewarded target position in primate supplementary eye field [Neuron, Reward]

Visually guided saccades in 4 directions, in when fixation disappears (overlap of 500-800ms). Small or big rewards in alternate blocks. 181 neurones in SEF classified according to activity during saccade (81) or reward (78): saccade-related, reward-related, directionally-biased-saccade-related, directionally-biased-reward-related, directionally-tuned-saccade-related-cell.

Findings: 50% of S/R cells were directionally biased. Main bias was at time of reward or saccade, not in delay period – i.e. explicit representation of reward.


Platt ML, Glimcher PW Nature 400:6741:233 1999

Neural correlates of decision variables in parietal cortex [Neuron, Reward]

Centrally colour-cued saccade to one of two targets, with asymmetrical reward. Neurones in LIP recorded and correlated with decision variables.

Findings: During the course of a trial, cells shift from being correlated with outcome probability, to being correlated with the instructed movement. In net, correlation was strongest with expected value.



Silvanto J et al. Cer Cor 19:2:327 2009

The perceptual and functional consequences of parietal top-down modulation on the visual cortex [Neuron, Attention]

TMS to PPC reduces phosphene threshold for TMS to V1/V2, only when asymmetrical.

Hikosaka, Takikawa, Kawagoe Physiological Reviews 80:3:953 2000

Role of the basal ganglia in the control of purposive saccadic eye movements [Theory, Saccade]

Long review of physiology (338 references). Role in attention, working memory, expectation, procedural learning. Two parallel mechanisms from cortex: caudate-nigra-colliculus, versus pallidum-subthalamic-nigra route. When active simultaneously, the pathways ‘focus’ spatial patterns of activity; when sequential they switch from suppression to execution of movement.



Melcher D, Colby C TICS 12:12:466 2008

Trans-saccadic perception (opinion) [Theory, Attention]

3 old models: scene-stitching (Jonides, Pollatsek, disproved by pattern fusion experiments), visual working memory (change detection 3 objects), no information retained between saccades (complex scenes, large changes).

5 principles: dynamic RF – remapping of remembered items or pre-saccadic remapping, prediction, intermediate processing stages (between pixel buffer and object memory), graded effect of saccade on representation, gist/general meaning contains much scene information but does not require remapping.





Cools R, Sahakian, Robbins Brain 124:12:2503

Mechanisms of cognitive set flexibility in Parkinson's disease [Disease, Executive]

Screen background colour determines task naming letter or number. Foreground is two characters, letter/number/symbol (neutral). PD patients have problems set-shifting compared to controls, but only when there the task-irrelevant item is competing (non-neutral).


Cools R, Barker, Sahakian Robbin Cer Cor 11:12:1136 2001

Enhanced or impaired cognitive function in PD as a function of dopaminergic medication and task demands [Disease, Executive]

Withdrawal of DA agonist causes decrement in set-shifting but improvement in reversal learning. Set shift: task of naming number or letter determined by background colour, two characters presented, letter/number/symbol; ‘cross-talk condition’ when the a number and letter are both presented; task switching slower in patients in off state, only when competition present. Probabilistic reversal: a red and a green target shown at two of four locations. 80% probability reward by colour (first colour touched was ‘correct’) for 40 trials, then reward probability reverses for 40 trials. All patients good off dopamine, but 50% are worse when on.

Propose two DA systems for DLPFC and OFC/ventral striatum, with different basal levels and therefore different points of U-shaped curve.

Everling & Munoz J Neurosci 20:387 2000

Neural correlates for preparatory set associated with prosaccades and antisaccades in the primate frontal eye field [Neuron, Executive]

Targets in the “movement field” of FEF neurones elicit prolonged responses when saccades are made, and attenuated responses when antisaccades are required. Error trials show higher responses.



Bergeron & Guitton J Neurophysiology 88:1726-1742 2002

In multiple-step gaze shifts: Omnipause neurones (OPNs) and superior collicular fixation neurones (SCFNs) encode gaze position error; OPNs gate saccades [Neuron, Saccade]

Single cell recording from cat pontine OPN and SC fixation neurones during visually guided saccades. Multiple hypometric saccades made. OPNs stop 40ms before each saccade. SCFNs however encode gaze position error throughout, firing below baseline from time of stimulus, increasing over time until last saccade.


Haggard, Poonian, Walsh Brain Research 1286:106 2009

Representing the consequences of intentionally inhibited actions [Behaviour, Awareness]

Free choice left or right keypress, free choice to inhibit the action. Tone 250ms after keypress, pitch corresponds to key pressed with 80% congruence. On inhibited trials, one of the tones presented randomly in a later time window. Subjects report perceived time of tone, and on inhibited trials, the intended direction of action.




Burgess & Hitch Psychological Review 106:551-558 1999

Memory for serial order: a network model of the phonological loop and its timing [Theory, Memory]

Timing/context layer of overlapping representations over time. Visual and auditory inputs interact at a ‘competitive cueing’ stage: strongest activity at each step is outputted then strongly inhibited. Baddeley & Hitch 1975 model accounts for articulatory suppression, limit of span, and phonemic similarity effect.

Seeks to explain

  1. bowed serial order curve
  2. ordering errors predominate, especially transposition
  3. interaction between phonemic similarity and serial position (e.g. alternating)
  4. presentation modality effect
  5. auditory suffix removes recency effect but not visual suffix
  6. span increase with familiarity of items
  7. span for visual nonwords increases with phonological similarity to words (BRANE > SLINT)
  8. Hebb effect: repeating the list improves recall, but not when temporal groupings are changed
  9. number of new errors on each rehearsal decreases with each rehearsal
  10. serial order intrusions: errors with items of the same serial position from previous list
  11. long-term learning


Liston & Stone J Neurosci 28:51:13866 2008

Effects of prior information and reward on oculomotor and perceptual choices [Decision, Reward]

Two bright discs presented on luminance pixel noise (d’ 5.5 and 4.2), 2AFC saccade task toward brighter disc (accuracy 70%) followed by 2-interval forced choice button press between the saccade-target and another test patch. Thus saccadic choice and perceived brightness could be both measured. Left-right probability (prior expectation) or reward were explicitly manipulated.

Perceptual noise increases linearly with perceptual gain, and both were correlated with saccadic bias, such that there was no improvement in d’ and no change in motor threshold. This indicates the bias occurs at a visual processing stage shared with perception, but after the dominant internal noise source. They interpret this as gain modulation in V4. Motor decision based on this raw signal, but perception receives a slower normalised signal.


Basso & Wurtz J Neurosci 18:18:7519 1998

Modulation of neuronal activity in superior colliculus by changes in target probability [Neuron, Decision]

Memory guided saccade task used to classify cells. Then saccade to dimmed item in an array, with or without fixation overlap, with or without repetition blocks.

Old findings – buildup cells have saccade-direction-sensitive increase.

  1. Fixation and burst neurones unaffected by probability.
  2. Buildup neurone activity generally lower when multiple targets present.
  3. High buildup activity correlates with shorter saccade latency
  4. High probability gives more buildup activity, even before target presented


Bendiksby & Platt Neuropsychologia 44:12:2411 2006

Neural correlates of reward and attention in macaque area LIP [Neuron, Reward]

Separate cue for saccade target, go signal, and irrelevant item. Go: 2nd light flickered on 80% of trials, 3rd light on 20%. flicker was v brief → Accuracy ~60%. Small or large reward blockwise.

LIP visual response (cue 1 in RF and cue 2 in RF) modulated by reward (null, R+ or R-)

Modulation by reward size was independent of which cue appeared in the RF.

Activity did not predict errors, but does predict RT independently of reward.

Interpreted as ‘reorientation of attention’ from flicker to saccade target.

LIP activity is negatively correlated with prosaccade RT (Roitman & Shadlen 2002), but positively correlated with antisaccade RT (Zhang & Barash 2000). → the amount of attention devoted to the cue determines the time required to reorient to the response target.


Leon & Shadlen Neuron 24:2:415 1999

Effect of expected reward magnitude on the response of neurons in the dorsolateral prefrontal cortex of the macaque [Neuron, Reward]

Recording from area 46 and FEF during memory-guided saccades. Exogenously precued location, and colour-precued reward size, in separate epochs, randomised order. Fixation offset was go signal.

  • Reward-related enhancement during gap, only in DLPFC not FEF, and greatest when in response field.
  • Monkeys broke off fixation twice as often after the small reward cue.
  • Enhancement partially depends on knowing the spatial cue beforehand


Rolls ET, Everitt B, Roberts A. Phil Trans Roy Soc B 351:1346:1433 1996

The orbitofrontal cortex [Theory, Executive]

Review of anatomy, neurophysiology and lesion studies.

  • Discrimination learning pointing to correct visual stimulus. Impaired when reward pairing reverses. Patients commit incorrect action despite being able to verbalise the correct response.
  • Dissociated facial and voice emotion recognition.
  • Experimental performance all correlate with behavioural problems.

Therefore OFC as pattern associator for reinforcement value. Correlates with experienced emotion. Outputs to striatum, which in turn decides action.


Schultz W. J Neurophysiology 80:1:1-27 1998

Predictive reward signal of dopamine neurons [Neuron, Reward]

Review of conditioning and nigrostriatal projection. Dopaminergic neurones do not covary with arm or eye movements, as would be expected from parkinsonism, nor with mnemonic or spatial delayed response tasks, as would be expected with a WM role. 75% apparently fired phasically in relation to rewarding characteristics of a wide variety of stimuli. 20% of these also respond to aversive stimuli. homogeneous scalar population signal

  • Dopamine response transfers instantaneously to CS. May be mediated by n. pedunculopontis (rat precursor to SN, now feeds into SNpr)
  • No response to an expected reward; depressed response to omission
  • Response generalisation to similar appearance in correct context (weaker with subsequent depression), i.e. appetitive events
  • Novelty response proportionate to salience of stimulus; gradual decay – 3-5 trials for low salience novelty, >1000 trials for loud noises or large images. May regulated by inhibitory projections from dorsal raphe.

Projection: 31m dopaminergic neurones. Striatum - “moderately topographical”, divergence factor 400. Also to area 4,6 & rest of cortex, layers I, V & VI. each cell 500,000 varicosities, 60% extrasynaptic, releasing ~1000 molecules per impulse, leading to [DA] 30-60 times baseline, lasting 200ms. Striatum 80% D1 (mostly in low affinity state, cells project to GPi, SNpr, enhances NMDA transmission but reduces EPSPs when membrane hyperpolarised) and 20% D2 (mostly in high affinity state, cells project to GPe, attenuates all depolarisations and AMPA & NMDA response). Posttetanic depression lasts 20min, and concomitantly increases corticostriatal connectivity. LTP of corticostriatal projections enhanced by 5ms pulses of dopamine.

Each striatal cell receives 10,000 cortical terminals and 1000 DA varicosities.

This suggests gating weak signals, and modulating Hebbian learning.



Roesch MR, Olson CR Science 304:5668:307 2004

Neuronal activity related to reward value and motivation in primate frontal cortex [Neuron, Reward]

Single cell recording from macaque OFC and postarcuate premotor cortex, in a 2AFC memory guided (3s) saccade task where target colour indicated size & polarity of reward. Interpreted as dissociating motivation from valence.



Gold JI, Shadlen MN TICS 5:1:10-16 2001

Neural computations that underlie decisions about sensory stimuli [Theory, Decision]

A difference in spike rates is proportional to logLR. This means the brain ‘does not need to know’ the mean firing rates for each cell in each hypothesis. It requires the activity of each neurone to be subtracted from that of its ‘anti-neurone’.


Earl Miller & Jon Cohen Ann Rev Neuroscience 24:167-202 2001

An integrative theory of prefrontal cortex function [Theory, Executive]

Review of WCST, Stroop, ToL, Desimone’s attentional ‘biased competition’. Theory of ‘’. PFC linked to & from higher-level association cortex. Reward signals foster formation of a task model, which reflects learned associations between task-relevant information. PFC thus recognises higher-level relevant patterns in association cortex. Excitatory signals from PFC feed back to enable task-relevant pathways. It therefore ‘guides’ / ‘modifies’ / ‘modulates’ other brain areas’ ‘prepotent tendencies’.

Used to explain attention, inhibitory control, action selection, WM, guided retrieval of LTM.



Asaad WF, Rainer G, Miller EK Neuron 21:6:1399-1407 1999

Neural activity in the primate prefrontal cortex during associative learning [Neuron, Decision, Learning]

Central object A/B as cue, then 1s delay, then two identical targets shown. Saccade L/R followed by reward. Pairings AL-BR and AR-BL alternated every 30 trials. Performance returned to 70% after 20 trials, reverted to 45% on trial 2 after change.

Recording from around principal sulcus of lateral PFC. 98% cells activated in at least one epoch (cue, delay, presaccade). Direction or object selective activity in 80%. Direction and object selectivity in 60%: linear additive in 16%; nonlinear interaction in 44%.

Direction selective cells became direction sensitive earlier in each trial, as block progresses.


Tremblay L, Schultz W Nature 398:704-708 1999

Relative reward preference in primate orbitofrontal cortex [Neuron, Reward]

Image above one of two levers determines target for reaching movement, identity predicts food/drink reward type; delay then go-signal (bilateral).

A. Recordings from sulcus between areas 12 & 13. 70% differentiate between expected reward, most strongly in the memory period; 3% discriminate left-right movements. They say cells don’t discriminate visual inputs [but they only show that individual cells fire the same amount for three images that give the same reward (i.e. no stimulus-reward reversal)].

B. Food preferences established before using 2afc. Given response preference A>B>C, when 2afc between A/B, a cell fires when A is on the left but not when B is on the left. However if 2afc B/C, the cell fires when B is on the left but not for when C is.


Bodi, Keri, Nagy, Moustafa et al Brain 132:9:2385 2009

Reward-learning and the novelty seeking personality: a between- and within- subjects study of the effects of dopamine agonists on young Parkinson’s patients [Disease, Reward]

Unmedicated vs medicated young PD patients on feedback-based probabilistic classification task. Reward or punishment for abstract visual shape A/B categorisation. Correlate performance with temperament and character inventory.

Conclude that unmedicated PD gives impaired reward processing, whereas dopamine agonsists impair punishment processing. Performance correlates with ‘harm avoidance’ in unmedicated patients but with ‘novelty-seeking’ in medicated patients. (Correlates with both in controls).


Swainson R, Rogers, Sahakian...Robbins Neuropsychologia 38:5:596-612 2000

Probabilistic learning and reversal deficits in patients with Parkinson’s disease or frontal or temporal lobe lesions: possible adverse effects of dopaminergic medication [Disease, Learning]

Comparison of frontal, temporal and PD, in

  1. ‘probabilistic discriminiation reversal’ – 2afc between 2 colours, feedback ‘CORRECT’ or ‘WRONG’ 80% consistent, reversing after 40 trials
  2. ‘concurrent reversal’ – 2afc with two pairs (pair A or pair B) of colours, intermixed trials, 100% consistent reward. After 50 trials, contingencies of pair B are reversed.

No evidence of perseveration, but PD, frontal and temporal patients all impaired on both.




Bechara, Damasio... Anderson Cognition 50:1:7-16 1994

Insensitivity to future consequences following damage to human prefrontal cortex [Lesion, Reward]

Iowa Gambling Task: $2000 loan, 100 trials. 4 decks, A +100 & 50% -250ish; B +100 & 10% -1250; C +50 & 50% -50ish; D +$50 & 10% -250.

Controls choose C/D, medial frontal choose B. Repeating with regular penalties and infrequent rewards (inverse task) suggests they are more influenced by immediate punishment than delayed reward (i.e they choose C’ and D’). They win-stay, lose-switch, and avoid punishment for several trials, but then return to bad decks. [is this working memory? interference?]


Rahman, Sahakian, Hodges JR, Rogers, Robbins Brain 122:8:1469 1999

Specific cognitive deficits in mild frontal variant frontotemporal dementia [Disease, Reward]

Cambridge gambling task: 10 blue or red boxes. Select blue or red, then select bet (ascending or descending sequence of bets shown 5% to 95% of bank, 5s each). fvFTD bet more, and adjust less to ratio, same up or down (ie they are risk-takers, not impulsive). Also: categorisation learning OK but reversal impaired. [they are impaired at unlearning?]

Hypothesise that vmPFC and OFC are selectively damaged in early fvFTD. [what about pathology?] and that reversal learning deficit is serotonergic dysfunction.


Peck CJ, Jangraw... Gottlieb J J Neurosci 29:36:11182 2009

Reward modulates attention independently of action value in posterior parietal cortex [Neuron, Reward]

300ms lateralised reward-cue (RC), identity heralds (100%) whether or not the trial is rewarded, but not direction predictive. Saccade target 600ms later, which is 50% congruent or incongruent to cue location. Incorrect response trials were repeated → optimal strategy is saccade to every target. Find that:

  1. Anticipatory licking for no-reward cues is extinguished after 8 trials. Not correlated with LIP activity, whose reward selectivity grew on longer timescales.
  2. RT shows more cue-location-congruence benefit when reward is predicted
  3. RT shows location-incongruence benefit when no-reward is predicted!
  4. LIP cells fire more for reward+ cues
  5. LIP cells only fire for cues when cue is in their RF; rate is lowered when cue opposite RF. Except: when an overlearned no-reward cue is presented there is lower firing for within-RF than in opposite side to RF.
  6. Probe trials (previously trained cues now presented in RF simultaneous with contralateral target) showed preserved reward modulation despite irrelevance.


Conclude: LIP encodes valence-specific attentional weight of stimulus. Reward learning occurring without operant associations, leading to plasticity of bottom-up visual responses (even in a novel non-reward-predictive context). Hypothesise this represents a change in salience of task-relevant objects.



Frank MJ, Seeberger LC, O’Reilly RC Science 306:1940 2004

By Carrot or by stick: cognitive reinforcement learning in parkinsonism [Disease, Reward]

Expt 1. Probabilistic learning to select one of each pair of abstract symbols AB, CD, EF. Reward/penalty matches correct/incorrect choice on 80% (A), 70% (B) or 60% (C) of trials. Then novel test combinations {AC AD AE AF} etc to discern between favouring A and avoidance of B.

Expt 2. Transitive inference learning AB BC CD DE. Better performance for AB when on medication, better for DE when off meds.

Used trial-to-trial data to show negative feedback on previous trial was used more by patients when off medication. [this could be modelled: estimate learning parameter independently for reward/penalty?]

Conclude: two-pathway model, direct Go (SNc-striatumD1-GPi/SNr-thalamus) and indirect Nogo (SNc-striatumD2-GPe-GPi/SNr-thalamus). But this does not explain patients’ superior performance. ??proposed differential modulation of excitatory and inhibitory pathways in VS by positive and negative outcomes??


Ekstrom LB, Roelfsema PR... Vanduffel J Neurosci 29:34:10683 2009

Modulation of the contrast response function by electrical microstimulation of the macaque frontal eye field [Neuron, Saccade]

Stimulated FEF at 50% of threshold (voltage where 70% of stimulations cause a saccade), and simultaneous fMRI of V1, V2 V3 V3A, V4 TEO TE MT MST FST LIP STP. Measurement of contrast response function for gratings within ‘movement field’. They fitted R=Rmax c2/(c2+c502). Conclusion: contrast gain is increased by subthreshold stimulation of FEF. [not possible to comment on spatial selectivity though]


Hazy TE, Frank, O’Reilly Phil trans Roy soc B 362:1485:1601 2007

Towards an executive without a homunculus: computational models of the prefrontal cortex/basal ganglia system [Behaviour, Executive]

RSVP AXBY task, with occational 1 or 2 that specifies whether AX or BY is the target. Respond L-no R-yes.

6 functional demands for WM: rapid updating, robust maintenance, multiple separate representations (for vs A/B), selective updating (1/2 on some trials, A/B on others), learning when to gate . Hypothesise BG as gate to WM, in or out; [what about repeated items?] and that BG learns to gate.



Bichot & Schall Nature Neurosci 2:549-554 1999

Effects of similarity and history on neural mechanisms of visual selection [Neuron, Decision]

FEF recording during monkey conjunction search. Fixed target shape & colour for each day.

  1. Monkeys show errors to distractors that were targets on previous day, and to colour match or shape match distractors.
  2. FEF learn over the day to discriminate distractors and target earlier in each trial.


Lueck CJ... Crawford, Kennard JNNP 53:284-288 1990

Antisaccades and remembered saccades in parkinson’s disease [Disease, Saccade]

  1. Step paradigm to one of 4 locations always 800ms after fixation, accompanied by 200ms tone. No difference in gain, latency, velocity.
  2. Memory saccade paradigm, with peripheral flash lasting 200ms, at 800ms after fixation, with a 200ms tone at 500ms after the flash to signal go. All subjects had lower peak velocity. PD patients gain 73% compared to 87%. Latency same.
  3. Antisaccade: as in step, but look to imaginary position opposite. Both patients and normals had reduced gain 75% and slower peak velocity (60deg/s), but no difference between groups. Errors were same velocity as step saccades but gain was reduced and latency prolonged (but not as great as the correct antisaccades); no differences between groups.



Della Libera, C & Chelazzi, L Psychological Sci 20:6:778 2009

Learning to attend and to ignore is a matter of gains and losses [Attention, Reward]

‘Attentional selection’ task: Red/green precue 400ms, 600ms gap, then 2 overlapping red and green nonsense shapes on left, and single black shape on right for comparison (3s). Comparison shape either same as target (50%), or completely different. Simultaneous same/different judgement. Reward 10p/1p presented for 600ms or error tone, depending not on performance, but on the shape. 5 shape types: T+ (high reward 80% low reward 20% when shape is target; 50/50 when shape is distractor), T-, D+, D-, and Neutral.

Expt 1: train then test (same experiment without rewards) after 2 days

Expt 2: cue 400ms target shape in black, then 300ms gap, then 2 black shapes on left and right 180ms then mask. One shape was previously neutral, the other was previously reward-biased. Report presence or absence of target shape.


Magno, Simoes-Franklin... Garavan J Cog Neurosci 21:12:2328-42 2009

The role of the dorsal anterior cingulate in evaluating behaviour for achieving gains and avoiding losses [fMRI, Reward]

Visual conjunction search with either 8 or 32 items. 4 responses: ‘absent’ (reward +1 or -1), don’t know ‘reject’ (reward 0), ‘present’ (reward +1 or -1), ‘definitely present’ (reward +5 or -5); or if no response within time interval, ‘missed’ (reward -1).

Results: behavioural results uninteresting; 7 of 26 subjects excluded because of too few of each outcome type. fMRI showed error-specific response in insula. They only compared all types of ‘error’ trials, ‘+5 trials’ and ‘reject trials’. Ant cingulate is more active in high reward and rejection trials, and less active in error trials.


Beck, Schlagenhauf... Wrase Biological Psychiatry 66:8:734-742 2009

Ventral striatal activation during reward anticipation correlates with impulsivity in alcoholics [fMRI, Reward]

Press button during a 200-1000ms target, presented 2250, 2500 or 2750ms after a reward cue. Adaptive target duration to get 67% accuracy. 7 reward cues: gain 3, 6 or 10 euros, or avoid losing 3, 6 or 10 euros, or neutral.

Healthy controls showed (anticipated gain > neutral) activation in bilateral ventral striatum, rt caudate tail, bilat thalamus, right insula. (anticipated loss > neutral) showed activation of bilateral ventral striatum, left MD thalamus, bilateral putamen & parahippocampal gyri, rt middle occipital gyrus & claustrum, left PCC, rt STG & cuneus.

Recently detoxified alcoholics: less ventral striatal activation than controls, other areas same.


Pearson, Hayden ... Platt Current Biology 19:18 1532-7 2009

Neurons in posterior cingulate cortex signal exploratory decisions in a dynamic multioption choice task [Neuron, Decision]

4-armed bandit – saccades to one of 4 coloured locations. Each arm’s payout is selected from a normal distribution about its value on the previous trial – random walk. Trials classified as explore/exploit using Kalman filter model on behavioural data.

Posterior cingulate neurones’ firing correlates with reward: positive, negative, U-shaped concave-up or concave-down, either in decision epoch or evaluation epoch = 8 types, equally probable. For trials with equal rewards, 12% of neurones showed sensitivity to expoit or explore, equally probable. 63% are target-location-sensitive.


Christopoulos, Tobler... Dolan, Schulz J Neurosci 20:40:12574-83 2009

Neural correlates of value, risk, risk aversion contributing to decision making under risk [fMRI, Reward]

Expt 1: Choice between 60-40 or 90-10, and certain values calibrated to equalise utility based on individuals’ risk aversion. Interspersed with ‘no-choice trials’ – same display and button press but a small arrow indicates which option they must choose!

Result: risky choice is worth 15 less, ranging from 2-31. Ventral striatum activity is positively correlated with certainty-equivalent-value for the low-risk>high-risk trials in which 1) subject chose the safe option and 2) no-choice trials, but not in which 3) subjects chose the risky options. i.e. correlation with risk but not utility.

Expt 2: 15/45, 10/50, 40/80 or 30/90. no outcome shown, and no correction for risk-aversion.

Result: ventral striatum also sensitive to high > low mean-expected-value gambles. Activity in dACC is seen in High-risk > low-risk trials where subjects 1) chose risky option but not in 2) chose safe option nor 3) no-choice trials. Also no correlation with risk-aversion. This implies dACC encodes objective risk and not EV or utility.

Objective risk is the spread of outcomes, but normalised for total expected value. Subjective risk (=risk aversion) is the deviation of a risky choice from its expected value, as measured by ‘certainty equivalent’ – the riskless value choice that balances the risky option.

Decision utility (ordering of preference) vs. experienced utility (pleasure from outcome). [choosing a less risky item with a low expected value is suboptimal, as subjects chose to have consistent (short-term) low reward than net gain despite some punishment (long term) – is this a form of impulsivity]


Rolls, McCabe, Redoute Cer Cor 18:3:652-63 2007

Expected value, reward outcome and temporal difference error representations in a probabilistic decision task [fMRI, Reward]

Left or right button press; left gives 10p with P=0.9, right gives 30p with P=1/6, 1/3 or 0.9 (blockwise). 30 trials per block, order randomised, subjects not told when change occurs. Modelled using ‘temporal difference’ reward prediction error as trial-by-trial deviation from EV [even though reward magnitude was all or nothing].

Results: they claim “stable response preference after 10 trials” [but for Rt EV=5p and 10p, responses were 50% left or right even for last 10 trials]. fMRI:

  1. Medial OFC and adjoining pregenual cingulate cortex activity correlates with both reward magnitude and EV.
  2. No regions represented probability of reward separately from EV.
  3. Ventral striatum and inferior frontal gyrus activity correlates with reward magnitude not EV.
  4. Bilateral insula activity negatively correlates with EV.
  5. Nucleus accumbens correlates with reward prediction error and reward magnitude but not EV.


Williams, Rolls, Leonard, Stern BBR 55:243-52 1993

Neuronal Responses in the ventral striatum of the behaving macaque [Neuron, Attention, Reward]

Cue tone, then showed foods and drinks to monkey through an aperture, before allowing them to taste. Some items were orange juice/saline; simultaneously interspersed novel/familiar items.

Recording of 1000 cells in nucleus accumbens & olfactory tubercle:

  1. 54% unresponsive. 16% cue-related responses
  2. responsive to appetitive visual stimuli (14%): 1% to aversive items, 4% to food, 1.8% to items associated with orange juice, 4% respond only to sight of orange juice vs saline. <1% for all arousing stimuli.
  3. response to interesting visual stimuli: 3% to novel>familiar, 1% to familiar , 5% to general interest, 1.7% face. 13% respond to peripheral visual and auditory stimuli
  4. 4.5% movement related, 6.8% to somatosensory stimuli. 1.5% task related but non discriminating.
  5. 5% during feeding

Interpretation: ventral striatum allows limbic structures (amygdala, hippocampus) and mesolimbic dopamine to affect output pathways, via ventral pallidum→STN→ventral thalamus. Compare with putamen (nigrostriatal dopamine→neostriatum). Segregation of function within basal ganglia.


Rangel, Camerer, Montague NRN 9:7:545-56 2008

A framework for studying the neurobiology of value-based decision making [Theory, Reward, Decision]

Stages: identifying internal state or external state, identify possible courses of action, assign each action a value, then compare values, perform action, then feedback learning.

‘Pavlovian’ systems – automatic, small number or responses, specific, and needs direct reinforcement. ‘Habit’ systems – assign value to each S-R association, using generalisation, context sensitive. ‘Goal directed’ system – computes outcomes and assigns value based on outcome.

Suggest dorsomedial striatum has role in learning/expressing action-outcome associations

OFC encodes outcome-value associations

Reviews literature on signals predicted by prospect theory, expected utility and moment-based signals, in insula, amygdala, OFC, ACC, ventral/dorsal striatum, and parietal cortex.

Temporal discounting: discusses dual-process model vs. single exponential/hyperbolic process.

Extension of race/diffusion models to value-based choices.

Feedback signals: positive medial OFC, negative ACC, insula

Models of learning: Q-learning, actor-critic learning, SARSA.


Bayer HM, Glimcher Neuron 47:1:128-141 2005

Midbrain dopamine neurons encode a quantitative reward prediction error signal [Neuron, Reward, Learning]

Monkeys shown peripheral target for 4s, and could look any time. Saccade rewarded according to how late the saccade occurs, up to a given cutoff time. Cutoff time varies blockwise 400-4000 ms.

Results: On each trial, monkeys adjust RT compared to previous trial.

50 midbrain dopamine neurones recorded. Baseline firing 5/sec, phasic responses to unpredicted reward. Burst activity at the tone signalling start of trial. Then phasic increase up to 10/s, independent of reward contingency. Phasic reward response greatest after several unrewarded trials. Repeated rewards lead to low responses. If reward higher than previous trial, activity is much greater – modelled as dependent on reward history:


There is no correlation when firing rates below baseline – therefore dopamine neurones only signal positive reward prediction errors. Firing rates not correlated with response time. Also did not predict future timing: no relationship between reward-related dopamine signal and change in reaction time on subsequent trial. No correlation between reward-related signal and RT error.

Conclude: in rewarded trials, information is encoded about reward history, but not in unrewarded trials (they always fire at 0Hz); but animals change behaviour most drastically in unrewarded trials. ?serotonin is the opponent system


Daw, Kakde, Dayan Neural Networks 15:4-6:603-616 2002

Opponent interactions between serotonin and dopamine [Theory, Learning]

Review of pharmacological results.

Tonic serotonin reports long-run average reward rate, tonic dopamine reports long-run punishment rate, and phasic serotonin might report ongoing prediction error for future punishment.

Dopamine: approach, arousal, reward

Serotonin: aversion, withholding reponse, differential reinforcement of low rates of behaviour



Montague, Dayan, Sejnowski J Neurosci 16:5:1936-47 1996

A framework for mesencephalic dopamine systems based on predictive Hebbian learning [Theory, Learning]

Review of physiological recordings from VTA and SN during reward, spatial and temporal tasks (mainly Schultz). Dopamine responses are multimodal, reward-related, CS-related, and can be spatiotemporally specific to predicted reward.

Model: highly convergent input from cortex to a subset of DA neurones, plus salience signal, and contributes to temporal difference model.

Used to predict monkey spatial task behaviour, human card choice behaviour, and neurophysiological responses.


Wilkinson, Humby, Robbins & Everitt European J Neurosci 7:2042-52 1995

Differential effects of forebrain 5-hydroxytryptamine depletions on Pavlovian aversive conditioning to discrete and contextual stimuli in the rat [Lesion, Learning]

5HT-depleting lesions specifically impair contextual conditioning to aversive stimuli. (sound paired with shock after 5s or 30s)


Schulz, Dayan, Montague Science 275:5306:1593 1997

A neural substrate of prediction and reward [Theory, Reward]

Multiple lines of evidence: addictive drugs, electrical self-stimulation, dopamine blockade reduces reward learning. Nondiscriminating phasic reward/novelty responses, homogeneous over 60-80% of cells.

Same as 1996 model! Acknowledge does not address attentional functions of accumbens / frontal cortex. ‘There is evidence however to suggest that the required attentional mechanisms might also operate at the level of the dopamine neurones. Their responses to novel stimuli will decrement with repeated presentation and they will generalise their responses to nonappetitive stimuli that are physically similar to appetitive stimuli’.


Knowlton BJ, Mangels, Squire Science 273:5280:1399 1996

A neostriatal habit-learning system in humans [Disease, Learning]

Amnesic and nondemented PD were tested on probabilistic classification. Predicting the rain or shine based on combination of 1-4 of 4 different cards, which independently probabilistically influenced weather (75%, 57%, 43%, 25%). Immediate feedback with correct answer + correct/incorrect sound + score bar.

Result: Double dissociation in declarative memory of learning episode vs. classification-learning performance. HD same as PD. Severity-dependent impairment.

Interpretation: neostriatum-neocortex supports habit learning, limbic-diencephalic-neocortical supports declarative. [control task is not comparable in difficulty]


Rushworth MFS & Behrens TEJ Nature Neurosci 11:389-97 2008

Choice, uncertainty and value in prefrontal and cingulate cortex [Theory, Decision]

Good review of cellular and imaging literature.

‘when the animal is uncertain about the environment, new information is more valuable’:

ACC is involved in tracking volatility of environment, and governs learning rate (a in temporal difference model). More active when information is being actively acquired. May allow sacrificing income for information.

‘OFC maintains a rich and detailed representation of various features of a potential reward’ – time, context, magnitude specific. But LPFC represents uncertainty when multiple options present.


Izquierdo A, Suda RK, Murray EA J Neurosci 24:34:7540-8 2004

Bilateral orbital prefrontal cortex lesions in rhesus monkeys disrupt choices guided both by reward value and reward contingency [Lesion, Reward]

Monkeys taught to discriminate pairs of visual objects, associated with food type 1 or 2. Took 3 years!

  1. selective satiation to food 1 or 2: measured change in choices of object
  2. selective satiation with no objects, just choice of foods: measured changes in choice of food. Effect of satiation:
  3. willingess to work for food task, before and after satiation. Monkey had to perform exponentially increasing numbers of repeated actions to obtain food, until giving up.
  4. object reversal learning: make many more errors but still learn slowly.


Roesch MR, Taylor AR, Schoebaum Neuron 51:4:509-20 2006

Encoding of time-discounted rewards in orbitofrontal cortex is independent of value representation [Neuron, Reward]

Trial-by-trial rats either forced to left well, right well, or given a choice of both. Odor precue for trial type. Blockwise manipulation of pre-reward delay and reward magnitude for left or right well (4 blocks). Results:

  1. choice exhibits preference for larger rewards (85% of the time), and preference for shorter delay (74% of the time)
  2. Reversal learning within 10-50 trials
  3. activity in OFC in increased during reward (53% of cells), is direction-sensitive, and modulated by time-delay (41% fire more for short delays), but not by reward magnitude.
  4. 17% of OFC cells increase firing in expectation of delayed reward
  5. OFC responses to cue do correlate with reward-magnitude

Conclude that: OFC does not encode value of discounted rewards in some common currency, at a single cell level. ? demonstrates different roles of OFC in guiding behaviour vs. learning.


Tobler PN, Fiorillo, Schultz Science 307:5715:1642 2005

Adaptive coding of reward value by dopamine neurones [Neuron, Reward]

  1. Visual cue specifies reward probability and magnitude. Awake macaques, recording from dopaminergic cells during cue. Activity correlates with expected reward. Cells’ sensitivity to reward magnitude is correlated with sensitivity to probability (?EV). Also respond to unpredicted rewards.
  2. Reward larger or smaller than predicted causes increased activity or suppression, respectively; no response when reward is exactly as expected. Thus a positive reward can still suppress firing.
  3. Scaling is relative to ratio not absolute difference between reward and expectation. Described as ‘adaptation’.

Tobler, Fiorillo, Schultz Science 307:5715:1642 2005

Adaptive coding of reward value by dopamine neurons

Reward cue indicates probability and magnitude of reward. Cells respond to a combination. Responses to the same reward can increase or decrease from baseline depending on expectation.

“the more sustained activity of dopamine neurons reflects a measure of reward uncertainty such as variance” ... this “may be achieved by subtracting the expected value from the absolute reward value and then dividing by the variance”




Yoshida W, Ishii S Neuron 50:5:781-9 2006

Resolution of uncertainty in prefrontal cortex [fMRI, Decision]

fMRI during maze navigation reveals uncertainty in the face of sequential decisions to reach a goal. 3D wireframe maze, only locally observable, in which subjects did not know where they were in a familiar map, and had to find a goal shown on the map beforehand (goal-search task). Control condition was subject making same movements to same stimuli, according to arrow on screen.

Used incremental Bayes HMM to estimate subject’s trial-to-trial belief of where they are. ‘Hiddent current position entropy’ (computational load for position estimation) and ‘backtrack probability’ were used as fMRI regressors. Results:

  1. RT spikes correlate well with estimated backtrack locations.
  2. Medial PFC active for goal-search > control
  3. Backtracking correlates with activity in medial PFC, MFG, bilateral PPC
  4. Current position entropy correlates with bilat MFG and SFG
  5. RT and distance to goal did not predict brain activity


Tanaka SC, Doya, Okada et al Nature Neurosci 7:887-93 2004

Prediction of immediate and future rewards differentially recruits cortico-basal ganglia loops [fMRI, Reward]

‘Markov decision task’: visual cue 1 of 3 shapes, then subjects choose left or right keypress, which delivers a cue-dependent reward, then cycles left or right to a different cue. Reward schedule: 4 block types (each 15 trials long)–

  1. ‘Long’ where optimum is always press right, getting two penalties then a large reward,
  2. ‘Short’ where optimum is to always press left, collecting multiple small rewards,
  3. ‘Random’ where small rewards or penalties occurred equiprobably, and
  4. ‘Zero’ with no reward.

r1=20 and r2=100 such that for optimal behaviour EV=60 in Long and Short, otherwise EV=0

Short > Zero: Lateral OFC, insula, striatum, medial cerebellum

Long > Short: VLPFC, insula, DLPFC, dorsal premotor cortex, IPC, striatum, GP, dorsal raphe nucleus, lateral cerebellum, posterior cingulate, STN

Long > Zero: union of the above areas – i.e. both future and immediate reward predictions

Activations greater in first two blocks (active trial and error)

TD learning model: reward prediction signal represented striatum only. Depends on discounting factor g for different times after current trial. Used reward predicted by different g as regressor, with striatum as ROI. Found ventroanterior-dorsoposterior gradient of peak activity with g.

Suggest: lat OFC and ventral striatum predict immediate reward, and dlPFC and dorsal striatum are involved in future reward prediction – i.e. in parallel. Dorsoposterior (granular) vs ventroanterior (agranular) insular cortex similarly. Dorsal raphe serotonin controls effective time scale of reward prediction, and is regulated reciprocally by medial PFC.


Duann, Ide, Luo, Li J Neurosci 29:32:10171-9 2009

Functional connectivity delineates distinct roles of the inferior frontal cortex and presupplementary motor area in stop signal inhibition [fMRI, Conflict]

Motivation: previous stop signal tasks shows IFC activity correlates with subjects’ faster SST. This could simply reflect attentional processing of stop signal, rather than inhibition.

Staircase procedure to find 50% success in stop signal task.

fMRI Grainger causality analysis: . Psychophysiological interaction.

Kawagoe, Takikawa, Hikosaka J Neurophysiol 91:1013 2004

Reward-predicting activity of dopamine and caudate neurons – a possible mechanism of motivational control of saccadic eye movement

1-direction-rewarded memory-guided saccade task.



Hauber W & Sommer S Cer Cor 19:10:2240-7 2009

Prefrontostriatal circuitry regulates effort-related decision making [Lesion, Reward]

Rats: stereotaxic lesion to nucleus accumbens or anterior cingulate. T-maze task. Large reward (4 pellets) requires climbing over 30cm barrier, vs small reward (2 pellets). 16d training (till 80% choose large reward), 3 days testing, then lesions, then 9 days testing.

Result: no effect of unilateral ACC or accumbens lesions, or of ipsilateral ACC+accumbens lesions. Impaired ‘effort-based decisions’ (i.e. fewer choices of large reward when the barrier is only on that side) with both bilat accumbens and contralateral ACC+accumbens lesions.

Postulate: mental effort discounting = basolateral amygdala → dopaminergic D1 modulation of ACC → accumbens


Paulus, MP Science 318:5850:602 2007

Decision-making dysfunctions in psychiatry – altered homeostatic processing? [Theory, Decision]

Suggest that decision-making problems in psychiatric illness result from ‘dysregulation of homeostatic balance’ rather than an information processing dysfunction.

  • Risk-taking behaviour is not maladaptive, rather it is due to need for high-variance reward when organisms are in a precarious balance.
  • Drug use problems due to different interoceptive afferent signals (i.e. related to somatic markers), rather than processing differences.
  • Depression impairs performance of IGT (Must A. et al, J Affect Disord 90:209 2006) – reduced activity in ventral striatum: interpret as impaired reward processing.
  • Anxiety patients behave differently to ambiguous stimuli (Blanchette I, Richards A, JEP-G 132:294 2003) – ACC and medial PFC overresponsive to errors; increased ant insula activity in neurosis and harm-avoidance: interpret as increased top down modulation of afferents biasing towards aversive outcomes.
  • OCD: decreased responsiveness in rt medial & lateral OFC. Assoc with higher response in rostral ACC to errors in IGT. ? DL/OFC/ACC disconnected from amygdala/BG: interpret as alteration in reward history and valuation of options
  • Schizophrenics have normal IGT but impaired risk judgement. Parietal cx more in assessment of uncertainty, less in success-related processing. More heterogeneous.


Daw, O’Doherty, Dayan, Seymour, Dolan Nature 441:876-9 2006

Cortical substrates for exploratory decisions in humans [fMRI, Decision]

Old hypothesis: Information from exploration is commensurate with reward from expoitation. (Kakade & Dayan, Dopamine: generalisation and bonuses, Neur Net 15:549 2002).

4-armed bandit each with different mean payoff, which vary trial-to-trial on random walk. 1<reward<100, drawn from Gaussian mi,t, so=4, mean changes each trial by Gaussian sd=2.8 (weighted towards 50 points). 300 trials in 14 subjects, button box choice in <1.5s, reward after 3s, lasting 1s, poisson ITI mean 2s. fMRI SPM of correlation with model parameters.

Results: Subjective 11/14 were ‘occasionally trying the different slots to work out which was best’. RT~430ms. 3 theoretical choice strategies:

  1. e-greedy: choose best action, or occasionally a random action
  2. softmax: relative expected value determines choice of explore/exploit
  3. uncertainty bonus, i.e. expected information + expected value.

Used Kalman filter (Bayesian TD-learning estimate of m + uncertainty tracking for so,sd, l, q), calculated likelihood of subjects’ choices given history. Softmax is best; parameters= prediction error (ventral + dorsal striatum), obtained reward (mOFC), probability of actually chosen action ~= expected reward (+medial and lateral OFC, extending to vmPFC, -- DLPFC). Exploratory>exploitative = right anterior frontopolar cx, bilat ant IPS.

Intepretation: exploration by overriding exploitation. Inconsistent with common currency hypotheses, and with dopamine phasic response to information and novelty. ?regulation of exploration by NA


Botvinick MM, Huffstetler, McGuire Cogn Affect Behav Neurosci 9:1:16 2009

Effort discounting in human nucleus accumbens [fMRI, Reward]

2AFC button press to visually presented coloured digits 12346789, 1.5s each. Yellow = parity judgement, Blue = magnitude </>5. Low-demand: blockwise task changes, High-demand: task alternates trialwise.

‘Probabilistic’ reward ‘$’ or ‘X’ at end of block, not dependent on speed, accuracy or block type. Subjects were told this! But framed as ‘pay’ so thought of as contingent on completing a block. Subjects told $20-50 net. Controls: instead shown ‘S’ or ‘K’ at end of block ‘for the information of the experimenter and irrelevant to your task, pay attention so that you don’t miss the next block’. Controls told $20 net.

Results: in both groups, high-demand blocks slower, less accurate, questionnaire rated as ‘more effortful’ and should be worth $1.89 vs $0.89. fMRI – high>low demand: dorsal ACC. $>X: accumbens, OFC, rostral medial PFC. Low>high demand: accumbens. No effect of block type in controls.



Aston-Jones G, Cohen JD Ann Rev Neuroscience 28:403-50 2005

An integrative theory of locus coeruleus-norepinephrine function – adaptive gain and optimal performance [Theory, Decision]

Effect of NA (local and paracrine): reduce spontaneous activity, enhance synaptically evoked excitatory or inhibitory activity. Main projections parietal, pulvinar, SC. Main inputs: subcortical, caudal OFC, ant insula, ACC.

Aston-Jones 1999: Tonic NA correlates U-shaped with arousal (high when awake, slow during slow-wave sleep, absent during REM-sleep). Correlates with lower response threshold, lower d’, wider RT distributions, more aborted trials. Phasic NA signal salient and arousing stimuli – target-related.

  1. filtering task-relevant events
  2. trade-off between decision complexity and efficiency: single-layered network can implement optimal diffusion model, multilayered networks introduce noise and delay.
  3. monitor task utility (from frontal lobe) and switch to exploration if dwindles

Monkey selective responding task shows selective phasic firing to target. Reversal of cues meanings gives reversal of firing pattern, and increased tonic firing during the period of mapping uncertainty. Time locked ~230ms before response.

‘Task processing is accompanied by rapid and dramatic pupil dilation, consistent with the occurrence of an LC phasic response’, ‘baseline (tonic) pupil diameter rose as the.. expected utility began to decline’.

LC microinfusion of clonidine (inhibit) vs pilocarpine (stimulate): in distractible monkey clonidine improves performance. In normal monkey, pilocarpine impairs performance.

Model: adaptive gain between decision layers. May be ‘important for sampling alternative behaviors and adaptively pursuing other tasks in a changing environment’

Yu & Dayan: Ach = estimate of expected uncertainty, NA = unexpected uncertainty (Bayes)



Laming D, Acta Psychologica 43:1990244 1979

Choice reaction performance following an error [Theory, RT]

Hick 1952: for easily discriminable targest, errors are faster than correct responses

Sequential probability-ratio-test model: (random walk between response A/B, walk probability given by probability ratio of targets) is inconsistent with this.

Rabbitt (1966) post-error improvement despite no feedback. Error correcting RT: actually shorter than normal correct RT

Link (1975) relative judgement theory: walk probability is asymmetrical (extra degree of freedom)

Laming (1968) the asymmetry is caused by change in prior, not response to target. Post-error slowing ~30ms for long response-stimulus interval RSI >200ms, but ~250ms if RSI>200ms. If post-error stimulus is a repeat, ↓P(error) but ↔RT. If post-error stimulus is different, ↑RT but ↔P(error).

Reanalysis of post-error data: 1. response criterion to error action is stricter. 2. preparation beforehand: there is an effect on both types of action. Before a stimulus, choice of response A, B, or ‘not-yet’. Anticipatory sampling of input causes increase in fast-incorrect responses; errors prevent this anticipation.


Haruno...Doya, Kimura, Samejima, Kawato J Neurosci 24:7:1660-5 2004

A neural correlate of reward-based behavioral learning in caudate nucleus: a functional magnetic resonance imaging study of a stochastic decision task [fMRI, Reward, Learning]

Task: 2 boxes L/R. ‘Move a green disc into the target box’. Left or right button press. Movement depends on a 2x2x2 probability matrix. Rule 1: left=no movement, right=switch. Rule 2: 0.8 chance ends up in the direction pressed. Rule 3: left=switch with prob 0.7, right=switch with prob 0.3. Rule 4: random position whatever. Reward +5 or penalty -5. Control block required same movements, signalled by arrow.

Behavioural parameters calculated: Short-term reward (over 6 trials), accumulated reward, learning-rate-index (amount of change in behaviour from previous trials, as ), learning-convergence-index (difference from target performance).

Test>control: R1: bilat IPS, sup parietal cx, cerebellum, R2+3: also BG, premotor, OFC, Rt PFC. R4: also Lt PFC, Rt amygdala, Rt superior temporal.

Learning rate: caudate, GP, OFC, PFC occipital, rt parietal, premotor, temporal, Cbm.

Learning convergence: dorsal premotor, parietal, SMA, Rt Cbm.

Short-term-reward: Lt caudate, bilat occipital, parietal.

Accum reward: bilateral PFV, premotor parietal, occipital, SMA, Rt OFC

Conclude: ventral caudate implicated in learning.


O’Doherty, Dayan, Schultz J..Friston, Dolan Science 304:5669:452 2004

Dissociable roles of ventral and dorsal striatum in instrumental conditioning [fMRI, reward, Learning]

Choose one of two objects, left or right button press, 2s gap, then fruit juice vs water reward. Reward trials: choose A or B, A gives reward probability 0.6, B probability 0.3.

Neutral trials: choose C or D, reward probability 0.

Control task: had reward and neutral trials, but computer made selection, subject have to indicate by button which object had been chosen.

Parameters calculated using advantage learning. Result:

  1. in reward trials Subjects choose high-reward object more.
  2. RT effect even in control task → Pavlovian learning occurred.
  3. fMRI: reward>neutral in control (Pavlovian) condition: ventral striatum
  4. reward prediction error in control condition: ventral striatum
  5. reward prediction error in operant task: accumbens, ant. caudate


dorsal striatum = actor: uses temporal difference prediction error to modify stimulus-response-reward associations, used for choice of action.

ventral striatum = critic: uses prediction error to update reward predictions.


Matsumoto, Suzuki, Tanaka Science 301:229-32 2003

Neuronal correlates of goal-based motor selection in the prefrontal cortex [Neuron, Decision]

Task: Recording from monkey medial PFC during visually cued, asymmetrically rewarded go/nogo task with reversals. 1 of 2 cues, followed by 0.5-1.5s delay, then change of fixation spot, followed by response – either pulling or just holding the joystick, depending on the cue. Then 0.5-1s delay, then reward if successful go (or nogo, in different blocks). 4 block types: cue mapping to go/nogo x reward for correct go/nogo. →Independent manipulation of visual, motor and reward.

Result: fixation broken more in reward– trials. Cell types:

  1. 25% medial and lateral cells respond to whether reward possible (+) or impossible (–), in response to cue, but not to task type or cue identity (‘R’)
  2. 11% medial respond only to one combination of reward and task type (e.g. unrewarded nogo), after the cue. Equal numbers for each of the 4 combinations. 15% lateral cells have this interaction but only during the late period.
  3. These interaction cells have increased activity at start of block during learning.

Conclude: ‘medial PFC contains memory for the action-outcome relation and... selects one out of multiple attempted actions based on the current task contingency’


Walton ME, Devlin, Rushworth Nat Neurosci 7:1259-65 2004

Interactions between decision making and performance monitoring within prefrontal cortex [fMRI, Executive]

1 of 3 cues, each maps to 1 of 3 button-press responses. 3 different S-R mapping rule sets to be learnt. Correct/incorrect feedback. After 12 trials, instructed ‘stay’ or ‘switch’, and on this trial, either

  1. subject chooses finger, and needs monitoring (‘GUESS’)
  2. experimenter instructs next fingerpress, but needs monitoring (‘FIXED’)
  3. subject chooses finger, and is told it will always be correct (‘GENERATE’)
  4. experimenter instructs next fingerpress, and subject is told it will always be correct (‘CONTROL’)

Switch cost (ms) high in Guess and Generate trials. In the Fixed and Control, RT higher on the following trial.

Switch>stay: dorsal ACC active when choosing responses, lateral OFC when response is given but result needs monitoring. dACC activity is high in Guess>Generate>Fixed>control


Rushworth, Behrens, Rudebeck, Walton TICS 11:4:168-76 2007

Contrasting roles for cingulate and orbitofrontal cortex in decisions and social behaviour [Theory, Decision]

  1. OFC lesions cause aggression, and reduced emotional responses e.g. fear of snakes.
    ACC lesions cause reduced interest to social stimuli e.g. other monkeys, but not fear stimuli.
  2. OFC lesions impair learning/generalisation to conditioned stimulus in rats
  3. in humans, ACC activated in interactive games e.g. trust game, prisoners’ dilemma (Amodio & Frith 2006)
  4. both ACC/OFC connect to ventral striatum and amygdala, but OFC receives sensory input from temporal lobe, whereas ACC receives motor and spatial input from premotor and parietal.


OFC important when reinforcement is associated with stimuli and for representation of preferences. Detailed outcome prediction & model of reinforcement environment.

ACC reward representation closely bound to action or task representation. Action history→action choice. Accounts for costs intrinsic to an action (effort).


Stalnaker, Franz, Singh, Schoenbaum Neuron 54:1:51 2007

Basolateral amygdala lesions abolish orbitofrontal-dependent reversal impairments [Lesion, Reward]

(See also: Two wrongs make a right: deficits in reversal learning after orbitofrontal damage are improved by amygdala ablation)

Rats learn to feed to cue A, and avoid B. Normal reversal learning is prevented by OFC lesions, and rats continue to try and feed to A, and avoid B. Basolateral amygdala lesions restore reversal learning.


Jones B, Mishkin M Experimental Neurol 36:362 1972

Limbic lesions and the problem of stimulus-reinforcement associations [Lesion, Reward]

Action selection according to rewarded location or rewarded object.

Monkey lesions to OFC, temporal pole of amygdala, fusiform-hippocampal gyrus + hippocampus.

  1. Fusiform/hippocampus causes impairment of place reversal learning.
  2. Amygdala & OFC impair reversal of object learning.
  3. OFC is involved later in learning than amyg – requires pre-existing tendencies.

Stage I: retaining previous pairings. II: respond at chance. III: acquired new pairings.


Hornak, O’Doherty, Rolls... JCN 16:3:463-78 2004

Reward-related reversal learning after surgical excisions in orbitofrontal or dorsolateral prefrontal cortex in humans [Lesion, Reward]

2AFC (same as O’Doherty et al., 2001) of 2 patterns, probabilistic reward either +ve or –ve expected value. Subjects told beforehand that a reversal would occur gradually.

Results: DLPFC could not answer the question “Which information was the most useful in choosing best response?” – denied that it was the amount won or lost. E.g. initially attended to amount, but then “other things seemed more important”, or noticed whether or not they had won but “paid very little attention to the amount”. Notably bilat OFC patients did attend to the correct information.

No problems with acquiring contingencies initially. Impaired performance in bilateral OFC and the DLPFC patients who failed to notice which stimuli were important. Medial PFC & unilateral OFC were unimpaired.


Rushworth, Mars, Summerfield Curr Opin Neurobiol 19:1-9 2009

General mechanisms for making decisions? [Theory, Decision]

Lateral/OFC: stimulus-reward; medial/ACC: action-reward (Glascher et al., 2008)

Lateral/OFC: negative reinforcement; Ventromedial: positive reinforcement (Kringebach 2005)

Medial PFC may be involved in making decisions, not just representing values!

Although striatal activity correlates with prediction error, this is not seen in single cell studies.



Plassmann, O’Doherty, Rangel J Neurosci 27:9984-8 2007

Orbitofrontal cortex encodes willingness to pay in everyday economic transactions [fMRI, Reward]

Hungry subjects bid for familiar foods in a Becker-DeGroot-Marshak auction (if bid >= randomly chosen value $0-3, they win). $3 spending, can keep whatever unspent, 100 trials, instructed that after expt, 1 trial is selected randomly & they would only be allowed to eat the food they bought for 30 mins. (i.e. each auction is independent). Instructed to ‘pay only what the item is worth’. [conflating utility (hunger) and worth?]. 50% trials ‘forced bid’ – computer instructed what to press.

Result: regress freedom, bid, surplus . Right mOFC, dACC and DLPFC correlate with willingness to pay in free trials.

mOFC and DLPFC both satisfy free>forced.


Rudebeck, Behrens, Kennerley..Walton, Rushworth J Neurosci 28:13775-85 2007

Frontal cortex subregions play distinct roles in choices between actions and stimuli [Lesion, Reward]

Monkeys lesioned in OFC or ACC.

  • Expt 1: 2AFC between 2 actions, rewarded or unrewarded, with action reversal. ACC lesions do worse after correct trials than error trials: inability to use positive reinforcement.
  • Expt 2: 2AFC between 2 actions, probabilistic high/low reward 1:0, 0.75:025, 0.5:0.2, 0.4:0.1, and 0.5:0.18, but if a reward was allocated to a particular action on a given trial, it stayed available until the macaque selected that action → optimal strategy may involve switching. ACC monkeys cannot perform this.
  • Expt 3: 2AFC between 2 stimuli, probabilistic high/low reward as above. Then repeated using different reward magnitude for each stimulus. OFC monkeys take 50 trials more to learn.

  • Expt 4: 3AFC between 3 stimuli, bandit-style with trial-by-trial payoff changes. OFC animals fail to track probabilities [but did not analyse exploratory behaviour].

‘If reward delivery is stochastic, then the best choice is dependent on the integrated history of reinforcement rather than just the last outcome.’ i.e. it’s a harder task. [but more complex than simple probability estimation]


Hare, O’Doherty, Camerer, Schultz J Neurosci 28:5623-30 2008

Dissociating the role of the orbitofrontal cortex and the striatum in the computation of goal values and prediction errors [fMRI, Reward]

Manipulate food preference, purchase price, and random bonuses.

Hungry subjects get to eat only what they buy after expt. Given $5 to spend/keep. 300 trials, 2afc whether or not to buy a familiar food item, given a displayed price (random from -3 to 3, only if they choose the item) and a gain/loss (-3 to 3 irrespective of choice). Additionally, 13% trials had food only with no price or bonus, 13% no food or bonus just price, 13% no food or price just bonus.

Becker-DeGroot-Marschak auction beforehand to measure willingness to pay = ‘goal value’.

Decision value = willingness to pay – price at which item is offered. – net benefit of purchasing an item [cf. effort]

Prediction error = the random bonus [what does this have to do with prediction error?]

fMRI Result: Goal value: mOFC, mPFC, amygdala. Decision value: cOFC. Prediction error: ventral striatum.


Amodio DM, Frith CD NRN 7:267-77 2006

Meeting of minds: the medial frontal cortex and social cognition [Theory, Conflict]

Review of fMRI in social tasks.

  • Monitoring action = ACC, monitoring reward/punishment = OFC, social cognition = ‘self-reflection, person perception, inferences about others’ thoughts’ = MFC + paracingulate cortex
  • 2 anatomical axes: round along cingulate, and crossing radially. Caudal MFC: cingulate hand/eye/mouth motor areas. Posterior rostral ACC: attention and error monitoring. Anterior rostral ACC: emotional ratings. Subcallosal ACC: visceral emotional responses.
  • Posterior rostral ACC correlates with: higher target-distractor conflict, larger penalty, reward variance, error probability prediction, regret (fictive reward), racism. Monitoring of self-selected>externally-guided action.
  • Orbital MFC: value of possible outcomes, not actions. correlates reward magnitude, regret. Active even for externally-guided action.





Wittmann, Bunzeck, Dolan, Duzel Neuroimage 38:1:194-202 2007

Anticipation of novelty recruits reward system and hippocampus while promoting recollection [fMRI, Memory]

5 images previewed 8 times for 1500ms. Cue predicting image novelty or familiarity, 75% valid, 1500ms then delay 0-4.5s, then picture 1500ms. 480 trials, 50% were novel. Button responses to familiarity, counterbalanced.

Subjects aware of probabilities. Separate experiement: old/new → remember/know/guess for old, sure/guess for new, for each picture; used 120 expected novel, 40 unexpected novel on day 1, then on day 2, these 160 pictures as ‘old’ and 80 further new [not as familiar though].

Results: ↑RT for unexpected and novel. fMRI:

Unexpected novel picture>expected novel: SN/VTA

Novelty-predicting cue>familiarity cue: SN/VTA

Novel picture>familiar picture: bilat hippocampus

Conclusion: dopamine novelty signal improves recognition memory


Bradley MM, Miccoli, Lang PJ Psychophysiology 45:602-7 2008

The pupil as a measure of emotional arousal and autonomic activation

Hess (1960): pupil constriction to aversive pictures, dilatation to pleasant. Unreplicable.

Steinhauser et al. (1983): pupil dilates to pleasant and unpleasant pictures.


96 pics to 27 subjects. Skin conductance=sympathetic, cardiac deceleration=parasymp.

Sympathetic activity to both pleasant and unpleasant pics. Parasympathetic activity to unpleasant pics only. Pupil correlates with sympathetics: with low arousal pics, pleasant pictures cause more constriction whereas unpleasant cause less. At high arousal, pupil does not change.

Steinhauer (2004): Mental load’s effect on pupil, though, is mediated by parasympathetics


Nobre, Coull JT, Frith, Mesulam Nature Neurosci 2:1:11 1999

Orbitofrontal cortex is activated during breaches of expectation in tasks of visual attention

Central precue predicts either time (300ms or 1500ms), spatial-location (L/R) or time+location of a target. Respond single right-hand button press to target. Cue validity was 100% or 60% in first and second block respectively (50% for uncued dimensions).

Result: all cuing conditions>baseline: activate frontoparietal network.

(100%>60% & 60%>baseline): rt lateral premotor, parietal foci, bilat ventral PFC around lateral orbital sulcus.


Freedman, Black, Ebert, Binns Cer Cor 8:18-27 1998

Orbitofrontal function, object alternation and perseveration [Lesion, Memory]

Wilson et al (1993): Monkeys store spatial WM in dorsal PFC, but object WM in middle, inferior and orbital frontal cortex.

6 bilateral frontal patients: ACA infarcts, leucotomy 39 yrs ago, traumatic contusion lobectomy, ACom surgery or bleed.

  1. Object alternation: find a penny in one of two wells; covered by objects. Reward-object pairing alternates after each correct response, but object location is random.
  2. Delayed alternation: same as above, but wells covered by identical lids. Reward-location pairing alternates after each correct response.
  3. Delayed response: experimenter places penny in 1 of 2 wells, cover, then response after 10-60s delay

Object alternation deficit correlates with area 10,11,24,32,47 – ventrolateral OFC. Others and WCST correlate with area 25, 33, 8.


Kawashima...Takahashi, Fukuda Brain Res 728:79-89 1996

Functional anatomy of GO/NO-GO discrimination and response selection – a PET study in man [fMRI, Conflict]

Drewe EA (Cortex 1975): human frontal lesions impair go-nogo learning.

Petrides (JN 1986): monkey periarcuate lesions impair go-nogo performance when nogo is also rewarded (symmetric), but not if nogo is unrewarded (asymmetric)

Response-selection task: Colour of go-signal indicates thumb- or finger-press. Followed by:

Go-nogo task: Colour of light signals thumb-press or nothing.

PET (go/nogo>control & go/nogo>response-selection): posterior lip of precentral sulcus, MFG lateral lip of superior frontal sulcus, Lt posterior insula, Rt SFG, IFG, bilateral ACC


Schmidt...Levy, Dubois, Pessiglione Brain 131:1303-10 2008

Disconnecting force from money: effects of basal ganglia damage on incentive motivation [Lesion, Reward]

Kirsch-Darrow (Neurology 2006): apathy is a feature of PD; Starkstein (2002) rating.

Jahashani...Marsden, Passingham (Brain 1995): self-initiated action vs externally guided action is impaired in PD

Compare 13 bilat striato/pallidal lesion patients with autoactivation deficit, 13 pre-DBS nondemented nondepressed PD withdrawn from dopamine for 24 hrs, control.

Instructed: ‘reach red line’ at 40%, 80%, 120% maximal, feedback ‘correct’/‘incorrect’. 4 trials of each force random order, executed before and after incentive trials.

Incentive: ‘the more you squeeze the grip the more you get’ of $1, $10 or $50, where maximal gives 50% of reward. 15 trials of each reward.

Results: MMSE in lesions was 24 (28 in PD). UPDRS 10 (32). Starkstein apathy 17 (8).

Apathy scores correlate with differential effect of monetary incentive on grip force.

Apathy in PD but preserved correlation of force with incentive: spared dopaminergic innervation to pathways from OFC/mPFC to ventral striatum and ventral pallidum.


Pessiglione, Seymour...Dolan, Frith Nature 442:1042-5 2006

Dopamine-dependent prediction errors underpin reward-seeking behaviour in humans [Drug, Reward]

2AFC abstract visual shapes A/B, C/D or E/F; button press selects upper, nogo selects lower. Probabilistic reward A=80% win, 20% nil; C=80% lose, 20% nil, BDEF give nil. L-dopa, haloperidol, placebo.

Results: L-dopa win more, lose the same. RT unchanged.

TD-model parameters: reward prediction error correlates +ve with bilat ventral striatum, left post putamen. Present in both gain and loss trials: avoided loss looks similar to reward. –ve correlation with Rt ant insula in loss trials only.

Gain>neutral: bilat ventral striatum; loss>neutral: bilat vent striatum + bilat ant insula.

Optimal top (Go)>Optimal bottom (nogo): left post putamen.

L-dopa enhances the prediction-error response to reward only.


Robbins, Everitt Curr op Neurobiol 6:2:228-36 1996

Neurobehavioural mechanisms of reward and motivation [Theory, Learning]

3 types of learning:

  1. Pavlovian
  2. reward driven associative action-reward pairing (depend on incentive state of anima),
  3. habit-like S-R associations – impervious to fluctuation in satiation

Salamone & Cousins (BBR 1994): dopamine depletion causes switch from effortful good food to effortless poor food, and reduces effort for lower densities of food.

Ventral striatum compartments – accumbens shell and core, for appetitive (dopamine) and consummatory (e.g. morphine) behaviour.


Redgrave P, Gurney K NRN 7:967 2006

The short latency dopamine signal: a role in discovering novel actions [Theory]

DA phasic signal: 70-100ms latency, 200ms bust duration.

  1. novelty, habituates rapidly
  2. sudden intense stimuli
  3. primary reward, attenuates when reward is predictable
  4. conditioned stimulus, and things physically similar to them.
  5. phasic suppression of DA (100ms) by nociceptors

         If reward value not known at time of a novel stimulus, how can it be reward prediction error?

         Rapid (presaccadic), homogeneous response to diverse stimuli → ?preattentive

         Decortication preserves, but collicular ablation abolishes → Tectonigral pathway.

         Identify which environmental stimuli are self-generated

         Novelty-inducing behaviour gets reinforced → exploration

         Allows discovery of which particular action /context causes a reward – ‘all short-latency signals are relayed to striatum to determine if they were caused by action’

         Post-gaze-shift evaluations must take place in amygdala, OFC


Gilzenrat, Cohen, Rajkowski, Aston-Jones SFN 2003

Pupil dynamics predict changes in task engagement mediated by locus coeruleus

2-interval forced choice, tone higher or lower, difficulty increases and reward+=5 each correct trial, reward-=10 if wrong. Between trials ‘accept’ or ‘escape’.

Pupil response to second tone: suggestion of decreased dilatation to tone as yield falls.

Pupil baseline before tones: gradually increases until escape, then falls over next 3 trials.

Pupil diameter correlates with expected value for next trial.


Fellows & Farah Cer Cor 17:2669-74 2007

The role of ventromedial prefrontal cortex in decision making: judgement under uncertainty or judgement per se? [Lesion, Decision]

Preference judgement task for food, people, colours. 6 items per domain, 2AFC with all possible pairs. Errors = nonassociative preference. Compared with perceptual judgements ‘which line is longer’, rate colours on ‘blueness’.

Compare vmPFC with DLPFC. Result: RT not impaired, but preference judgements are.


McClure, Li, Tomlin, Cypert, Montague Neuron 44:379-87 2004

Neural correlates of behavioural preference for culturally familiar drinks [fMRI, Decision]

Coca cola or pepsi in scanner. Labelled, semianonymous (one labelled, the other unlabelled but the same) or anonymous.

Find: VMPFC active for perceptual preference discrimination, and DLPFC/hippo/midbrain active for cultural information.


Huettel, Stowe, Gordon, Warner, Platt Neuron 49:5:765-75 2006

Neural signatures of economic preferences for risk and ambiguity [fMRI, Reward]

Choice between two roulettes with same expected value. Can be ‘certain’, ‘risky’ (probability of each reward known) or ‘ambiguous’ (possible rewards but not probabilities known).

fMRI ambiguity>risk: posterior IFS, ant insula, PPC, specific to decision phase.

(AR | RR) > (CR | CA) i.e. where there are two probabilistic alternatives: posterior IFS, specific to reward phase. Behavioural risk-taking and ambiguity preference estimated, correlate with risk (u=xb) and ambiguity effects (AC+AR-RC+RR) in pIFS. Barratt impulsiveness correlates negatively with ambiguity effect in pIFS.



Warren, Olanov, Sethi Neurology 72:S1-136 2009

The scientific and clinical basis for the treatment of Parkinson disease [Disease]

Old and new models of dopamine function

Ropinirole and Pramipexole are D2+D3 agonists, rotigotine and pergolide are D1+D2+. Bromocriptine inhibits D1. Cabergoline and lisuride are D2+ specific, and apomorphine is D123+.

They act directly: and not dependent on dopaminergic neurones being in tact, receptor specific. Reliable absorption, longer half life, no free radicals. Ergots activate 5HT2b.

Surgery: thalamotomy, pallidotomy, subthalamotomy; DBS ventral intermediate thal, GPi, STN, PPN.

Dementia hallucination delirium is leading cause of nursing home placement. Impulse control 7%. Sleep: 70%, insomnia, RLS, PLMS, akathisia, parasomnia (nightmare, somnambulism, terrors, vocalisations, hallucinosis, panics, RBD)


Kvernmo, Hartter, Burger Clin Therapeu 28:8:1065 2006

A review of the receptor-binding and pharmacokinetic properties of dopamine agonists [Drug]




































































56.2 (PAg)

1.2 (Ag)

7.1 (Ag)





















Dose (mg/d)






Tmax (h)






t (h)







Everitt & Robbins Nat Neurosci 8:1481-9 2005

Neural systems of reinforcement for drug addiction: from actions to habits to compulsion [Theory, Learning]

Striatum + amygdala:

  • Unexpected drug-CS pairs → dopamine in accumbens core.
  • Lesions of accumbens core impair Pavlovian approach responses (& consolidation of these responses, & pavolvian-instrumental transfer).
  • Lesions of accumbens core or amygdala increase choice of small immediate reward over large delayed reward (impulsivity)
  • CS can act as a reinforcer, but only when presented expectedly based on animal action.
  • Taste, smell and CS-mediated reinforcement → accumbens shell
  • Habit then develops in dorsomedial striatum-medial PFC, or dorsolateral striatum (food)
  • Flupenthixol in dorsal striatum but not accumbens abolishes established cocaine habit.
  • Agrees with ventral-pavlovian/dorsal-instrumental, actor/critic, reward-prediction/action model


  • Lesion prevents context-induced reinstatement of habit.
  • Theta-burst stimulation of HC reinstates habit; glutamatergic blockade at VTA prevents.
  • HC and amygdala compete – context vs. immediate cues.

Prefrontal: drug abusers lack inhibition; reduced OFC activity. Habit forming devolves control from prefrontal cortex to dorsal striatum.


Seymour, Daw, Dayan, Singer, Dolan JNeurosci 27:18:4826-31 2007

Differential encoding of losses and gains in the human striatum [fMRI, Reward]

Pavlovian conditioning: 5 abstract visual cues each predictively paired with 2 outcomes probability 50%: 0/0, -1/0, +/+ , +1/0, +1/+1. Subjects did nothing! Then preference measured – 2AFC for all possible pairs of cues, no reward.

Results: pupil dilates for loss or gain

‘positive-reward prediction error’: (win 1 for +1/-1) > (win 1 for +1/0): ant. VS activation

‘positive-loss prediction error’: (lose 1 for +1/-1) > (lose 1 for -1/0): posterior VS activation

‘negative reward prediction error’: (win 1 for +1/0) > (win 1 for +1/-1): no effect

‘negative loss prediction error’: (lose 1 for -1/0) > (lose 1 for +1/-1): no effect

TD model a=0.3, prediction error as regressor [assumes –ve loss is a gain] on bivalent trials only: TD loss prediction = posterior mid putamen, TD gain prediction = anterior ventral striatum.

Goudriaan, Oosterlaan...van der Brink CBR 23:1:137-51 2005

Decision making in pathological gambling: a comparison between pathological gamblers, alcohol dependents, persons with Tourette syndrome, and normal controls [Disease, Risk]

IGT (bechara/damasio ’94): has only reward or reward+loss trials.






Reward size





Loss Size





Loss frequency





High rewards high infrequent losses –ve EV, High rewards low frequent losses –ve EV, Low rewards high infrequent losses +ve EV, Low reward low frequent losses +ve EV.

‘Card-playing task’ (Newman 1987): deck of 100 cards, gives +1 or -1. Decide play or quit. Initially reward 1:9, decreasing every 10 cards, later majority loss. Optimum is to take 33-52 cards (giving~ +20). Known impairment in psychopaths, behaviour disorders.

‘Go-nogo discrimination task’ (Newman 1985): learn to discriminate between 8 2-digit numbers – go or nogo in 2s, 80 trials, ‘correct’ or ‘incorrect’ feedback. Reward condition: gain 1 for each correct response, vs Response cost condition: start with 50, each error costs 1.

Results: Gamblers are faster at IGT, don’t change deck after loss but do have post-loss slowing. Card-playing: pathological gamblers either stop too early or too late. Alcoholics stop too soon, Tourette’s stop too late. Gamblers and alcoholics have no post-loss slowing. Go-nogo learning: All groups are worse than controls and have post-reward slowing (reverse normal pattern), gamblers learn better with losses (reverse normal pattern). Slot machine gamblers worse than casino.


O’Doherty, Kringelbach, Rolls, Hornak.. Nature Neurosci 4:1:95 2001

Abstract reward and punishment representations in the human orbitofrontal cortex [fMRI, Reward]

Reversal learning, 2AFC abstract images; 50 trials 2 reversals before scanning, new stimuli in scanner. S+ 70% reward 80-250, 30% loss 10-60, S- 40% reward 30-60, 60% loss 250-600, uniform distribution. After 5–8 consecutive S+ responses, contingencies reverse. Control subjects had to simply select 1 predetermined stimulus of the 2, no feedback. [maybe optimum strategy is to purposely choose S- every 5 trials!]. Subjects always choosing S+ before the reversal (acquisition phase) and then persist with the same stimulus after revesal (now S-, reversal phase); 30% of the acquisition (S+) trials are losses, 40% of the reversal (S-) trials are rewarded.

  1. rewarded acquisition > punished reversal: mOFC, med PFC, L>R
  2. rewarded acquisition > punished acquisition: mOFC, med PFC
  3. punished reversal > rewarded acquisition: lat ant OFC, Rt VLPFC, dACC, R>L
  4. punished acquisition > rewarded acquisition: rt posterior inf prefrontal sulcus, dACC
  5. rewaded reversal > rewarded acquisition: rt posterior inf prefrontal sulcus, dACC
  6. (rewarded acquisition or reversal) > control: mOFC
  7. (punished acquisition or reversal) > control: lat OFC

Conclude: mOFC responds to receipt of reward correlates with magnitude. Lat OFC responds to loss, correlates with magnitude. [4 should be neg prediction error].

Inferior prefrontal sulcus: ambiguity in whether reward/penalty signals contingency changes; may be same as nogo-inhibition region or set-shifting area in WCST?

Iverson & Mishkin (1970) monkey lat OFC lesions persist responding when reward is stopped. med OFC lesions difficulty in acquiring associations, do not perseverate.


Rolls ET Cer Cor 10:3:284-294 2000

The orbitofrontal cortex and reward [Theory, Reward]

Big review

  • Monkey lesions: Errors of commission of go-nogo task (inferior convexity). Reversal and extinction tasks perseverate on previously rewarded objects (caudal area 13). Decreased object WM in DMTS/DNMTS (inferior convexity/lateral; cf DLPFC spatial WM). Altered preferences and emotions (caudal).
  • Physiology: Taste and smell. Visual responses to predicted taste reward (shown by S-R reversal experiment). Face responses – face, movement and gesture selective ?expression. Somatosensory response to oral and hand pleasant/aversive touch.
  • Human lesions: WCST deficit in both acquiring initial principle, and shifting. Stylus maze - difficulty changing direction in response to feedback sound. Can verbalise rule but cannot correct behaviour. Euphoria, flat affect, indifference. Rewarded visual discrimination reversal learning. IGT – insensitive to rare heavy penalties. Facial expression discrimination (except happy). Voice discrimination.
  • ‘disinhibited, socially inappropriate, misinterpretation of other peoples’ mood, impulsiveness, unconcern for their own condition, lack of initiative’ (Rolls 1994)

Theory: WM continuous attractor networks input from temporal lobe, plus rapid synaptic S-R learning from limbic structures. Output maybe to ventral head of caudate and ventral striatum. Amygdala does not learn quickly, if at all.


Presuschoff, Bossaerts, Quartz Neuron 51:3:381-390 2006

Neural differentiation of expected reward and risk in human subcortical structures [fMRI, Risk]

Manipulate risk orthogonally to expected reward: variance = p (1-p). Betting task: 2 random cards drawn value 1-10, 2AFC ‘second card higher’ or ‘lower’. Win +1 or -1, first card then second card shown, then ask to report whether they won or lost. (penalty for misreporting, so that subjects have to notice winning!)

On placing bet, reward probability is 50%. (no strategy possible) After seeing card 1, this becomes a linear function of the card value.

After reward > after loss: caudate, GP, thalamus, putamen, midbrain, cingulate. Loss>win nil

Early correlation with expected reward probability after first card: putamen, VS, GP, ACC, midbrain, medial orbital gyrus, medial temporal gyrus

Late correlation with risk after first card: VS, STN, MDthal, midbrain, ant insula. (Some early response in ant insula and OFC.)

Overlap between probability and risk: left VS.

No evidence of switching after a loss, & no correlation with prediction error (→ no learning).

Conclude: distinctly represented in time and space, but may overlap.

Zink 2004: dopaminergic activation of basal ganglia may represent salience


Iriki, Tanaka M, Iwamura Neurosci Research 25:173-81 1996

Attention-induced neuronal activity in the monkey somatosensory cortex revealed by pupillometrics [Autonomic, Attention]

Pupillary dilatation observed in:

Hess & Polt 1960, 1964, 1972 – attractive photos, mental calculations

Kahneman & Beatty 1967 – auditory discrimination

Richer & Beatty 1985 – complex finger movements

Paivio & Simpson 1966 – creating an abstract image

Jampel 1960: stimulation of dorsal limb of arcuate sulcus (FEF and dorsally adjoining premotor area) and occipitotemporal area of prelunate gyrus → pupil dilatation

Red light warning < 12s. Once monkey’s hand stationary over button, light turns green. Button press delivers reward.

  • Pupil dilates 20% during red. Dilates more during green. Returns to 0% after green.
  • 79% S1 cells had glabrous skin tactile RF or finger joint position sense
  • 12% cells showed exitation – insensitive to passive touch before red light, become responsive during red light. 10% inhibition – touch-sensitive responses before and after light, but muted during the lights. Posterior cells also fired in response to light.
  • More cells had tactile RFs during the lights. Modulation of baseline activity and sensitivity to tactile stimulus was greatest in in posterior region (anterior bank of IPS)
  • Through the day, success drops from 100% to 0% (250 trials), and dilatation drops from 40% to 10%. Note it goes up to 50% just when start to make errors.
  • Cell excitation/inhibition correlates better with pupil dilation than with the light.

Interpret as subthreshold modulation. Possibly mediated by pulvinar (input from periarcuate cortex, output to postcentral gyrus). Light responses could be ‘corollary discharges’ from preparatory motor cortex activity.


Satterhthwaite, Green, Myerson...Buckner Neuroimage 37:3:1017-31 2007

Dissociable but inter-related systems of cognitive control and reward during decision making: evidence from pupillometry and event-related fMRI [fMRI, Decision]

One card face-up, one face down, randomly on L & R. 2afc ‘which will be higher?’ for 2s, with fixation cross; then button response L/R.Win/lose 50; no total score. $1 penalty for not responding in 2s. 13 values, never equal. [fixed reward, variable probability. Risk and EV are completely conflated]

First card 1,13=certain (n=20) / 234,10,11,12 =83% probable win (n=120) / 56789 =60% uncertain win (n=60). There were more ‘probable’ trials, so as to give enough ‘probable but lose’ trials. → total reward 78%.

Results: Subjects almost always selected correctly [so is there any sense of choice or risk?] but took 1100ms for ‘8’, 740ms for ‘1’ & ‘13’. [note there are no ‘unexpected wins’ to compare]

Pupil dilates before uncertain decision and after unexpected loss. ?’cognitive effort’ (Kahneman 1973) or conflict (Kerns 2004), ?prediction error

Uncertain>probable:PFC, pre-SMA

(Uncertain > certain & Lose > win): PFC, dACC, ant insula

Win>loss: dorsal striatum

Conclude: distributed frontal network for cognitive control, gated by dopamine which is phasically suppressed by unexpected negative outcomes.


Critchley, Mathias, Dolan Neuron 29:2:537-45 2001

Neural activity in the human brain relating to uncertainty and arousal during anticipation

GSR and fMRI [fMRI, Risk]

Higher-lower 2-card game. 1-10, no ties. 1s cue card, then higher-lower button press, then 8s delay, then feedback card 1s. No other feedback signals, told beforehand +50p win, -1 lose.

Results: Post-decision, Pre-feedback: bilateral ventral/medial PFC (OFC, ant insula, IFG), lateral anterior temporal lobe, right DLPFC, inferior parietal lobe. Suggest WM of card and response, vigilance/anticipation for feedback, cumulative progress, non-specific eye movements.

  • Correlation with uncertainty=risk: bilat ACC, lateral OFC
  • Correlation with GSR (4s delay): right DLPFC, right ACC (arousal)
  • Bilateral ACC regions correlate with both risk and GSR in all subjects. No interaction.

Conclude: Ant cingulate integrates uncertainty with adaptive changes in bodily state & arousal.



Lebreton, Jorge, Michel.. Pessiglione Neuron 64:431-9 2009

An automatic valuation system in the human brain: evidence from functional neuroimaging [fMRI, Decision]

VAS rating of pleasantness or age of pictures from different catevories (face, house, painting) during fMRI. Afterwards 2afc ‘preferred’ easy (ratings 50% apart) or hard (adjacent ratings). Done both immediately and 1 month after scanning. Pictures chosen to ‘allow considerable intersubject variability in preference’

Ratings: intersubject SD =~ interstimulus SD. Agreement between subjects in 2afc ~60% hard, 70% easy. Prediction of 2afc from rating (within subject) ~ 71% hard, 75% easy even after 1 month. 2afc RT correlates negatively with distance between ratings of two options.

Face>(House | Painting): temporoparietal junction, inferior temporal cortex, caudate tail. House: medial occipital, medial temporal , medial parietal lobe

Paintings: medial and dorsal occipital

Correlate with pleasantness rating but not age estimate: VMPFC, OFC, ACC, VS, amygdala, hippocampus, posterior cingulate, V1. Correlate with age not pleasantness: nil.

(Pleasantness > age judgement) & (pleasant > unpleasant as judged in easy 2afc afterwards): vmPFC, VS, HC, post cingulate, V1.

(pleasant>unpleasant as judged in 2afc) activations same for high-agreement stimuli and low-agreement stimuli, and for all stimulus categories. Significant both in age-estimation and pleasantness-rating tasks.

Conclude: “brain valuation system,” active in explicit and non-explicit tasks. Reflects both ratings and binary choice preferences.


Critchley Mathias...O’doherty Cipilotti Shallice Dolan Brain 126:10:2139-52 2003

Human cingulate cortex and autonomic control: converging neuroimaging and clinical evidence [fMRI, Executive]

Expt 1: Cognitive: 1-back or 2-back task, 1s consonant letters, 1s ISI, handgrip response to repeats. Motor: maximal handgrip squeeze for 6s or 11s, then equal rest period.

Blocks interleaved, instructed at start of each block.

Heart rate: variability regressor; high-freq and low-freq RR-interval orthogonalised regressors; individual trial type regressors. Results:

  • fMRI Correlation with HR variability independent of cognitive & motor: dACC, mOFC, insula, hypothalamus.
  • Correlation with HR variability & (cognitive>motor): OFC, retrosplenial, lateral parietal cortex.
  • Correlation with Low freq (sympathetic): bilat dACC, insula, hypothalamus, inf parietal, somatosensory cx. High-freq (parasymp): left somatosensory cx, precuneus, left dorsal cingulate, left ant insula, cerebellar vermis.

Expt 2: 3 bilat ACC patients (2 tumours, 1 traumatic bleed), ‘normal speed & attention’ but 2 pts worse on frontal tests. Tests: maximal grip strength, serial 7’s from 400.

Result: Focal ACC lesions give blunted HR responses to effortful cognitive task (consistent) and motor task (inconsistent).

Conclude ACC mediates contex-driven modulation of arousal


Micieli, Tassorelli...Magri, Nappi Clin Autonomic Res 1:55-58 1991

Disordered pupil reactivity in Parkinson’s disease [Disease]

Lowenstein & Westphal: Light reflex→long-latency, reduced amplitudes, fast recovery (diencephalic), and slow edge-light cycle time (E-W)

23 PD off meds with normal ophth exam, VEP, EMG. Soundproof, adapt 10 min to dark.

1. light reflex: Collimated beam 1400lux 500ms, intervals 5s.

2. near reflex: 5m → 10cm

3. nociceptive vs non-nociceptive: sural nerve stimulator

Results: in PD: No anisocoria. Resting pupil is larger. PD have longer latency and contraction time, reduced magnitude. Redilation time and Near reflex parameters all normal. Impaired reaction to sural nerve stimulation for twice-threshold stimuli. No difference between PD subtypes.

Conclude impaired parasympathetic modulation, with some latent sympathetic involvement.


Schmidt, Herting, Prieur,...Ziemssen Movement Disorders 22:14:2123-6 2007

Pupil diameter in darkness differentiates progressive supranuclear palsy (PSP) from other extrapyramidal syndromes [Disease]

PSP, PD, MSA patients underwent pupillometry to 20 mins dark adaptation, then light reflex 104 cd/m2 for 200ms.

PSP had significantly smaller pupil diameter in darkness. Light reflex showed smaller absolute constriction magnitude in PSP, but no other changes (latency, half-time, velocity, relative constriction magnitude).


Fotiou, Stergiou...Karlovasitou Int J Psychophysiol 73:2:143-9 2009

Cholinergic deficiency in Alzheimer’s and Parkinson’s disease: evaluation with pupillometry [Disease]

Maximum constriction acceleration and maximum constriction velocity reduced in PD with and without cognitive impairments. Patients with cognitive deficits also had lower amplitude, and slower vel/accel.

“The evidence concerning the pupil's reaction to light in PD is limited and controversial.”

Intepret as central cholinergic deficit, and pupillary abnormality correlates with cognitive deficit. [used median split to define ‘no cognitive impairment’]


Kakade S, Dayan Neur Net 15:4-6:549-59 2002

Dopamine: generalisation and bonuses [Theory, Learning]

TD error: used for 1) training predictions themselves, 2) finding actions that maximise reward

Dopamine neurones fire to a cue- that does not predict reward (generalisation response): short lasting, phasic immediate depression ~100ms, no depression at time reward would have been expected on cue+ trials. Unexpected rewards at that time elicit large responses.

  • Generalisation modeled as three cue types, including cue, which has an intermediate reward expectation between cue+ and cue-. Milliseconds after a cue appears, it starts as cue, but then becomes disambiguated.
  • Novelty modelled as bonus, favouring exploratory behaviour in instrumental conditioning. Use a ‘shaping bonus’ (Ng 1999) where novelty is a potential function of prediction, to avoid distorting the prediction.
  • Alternatively exploration bonus – adds on number proportional to time since last in that state; no evidence although not tested in explore-exploit paradigm.
  • Aversive phasic decrements are only suppression. Need for opponent system ?dorsal raphe serotonin


Gesi, Soldani, Giorgi...Fornai Neurosc& Biobeh rev 24:6:655-68 2000

The role of the locus coeruleus in the development of Parkinson’s disease [Disease]

  • Rat noradrenergic cells: medullary ventrolateral reticular formation, dorsal vagal complex, NTS, and pontine LC (nucleus pigmentosus pontis) = LC sensu strictio, nucleus subcoeruleus, some in parabrachial nucleus. Present in all mammals.
  • in human: tube shape, 2 cell types:
    • fusiform medium-sized NA cells, coarse melanin particles (caudal), multipolar arborisation, 60,000 decreasing to 40,000 in elderly
    • smaller non-catecholamine dendritic cells, unaffected by PD
  • Whole brain: dorsal bundle, rostral limb of dorsal periventricular pathway; also cerebellum, lower medulla and cord. Terminals do not contact postsynaptic cells, ‘boutons en passage’. Pontine cells - small terminals, more sensitive to insult.
  • Sleep/wake (anticipates EEG), state of vigilance, monitoring environmental stimuli, response to novelty with rapid habituation, orienting and attending, autonomic control
  • Expression of c-fos, NGF-induced-A, tis, zif during waking, inhibited by clonidine
  • Caudal cells spared in PD, LC affected (70% loss compared to age-matched controls). PD without dementia & Pick disease: homogeneous loss; PD with dementia & Alzheimer: pronounced in rostral LC.
  • Loss of DA fibres involves cortical layers I, II; NA loss in all layers
  • LC loss in PD may cause akinesia, freezing, on-off, tremor, dementia, depression
  • LC NA lesion sensitises nigrostriatal DA neurones to neurotoxic insults


Parkinson’s < controls
















N Accumbens




















Motor cortex




Premotor cx












Bear, Malenka Curr Op Neurobiol 4:389-99 1994

Synaptic plasticity: LTP and LTD [Neuron, Learning]

  • LTP blocked by NMDA and calcium blockers. Can potentiate neighbouring cells (within 150 microns), so not synapse-specific. Effect shorter-lived if generated by infusion of calcium alone – therefore maybe dependent upon mGluR as well. Evidence for increased glutamate release – ?presynaptic component.
  • Heterosynaptic depression: LTP in one synapse depresses all other synapses to same cell.
  • Initially Sejnowski described homosynaptic LTD with alternating inputs; disproved.
  • LTD (homosynaptic) as opposite of LTP: 1-3 Hz stimulation gives sustained reduction of EPSP. Same number of pulses at 10Hz produces no effect, 50Hz gives LTP. NMDA and calcium dependent, and able to unsaturate a saturated LTP effect – therefore same mechanism. Protein phosphatase dependent.


Kerr & Wickens J Neurophysiol 85:1:117-24 2001

Dopamine D-1/D-5 receptor activation is required for long-term potentiation in the rat neostriatum in vivo [Drug, Learning]

Corticostriatal inputs to neostriatum synapse directly on spiny projection neurons (output). Spiny neurons usually quiescent. LTD with high-frequency stimulation in vivo; LTP only in Mg-free bath, and blocked by NMDA blockers. LTP induced by dopamine in bathing solution, pulsed coincident with the cortical input. Also by SN stimulation in vivo (Wickens 2000).

Rat brain sliced 400um to preserve cortex, neostriatum and connecting fibres. Intracellular electrode in spiny cell, extracellular bipolar stimulator in deep cortex/white matter. High-frequency cortical stimulus below threshold of exciting striatal cells (50 pulses 100Hz every 10s, 6 times), paired with suprathreshold intracellular current pulse to give APs (520ms 1nA). Test response 20min later; 1 cell tested per slice.

Mg-free preparations showed LTP. Picrotoxin (GABAA blocker) to ensure not due to reversal of IPSP. D1/D5 blocker abolishes LTP. D2 blocker no effect. Dopamine depleted preparations (alphamethyl paratyrosine) did not show LTP, unless D1 agonist added.


McClure, Daw, Montague TINS 26:8:423 2003

A computational substrate for incentive salience [Theory, Reward]

‘Dopamine release encodes a measure of the incentive value of a contemplated behavioural act’: reward prediction error or incentive salience?

Dopamine blockade does not change appetitive value of reward, but prevents initiation of actions to acquire rewards (Ikemoto & Panskepp 1996 rat maze with flupenthixol in striatum).

Dopamine ‘maps liked objects to wanted objects’, characterised by ‘working to acquire’ them (Berridge & Robinson 1998).

Propose: Incentive salience is expected future reward, and this signal biases action selection: i.e. softmax, and dopamine d reports relative usefulness of contemplated actions.

Explains gradual extinction with low-dose dopamine blockade: no actual prediction error, but inhibition of postsynaptic receptors (negative prediction error) at time of reward causes unlearning.


Corbetta & Shulman NRN 3:201-15 2002

Control of goal directed and stimulus-driven attention in the brain [Theory, Attention]

(gratuitous bosch painting). Dorsal vs ventral parietofrontal network: alerting vs. task-set

Top down signals: precue for directing visuospatial attention gives sustained activity in dorsal PPC, along IPS, and in FEF; insensitive to cue, stimulus, attended location, or action. Ventral IPS and FEF also have lateralised modulation of activity.

  • Location orienting = same areas as expected direction of motion, almost same as area for attending to a colour (does not activate posterior IPS). The posterior area also active for switching attention between 2 objects at same location (Serences Schwarzbach & Yantis 2001). Relates to expectation of moving occluded object reappearing (Assad & Maunsell 1995)
  • WM for attentional set: overlaps with top-down control areas – FEF, IPS.
  • ‘Attending’ to effector (preparing movement): LIP, FEF, IPS, Parietal reach region.
  • Areas for eye movements = subset of areas for covert attention
  • S-R mapping separate from stimulus selection in intraparietal regions (Rushwth ‘01)
  • Macaque PFC neurones ‘code’ for S-R conjunctions and ‘rules’ (Miller & Cohen ‘01).
  • Left PPC (junction of IPS and IPS) active during task switch trials: number or letter classification (Kimberg, Aguirre, D’esposito), colour or motion categorisation.

Right ventral network for ‘circuit-breaking’ to unexpected salient events: Rt TPJ, Rt ventral frontal cortex (MFG, IFG, frontal operculum). Not influenced by cue or task set, but active at target detection, esp unexpected, low-frequency.

  • Stimulus-driven orienting is contingent on task set (Folk Remington Johnston JEPHPP 92)
  • FEF activity for oddballs: colour, shape etc.; LIP active for transients → salience map
  • Ventral – low spatial resolution, damaged in neglect. Active during vigilance ?LC/NA
  • Dorsal – goal directed orienting, sensory-motor mapping.


Daffner, Mesulam, Scinto...Holcomb Brain 123:5:927-39 2000

The central role of the prefrontal cortex in directing attention to novel events [Lesion, Attention]

9 nondemented frontal infarct patients 1.5y post-stroke. 6 blocks of 50 line drawings, 70% up triangle, 15% down triangle (target), 15% novel unusual objects shown once each. ‘Look at picture however long you like’, button press to move to next image, respond to target with foot pedal ‘sequence marker - to let experimenter keep track of where you are’. All displayed for >600ms; ISI 1-1.5s.

EEG: reduced P3 amplitude at Cz to novel drawings. Both RT and P3 amplitude correlate with informant-based and self-reported apathy score (Starkstein).


Gottlieb J in press 2009

Attention, information seeking and decision making: on the distinction between action selection and information selection [Theory, Reward]

Decision: choice to get reward; Attention: selection to get information.

LIP neurones ramp to threshold with motion coherence, and increase monotonically with expected reward.

However, they also fire to salient stimuli that do not mark a decision alternative, e.g. a cue.

Expt 1: Monkeys grab two bars to start trial, and shown visual search for E or $ amongst other symbols. Rewarded if they released the bar in L or R hand depending on orientation of target. Location of target dissociated from response side.

Result: Response to target in RF; firing rate correlates with performance (display size). Muscimol (GABAA agonist) injection to L/R LIP affects target location not response hand. Cell response to target presence is modulated by action hand. Crossing arms crosses responses (so not spatial location), and no effect of switching cue meanings.

Explained as action ‘feedback’ into attentional area.


Expt 2: Visual coloured reward precue (predicts 100% whether trial rewarded) on L or R, delay, then white saccade target at unpredictable location. Monkeys had to respond correctly to all trials, or else repeat trial.

Result: LIP cells respond +/- to reward cue valence (relative to contralateral cells) during delay. Correspondingly, saccades inaccurate and slow when target is congruent with a negative reward cue.

Explained as ‘automatic repulsion of attention’ from cue ‘bringing bad news’; automatic reward ‘feedback’.

Action and reward feedback in LIP = teaching signal → information value of a sensory cue

Necessary as attending itself is not rewarded. Could it be dopamine?


Gold & Shadlen Ann Rev Neurosci 30:535-74 2007

The neural basis of decision making [Neuron, Decision]

Signal detection theory, sequential analysis, sequential probability ratio test.

Two-interval discrimination of vibrotactile stimulus (f1>f2?). Signals in S1: no interval memory. average firing rate correlates with behavioural sensitivity. Signal in ventral premotor cortex: memory during interval, and during f2, increased activity if f1>f2 – represents a decision variable. Also similar information in S2, medial premotor cortex, dlPFC.

dlPFC represents f1 in interval, and not expectation of f2 being higher or lower (in experiment when f2 is unpredictable from f1).

Random dot motion: FEF stimulation-evoked saccade deviates from expected direction by an amount consistent with an evolving decision variable.

Stimulating rightward MT neurons increases rightward choices, and RT for leftward motion. (i.e. adds momentary evidence over time). Stimulating rightward LIP neurons does not alter choice, only RT. (i.e. adds to cumulative decision variable).



Also heading discrimination – where is DV? Disparity discrimination – also in MT. Face/object discrimination – microstimulation of IT biases decisions, but IT does not contain decision variable; DLPFC may. Olfactory discrimination.

Motion detection: MT signal ramps up on hit and miss trials, but not on false alarms. Ventral intraparietal neurones were more related to response (i.e. may be decision variable).

Discussion of evidence for LATER. Randomness and procrastinating for exploration.

OFC=value independent of evidence, choice or action. ACC=negative value. DLPFC =dynamic recent history of choice & consequence. They are not sure value can be used as a decision variable; ? randomness for game theory.


Pessiglione, Schmidt... Dolan, Frith Science 316:904-6 2007

How the brain translates money into force: a neuroimaging study of subliminal motivation [fMRI, Reward]

Pre & post-masked pound or penny coin at start of trial 17ms, 50ms (unaware), 100ms (aware). Hand-grip force displayed as visual gauge. Control experiment 4afc pound/penny.

  • Grip strength: increased for high stake>low stake even at 17ms p<0.01
  • Skin conductance: increase for high stake>low stake; effect at 100ms & 50ms
  • High stake>low stake (100ms): v striatum, bilat ventral pallidum, amygdala, Meynert
  • Correlation with force: SMA & left M1, not pallidum
  • High stake>low stake (50ms): ventral pallidum
  • Control 4afc: saw penny only at 100ms [but guessed 65% at 50ms]

Conclude: ventral striatum process subliminal reward cues and translates into motivation.


Rabey JM, Burns RS Random book ?

Corpus striatum ~108 neurones, 75% medium spiny. Main output of caudate and putamen is GABAergic to GPi+SNpr. GPi 700,000 cells (convergence >100:1)

SNpr is developmentally part of globus pallidus, separated by internal capsule

Corticostriatal and striatopallidal projections: both terminate in rostrocaudal bands, mapped according to locations of source.

GPi & SNpr main output is GABAergic to UMNs: motor cortex via VA & VL thalamus, or superior colliculus.

GPi and SNpr have high levels of spontaneous activity that tend to prevent unwanted movements by tonically inhibiting SC and thalamus.


SNpc dopaminergic, D1 (direct pathway, LTP & LTD) and D2 (indirect pathway, LTD) postsynaptic receptors.

Corticostriatal pathway: glutamatergic, NMDA+AMPA

Fuente-Fernandez, Stoessi TINS 25:6:302-6 2002

The placebo effect in Parkinson’s disease [Drug, Disease, fMRI]

Kinesia paradoxica – in unusual circumstances, PD patients can act fast. Arousing / emotional stimuli generate dopamine.

Strong placebo effect in PD ~23% (9-59% that of active drug). Bradykinesia and rigidity the most.

PET with 11C-raclopride during saline gives change in binding potential of 17% caudate and 19% putamen – same as therapeutic dose apomorphine, and amphetamine administration in normals. All patients showed biochemical placebo effect, only 50% reported benefit (with no difference in their chemical effect).


Kaye & Nicholls Clin Pharmacokinet 39:4:243-54 2000

Clinical pharmacokinetics of ropinirole [Drug]

Selective D2-agonist

Rapid absorption, 50% bioavailable, low protein binding

CYP1A2 liver metabolism, no active metabolites

t = 6h


Bamford...Sulzer Neuron 2004

Heterosynaptic dopamine neurotransmission selects sets of corticostriatal terminals [Neuron, Reward]

Mouse slices, simultaneous recording of Glu release, DA release and EPSP.

  • Cortical stimulation produces glutamate release, causing EPSP.
  • Amphetamine or striatal stimulation to evoke endogenous DA release.
  • DA inhibited Glu release from presynaptic corticostriatal neurone
  • This effect was mimicked by quinpirole, and prevented by sulpiride & in D2 receptor knockout mice.
  • Low-frequency corticostriatal stimuli: no block of Glu release; Hi-freq: attenuation. Dopamine as low-pass filter

Conclude “The most active corticostriatal inputs are selected by filtering out the activity of the less active inputs.”

“DA release associated with salience interacts with a highly activated cortical input during motor learning to reinforce specific subsets of corticostriatal connections.”

Amphetamine – “may dissociate the time-coding characteristics of behaviourally relevant activation of DA input, leading to aberrant filtering of less active corticostriatal inputs” → reinforcement → dependence


Giakoumaki, Roussos, Frangou, Bitsios Psychopharm 2007

Disruption of prepulse inhibition of the startle reflex by the preferential D3 agonist ropinirole in healthy males [Drug, Learning]

Role of D3 in schiz. Ventral striatum 30% neurones express D3, 75% express D2 (Gurevich & Joyce 1999). Ropinirole has 20-fold selectivity for D2 over D3.

Prepulse inhibition:

  • 70dB binaural white noise background, 20ms 75dB / 85dB prepulse, then 50ms / 80ms gap, 40ms 115dB pulse, ITI ~ 15s random.
  • 46 trials: 5 pulse-alone + 36 mixed pulse / prepulse-pulse + 5 pulse-alone.
  • EMG of orbicularis oculi (eye-blink component of acoustic startle response).
  • 1.5h after ropinirole 0.25,0.5mg or placebo.
  • Subjects selected for high startle and high prepulse inhibition

Present in mice/normals, absent in schiz, restored by atypical antipsychotics. Caffeine / nicotine withdrawal alters it. Note made of D3 ser9gly polymorphism (Jeanneteau 2006)


Pizzagalli, Evins...Frank, Culhane Psychopharm 196:2:221-32 2007

Single dose of a dopamine agonist impairs reinforcement learning in humans: Behavioral evidence from a laboratory-based measure of reward responsiveness [Drug, Learning]

D2 receptors likely to be reward-related. D2 & D3 both located pre & postsynaptic; D2 agonists reduce phasic DA release in striatum, suppress VTA/SNc firing rates in rats (Schmitz 2003). 0.5mg pramipexole reduces alertness, causes pupillary dilatation, increase HR (Samuels 2006).

Vigilance: simple RT index finger button press to red square. VAS mood scale. Rapid alternation task: right index finger to alternately press 2 keys as quickly and accurately as possible (Giovannoni 1999).

Probabilistic reward: 3 blocks of 100 trials, mouthless line drawing of face shown for 500ms, then 100ms mouth long/short added. Keypress judgement, correct wins 20c (60% for one face, 20% for the other).

With placebo, response bias in block 2&3. Pramipexole abolishes response bias, and effect of previous trial reward. No effect on simple RT. Mood scale: subjects do not become sociable as much, feel mentally slow, and feel more tense. Reduced kinesia score on alternation task.


Samuels, Hou, Langley...Bradshaw Psychopharm 2006

Comparison of pramipexole and amisulpiride on alertness, autonomic and endocrine functions in healthy volunteers [Drug, Autonomic]

Pramipexole activates inhibitory D2 autoreceptors on the VTA → removal of meso-coerulear pathway and putative mesopupillomotor pathway.

Amisulpiride D2/D3 antagonist 50mg (selective for VTA mesolimbic and mesocoerulear pathway blockade as opposed to SN nigrostriatal pathway) vs pramipexole 0.5mg.

Pupillographic sleepiness test: fluctuations over 11 mins in the dark, take total distance traversed by pupil edge (pupillary unrest index) and power of fluctuations (from FFT) – both increased by pramipexole not amisulpiride. Flicker fusion frequency – no effect. VAS, light and dark reflex, HR/BP

Pramipexole causes sedation and pupil dilatation. Reduction of light-reflex amplitude.

Amisulpiride caused pupil constriction, increased light reflex amplitude.

Also pramipexole increase in GH, reduce TSH and PRL. Little effect on BP, pulse, Temp.


Samuels, Hou, Langley...Bradshaw Psychopharm 20:6:756-770 2006

Comparison of pramipexole and modafinil on arousal, autonomic and endocrine functions in healthy volunteers [Drug, Autonomic]

0.5mg pramipexole vs 200mg modafinil. Pramipexole lowers fusion frequency by 1.5Hz, modafinil reverses this. Pramipexole increases pupillary unrest index & power of pupillary frequency spectrum; not reversed by modafinil. Pramipexole decreases VAS alertness and contentedness, does not affect anxiety. Modafinil reduces anxiety.

Pramipexole increases heart rate when standing. Modafinil increases systolic and diastolic blood pressure when standing. Both drugs increase baseline pupil size in bright ambient light, but no difference in dark. Pramipexole reduces amplitude and speed of light reflex, and increases the magnitude and velocity of the dark reflex.


Krain, Wilson, Arbuckle...Milham Neuroimage 32:1:477-84 2006

Distinct neural mechanisms of risk and ambiguity: a meta-analysis of decision-making [fMRI, Decision]

OFC ‘hot’ affectively laden, DLPFC for ‘cool’ cognitive, executive function.

Bechara 2005 distinguishes risk (known probabilities e.g. IGT, Cambridge risk task) vs ambiguity (unknown probabilities or probabilities close to chance, options have the same expected value, e.g. 2-choice prediction).

Meta-analysis of 27 papers with stereotactic coordinates of activation in decision vs. non-decision control task. Montral neurological institute (MNI) coords transformed to Talairach. ‘Activation likelihood estimate’ for each voxel; total of 287 foci of activation including OFC DLPFC indula, ACC, precuneus, SPL, thalamus, caudate.

Risk>ambiguity: OFC, rostral dACC (BA8), bilat superior parietal, inf parietal

Ambiguity>risk: DLPFC, caudal dACC (BA32), subcallosal BA24, precuneus, right parietal.


Nachev, Kennard, Husain NRN 2008

Functional role of the supplementary and pre-supplementary motor areas [Theory, Executive]

Anatomy: SMA + pre-SMA medial surface, just dorsal to cingulate sulcus, in SFG, BA6c (6aa,ab). SEF at border of SAM and pre-SMA, close to paracentral sulcus, on the vertical comissure anterior line.

Main output = putamen + caudate + STN. SMA also contributes ~10% of corticospinal tract, probably onto aMNs, plus reciprocal connections with M1. Pre-SMA & FEF project to DLPFC. Main input = thalamus: VL, VPL, VA pars parvocellularis. Mainly from GPi.

Stimulation: SMA → complex movements, urge to move, inhibition e.g. speech arrest. Somatotopic. SEF eye + head movements. Pre-SMA variable, may not evoke movements.

Microstimulation of SEF decreases success on stop-saccade task.

Recordings: SMA fires before movement, effector-specific; SEF fires before eye movement. May relate to Bereitschaftspotential. SMA fires to both internally and externally guided movements (Kurata & Wise 1988, Rushworth et al TICS04), or can be cue-and-action-conjunction-specific. Some cells show ramped activity to action-cue suppressed by nogo cue; others phasically respond to nogo signal (Schall 1991).

Tanji & Shima Nature94:

  • Some cells fire before action sequence ABC but not ACB – sequence specific;
  • some fire between AB but not between CB – conditional on prior movement;
  • others fire to serial position of action irrespective of the specific action

SMA encode number of movements remaining in sequence (Sohn & Lee JN07). Pre-SMA encode odd/even sequential order (Shima & Tanji JN06).

Some SEF cells more active during learning stimulus-saccade pairings; preferred saccade direction evolves during learning (Chen & Wise 95). Pre-SMA cells may learn stimulus-keypress pairings (Hikosaka). Some SEF cells respond only when eye movements and accompanying hand movement (Tanji JNph02).

Pre-SMA respond in change of plan & stop-signal (SEF in countermanding). Associated both with successful & unsuccessful withholding.

Isoda & Hikosaka NN07: 2 coloured targets L/R, then central colour cue (same for few trials)

Same cells responding to switch>noswitch also respond in go/nogo task to nogo signal.

fMRI: Free>instructed choices: pre-SMA but not SMA (Cunnington 2004). ‘Paying attention to intentions’ activates pre-SMA (Lau, Haggard & Passingham 2004). Observing graspable objects activates SMA (Grezes & Decety 2002). Activity during internally simulating motor sequences as well as executing. SMA activity during learning visual-buttonpress associations > visual passive learning. PET shows dopamine release ↑in pre-SMA but ↓GPi during learning.

Lesions: SMA → motor neglect, akinetic mutism (Talairach 1977). Monkeys bimanual problems. alien limb, utilisation, can be effector specific.

Monkeys taught to initiate an action to get a reward; impaired after lesion. ?internally generated.

Monkeys taught go=AC or BD, nogo=AD or BC (ABC=tones, D=touch). Omission and commission with cooling of SMA (Tanji EBR85).

Muscimol into SMA or pre-SMA bilaterally – memory guided sequences impaired, visually cued sequences OK, internally guided single movements OK. Unilateral lesion prevents learning but not old sequences.

Lesions to SMA ↑stop-signal time (TMS too), Eriksen flanker task, change-of-plan task.

PD: decreased activity in SMA, improved by L-dopa, STN DBS, local TMS.


Supplementary areas ‘link conditions to actions’. Complexity of environment-action mapping is greater for ‘internally generated movements’, for ordering operations, during learning new mappings, and in conflict. [but is complexity is a feature of many other tasks, which do not activate SMA?] SMA not ACC may be conflict resolver. SMA and SEF normally suppress primed motor programs. Areas are not modular.





Bijleveld, Custers, Aarts Psychol Sci 20:11:1313 2009

Pupil dilatation reveals strategic recruitment of resources upon presentation of subliminal reward cues [Autonomic, Reward]

Digit retention 3 or 5 items, for reward 1c / 50c. Pre- and postmasked reward cue 17ms / 300ms (50%). Then number of digits displayed 2s, then 3s gap, then aural digits 1s each, then 4s retention interval, then tone cues verbal report, then feedback displayed 2s, then cumulative cash 1.5s.

97% accuracy. Valuable rewards led to recruitment of more resources (pupillary dilatation), but only in the high-demand task. Awareness is not necessary for this strategic recruitment. ‘analyses of costs (effort) and benefits (value of rewards)...can occur outside of awareness’


Lee, Dolan, Critchley Cer Cor 1093/bhm035 2007

Controlling emotional expression: behavioural and neural correlates of nonimitative emotional responses [Behaviour, Executive, Reward]

‘Emotion expression interference’: Stroop-like effect. Dynamic 700ms expression videos, 4 intensities of happy or sad. Auditory cue ‘smile’ or ‘frown’ 200ms. Then emotion intensity judgement of viewed face VAS 2600ms later. EMG zygomaticus major vs corrugator. Empathy quotient & regulation questionnaires.

Compared with Simon task: white dot at 4 distances to L/R of fixation, then auditory instruction ‘left’, ‘right’ 20ms later, then button press response, then ‘VAS’ of distance!

Overtrained. Performance >95% face; 98.7% VAS, but no EMG in scanner.

EMG latency: concordant - decreases with increasing emotion VAS, incongruent increases.

Emotion incongruent>congruent: motor cx, VLPFC, lingual gyrus, rt ant insula. Not dACC.

Simon incongruent>congruent: frontostriatal, extrastriate, left post insula, cbm. No overlap.

Simon error>correct: dACC. Dot position did not modulate interference! [poor control]

Correlation with VAS incongruent>congruent: IFG, rt ant insula, STS, bilat MFG, left OFG

Conclude: Right IFG inhibits automatic emotional expression, by projection to BA44.


Mula, Cavanna, Critchley, Robertson... J Npsych clin neurosc 20:223-6 2008

Phenomenology of obsessive compulsive disorder in patients with temporal lobe epilepsy or Tourette syndrome [Disease, Reward]

OCD symptoms in 20-60% of GdlT, and 14-22% TLE.

ADHD prevalent in GdlT. 50% in both groups had affective disorder.

OCD scales not different, but significant differences in thematic content of obsessional symptoms: TLE = existential attitudes toward religion, intrusive thoughts about ageing, time passing, need to atone for imaginary sins e.g. denial of food or relaxation; GdlT = sexuality, gambling, play with fire, provoke accidents, aggression – but did not fulfil DSM criteria for impulse control disorder kleptomania, pyromania, pathological gambling.

Impulsiveness correlated with OCD score only in GdlT. Shared traits: indecision, checking behaviours, emotional control, precision, order, symmetry. Both groups have egosyntonic (comfortable) rather than egodystonic symptoms that characterise primary OCD.

[DSM: axis 1: all major clinical disorders, 2: personality & retardation, 3: acute medical & physical, 4: psychosocial & environmental contributing factors, 5: children global function]


Hup, Lamirel, Loreceau Vis Res 9:7:10:1-19 2009

Pupil dynamics during bistable motion perception [Autonomic, Attention]

Anatomy: Pupil is a non-linear closed-loop feedback system. Intrinsically photosensitive retinal ganglion cells → pretectal olivary nucleus → EWN → ciliary g; PON receives cortical and subcortical inputs; EW receives cortical and locus coeruleus inputs. Sympathetic frontohypothalamic projections; Loewenfeld 1999: Any sensory stimulus, spontaneous thought, emotion → dilatation.

Expt: ambiguous plaids. Press L/R mouse button for transparent vs coherent percepts. 36-68%. Ambiguous diamonds. Induced switching by stimulus or a sound.

Confounds controlled for: Post-blink meiosis, simple motor response to sound, endogenous choice to press button (together account for 70% of pupil response), surprise from change in appearance (controlled for using a physical change), difficulty/attentional load.

Conclude: change is small, cannot be used as an indicator of perceptual state.


Kennerley WS, Wallis J Neurosci 11:29:10:3259 2009

Reward-dependent modulation of working memory in lateral prefrontal cortex [Neuron, Memory, Reward]

Recording from monkey – simultaneous DLPFC and VLPFC, plus ACC and OFC, using 24-electrode arrays. Memory-guided saccades with reward-cue: 500ms cue1, 1s gap, 500ms cue2, 1s gap, then saccade go signal. Cues: reward-cue 10 pictures predicting 5 levels of juice, and saccade target dot 24 locations.

Random sampling of cells. Pupillometry. Analysis in 6 x 500ms epochs

% cells


Cue 1

Delay 1a

Delay 1b

Any epoch



























Reward & space










  • stronger encoding of spatial and reward in VLPFC than DLPFC.
  • Don’t encode object identity → use of object information to guide goal-directed behaviour, rather than what/where.
  • Reward increases spatial selectivity → attention rather than reward.
  • Inconsistent with Goldman-Rakic (1996) domain-specific WM.


vdMeer, Beyer, Horn...Wartenburger Psychophysiol 2009

Resource allocation and fluid intelligence [Autonomic, Memory]

Psychometric intelligence correlates with reaction speed on simple and complex tasks – reading, classifying numbers/letters, simple choice RT. Just & Carpenter 1992: resources = ‘amount of activation available for information storage and processing’ = structural connectivity + neurotransmitters & metabolism. Spearman 1904: intelligence as ‘general mental energy’ available.

Geometric analogical reasoning vs simple speeded L/R button press to dot on L/R of screen.

Dilatation with task difficulty only seen in high-intelligence group. Maybe intelligent people automatise the easier task, and stupid people are maximally attending throughout. Tonic baseline, and pre-trial baseline, pupil size larger for higher intelligence – plus tendency towards task-free exploration (cf. Aston-Jones).

Pupil is resource dependent, not effort → Inconsistent with Ahern & Beatty (1979)


Morsella, Gray JR, Krieger, Bargh Emotion 9:5:717-28 2009

The essence of conscious conflict: subjective effects of sustaining incompatible intentions [Behaviour, Executive]

Mayr 2004, Mayr 2003, Mulert 2005, Rosen 2007.

Unconscious conflict e.g. McGurk, ventriloquism, vs. conscious conflict e.g. Stroop. Explained as: all conscious conflict is for ‘skeletomotor’ control, and it is ‘only through consciousness that parallel systems can “cross-talk”’ – cf. blinking/breathing/micturition vs. pupillary, peristaltic.

[eye movement is not quite skeletomotor then!]

Stroop: trialwise answer ‘how strong was the urge to make a mistake?’ 1-8. Note that in congruent trials, one is unaware that two processes occur because output is identical.

Then task where 2 possible arm movements are co-expressible (compatible) or physically impossible to perform simultaneously (incompatible). Control was pupillary light reflex!

Subjects were able to transfer this concept of conflict to another task without it being explained.

Incompatible: mean 4.48 (sd 0.28), Control: 2.84 (0.25), Congruent: 1.47 (0.13).

Urge correlates with RT in all conditions (just about). Not just difficulty, as one of the responses was rated more difficult, but has same conflict ratings.

[incompatible and incompatible should be as similar as possible; conscious and unconscious conflict should be as similar as possible, e.g. a conscious perceptual conflict?]


Critchley, Tang, Glaser, Butterwth, Dolan Neuroimage 27:4:885-95 2005

Anterior cingulate activity during error and autonomic response [fMRI, Autononic, Decision]

Numerical stroop. L/R hand button press for higher number, ignoring physical size (or vice versa).

Post-stimulus redilatation following light reflex: parasympathetic blunting of reflex, onset of sympathetic influence – dilatory, manifest within steeper gradient of recovery and rebound.

RT 40ms incongruence effect, reverse-distance effect – closer L/R disparities → longer RT. Error trials had greates variability in pupil response; not all errors produced a distinct arousal reaction. Used modulation of pupillary reflex as a regressor. Correlates with bilateral rACC, distinct from pre-SMA.

Incongruent>congruent: Rt SMA, premotor cx, IFG, dACC.

Error>correct: bilat insula, right parietal, right IFG, medial thalamus, dorsal pons, rACC.



vEimeren, Ballanger...Lang, Strafella Neuropsychopharm 34:2758-66 2009

Dopamine agonists diminish value sensitivity of the orbitofrontal cortex: a trigger for pathological gambling in Parkinson’s disease? [Drug, fMRI, Reward]

fMRI during risk-taking on and off medications. Below-average risk-taking behaviour (off meds). PG: no dose effect, but in individuals, a threshold may be evident. Hypothesis: agonists prevent phasic pauses, impairing negative reinforcement learning by losing. Levodopa enhances pulsatile stimulation of receptors.

Balloon analogue risk-taking task: counter increases for each button press, until balloon bursts (no other details!)

Probabilistic reward task: 16-pocket roulette, 50% of trials choose 1 of 4 equiprobable colours (25% wins), or 50% choose 3 of 4 equiprobable colours (75% wins). Stakes $1 or $5 and options presented for 2s, then 3s choice, then 8s roulette, then 3s outcome.

UPDRS improved on Ldopa/pramipexole. No effect on risk-taking score, or gambling RT.

Feedback epoch>baseline: visual cx, cerebellum, putamen, cingulate motor, ventral premotor

Dopamine agonist>(LD or off) during feedback: OFC. OFC correlates with risk-taking

Correlation with reward prediction error: ventral striatum, left lat OFC. Dopamine agonists diminished reward effect throughout, L-dopa diminished effect only in striatum.


Koepp, Gunn, Lawrence...Brooks, Grasby Nature 393:266 1998

Evidence for striatal dopamine release during a video game [Drug, Reward]

(Hammersmith) C11-raclopride PET during tank game played using mouse in Rt hand, to collect flags and destroy enemy tanks, 3 lives. 7 for each level completed in 50 mins.

Binding less during game than during blank screen, local to striatum, ventral>dorsal.

Correlates with task performance.



Goller, Otten, Ward J JCN 21:10:1869 2009

Seeing sounds and hearing colours: an event-related potential study of auditory-visual synesthesia [EEG, Attention]

Tone-colour and bidirectional synaesthetes. Photisms were either from location of sound, or from behind or in front of them. Never affected by gaze/head direction. ‘test of genuineness’ – compare repeatability of colour matching with controls who were encouraged to guess.

ERP oddball paradigm, compare synaesthete detecting differences in pitch of tone, or in colour of tone, or in colour of visual patch. Bidirectional also did pitch of colour patch. Controls just attended to pitch or colour. Analysed ERP of non-oddball stimuli.

Synaesthetes have early differences to controls, unaffected by attention.


Fleming, Mars, Gladwin, Haggard Cer Cor 19:2352-60 2009

When the brain changes its mind: flexibility of action selection in instructed and free choices [EEG, Decision]

Left, right, or bidirectional arrow 250ms signifying instructed or free-choice. Then 1450ms gap, then either square /diamond 250ms signifying stay 28% or change 28% then 950ms gap (present 56% of trials only, 44% shorter trials without stay/change signal served to ensure preparation occurred), then green/red circle for go 80% or nogo 20%. Keypress L/R. ITI 2.5s.

Results: 2% errors, no RT difference. ERP at 520-540ms after change/stay cue shows increase only when instructed

“consistent with the idea that when people freely choose between action alternatives, they do not in fact strongly commit to one action over another” [yes - why should they in this expt?]


Haber SN, Knutson B Npsychophrm rv 1-23 2009

The reward circuit: linking primate anatomy and human imaging [Theory, Reward]


LaBerge S (in Bootzen Kihlstrom, Schachter, ‘Sleep & cognition’) 109-126 1990

Lucid dreaming: psychophysiological studies of consciousness during REM sleep [EEG, Sleep]

  • Aserinsky, E., & Kleitman, N. (1953). Regularly occurring periods of eye motility and concomitant phenomena during sleep. Science, 118, 273-274 – first described REM
  • momentary intrusions of wakefulness common in REM (Schwartz and Lefebvre 1973)
  • “a necessary part of the experience we call 'sleep' that we lose a directive and reflective self.” (Foulkes D, Dreaming: A cognitive-psychological analysis, 1985 p42)
  • awakenings following REM periods with high alpha yield reports of "thinking". awakenings from low-alpha REM periods yield "dreaming" reports (Antrobus, Dement & Fisher (1964).
  • LaBerge ‘perception of the external world as a criterion of being awake’; but what about inclusion of electric shock into dream-world without subjective or physiological arousal?
  • Antrobus, Antrobus and Fisher (Discrimination of dreaming and nondreaming sleep. Archives of General Psychiatry, 12, 1965): gradations of sleep and wake, also some systems may be more awake or asleep than others.
  • Piaget (the childs conception of the world 1927): developmental stages 1. external 2. mixed 3. internal conception of dreams. LaBerge: adults remain in stage 1 during sleep, out-of-body exp = 2, lucid dreaming = stage 3.
  • Malcolm 1959: when we think in a dream, we only dream we think.
  • Mandell (towards a psychobiology of transcendence: god in the brain, 1980): serotonergic activity inhibits vivid images i.e. hallucinations, but this 5HT mechanism is inhibited in REM sleep.

Lucid dreams

  • Lucid dream definition: “able to freely remember the circumstances of waking life, to think clearly, and to act deliberately upon reflection, all while experiencing a dream world that seems vividly real” Incidence: most people report 1 in life, 20% report >1/month (Snyder & Gackenbach, in ‘Conscious mind, dreaming brain’ 1988)
  • subjects aware of being asleep, but unable to feel bedsheets or hear ticking clock. ‘therefore, on empirical grounds, they conclude that they are asleep’
  • Learning to lucidly dream: conditioned stimuli, words or light to remind.


  • EEG shows REM immediately prior to reported lucid dreams (Ogilvie 1978)
  • Method of lucid dream induction (LaBerge 1980), instruct to signal with fist clench / LRLR eye movt / breathing. Co-occurrence of report & signal 90% during REM
  • Lucid dreams 90% in phasic rather than tonic REM, with ↑HR+RR, ↑scalp artefact, ↑saccades, lower H-reflex amplitude. Followed by 2min of uninterrupted REM. Occur in later sleep cycles.
  • 72% dream-initiated, 21% wake-initiated (dream becomes bizarre enough to elicit reflectiveness)
  • tracked the tip of their fingers moving slowly L/R. 4 conditions: 1) awake, eyes open; 2) awake, eyes closed mental imagery; 3) lucid dreaming; and 4) dream+imagination ("dream eyes closed"). 2 & 4 → saccadic eye movements; 1 & 3 → smooth pursuit
  • EMG qualitatively reflects motor experience but intensity not correlated. Counting seconds → no time distortion. EEG lateralisation of alpha to singing vs counting is preserved.
  • LaBerge, Greenleaf, and Kedzierski (1983): ↑genital EMG during lucid sexual dreams

Anderson (the architecture of cognition 1983) ACT* model of cognition to dreaming

Alfred Maury’s guillotine dream – recounted in Freud Interpretation, as rapid construction of lengthy dream at moment of waking. ‘Hall, 1981’


Domhoff GW Dreaming 11:13-33 2001

A new neurocognitive theory of dreams [Theory, Sleep]

Solms 1997: 29 controls, 332 lesions, 73 previous case reports: asked about dreaming changes; 60% had. Parietal lobe lesions, some bifrontal lesions.

Cf leucotomy in schiz → 80% loss dreaming. REM not affected. Patients also had decreased motivation, affect and fantasy.

Some focal frontal-limbic lesions caused vivid dreams, plus dreamlike thoughts in waking life, evidence of confusion of dream and wake. Temporal lesions → increased repetitive nightmares. Visual cortex → loss dream images, or static images, also waking loss of imagery.

Braun, balkin et al (science 279p91, 1998): PET activity during REM: limbic, paralimbic, inf parietal, occipitotemporal activity.

Foulkes (childrens’ dreaming and the development of consciousness 1999): longitudinal dream for 9 concurrent nights every 2 years, plus psychological measures. Before age 9, 30% dreams in NREM, after age 11 =80%. Recall from NREM age 6 =6%, age 12 =40%. Recall of events later in night comes first, then recall earlier in night develops. Age 5: ‘static bland images’ eg animal/action; age 7 simple narratives; later develop gender differences. Few negative emotions. Visuospatial skills correlate with dream parameters, but none of personality measures or linguistic ability correlates with dream content until adolescence.

Day emotions are carried over to dream.



den Boer JA, Reinders AAT, Glas G Theory &Psychol 18:3:380-403 2008

On looking inward: revisiting the role of introspection in neuroscientific and psychiatric research [Theory]

“there is a disparity between the rhetoric of psychiatry as a discipline that is based on ‘objective’ evidence and the factual reliance on introspection and subjective reports.” [but mistaking the meaning of subjective- experimenter or subject?]

Titchener vs Wundt 1910 – imageless thoughts exist. → demise of introspectionism

Searle 1992 re spatial metaphor of introspection: for “there to be something to which I have privileged access, I would have to be different from the space in which I enter” [but introspection is just that, like an external observation of the self] den Boer: “Second-order introspection... is not identical to a third-person type of approach, even if we adopt a third-person attitude to our experiences. Nor is such a third-person attitude identical to the adoption of atheoretical stance.”

Nisbett & Wilson 1977: introspection unreliable for causes of behaviour.

Even if memory vividness != veracity, phenomenal memory still useful for correlating with neural states.

Hurlburt 1997: descriptive experience sampling: bleep indicates to report what you are thinking at that moment – includes visual, sensory and inner speech

Regan & No 2001: ‘neural activity alone is not sufficient to produce conscious experience’

Chalmers 1996: unlikely to be a 1:1 relation between neural state and conscious experience

‘Guided vs unguided’ introspection – i.e. with regard to ongoing stimulus. Unguided = free report.

Gallagher Jack Roepstorff & Frith 2002: play scissors/paper/stone against computer, but told it could be either human or computer. Subjective report vs behaviour vs MRI

Varela 1995: trained subjects to report the Husserlian intersubjective appearance of the ‘object-side of the act of knowing’. Delineated invariants (phenomenological clusters): distractions, inattentive moments, cognitive strategies, daydreaming, etc.: Neurophenomenology – ‘open to experience without specific expectations or an interpretative stance’.

Dennet 2003 JCS heterophenomenology

Parvizi, J., & Damasio, A. (2001). Consciousness and the brainstem. Cognition, 79(1–2), 135–160.

Rodriguez...Varela Nature 1999 Perception’s shadow: mooney faces and EEG → synchrony

Introspection is heterogeneous (Prinz JJ 2004 JCS the fractionation of introspection): intensification of experience without language, second order linguistic thought, replay of experience


Van Der Heijden, Bem Consc & Cogn 1997

Eye movements and attention [Theory, Saccade, Attention]

Discussion of why eye movements and covert attention fulfil the same purpose, using different mechanisms (posed by Jonides 1983 “Further toward a model of the mind’s eye’s movements”). 2 old arguments – small brain or big world; they claim that “Of relevance is only the amount of information picked up by the receptors in the senses relative to the information processing capacity of the brain.”


Sheliga, Riggio & Rizzolatti EBR 98:507-522 1995

Spatial attention and eye movements [Behaviour, Saccade]

Vertical saccades curve away from lateralised attention. Instruction for saccade (‘imperative cue’) presented to left or right; ?80% expected. True for exogenous flash precue, central precue, and centrally post-cued items. Curvature greater when imperative is contralateral to attention.


Nobre, Gitelman, Dias, Mesulam Neuroimage 11:3:210-6 2000

Covert visual spatial orienting and saccades: overlapping neural systems [fMRI, Attention, Saccade]

Saccade task: 22 saccades to 2 randomly timed (400-1200ms) alternating targets at 12 degrees in periphery. Attention task: 10 trials, central arrow for 100ms 80% predictive then detection of 50ms targets SOA=200-800ms, 1.5 degrees in size, at 7.5 degrees in periphery. Button press L/R.

fMRI common areas = lateral & medial premotor area, FEF+SEF, ACC, ant insula, intraparietal sulcus, IPL, right STS, inf/MT, left ventral extrastriate cx, putamen, cerebellum.

Sacc>Covert = medial occipital, striate. Covert>Sacc = ventral extrastriate, lateral motor/premotor, FEF, post IPS + IPL; plus DLPFC (not active during saccades)

Suggest common mechanism, and FEF engaged independently of eye movts during attention.



Ophir, Nass, Wagner PNAS 106:37:15583-7 2009

Cognitive control in media multitaskers [Behaviour, Attention]

Created ‘media multitasking’ index from questionnaire of times and concurrency matrix (printed, television, youtube, music, computer games, phone, email, txt, web, computer apps). Took top and bottom 20 subjects of 260. Change-detection 2afc task with 1s gap, 45 degree rotation of 1 red line segment among 2 red lines plus 0-6 blue lines: high-multitaskers worse with distractors. ‘AX-continuous performance test’: cue ‘A’→probe ‘X’→respond ‘yes’, respond ‘no’ for AY or BX; with coloured distractor between cue and probe: high multitaskers same as low multitaskers without distractors but slower when distractors. Stop-signal buttonpress word-categorisation (animal/nonanimal), with tone on 25% trials 225ms before mean RT. 2-back & 3-back: serial letter presentation with 3s gap. 2afc ‘matches letter 2 trials ago?’: high multitaskers more false alarms than low multitaskers, on 3-back. Task-switching: cued ‘NUMBER’/’LETTER’ 200ms, then ‘A2’→vowel/consonant or odd/even. Equal number of 1, 2, 3, 4 same-trial sequences = 40% switch trials. High>low multitaskers RT: 167ms slower on nonswitch, 260ms slower on switch.

Conclude: mutitaskers have greater difficulty filtering out irrelevant information.


Churchland AK, Kiani, Shadlen Nat Neurosci 11:693-702 2008

Decision-making with multiple alternatives [Neuron, Decision]

Macaques motion-direction-detection task, comparing 2AFC vs 4AFC saccade response. Also a 2AFC ‘90-degree’ control task with nonorthogonal targets.

LIP cells selected for spatially selective persistent activity 15 deg from saccade target, in memory-guided saccades, and no response to central motion cue.

Result: rate of increase proportional to motion strength, but correlation vanished before saccade. Rate of increase that correlates with motion strength is the same for 2 and 4 choices. Common firing rate close to end of trial. Lower baseline firing with 4 choices, but same ‘threshold’. Cells only weakly negatively affected by motion in directions orthogonal to target. Plus a prominent buildup that is not correlated with either motion or choice: ? RHSC’s “sense of urgency”. This motion-independent component of buildup (urgency) is much slower in 4-choice task. Buildup rate (spikes/s/s) inversely correlated with RT, but not strong!

Conclude: the decrease in firing before 4afc, and lower urgency buildup, is to reduce uncertainty at the expense of longer decision time. Model: behavioural RT fit to model with bound height and urgency component taken from cellular responses, and fit the evidence scaling, evidence SD, and nondecision time (constrained to be same for 2 or 4). Hyperbolic urgency = U*t/(t+t). SD and scaling fit using MLE were the same for 2 or 4.

What happens in continuous choices? More accumulators?


Maunsell JHR TICS 8:6:261-5 2004

Neuronal representations of cognitive state: reward or attention? [Theory, Reward, Attention]

Platt & Glimcher Nature 99: LIP modulaton by reward size and frequency.

Bendiksby & Platt 2003: increased neuronal selectivity with reward.

Kawagoe NatNeurosci 98: compare rewarded with unrewarded task: purely ‘reward-related’? no because could be a task that requires less attention.

Stanford, Shankar...Salinas Nat Neurosci 13:379-85 2010

Perceptual decision making in less than 30 milliseconds [Neuron, Decision, Saccade]

Gap saccade 2-choice paradigm, where targets appear at go signal, but peripheral cue for correct target is delayed. FEF recording from regions where low currents evoke saccades, and cells fire during single-target delayed-saccade task.

Behaviour modelled as race between where

Before cue, rates of rise are constant, chosen randomly (correlated) about a guessing value

Integration stops at time of cue + [how can this stop before the cue?] and is interrupted for discarding any negative TI.

Then, rates of rise approach rT and rD respectively over time t,

, then remain constant at .

where decisiontime (ePT) is time from cue-onset to threshold.

Fitted to predict 6 datasets: [%correct, mean(RT) and SD(RT)] for each gap, distribution of ePT for correct and incorrect trials, and %correct for each ePT.

Hikosaka & Isoda TICS 853 2010

Switching from automatic to controlled behaviour: cortico-basal ganglia mechanisms [Theory, Executive]

Retroactive (ACC) vs proactive switching (preSMA)

Evidence from imaging and EEG that ACC active when error or unexpected reward influences next trial. Input from LPFC or striatum. Monkey ACC task-selective neurones most active after a switch trial, but only on successful switch; other cells active when reward absent.

fMRI shows preSMA active in response to task-switch cue, or nogo signal. Lesions ad TMS to preSMA impairs nogo task. TMS affects only task-switch trials. preSMA Neurones fire on first trial of cued switching only.

Integration of ACC and SMA inputs in striatum (abstract rules) and STN (prepotent responses). Note STN improves motor system but interferes with ability to slow down under conflict (Frank 2007).


Bromberg-Martin E, Hikosaka Neuron 63:1:119-26 2009

Midbrain dopamine neurons signal preference for advance information about upcoming rewards [Neuron, Reward, Decision]

Information as value. Monkeys choose to get information about reward, independent of choosing reward. Cells selected from midbrain if they signalled the value of water rewards. These cells phasically fire in response to high-reward cue, and inhibitied by low-reward cue. No response to uninformative cues. Strong response to reward outcome – inhibited by small reward. Same cell also responds to targets indicating availability of information: excited by information-predicting item, inhibited by item predicting uninformative cue. Also excited by free-choice of information (monkey always chose information).


Ding & Hikosaka JNeurophys 97:57-61 2007

Temporal development of asymmetric reward-induced bias in macaques [Behaviour, Saccade, Reward]

Visually guided saccades with asymmetrical reward; 2s waiting period. Probe trials early target, 16% (4% at each foreperiod). Bias in RT measured.

Ding & Hikosaka J Neurosci 26:25:6695-703 2006

Comparison of reward modulation in the frontal eye field and caudate of the macaque [Neuron, Reward, Saccade]

Asymmetrically rewarded memory-guided saccades.

Both FEF and caudate exhibit similar reward sensitivity in the memory period, specific to location. However only caudate shows visual responses that were specific to reward size but not spatial location.


Lo & Wang Nat Neurosci 9:7:956 2006

Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks [Theory, Decision]

Network model of motion direction discrimination.

‘Chaotic’ cortical firing in 2 networks accumulates information about opposing directions.

Superior colliculus burst cells with nonlinear (either silent or bursting) responses have a sharp threshold ~250Hz. Recurrent NMDA synapses.

SC under tonic GABAergic inhibition from SNpr (50-100Hz baseline). This inhibition is switched off by caudate, and allows corticocollicular stimulus to cause a burst.

Corollary discharge resets the cortical network.

Speed accuracy tradeoff explained: lower cortico-caudate synapse efficacy causes higher threshold [but is this actually the case?]


Wang, Vijayraghavan, Goldman-Rakic Science 303:5659:853 2004

Selective D2 receptor actions on the functional circuitry of working memory [Neuron, Memory]

In cortex D1>>D2 receptors. D2 localised to layer V (output). Macaques, 8-target spatial oculomotor delayed response. Iontophoresis of D2 agonist/blocker in DLPFC. 100 cells: 10 pure cue-related, 47 sustained delay-related, 45 pure saccade-related (10 pre and 35 postsaccadic).

  • D2 blocker reduces saccade-related activity for trials in the preferred-direction. Agonist reverses this, and increases saccade-related activity.
  • D1 affects memory period activity but not saccade-related activity.

Could pre/post-saccadic activity be a corollary discharge from MT thal, or PPC/FEF? D2 related could explain schiz.


McCoy, Crowley, Haghigian...Platt Neuron 40:5:1031-40 2003

Saccade reward signals in posterior cingulate cortex [Neuron, Reward, Saccade]

Saccade overlap task (go signal is fixation offset) with 1 or 2 choices. Blockwise asymmetric reward (size, or probability). PCC cells correlate with reward (negative or positive). 50% spatially selective but broadly tuned, generally to contralateral movements.

Probabilistic condition reveals cells with increased or decreased firing after omission of expected reward. Modulation by reward size does not predict modulation following reward omission.


Roesch & Olson JNeurophysiol 90:1766-89 2003

Impact of expected reward on neuronal activity in prefrontal cortex, frontal and supplementary eye fields and premotor cortex [Neuron, Reward, Saccade]

Could ‘reward-related’ activity be explained as “motivation-dependent variations in the monkey’s level of motor preparation or motor output”? this would predict more reward activity in motor areas, rather than limbic areas.

Lateral: PFC < FEF < premotorFEF < PMC, and medial: SEF < SMAr. Increased EMG activity in masseter etc during reward anticipation.


Deijen, Stoffers...Theeuwes BMC Neurology 6:43: 2006

Abnormal susceptibility to distracters hinders perception in early stage Parkinson’s disease: a controlled study [Disease, Attention]

Previously, PD has decreased visual sensitivity and spatial contrast sensitivity, decreased temporal sensitivity, slower saccadic latencies, smooth pursuit gain, and deficits in predictive saccades (Bronstein & Kennard 85). Memory-guided saccades (Kennard, Rudge) and antisaccades impaired, and inability to suppress reflexive saccades (Chan...Munoz Neuropsychologia 05).

Expt: use perceptual discrimination RT, rather than eye tracking, which is hard in PD. 6 grey discs containing figure ‘8’, 5 turn red & contain a letter, subjects need to foveate the disc that remains grey to determine if it contains ‘c’ or backwards ‘c’, button press 2afc.


Theeuwes, Kramer, Hahn, Irwin Psycholog Sci 9:5:379 1998

Our eyes do not always go where we want them to go: capture of the eyes by new objects [Behaviour, Attention, Saccade]

Perceptual discrimination of a ‘c’ vs backwards ‘c’ in the circle that remains grey.

Capture is overcome by location precue. Erroneous fixations last 25-200 ms, mode ~120ms.


Ludwig & Gilchrist P&P 65:8:1243-51 2003

Goal-driven modulation of oculomotor capture [Behaviour, Saccade, Attention]

When subjects knew colour of target (‘always green’, ‘always red’) there was high capture by similar but not dissimilar coloured distractors. When the target colour was unpredictable (‘singleton’), there was slightly more distraction by dissimilar items. Thererfore “unlikely that the effect of target similarity in the constant target blocks was completely due to inhibition of all elements in the distractor colour.” Also tried a blue distractor (onset or non-onset) which creates a ‘colour discontinuity’ but this also does not cause as much distraction when expecting a red target among green distractors.

Theeuwes 1999 argues for parallel programming of second saccade (2nd SRT~150ms). ‘vincentised’ RT distribution of erroneous and corrective saccades agrees with this: correlation trial-to-trial of SRT1 and SRT2 ~ -0.5.


Rouder & Speckman Psychonom bull rv 11:3:419-27 2004

An evaluation of the Vincentizing method of forming group-level response time distributions [Theory]

Comparison of Vincentizing + least squares, least-squares + parameter averaging, max likelihiid + parameter averaging. Tested ex-Gaussian (μστ), Weibull (shift, scale, shape) and shifted-Wald (shift, drift-rate, bound) distributions; results compared in how well they reconstructed the original parameters!

Finding: Vincentization more biased for shifted-Wald and ex-Gaussian, but is the best for Weibull data. For ex-Gaussian, LS+PA is very good, and for Wald, LS+PA is best.


Gehring, Goss, Coles, Meyer, Donchin Psychol Sci 4:6:385 1993

A neural system for error detection and compensation [EEG, Decision]

Used zero-displacement dynamometer + EMG + ERP on 2afc letter discrimination ‘H’ or ‘S’ for the central letter in an array, with flankers that were all compatible vs incompatible. Financial reward and penalty with time deadline; 3 conditions ‘speed’, ‘neutral’, ‘accurate’ blocks counterbalanced across subjects. Response criterion 25% maximal squeeze.

Results: Accuracy was 67%-89%, Latency 236-304ms. Stepwise discriminant analysis of ERP showed post-response Cz electrode predicts correct vs incorrect trials.

Large ERN correlates with small error-squeeze. Probability of correction increases with size of ERN. Larger ERN correlates with slower RT on the immediately following correct trial. (like rabbit 66 and laming 68).

They found ERN present in correct trials in which the correct response was followed by an error – unclear whether ERN is due to error monitoring and compensation, or due to the second erroneous response itself.


Los & Agter P&P 67:7:1161-70 2005

Reweighting sequential effects across different distributions of foreperiods: segregating elementary contributions to non-specific preparation [Behaviour, Decision]

Warning then go signal (non-specific preparation). RT vs foreperiod is downsloping and negatively accelerated (Woodrow 1914); could be due to conditional probability of go signal (Luce). Consistent with this, Ntnen (1971 Acta Psy, ‘non-ageing foreperiod’) showed exponential foreperiod distribution gives flat RT. But: even Woodrow 1914 and Karlin (JEP 1959) show effect of previous trial foreperiod. At shortest previous foreperiod, RT vs FP graph is flat. At longest, the graph is steep.

Compared exponential, uniform, peaked distributions, with variable foreperiods 300, 600, 1200. Separated out trial-to-trial effect using reweighting; but there was still an effect of distribution type on shape of FP-RT curve.


Smith, E Psych bull 69:2:77 1968

Choice reaction time: an analysis of the major theoretical positions [Theory, Decision]

Reviews criticisms & developments to Donders’ sequential model.

One-to-many and many-to-one mappings: RT = a + b log n. (Hick QJEP 1952), as information when probabilities are varied (Hyman JEP 1953)

Degree of parallelism measured using by perceptual previews (Crossman1953)

Are subcomponent tests in a single categorisation judgement performed in serial (Neisser 1963) or parallel (Sternberg 1963)?

Card sort many-to-one-mapping showed RT number of responses or its log, but not number of stimuli if < 6 (Rabbitt Nature 1959).

One-to-many mapping where subjects should ‘unsystematically’ emit one of several possible responses: ↑RT with number of responses confounded with memory of previous responses.

Hick’s law not due to frequency alone: RT to a stimulus occurring 75% of trials increases if there are 1 vs 3 other stimuli (Broadbent & Gregory 1965)

Fitts: payoff matrix in 1:1 mapping: high value → ↓RT. Laberge 1964 replicated this with a valued and unvalued stimulus mapping to the same response.

Repetition of stimulus →↓RT, and eliminates S-R incompatibility effect (Bertelson 1963). 4 stimuli 2 response task: ↓RT similar for repetition of response vs repetition of stimulus-and-response, therefore repetition of response accounts most of the effect.

Could Hick’s law be due to repetition effects? No because effect is seen even when only first occurrence of each stimulus is analysed (Smith 1967)

Noisy stimulus →↑RT but this does not change the slope with N.

Chase & Posner (1965) stimulus similarity ↑slope of RT vs number of possible stimuli.

Sternberg 1964: degradation of stimulus ↑intercept of RT vs number of possible stimuli.


Hick: ‘replication of stimulus’ for each of the possible ‘stimulus templates’, before a parallel comparison stage. Replication could be simultaneous, serial, or geometric (giving RT ~ log N). Rapoport (Behav Sci 1959): although comparisons may be simultaneous, the winner does not terminate the search, rather, we wait for all comparisons to be complete. Sternberg 1963: Agrees, but specifically ‘Not a target’ responses definitely require exhaustive comparison. All these models fail for many-to-one mappings.

Hick: comparisons are sequence of tests, serially more accurate classifications – e.g. dichotomising; if tests are “like of kind & take about the same time” gives log. Also explains probability/value, as dichotomising is optimal. Bertelson 1963: first test is always “is it identical to the previous trial?”, & S-R incompatibility causes ↑number of tests needed (explains why incompatibility effect not seen for repeated trials).

Stone (Psychometrika 1960) cumulative log likelihood ratio model.

Edwards 1965: normative Bayesian model for reward/payoff causing speed-accuracy tradeoff.

Neisser ‘63: Hierarchical cumulative log-odds model of recognition, starting w basic features



Donders 1868

On the speed of mental processes [Behaviour, Decision]

Compare simple reaction time (a) to 1 stimulus 1 response, with choice reaction time (b) to one of 2 stimuli 1 response, with go-nogo (c) to one of 1 stimuli, where response is to be withheld to one stimulus. Theorised that c = a + stimulus-categorisation time, and b = c + response-selection time. Assumes band c are compound processes, and components do not overlap, and that same process as in a is being used in b and c.


Engbert, Krampe, Kurths & Kliegl Brain & cog 48:107 2002

Synchronizing movements with the metronome: nonlinear error correction and unstable periodic orbits [Theory, Time]

Models of synchronized (metronome) tapping:

Wing & Kristofferson P&P 1973: influential model.

Ivry & Keele (JCN 1989) ‘timekeeper’ generates Gaussian uncorrelated intervals .

Vorberg & Wing 1996’s model: Uses Box & Jenkins 1976 techniques of time series analysis: Linear stochastic process: autoregressive first-order error correction (t = metronome interval) for motor errors given by : stable for .

But: Ding (B&C 2001) suggests noise in synchronized tapping.

Cvitanovic 1988 Detection of unstable periodic orbits useful for controlling chaotic systems; also is evidence for determinism and nonlinearity.

So: use a sigmoid error correction function:: larger errors saturate the system. Used by Engbert 1997 “Tempo induced transitions in polyrhythmic hand movements”. 4 parameters: variance of x, m, and corrections a, k. Then show subjects satisfy criteria for deterministic chaos from Pei & Moss Nature 1996.

[intervals cannot be gaussianly distributed!]



Eimer, Forster, van Velzen, Prabhu Neuropsychologia 43:6:957-66 2005

Covert manual response preparation triggers attentional shifts: ERP evidence for the premotor theory of attention [EEG, Attention]



Smit AC, van Gisbergen J, Cools AR Vis res 27:10:1745-62 1987

A parametric analysis of human saccades in different experimental paradigms [Behaviour, Saccade]

Visually-evoked, memory-guided, and antisaccades compared.

Antisaccade had 1s flickering fixation → red ‘anticue’ 200ms at fixation offset → saccade during gap → green target at 1450ms as feedback.

Memory-evoked was 200ms peripheral flash → fixation spot for 100, 400, 800 or 3200ms → fixation offset and saccade → green target at 1450ms as feedback.

Duration x peak vector velocity

Peak velocity =

Latency of the prosaccades in mixed blocks of (pro- and anti-saccades) was bimodal, and in blocks of (pro- and memory-saccades) was longer.

Gain histogram, latency histogram, (peak velocity) vs amplitude, (vel x duration) vs amplitude, and (skewness of velocity profile) vs duration

Evidence shows that variability in memory- and anti-saccades arises in colliculus not in burst cells or integrator.


Bendixen, Grimm...Schroeger Neuropsychologia 48:7:2130-39 2010

The time-course of auditory and visual distraction effects in a new crossmodal paradigm [EEG, Attention]

2afc L-hand or R-hand = stimulus same or different to previous one. Instructed to attend to visual or auditory stream. No feedback. The task-irrelevant-modality stimulus appeared 400, 200 or 0 ms before the target stimulus, manipulated blockwise, and all stimuli lasted 500ms with 500ms gap.

EEG shows visual modality is ‘contrained in terms of a critical time-range within which distraction effects could be elicited, whereas the impact of auditory stimuli on task-related processing extended over a longer time range’.

Wais, Rubens, Boccanfuso, Gazzaley J Neurosci 30:25:8541-50 2010

Neural mechanisms underlying the impact of visual distraction on retrieval of long-term memory [fMRI, Memory]

Encoding: 1-4 equal-sized copies of each object that vary in ‘real size’. 3s each. First run: ‘does it fit in shoebox?’; second run ‘caould you carry all at once?’. Then 1 hr break.

Surprise memory test: Presented auditory name of an object (old or new), required to press 1,2,3,4 or NEW, as rapidly as possible. Either eyes closed, blank screen, or unrelated distracting natural images. independent measures of recollection accuracy, recognition without recall, forgotten items, false familiarity.

fMRI: (LIFG, L hippo, LOC ) reduced functional connectivity when distractor image shown.

Highest functional connectivity in Incorrect trials with distractor


Halliday & Carpenter Perception 39:41-50 2010

The effect of cognitive distraction on saccadic latency [Behaviour, Saccade]

Go-nogo saccade task, under no interference, pink noise, and cognitive load (antonym task)


Cognitive load increaseserrors from 20% to 45%, and increases sE without affecting mean latency. Proportion of short latency saccades is increased. I.e. reduced inability to inhibit reflexive saccades. Cf Mitchell 2002 who argue suppression of reflex saccades depends critically on working memory.


Mobbs, Seymour Marchant, Dolan, Frith Psycholog Sci 20:8:955 2009

Choking on the money: reward-based performance decrements are associated with midbrain activity [fMRI, Reward]

Competition, audience, high reward can be detrimental to performance = ‘choking’.

Pacman task, calibrated to >50% success, but also measuring near misses. Before trial: passive movement of prey, then reward cue ‘5’ or ‘0.50’, then chase.

  • fMRI: approaching prey = dorsolateral striatum and rt mOFC.
  • Proximity of prey = ventromedial striatum and rostral ACC.
  • Proximity x reward magnitude = left ventral midbrain (VTA + SN) and bilat ventral premotor area
  • Proximity x (Low reward < high reward) = right ACC, medial PFC, dorsomedial striatum
  • Subjects’ money motivation (subjective rating of wanting the money) correlates with effect of proximity on midbrain activity.

‘behavioural results show that high-reward contingencies result in less-than-optimal performance’

medial and lateral PFC = predicts better performance and reduced errors in high incentive condition.

  • ‘Why incentive-based motivation causes performance deficits is open to argument; however, the anatomical location of the activity we observed (i.e., ventral midbrain and striatum) is consistent with a dopaminergic basis. Dopamine is implicated in increased motivational vigor and increased sensitivity to positive outcomes, yet can impair performance (Frank, Seeberger, & O’Reilly, 2004; Murphy, Arnsten, Goldman-Rakic, & Roth, 1996).’
  • One possibility that could be explored further is whether incentive-based motivation results in increased attentional narrowing (Easterbrook, 1959)
  • or simply actions without foresight (Robbins, 2002).
  • Alternatively, it is conceivable that high rewards are framed in terms of losses in some situations’, and if performance is driven by fear of loss, anxiety can reduce performance (Ashcraft & Kirk, 2001). – serotonin opposing dopamine (Daw, Kakade, & Dayan, 2002)

Mobbs...Seymour, Dolan, Frith Psychol Sci 20:8:955-62 2009

Choking on the money [fMRI, Reward]

Theories: top-down attentional-distraction or explicit-monitoring theories, vs. incentive-based overmotivation/overarousal theories

Capture pacman in a maze, speed precalibrated to 50% success. Reward precue 50p or 5. Midbrain activity correlates with performance decrements and near-misses induced by high rewards.


DLS activity correlates with approach to prey in both conditions. VMS and rostral ACC correlates with proximity to prey. (High – low reward) = VTA+SN, right DLS, bilateral ventral premotor area. (Low – high reward) = right ACC, mPFC, dorsomedial striatum.

Interindividual differences in motivation (performance in hi vs lo reward conditions) = midbrain; inverse = rostral ACC.



Shin & Sommer J Neurophysiol 103:1874-87 2010

Activity of neurons in monkey globus pallidus during oculomotor behaviour compared with that in substantia nigra pars reticulata [Neuron, Saccade]

Memory guided saccade (500-800ms fixation, 50ms target, 500-1000ms delay, then remove fixation → saccade, after 300ms target reappears and reward delivered)

Pallidal neurones show high spontaneous firing rate with ‘streaks of intermittent pauses’ (DeLong 1971). Cells at posterior and mediodorsal GPe and GPi (a volume of a few cubic mm) were task-modulated.

SNr neurones had high baseline rates (50-90 Hz)

Number of cells in GPe, GPi with signals modulated by each factor.

Spatial tuning curves narrow during visual cue and saccade, but broader during reward.


Yoshida & M Tanaka Cer Cor 19:1:206-17 2009

Enhanced modulation of neuronal activity during antisaccades in the primate globus pallidus [Neuron, Saccade]

Pro, anti and memory saccades, plus smooth pursuit. Modulation index calculated from preferred minus opposite direction trials, in bin 100ms before to 50ms after saccade.

  • Firing rates: 55% increased, and 45% decreased, in response to saccades.
  • More modulation in antisaccade condition, for all cell types; both during instruction period and saccade.
  • More modulation during instruction period in correct trials than incorrect trials.

Inactivated left GPe with muscimol (gabaA angonist) vs saline. Errors 15% → 70% for rightward antisaccades.

GP neurones don’t provide immediate drive as they don’t show modulation before saccade

Unlike PFC inactivation (Condy et al 2007) ‘the impaired direction is different from site to site’

  • inactivation of the SNr increases the occurrence of reflexive saccades in the memory-guided saccade task (Hikosaka and Wurtz 1985)
  • lesions in the medial frontal cortex shorten the latency of antisaccades (Boxer et al. 2006)
  • stimulation in the medial frontal cortex prolongs saccade latencies (Isoda and Hikosaka 2007)
  • rostral SC fixation neurons are elevated during preparation for antisaccades as compared with prosaccades, but are strongly suppressed around the time of saccades. (Everling et al 1999)
  • Task-instruction-related activity seen in PFC, ACC, SEF

Therefore hypothesise some of the signals from GPe go via thalamocortical pathway.


Jiang, Stein, McHaffie Nature 423:982-6 2003

Opposing basal ganglia processes shape midbrain visuomotor activity bilaterally [Theory, Executive]

  • Crossed and uncrossed gabaergic pathways from SNr to SC.
  • Visual stimuli inhibit uncrossed pathway neurones, but excite the crossed neurons.
  • Crossed neurons lower spontaneous activity (absence of strong tonic crossed nigrocollicular influences).
  • Crossed SNr neurons activated by the contralateral hemifield to the corresponding SC cell. Uncrossed neurones were congruent with their SC target.
  • Crossed pathway is fewer in number, broader projections, and establish direct synapses with contralateral colliculoreticulospinal neurons. Stimulation leas to profound long-duration suppression of visually evoked activity in the opposite SC.

Explained as push-pull mechanism, with uncrossed neurons locally disinhibiting i.e. acquiring a selected target, versus crossed neurones globally suppressing activity associated with competing distractors.

Could explain Sprague effect: large unilateral visual cortex lesion produces ‘neglect’, but lesion to crossed nigrocollicular neurons removes widespread inhibition to cortically deafferented SC, reinstating responsiveness.

‘Basal ganglia a re critical stsructures in the manifestation of human hemineglect syndromes’.


Jonides, ...Nee, Lustig, Berman... Ann Rev Psychol 59:193-224 2008

The mind and brain of short-term memory [Theory, Memory]

STM as activation of LTM representations

  • LTM and STM retrieval overlap neurally – frontal
  • HC binds items to context – common to STM and LTM, important for novel items

Decay and interference

  • McGeoch 1932: “Rust does not occur because of time itself”
  • Malmo 1942: monkey with frontal lesions able to perform delayed response task 97% accurate if no visual stimulus or motor activity during 10s delay; 25% accurate when unrestricted.
  • McKone 1995: nonwords decay faster than words for post-LDT recognition
  • Interference: at encoding or retrieval, at cue representation or response level, retro- or proactive
  • Proactive interference is similarity-sensitive, affects both encoding and retrieval of new items, and increases over time, unlike retroactive interference
  • Decay and interference interact
  • ‘Temporal synchronisation of neuronal activity is an important part of the representation’: Deiber et al 2007, Jensen 2006, Lisman & Idiart 1995, Lustig 2005

Keppel and Underwood 1962: poorly learned items result in retention interval effects even on first trial

Rehearsal? ACT-R (Anderson) stores 1 active chunk, EPIC (Myer and Kieras 1997) has a special auditory store, and Burgess & Hitch model has a winner-take-all network.

Nairne 2002, Oberauer 2006: bundles of features for stored information

Zucker & Regehr 2002: synaptic plasticity over 10ms. Rapid decay of active firing pattern → short-term potentiation


Zedelius, Veling, Aarts Consc & Cognition 2010

Boosting or choking – How conscious and unconscious reward processing modulate the active maintenance of goal-relevant information [fMRI, Reward]

Remember 5 words & Recall in arbitrary order after delay with distractors, and preceding reward cue.

Expt1: 1s fixation → Reward cue - coin image 1c / 50c (50%), either 300ms supraliminally or 17ms subliminally (50%) with 1s premask and 600ms postmask → 1s fixation →5 words visually presented 600ms → delay period with 8 visual distractors 1300ms each, different colour: unrelated meaningful words (‘high interference’) or meaningless letter string (‘low interference’) 50% → recall cue → repeat words in any order → feedback

Expt2: same except reward cue after all target words



  • high reward (conscious or unconscious) boosts maintenance even if presented after encoding
  • conscious perception of high reward impairs performance in recall
  • reward-dependent increase in performance was due to an improvement of the active maintenance process


Olivers & Nieuwenhuis Psychol Sci 16:4:265-9 2005

The beneficial effect of concurrent task-irrelevant mental activity on temporal attention [Behaviour, Attention]

Attentional blink – previous subjective report of better at detecting T2 when ‘defocused’

‘Concentrate on task’

‘think about holidays’ or ‘shopping plans’ – free association

Continuous rhythmic tune at 120bpm – ‘listen to the beat’

rhythmic tune plus occasional yell that was part of the music – if a yell occurred (15% of trials) then press ‘XX’ instead of response (those trials were excluded)

Conclude ‘more diffuse attentional state’, or positive affective state, or yerkes-dodson effect


Olivers & Nieuwenhius Psychol Science 16:4:265-69 2005

The beneficial effect of concurrent task-irrelevant mental activity on temporal attention [Behaviour, Attention]

Attentional blink: RSVP 88ms + 32ms blank. Task: identify the 2 digits in a stream of letters, 0.8 degrees size, at fixation.

4 groups:

Standard-instructions, to concentrate and report as many targets as accurately as possible

free-assoctiation, to think about holiday or shopping for an imaginary dinner

listen-to-music, to passively listen to nonverbal music, +/- report a yell in the tune

reward, paid dependent on accuracy.

Low level of distraction improves performance.

Diffuse attentional state – could be arousal curve (Yerkes & Dodson 1908), positive affective state, increased interest with unusual instructions. Could ‘widening’ the state of attention also widen the temporal properties?



Mulckhuyse, vdStigchel, Theeuwes JNeurophys 102:1451 2009

Early and late modulation of saccade deviations by target distractor similarity [Behaviour, Saccade, Attention]

2 possible target locations 7.7deg ecc. Equiluminant grey→red change indicated target; fixation removed at SOA of -150 to +150ms. Expt 1: 1 block with, 1 without distractor. Distractor identical to target, at 6.5 deg to left/rt. Warning beep if RT<80 or RT>600ms. Expt 2: trialwise distractor present / absent; distractor square (nonidentical to target).


Van Zoest, Donk Perception 33:927 2004

Bottom-up and top-down control in visual search [Behaviour, Attention]

Independently varied salience of distractor compared to target, and T-D similarity.

Theeuwes 1991,1992,1994: a salient distractor disrupts singleton search

Kumada 1999: 45 L-tilted line target among vertical, disrupted by 1 R-tilted distractor knowing target identity doesn’t help

Folk 1992: involuntary orienting only if distractor is uniquely the same colour as the target

Egeth 1994: modes of search – for elements that differ from their background, singleton detection mode, feature detection mode.

Expt: Target present judgement on array. Nontargets + 1 target + 1 distractor, or Nontargets + 1 distractor1 + 1 distractor2. RT for target-present trials calculated when target most salient vs distractor most salient (different from background nontargets).

Conclude: independent effect of relative salience and similarity


Sperling & Melchner Science 202:315 1978

The attention operating characteristic: examples from visual search [Theory, Attention]

Report 2 targets in an inner and outer grid of characters. Target: letter amongst numbers or number amongst letters. Only 1 out of 24 grids contains the 2 targets; task report identity, location of targets and confidence rating.

Conditions with ‘noise’ (one grid has squiggles) and ‘reversal’ (one grid has reversed target-distractor relation. All blocked design. Control blocks: report only one target


Jazayeri & Shadlen Nat Neurosci 13:1020-6 2010

Temporal context calibrates interval timing [Behaviour, Time]

Two flashes (ready, set) followed by button press (go). The sample interval was given by a flat prior in 3 blocks after adapting to >500 trials each: short = [490-850], medium = [670-1024], long = [840-1200] ms. Production time measured from second flash till button press. Feedback green light if within an adjusted window around the interval.

Shows observer is Bayesian rather than using a simple estimation mechanism (noisy measurement → estimate of interval). Modelling shows Bayes least squares fits better than MAP or MLE (last diagram, showing performance as solid, and model as dots). MLE and MAP significantly underestimate the bias. BLS uses 2 noise sources, measurement and production, both with different Weber fractions.


Baumeister & Showers E J Social Psychol 16:4:361-83 1986

A review of paradoxical performance effects: Choking under pressure in sports and mental tests [Theory, Reward]

Is it self-awareness or attention

5 models of how self-awareness can harm performance:

  1. performance-contingent reward causes performer to imagine already having won reward → distraction
  2. worry from self-focused fear of failing → distraction
  3. attention to self is per se a distraction (though process-oriented models of self-awareness don’t require attention to task or self to be mutually exclusive)
  4. skilful performance is best if executed automatically. Conscious control therefore disrupts.
  5. conscious processes are slower and the extra level of control causes an extra time cost to reject other nonoptimal responses


Sarter, Gehring, Kozak Brain Res Reviews 51:2:145-60 2006

More attention must be paid: The neurobiology of attentional effort [Theory, Attention]


Evidence from sustained attention paradigms: ACh from basal nuclei of Meynert modulate attentional gain of thalamic inputs to cortex.


Beatty Psych Bull 91:2:276-92 1982

Task-evoked pupillary responses, processing load, and the structure of processing resources [Autonomic, Attention]

Kahneman (1973)’s 3 criteria for a measure of mental effort: sensitive to

  1. within-task variations produced by changes in task parameters,
  2. between-task differences in processing load elicited by qualitatively different cognitive operations,
  3. between-individual differences in processing load as individuals of different abilities perform a fixed set of cognitive operations


In digit span, pupil peak dilatation is a monotonic function of string length

Manipulations that increase span (make it easier, like grouping or rehearsal) reduce effect

Dual task decrement due to difficulty of primary task correlates with pupillary dilatation

Dilatation asymptotes once the capacity limit of memory is reached

Same-different sequential ‘A/a’ gives more dilatation than ‘A/A’

Same-different word meaning task: words chosen from easy or hard vocabulary (Ahern 1978)

Sentence complexity in Baddeley’s Grammatical Reasoning Task: “a follows/precedes b”, then “a-b” or “b-a” is an exemplar? Active vs passive, positive vs negative sentences.

Sentence comprehension vs repetition: dilatation pattern differs if question asked before/after

Repeat 6-word sentence – normal, grammatical but meaningless, or random words.

Mental multiplication/division difficulty (Payne et al 1968) – pupil nonlinear with respect to %correct, solution time, subjective difficulty rating; Pupil peaks rapidly.

Detection of luminance increment / tone in noise: dilatation only iff target detected.

Subjective certainty rating for tone detection proportional to amount of dilatation

Difficulty of tone detection does not modulate pupil, despite impairing performance: “evidence this is a data-limited not resource-limited process” (Norman & Bobrow 1975)

Difficulty of tone pitch discrimination (higher/lower?) does modulate pupil

Sustained attention / vigilance (45 min): pupil smaller and performance deteriorates over time

Selective attention to 1 of 2 channels of tones: tiny evoked pupil dilatations seen for nontargets on attended channel but not unattended channel.

“Magnitude of pupillary evoked responses... is independent of baseline pupil diameter over a physiologically reasonable but not extreme range of values”

Students divided according to Scholastic Aptitude Test; Pupil responses on 3 types of task consistently smaller for more intelligent subjects. (whereas light reflex showed no difference)

Kahneman Peavler and Onuska 1968: short term memory, trial-to-trial varying monetary incentive for performance: incentive did not affect pupil or performance. Also: digit span: repeat string twice, or rehearse once then repeat once. (rate 1/sec). Silent repetition gave same form but decreased amplitude of pupillary response. [could this be reward/feedback related?]

Pupil is “a measure of aggregate task-induced utilisation of multiple processing resources”.


Beatty Psychophysiology 19:2:167 1982

Phasic not tonic pupillary responses vary with auditory vigilance performance [Autonomic, Attention]

Parasuraman 1979 using SDT showed increased conservatism and decreased sensitivity over time; sensitivity decline only if task required ongoing comparisons of event with WM.

Lowenstein & Lowenfield 1962 thought tonic pupil diam = general ongoing arousal level.

P.Lavie 1979: ultradian and circadian rhythm of basal pupil diam.

Yoss et al., 1970: Tonic pupil diam decreases at sleep onset, rapid redilation on waking.

Tone bursts every 3.2 sec; detection of a target tone probability=0.12, over 48 minutes.


Luria The working brain 1973

The working brain (monograph) [Theory]

Reticular formation plays a role in activating the cortex and regulating the state of its activity. The higher levels of the cortex, which form plans, “recruit the lower systems of the reticular formation of the thalamus and brain stem, thereby modulating their work and making possible the most complex forms of conscious activity.”



Page & Norris Psych Rev 105:4:761-81 1998

The primacy model: a new model of immediate serial recall [Theory, Memory]

  • Initial activations of memory items decrease with serial position: ‘primacy gradient’, (x1-x2)=(x2-x3)=(x3-x4)...
  • Recency is confined to the last position
  • During recall, the activations decay exponentially over time
  • Initial activations all modulated by a factor that decays exponentially from list start time (to compensate for decay before recall)
  • Rehearsal as re-presentation. (Explains word-length, list-length, and delay effect interactions)
  • Rehearsal is cumulative, not repetitive (last item) or associative (connections)
  • Noise only in selecting the item from memory. Errors are mainly transpositions. Primacy and recency end-effects come about because of this!

Old model doesn’t explain errors due to omissions, intrusions, repeats: these errors should increase monotonically with serial position.

  • New model: Activation is thresholded before recall. Subthreshold → guessing.
  • Henson 1996: repetitions are “literally few and far between” → decaying response suppression; not modelled here.

[doesn’t explain intrusion serial positions, nor temporal grouping effects]


Bradski, Carpenter GA, Grossberg Biol Cybernetics 71:469-80 1994

STORE working memory networks for storage and recall of arbitrary temporal sequences [Theory, Memory]

Model of event storage (Grossberg 1978 J Math Psychol):

  • uses 2 gated layers to create a WM that records item and order information
  • Each possible event corresponds to a unit.
  • WM unit: arbitrary sized chunk that maps (via weights) to LTM units
  • Magnitude of unit activity represents recall order, highest first. (Also explains %correct, order in free recall, RT data)
  • Readout (=recall, rehearsal) = wave of increase in activity.
  • ‘Invariance principle’: storing an item multiplies all currently active items by a common factor → temporal order
  • ‘Normalisation rule’: total network activity can increase up to a finite maximum → WM capacity

bowed serial order curve for 0<A<1

y relaxes to a steady state after each successive input is switched off. This time determines fastest storage speed.

This paper:

  • using only one layer results in sensitivity to timing of input presentation (amount of decay of previous items becomes dependent on presentation duration of new item)
  • add a decay term in dx/dt := -Bxi. parameter B determines bowing position.
  • attempt to allow repeated items: uses a ‘preprocessor’ that “represents repeated items in spatially separate channels”. requires a 2D array of (items x repeats)!
  • variant of dx/dt equation that allows simultaneous storage

[concludes that, if curve is bowed, serial order is only partially stored]


Jones, Macken, Nicholls JEP:LMC 30:3:656-674 2004

The phonological store of working memory: is it phonological and is it a store? [Behaviour, Memory]

Phonologically similar items → harder to recall; however it occurs even for visual stimuli.

Ignored auditory items → interference; however it occurs even for visual stimuli

Premises: (from baddeley)

  • Visual-verbal stimuli only enter phonological store after grapheme-to-phoneme conversion via articulatory loop
  • Phonological store decays without rehearsal in articulatory loop
  • Auditory-verbal stimuli have obligatory and direct access to phonological store

Prediction: “the effect of irrelevant sound and the effect of phonological similarity each survive the action of articulatory suppression, but only when the presentation is auditory”.

Hanley & Broadbent 1987: “irrelevant sound effect was abolished by articulatory suppression with auditory presentation.” – but synchronous distractors →may be due to encoding effects.

Target digits 0-9 presented 250ms each, 750ms ISI, either auditory or visual: blocked.

Distractor letters a-h (auditory only; at lower pitch & 60ms asynchronous to aid streaming) every 250ms. 3 distractor conditions, trialwise: none, repeated or random.

Suppression: whisper ‘x,y,z’ each second, from 2s before presentation, until recall: blocked.




Morris Moscovitch in Schacter & Tulving 1994

Memory and working with memory [Theory, Memory]

Modules are ‘stupid’ without a frontal executive. Memory consists of operation of modules and central systems


  • reduces interference in memory.
  • Is it conscious or spatial? Both if consciousness always has a spatial component.
  • Reinterprets Tulving 1983’s encoding-specificty principle as “consciousness in, consciousness out”. [Trying to say that hippocampus is the seat of consciousness?]
  • [cf Atkinson and shiffrin 1971 “we tend to equate the short-term store with ‘consciousness’”]

Frontal lobes convert remembering from “stupid reflexive act” to an “intelligent goal-directed activity”.

Craik 1971’s “release from proactive interference”:

  • A-B, A-C paired-associate learning negative transfer. Implicit recall (first word that comes into your mind) gives more interference from B-list than explicit (recall paired-C given A).
  • Opposite result by Schacter: explicit (complete paired-C given A) stem completion for C gives more interference from B, than implicit (first completion of C that comes into your mind)


Burgess N & Hitch TICS 9:11:535-41 2005

Computational models of working memory: putting long-term memory into context [Theory, Memory]

WM vs LTM:

  • Sensitivity to phonological vs semantic similarity
  • Maintained firing / short-term potentiation vs LTP
  • Chaining (omitted items unlikely to be retrieved until end of recall) vs independence (alternating phonemic similarity ‘QDRBNP’ does not affect intervening items QRN)
  • Hippocampus vs neocortical damage

Hebb: immediate serial recall – repeating list, even with intervening lists → improvement

Milner: this depends on hippocampi

Estes: perturbation model – serial order can change by 1 on each time step.

Evolving context signal

  • Burgess & Hitch 1999 moving window of activation, Houghton 1990, Henson 1998 – increasing and decreasing nodes
  • encoded with items, with noisy retrieval by context.
  • supplementary motor cortex has cells that fire between 2nd and 3rd action irrespective of identity of actions.

Competitive queueing

  • Grossberg 1978, Page&Norris, Houghton. most active item is output then suppressed.
  • Consistent with transposition errors.
  • PFC neurone activity corresponds to order of queued actions (Bullock 2004 TICS)

Conrad 1960: intrusions from previous lists occur at same serial position. Henson: this is maintained relative to both start and end of variable-length lists!

Temporal distinctiveness

  • Murdock 1997, Minsink 1988, Neath 1993
  • recency effect even if WM erased → more retrievable if in a distinctive context
  • “captures continuity between STM and LTM but ... tends to ignore the many dissociations”

LTM: Distinctiveness + chaining → temporal context model (Howard & Kahana 2002 J Math Psychol). Explains forward bias in retrieval – given an item, more likely to recall item after than one before. Recall as random walk of context units.

? same as chunking

Gathercole 2001: Serial recognition – no effect of lexicality. Serial recall – strong effect.

(Nonwords > words) have effects of word length, phonemic similarity, articul suppression

Hebb repetition is sensitive to temporal grouping, but not articulatory suppression or phon similarity

Botvinick & Plaut Psychol Rev 2005: bigram frequency effect (neural network model)


vd Stigchel, Arend ... Rafal Neuropsychologia 48:3497-504 2010

Oculomotor integration in patients with a pulvinar lesion [Disease, Saccade]

  • Thalamic stimulation can evoke saccades; thalamic neurones active during saccades (Crommelinck 1977, Schlag-Rey & Schlag 77) in retinotopic manner
  • Rafal 2004: thalamic lesions → absence of fixation offset effect for visually evoked saccades
  • Antisaccades slower for saccades away from contralesional target (Arend 2008)

Expt 1: distractor always in opposite field to target, at 150 degrees. 344 trials + 24 practice. Include 80<RT<1200ms. t-test between conditions (distractor present? ipsi/contra?)

Saccades slowest and distraction greatest, if distractor in good field.

Conclude: pulvinar contributes to control over reflexive saccades to the contralesional field

Expt 2: single target disc vertically above fixation, 33% with distractor diamond above and to right or left of fixation (~53deg away), simultaneous with target. Fixation offset: -150ms, -50, 0, +50, +150ms relative to target.

Trajectories deviated away from ipsilesional distractor, but not from contralesional distractor.


  • Inhibition stronger for ipsilesional field.
  • Since contralesional distractors did not give deviation towards a saccade, there was still some inhibitory control – pulvinar not plays a role in integration of high/low order information, but not crucial.
  • No effect without distractor → pulvinar not involved in target selection, but in inhibition.


Snow, Allen, Rafal, G Humphries PNAS 2009

Impaired attentional selection following lesions to human pulvinar: evidence for homology between human and monkey [Patient, Attention]

Ventral pulvinar ←→ ventral stream of processing; dorsal → frontoparietal network

SC projects to ventral pulvinar (Shipp 2003)

Visual areas: posterior → rostrolateral pulvinar; anterior temporal → caudomedial pulvinar

Expt1: orientation discrimination threshold in each hemifield: gabor to L/R of fixation, tilted to L/R of vertical. L/R 2AFC. Distsractors: Gaussian bright/dark discs.


  • Not attributable to visual acuity. Filtering deficit in both hemifields, but contralesional visual field worse.
  • Equivalent to increased crowding, e.g. pooling of luminance without mediation of pulvinar.
  • Resembles monkey lesions of extrastriate ventral stream.
  • But discrimination not impaired by high contrast distractors. ? pulvinar increases temporal synchronicity


Jensen, Idart & Lisman Learn Mem 3:243-56 1996

Physiologically realistic formation of autoassociative memory in networks with fast theta/gamma oscillations: role of fast NMDA channels [Theory, WM]

  • Theta 5-8Hz subdivided into 7 gamma cycles (20-60Hz): ‘multiplexed STM buffer’
  • Fast NMDA channels (neocortex) maintain autoassociative STM over gamma subcycles (15-50ms): components of a single memory are bound – single novel items
  • Slow NMDA channels (hippocampus) maintain heteroassociative memories (150ms): several items in WM are bound together – several non-novel items
  • Afterdepolarisation ‘sufficient duration to trigger firing on subsequent theta cycles’
  • “stands in contrast to reverbaratory...loops of interconnected cells (Hebb 1949)” [but does it though?]

Sternberg 1966’s temporal duration of memory ?corresponds to period of gamma oscillation

O’Keefe: Hippo place cells have different phases during theta cycle

This paper: how can LTM be generated?

  • Hopfield 1982: recurrent collaterals with modifiable synapses can lead to distributed LTM
  • Cyclic repetition of STM gradually causes NMDA-dependent LTP/D
    (requires >1s stimulation, >10s to be expressed → buffering in STM is needed)

Qualitative difference between STM for novel and prelearned stimuli:

Known items: single presentation is enough to learn; ‘episodic’

Novel items: noise → some cells fire in wrong gamma cycle → needs error correction, e.g. by repetitive presentation + LTD mechanism to remove erroneous learning... or presentation of items in different orders.



Terms: After-depolarisation 200ms, afterhyperpolarisation 5ms, beta drive, inhibitory feedback, AMPA/NMDA synapses, and input to memory.

  • Theta oscillations: septal cholinergic and GABAergic input to hippocampal/entorhinal oscillator. Medial forebrain nuclei similarly drive cortical areas.
  • GABA feedback: interneuronal network rapidly provide generalised inhibition after a subset of pyramidal neurones fire 4ms, aN=network size=5.
  • Plus model of LTP, dependent on bound fraction of NMDA

Memories are inserted every 340ms – after theta troughs, in between regeneration of the afterdepolarisation. However they are sequentially reinstantiated with 12ms separation → ‘time compression’. Moment (theta phase) of insertion does not matter; item always occupies last vacant gamma cycle.



Consider situation where network already contains autoassociative representations of each item in LTM (pre-existing synaptic connections).

  • Timing noise is corrected. Degraded patterns get completed.
  • An 8th item will erase a previous item
  • This network itself will not store the list into LTM (heteroassociation)
  • Too much overlap (eg >20%) corrupts the code; Rolls & Treves 1990 suggest the brain has 2% sparseness.
  • Cannot store 2 identical memories at once. ?explains repetition blindness

Cooling frontal lobe (but not other cortical regions that show sustained firing during intervals) abolishes WM in delay tasks.



Wallenstein, Eichenbaum, Hasselmo TINS 21:8:317 1998

The hippocampus as an associator of discontinguous events [Theory, WM]

         Hippocampal amnesia: Intermediate STM loss, but also impairment in rats “performing tests of associative learning such as time interval discrimination” [sic]

         Fornix lesion: able to perform discrimination OK, unless brief interruptions of the period to-be-estimated. Normal rats summate the total interval, fornix lesions initiate timing afresh. (Meck 1984)

         Rawlins 1985: Hippocampal tasks are those with temporal discontiguity – air puff conditioning in hippocampal rabbits is impaired only when CS and US are separated by >500ms. (Thompson 1982)

         Hippocampal rats can learn spatial cues when they are close to one another, but not when distributed spatially around maze

         Hippocampal rats learn A-B and B-C pairings normally, but fail to make transitive inference (infer A-C) (Bunsey & Eichenbaum Nature96)


100ms window in which postsynaptic cell must be depolarised for LTP to occur.

Noncontiguous events ‘bound’ by CA3 recurrent excitatory connections on pyramidal cells (Levy 1989)

Context fields:

  • single item = unique group of cells firing simultaneously; each cell fires for a few consecutive items in a sequence, signalling a temporal context. (entorhinal input via perforant path)
  • Initially cells fire randomly in background due to recurrent activation from other cells
  • When exposed to repetitive sequence, these undergo LTP, due to the overlap in time, & generate overlapping temporal fields – “like glue”

May also occur in PFC, parahippocampal cx, +?medial thalamus/lateral amygdala. Arise robustly – even in simple integrate-and-fire models. Requires:

  • Asymmetric connections among excitatory synapses (→order-sensitive). Cells in hippo contact 1-5% of neighbouring cells.
  • Low diffuse background activity during learning – contribution from cholinergic basal forebrain input. In HC, 10-20% cells active at a time.
  • Periodic switching between afferent and intrinsic sources of activity: brief alternating periods of ‘input’ and ‘association’. GABAB inhibits Schaffer collateral EPSPs (stratum radiatum) in CA1, but not perforant path transmission. ? synchronous with 4-10Hz theta.

Can solve sequence-learning/time-delay tasks, pattern completion, transitive patterning and inverse, spatial learning.


Cho, Ko... Van Eimeren, Cilia, Strafella Brain Stim 3:170 2010

Continuous theta burst stimulation of right dorsolateral prefrontal cortex induces changes in impulsivity level [TMS, Impulsivity]

“impulsivity measured by go/nogo, stop-signal, the [iowa] gambling task”.

“PFC neurones reflect reward preference in delay tasks” [- but cites the 2005 pigeon study on nidopallium caudolaterale...]


80% FDIO threshold, over right DLPFC.

Continuous theta-burst stimulation = 200*(3*(pulse+20ms gap)+140ms gap) = 3 min

Intermittent TBS = 20*(10*((3*pulse+20ms gap)+140ms gap)+8s gap) = 40 sec

Sham = same but perpendicular to region.


2AFC delay discounting inventory, immediate vs delayed reward: 1wk10yrs, $1-1000. 120questions x 5s each = 10 min. Gap of 45 min between TBS conditions.

Hypothetical rewards.


Petrides, Alivisatos, Evans, Meyer PNAS 90:873-7 1993

Dissociation of human mid-dorsolateral from posterior dorsolateral frontal cortex in memory processing [fMRI, WM]

Pointing to a symbol on cards, each card contains same 8 symbols in different positions. Control: always point to a fixed previously demonstrated symbol. Self-ordered: point to a different design on each card until all 8 have been selected, then white card →reset. Conditional task: point to a symbol depending on the colour of the card, previously trained to associate choices by trial and error until no errors. PET:

(Self-ordered – Control)=right mid-dlpfc

(Conditional – Self-ordered)=right mid-dlpfc (areas 9, 46; middle sfc+ifg ~ granular cx)

(Conditional – Control)=posterior dlpfc (Lt area 8, deep superior frontal sulcus)

[maybe component of inhibiting their learned responses to card colour?]

Conclude: 2 functional systems: mid-dlPFC = WM; posterior dlPFC = higher-order control of behavioural responses involving selection dependent on environmental contingencies


Droll, Gigone, Hayhoe J Vis 7:14:6 2007

Learning where to direct gaze during change detection [Attention, Behaviour]

Rensink 2002: “change simultanagnosia”, unable to see more than 1 change at once

  • 8 objects in different grid locations, jitter 3 deg, 400ms, blank 100ms, mask 250ms, blank 100ms, second array 400ms, with zero (~10%), one or more objects had different orientation
  • repeating until subjects “confident they knew whether there is a change or not”
  • Then ‘inquiry array’ where subjects mouseclick the “location of the object they first detected as having changed” or no-change icon.
  • Then feedback: showed answers with red circles.
  • Unknown to subjects, each object shape had a fixed probability of changing: 60%, 30%, 10%, 0%

More fixations on items likely to change.

Expt2: 2 border colours, indicating probability of change = 18% vs 2%. Subjects unable to learn probabilities!

Conclude: reward-related learning.


Hengquing, Todd, Beilock, Lleras J Vis: VSS 10:7:251 2010

Endogenous attention control “chokes under pressure” [Abstract only] [Reward, Behaviour]

Coloured dot precue 80% predictive of target location. SOA 200, 600, 1000ms.

Alternated with mental arithmetic.

Second half: “you will be videotaped”, “you can double your money if you and partner improve by 20%, and partner has already improved”

Cueing at long SOA is worsened.


Baldassi & Simoncini “Nature precedings” hdl:10101/npre.2009.3023.1 2009

Reward sharpens orientation coding independently on attention [Reward, Behaviour]

Cue line 45 L/R indicates which orientation has high reward probability (90%).

Then dual task:

1) count a foveal disc flashing 100ms + gap 400-4000ms x 3-14 times, under noise, varying contrast to yield 60% or 90% accuracy = varying attentional load

2) synchronous with one of the discs, a target gabor appears 7 to L/R 100ms, angle random but near 45 L/R. Subject mousclick to select number of discs counted, and then identify orientation.

  • 3 reward-probability levels: If correct report then 0%, 10% or 90% P(reward). Wrong counting or discrimination = always 0%.
  • 2 attentional load levels, 60% vs 90% disc-counting accuracy

reported tilt for 4 & 8


  • orientation threshold improved between 10% and 90% reward conditions, “unaffected by attentional load to central counting task” → “difference obtained in light load condition could not be attributed to residual attentional resources allocated peripherally in the high-reward condition”
  • Effect obtained when target is in the rewarded direction.
  • High-reward: narrower error, and more veridical (m closer to stimulus value).
  • Mispositioning of peaks to higher tilt values = ‘off-orientation looking’, relying on more-tilted channel to optimise performance. This effect is not influenced by attentional load.

As the stimulus itself (in conjunction with the cue) indicates the reward probability on each trial, unlikely to be due to motivational state

P(reward) at each orientation governs SNR of units, independently of attention.

?mediated by D1 receptors in V1


Mller, von Cramon, Pollmann J Neurosci 18:7:2720-8 1998

D1- versus D2-receptor modulation of visuospatial working memory in humans [Drug, WM]

Goldman-Rakic 1995: Monkey PFC neurons around sulcus principalis show delay activity; iontophoretic D1 blockers enhances activity in spatially tuned fields. Region has 3x as many D1 than D2 receptors. D1 localised on pyramidal neurones.

Luciana et al 1992: bromocriptine 2.5mg (D2) improves a delayed visuospatial task.

0.1mg pergolide or 2.5mg bromocriptine or placebo (1 day, different weeks). Prior practice to minimise repetition effects.

Letter cancellation, Trailmaking, Adjective mood scale, State trait anxiety inventory performed before and after drug.

Location of a dot-pattern, compare after 2/8/16s delay which is same or shifted 3mm L/R within a frame rectangle.

Pergolide (D1) improves WM but not bromocriptine (D2)


Hinson, Jameson, Whitney JEP:LMC 29:2:298-306 2003

Impulsive decision making and working memory [Behaviour, WM, Impulsivity]

  • Delay discounting greater under WM load.
  • Impulsiveness (BIS), dysexecutiveness (DEX), and delay discounting all correlated between subjects.

Frontal patients have ↓WM and poorer impulse control – decreased inhibition systems, poor planning and evaluation; but may reflect emotional (vmPFC) rather than cognitive (dlPFC) impulsiveness.

Expt1: remember digits “25341”, cue: “press the number to the right of 3”. Simultaneously $100-$900 immediate, vs $1100-$2000 at 1wk to 2 yrs. Compared with ‘pick random number’, or ‘press the displayed number’ control tasks.

Expt2: increasing number of options: immediate vs 1/2/3 other delayed sums.

Conclude: “person may come to adopt an impulsive syle” – “make quick decisions on the basis of the simplest or earliest information”. ? selecting immediate reward as a “simplifying strategy”



Boucher, Palmeri, Logan & Schall Psych Rev 114:3:376-97 2007

Inhibitory control in mind and brain: an interactive race model of countermanding saccades [Theory, Saccade]

Paradox: psychological model of 2 independent processes, vs. physiological model of closely interacting inhibitory and movement-initiating cells.

SST: model Logan & Cowan 1984 race model, explains keypress, verbal, arm, squeeze, saccade and head movement RTs. Vary with age.

Linking proposition”: to link “cognitive processes with neurophysiology, researchers must identify the population of neurons that carry out a particular cognitive process”.

bridge locus” FEF/SC movement neurones=go? FEF/SC fixation cells=stop? (i.e. not motorneurones)


Independent model: has a ‘leakage’ k to ensure a is bounded. Interactive model: has mutual inhibition b. q=1000. x is Gaussian noise with different sd s2go and s2stop. Activations rectified (Usher ^ McClelland 2001). 10ms ballistic onset time (omnipause cessation). Separate model parameter for fixed delay of units Dgo (manually set for each monkey based on mean neuronal onset time) and Dstop (fitted). RT = Dgo + t + 10ms; SSRT=Dstop + t + 10ms

Tried with Dstop=0, mstop=mgo, bstop=bgo, and both m and b same.

If fitted, bstop >> bgo. Dstop=60ms.


Hintzman & Block JEP 88:3 1971

Memory for lists: each trace has a time tag. [Behaviour, Memory]

Repetition and memory: evidence for a multiple-trace hypothesis

If an item is repeated, does it increase the strength of the memory trace that was previously associated with the item, or does it create a new memory trace on each instance? Assume 1) each repetition leaves its own treace, 2) each trace coexists with other repetitions of the event, and 3) each trace can be identified by its time tag.

Yntema & Trask 1963: “which X and Y occurred more recently?”

Hinrichs & Buschke 1968: “how many intervening items since X?”

Some tasks may not discriminate between ‘AACD’ and ‘BCDA’, but that doesn’t mean the information is not in memory.

Expt 1: “write down serial position of item X, one number if it occurred once, two if it occurred twice.”: 50 visually presented 3-letter nouns, 5 sec per item. Then pen/paper task –which serial ‘dectile’ was it in. No repeats.

Expt 2: 4 zones A=3-8, B=9-14, C=15-20, D=43-48, chosen to maximize ability to discriminate A/B and C/D. Words were either ‘never occurred’, ‘once’ A/B/C/D, ‘twice’

No repeats: serial position judgements show “primacy effect” (‘increased sensitivity’ for position of earlier items) that is not explainable as trace strength.

With repeats: 2 positions of an item can be remembered independently.

Expt 3: Frequency judgement – to show n repeats is not the same as increasing the strength.

2 lists of 104 words, separated by 5 min. frequency in each list = 0/2/5 repetitions; order such that the test items occurred between position 26 and 84 – to keep recency/primacy effect similar. Spacing: >3, mean 13 items. Subjects write freq in list 1 & list 2 of 36 test words.

Result: subjects have some memory of ‘overall’ frequency, combined with a list-specific frequency (= memory trace is not path-independent).

Conclude: Most parsimonious explanation for 2 & 3 together, is multiple-trace hypothesis.

Does not help decide

  • whether memory is associative or nonassociative
  • whether memory traces are unitary (acquired and lost as a unit) – there could be a shared component
  • whether traces pass through multiple stages, or even whether multiple traces can merge at a late stage to form a strength-only trace.


Watanabe, Kodama, HikosakaK J Neurophys 78:2795 1997

Increase of extracellular dopamine in primate prefrontal cortex during a working memory task [Neuron, WM]

Delayed alternation task: hold bar 5s → go-light → L/R keypress, alternating reward side

Sensory guided task: hold bar 5s → signal-light L/R → press corresponding key

i.e. same motor and motivational requirements, but differ in needing WM for previous trial

extracellular microdialysis → liquid chromatography → DA measurement

  • Dorsal and ventral DLPFC: significant difference between WM/noWM tasks; no difference in principal sulcus, premotor cx, OFC, arcuate area.
  • [other areas may also be sensitive to WM demands but may reach ceiling DA levels?]
  • Supportst Schultz 1992 – phasic DA for delayed alternation but not simple RT tasks.
  • Does not distinguish between WM demand / attentional demand / task difficulty / goal-directed behaviour / instructional component of reward delivery


Braver & Cohen in Monsell & Driver – Control of cognitive processes 2000

On the control of control: the role of dopamine in regulating prefrontal function and working memory [Theory, Executive]

Control =

  • Selecting appropriate context
  • representation and maintenance of context for length of time
  • protection of context against interference
  • updating context at appropriate junctures

2 similar mechanisms used to model AX-CPT: RSVP letters, black AX=right button, BX or CX=left button, but ignore all intervening letters that are red (distractors).

  • Phasic DA gates the PFC attractors: by transient potentiation of excitation / inhibition
  • Strengthens association of reward predictors→DA: by temporal difference learning



  • Reward-predicting-gating units match dopaminergic firing.
  • Explains how multiple reward-times can be learnt: context is maintained and provides continuous reward prediction signal – more plausible than ‘complete serial compound’ representations of stimulus over time.
  • Does not explain ontogeny of active maintenance circuits ?prewired
  • Gating is not strictly necessary, as long as task-relevant information can be selected.


Watanabe Nature 382:629 1997

Reward expectancy in primate prefrontal neurons [Reward, Neurone]

3 tasks: visible food, stimulus associated with food, stimulus associated with drink.

Principal sulcus: 124 delay neurones. 66% spatially specific. 33% managed types of reward.

  • 50% were reward-type modulated, whether or not spatially specific.
  • Associated stimuli = similar to rewards.
  • Preferred rewards gave larger reward-modulation (banana>cabbage)
  • Reward modulation was independent of L/R stimulus location, even in cells spatially selective for the stimulus (= reward represented independently of location)
  • 77% showed task-dependent differential activity, 27% required visible reward
  • One had spatial selectivity in visible food task but not in stimulus-association
  • 8 showed different pattern between food and liquid, irrespective of visibility
  • 3 showed distinct responses for all 3 tasks
  • Unexpected rewards during food-visible trial: monkey upset, neuronal activation for 1 minute with no delay period activity!
  • Unexpected reward during associated-stimulus trial: cells learn to predict

Conclude: reward expectancy coded in DLPFC. In parallel with spatial working memory.


Thut, Schultz,...Leenders Neuroreport 8:5:1225 1997

Activation of the human brain by monetary reward [fMRI, Reward]

PET during Go-nogo.

1 of 12 fractal images (instruction, 6 go, 6 nogo, pretrained) 1s → 2.1 to 3.1s delay → red square trigger → mouse button press <1.4s → trigger vanishes after 1.4s or after response → reinforcement only if correct. Reinforcers: (on different scan blocks) either ‘OK’ screen or ‘$1’, both with accumulated total. Semirandom: no more than 3 repeats.

Errors: false-go = 1%.

(money – OK) = Left DLPFC, left lateral OFC, right occipital, left thalamus, left midbrain, slightly anterior to STN.

Conclude: DLPFC encodes reward too.



Smith JNNP 26:535-9 1963

Cerebral pathology in subarachnoid haemorrhage [Patient]

5 fatal cases: Rt ACA, Lt PCom, Acom, Lt ACA, Rt Sylvian point, and basilar bifurcation.

  1. Lesions maximal in the territory supplied by the aneurysm-bearing vessel. Pallor of myelin ending abruptly at the edge of a vascular territory. Ischaemic lesions with sharply circumscribed patches of necrosis, becoming confluent in the aneurysmal territory.
  2. Oedema and eosinophilia in less-affected areas. Presumed ischaemia without infarct. Present throughout the cortex; note that widespread vasospasm throught vascular tree is common at angiography.
  3. No preferential layer or cortical area found. Sulci and gyri equally affected. No predilection for watershed areas.
  4. Overall, maximal damage at cingulate, hippocampus, insula. Minimal damage to hypothalamus, thalamus, and basal ganglia.

Perhaps later on, vasospasm becomes localised to the neighbourhood of the bleeding vessel. Or perhaps lesions relate to arteries too small to see radiographically.


Passingham & Sakai Curr Op Neurobiol 14:163-8 2004

The prefrontal cortex and working memory: physiology and brain imaging [Theory, Memory]

Sustained activity in humans vs monkeys?

  • Monkey 18% of cells on area 46 show delay activity. fMRI failed to find delay activity – not sensitive enough?

What sustained activity represents?

  • Posterior 1/3 of principal suclus: code for remembered location not action location (varying the response). Same is found in fMRI of area 8. In this situation, small load → very little area 46 activity. However area 46 active in face / house / pattern WM. “it could be that memorising such stimuli requires active rehearsal or recoding of the items” – Curtis & DEsposito TICS 2003.
  • Cells in principal sulcus 46 do code for future saccadic response (only when you can prepare response). There is transformation of activity during the delay period, from cue-location to response location (e.g. when some trials require a response 90 deg rotated). fMRI shows area 46 when reordering items during the delay is required.
  • Premotor cx sensitive to task rules (e.g. match vs nonmatch-to-sample). fMRI → polar & ventral PFC. Sakai & Passingham NN2003: (verbal/spatial material) x (forward/backward memory task) → frontopolar activity after task instruction but before items; reversal increases connectivity (correlation) of frontopolar ↔ area 8 in spatial, and frontopolar ↔ area 44 broca in verbal trials.
  • Expected reward-sensitive in OFC, but in lateral PFC individual cells can code simultaneously for spatial location and reward, or have reward-dependent delay activity. fMRI shows greater polar activity under higher reward-expectation.

Why sustained activity in PFC is different to that in other areas like PPC?

  • Pochon et al 2001: fMRI shows Area 46 sustained activity during delay only when subjects can prepare the response; parietal regions sustained activity even if they cannot.
  • Sakai et al NN2002: fMRI WM for spatial sequences, with spatial distractor task before recall. Delay activity in area 46 but not parietal cx predicts accuracy. When area 46 active, there is closer correlation of area 8 and parietal cx activity. Conclude: DLPFC activity = reorganising/rehearsing to resist distraction.

PFC = Prospective use of sensory information


Bo & Seidler J Neurophysiol 101:3116-25 2009

Visuospatial working memory capacity predicts the organization of acquired explicit motor sequences [Behaviour, WM]

Explicit sequence learning task: 12-element repeating sequence of 4 colours, corresponding to finger movements on 2 hands. 1 second fixed ISI, 30 repetitions then more until 90% accurate. Then instruction ‘generate entire sequence’, after viewing once at 500ms ISI. Then generation without visual cue. Then ‘transfer’ task to unimanual responses.

Visuospatial working memory task: (Luck & Vogel 97) 100ms array of 2-10 coloured squares from 7 colours, 900ms blank, 2s test array with 1 changed colour on 50% of trials, same/different judgement.

Continuous tapping: 500ms, 1000ms, 1500ms periodic tones, synchronised tapping over 12 beats, then to continue tapping for 30 more beats.

Chunking: assume chunk length >=2 items, chunk starts with 1st and ends with last item, longer RT should be at start of chunk (Kennerley 2004)

Memory capacity = array size x (hit rate – false alarm rate)

25 subjects. Fitted a power function to learning curve [didn’t give any params or pictures] but said that the ‘decay constant’ [exponent] correlated with memory capacity. Chunking pattern transferable to unimanual task only in some subjects. “Ss who formed their chunking patterns earlier in the experiment exhibited better transfer of the pattern to the new effector”. No correlation of sequence learning with tapping variance.



Kennerley Rushworth


ACC lesions on reversal learning. Monkeys choose between 2 actions. Expt 1: reward for one action only, switches every 25 trials. Expt 2: dynamic foraging, rewards assigned with unequal probabilities, and remain until obtained, 0.4:0.1, 0.5:0.2, 0.75:0.25, 1:0. Expt 3: Work-related – size of reward vs number of responses needed to elicit reward.

Used ‘ECn+1’ analysis: find an error followed by n correct trials, then what is the performance on the next trial?

Used logistic regression to explain choice in dynamic foraging:

ACC lesion “impaired their ability to sustain rewarded responses” leading to impairment on switching task and foraging task, but not work task.


Yaari ME Econometrica 55:1:95-115 1987

The dual theory of choice under risk [Theory, Reward]

Preference is a product measure, factorisable into two marginal measures. Expected utility theory takes the marginal measure along the probability axis as Lebesgue . In the dual theory, the payment axis is Lebesgue. Quiggin 1982 studies the perception of risk in this way, but assumes neither is Lebesgue and.

Assumptions: Neutrality, complete weak preference order (transitive reflexive and connected), continuity, monotonicity, and independence of distributions.

Dual independence implies comonotonicity – i.e. the two measures are not hedged.

Explains common ratio paradox ( [0.3;1] ≥ [0.4; 0.8], but [0.4; 0.2] ≥ [0.3; 0.25] ), Allais paradox (1953), and the Newbery 1979 paradox where there does not exist a vN-M utility corresponding to the Gini coefficients.

Risk aversion as concavity of vN-M utility ≡ convexity of probability perception function. “Adding noise never improves preference.”


Allais, Maurice Econometrica 21:4:503 1953

Le comportement de l’homme rationnel devant le risque: critique des postulats ex axioms de l’ecole americaine [Reward, Theory]

  • Cardinal utility as equivalent with Weber-Fechner minimum sensible threshold.
  • Distorted appearance of subjective probabilities.
  • Risk as “the dispersion (variance) and general properties of the form of the probability distribution of psychological values.”
  • Rationality should be defined as either “in the abstract by referring to a general criterion of internal consistency employed in the social sciences... [or] experimentally by observing the actions of people who can be regarded as acting in a rational manner.” –i.e. looser criteria than the Bernoulli principle.
  • Counterexamples to Bernoulli: very prudent people gambling small sums, choice of risks bordering on certainty that contradicts the independence principle of Savage, and the substitutability principle of Samuelson, and the behaviour of entrepreneurs when great losses are possible.

Bernoulli requires Independence axiom: i.e. indifference between gambles L1 and L2 implies indifference between mixtures (L1+L3) and (L2+L3).

A: [1m; 1] > {[1m; 0.89] [5m; 0.10] [0; 0.01]}

B: {[1m; 0.11] [0; 0.89]} < {[5m; 0.10] [0; 0.9]}

Counterexample to independence axiom: 89% of the time, both gambles of A are identical (1m) and of B also (0). The remaining 11% is the same for A and B (1% nothing and 10% 5m).


Camerer, Loewenstein, Rabin Behavioural economics

Bernoulli 1738: defined expected utility. St. Petersburg puzzle: game with 2n payoff where n is the number of coin tosses until a head appears. , but “most people would be prepared to pay only a relatively small amount to enter it”. Assumes cardinal utility.

Uncertainty: some of outcomes or their probabilities are unknown.

Prospect: list of outcomes and their probabilities → Risk.

EU requires preference ordering (completeness + transitivity), continuity, independence.

‘Conventional’ strategies allow violations of independence but retain monotonicity.

Preference Reversal (Lichtenstein & Slovic 1971): two prospects – $-bet (unlikely high reward) and P-bet (probable small reward). In 2 different tasks, the same subjects will choose P-bet, but will assign higher monetary value (minimum selling prices) to the $-bet. Is this a failure of ‘procedure invariance’ (different neural processes in each situation) or failure of ‘transitivity’? (Loomes moffat & Sugden 1998)


Walker, Deubel, Schneider, Findlay J Neurophysiol 78:1108-19 1997

Effect of remote distractors on saccade programming: evidence for an extended fixation zone [Saccade, Behaviour]

Fixation, then 0-gap target (cross), with simultaneous distractor (circle) on ‘a majority’ of trials. Direction of target was always same in one block, so effect is not due to discrimination of target.



Pouget, Dayan, Zemel NRN 1:125-132 2000

Information processing with population codes [Theory, Decision]

e.g. motion direction (circular)

firing rate = Gaussian(stimulus-s0) + noise

Bayesian posterior: , and if noise is independent across neurones, then P(r|s) is the product of each neurone’s response function. How do we retrieve a single estimated value of s? MAP, mean posterior, MLE.

If prior P(s) = constant, MLE = MAP. But these methods need to know the empirical tuning curve of each cell.

  • Population vector estimator (Georgopoulos) = sum of votes. i.e. can be used without knowledge of each neurone’s tuning curve shape. Firing rate = cos(stimulus-s0)
  • Estimate is most sensitive to neurones who are on the flank of their tuning curve, i.e. point with highest derivative.

‘Standard model’

  • Nonlinear cells with lateral connections (model of hypercolumn) – smooth hills are network attractors. Noise reduction or smoothing of an activity hill corresponds to MLE, or optimal inference.
  • Set of tuning curves form basis functions for arbitrary transforms. Motor command = some weighted combination of coefficients of eye-position-basis-functions and retinal-location-basis-functions.
  • Basis functions can be learned using self-organising / unsupervised algorithms, and weights can be learned using back-propagation.
  • to allow transparency (e.g. 2 stable peaks) need to tunings with all consistent stimuli in scene. However decoding is v complex.
  • Doesn’t allow uncertainty e.g. about horizontal component in aperture motion.

‘open question’.





Cowan & Morey Psych Sci 18:686 2007

How can dual-task working memory retention limits be investigated?

WM for 3 visual items +/- 5 verbal items. Dual-task memory for similar vs dissimilar domain sets shows greater interference for similar sets, and postcues to retain only one set improve performance.

“The domain-specificity of interference disappeared if we instead compared memory on trials in which two sets were encoded and the postcue called for retention of only one of them with memory on trials in which two sets were encoded and the postcue called for retention of both sets.”

Dotted line = single set performance

Therefore, “the additional conflict was between encoding a second set and retaining the first, and was not due to dual-set retention”.


Sawaguchi & P Goldman-Rakic Science 947 1991

D1 dopamine receptors in prefrontal cortex: involvement in working memory

DA concentration is highest in PFC in monkey.

  • Iontophorically applied DA augments delayed-response performance
  • Memory guided (oculomotor delayed response) vs visually guided saccades
  • Injection of selective D1 blockers into monkey PFC → ↑latency ↓accuracy

? affects deep neurones projecting to thalamus, caudate, SC.


Vogel Woodman Luck JEP:HPP 27:1:92-114 2001

Storage of features, conjunctions, and objects in visual working memory

Binet 1909: immediate memory = window into intelligence

Atkinson & Shiffrin 1968: modal model, STM = immediate store & also gateway to LTM

Petersen & Petersen 1959: rapid decay without rehearsal

Wickens Born & Allen 1963: actually, decay only due to interference

Shallice & Warrington 1970: patients with STM deficit can still form LTM

Craik Gardiner Watkins 1970: prolonged storage in STM is insufficient to create a LTM

Baddeley & Hitch 1974: WM to emphasise holding in an active form

Jonides 1993: functional imaging dissociates VSS and PL

Baddeley Thompson & Buchanan 1975: longer words consume more memory; ? phonological code, capacity characterised by time

Sperling 1960: VSS 6 items, but: letters/numbers, report by writing down → some verbal

Ellis & Hennelly 1980: bilinguals have shorter span in welsh – vowels are longer!

Duncan 1984: if 1 feature is attended, all features of object are available

Set size effect: falloff after 3 items. Small effect of PL load for >4 items.

Effect of partial report on change detection: no effect. suggests it is not due to accumulated errors

500ms encoding time didn’t eliminate set-size effect

A simple target-colour-matching task shows no set-size effect, indicating memory comparisons cannot account for the effect

Retention interval has a small effect, only at larger set sizes

Orientation and colour tasks show same curve shape

Conjunction of the two has same curve. 4-feature conjunction also same!

Square with 2 colours in/out: conjunction is same as for single-coloured square. Shifting the inner square so it appears as a separate object → much worse performance.

Spurious synchronisations as mechanism of misbinding.



Hikosaka O, Nakamura, Nakahara J Neurophysiol 95:567-84 2006

Basal ganglia orient eyes to reward

Evidence that BG controls reward-related behaviour:

  1. motor function: BG → SC; BG → thalamus → motor cortex
  2. reward-related information: Amygdala/limbic system → v striatum (eg n acc) + d striatum (eg caudate/putamen) + SN
  3. sequential and parallel inhibitory connections within BG = selecting optimal behaviour
  4. habit learning in BG
  5. cognitive signals: cortex → BG → cortex

Dorsal striatum sensitive to sensorimotor events (somatotopic in putamen, oculomotor in caudate) and reward → integration of reward into cognitive processes (O’Doherty 2004)

Caudate projection neurones: medium spiny. Low spontaneous activiy, deep resting potential. Bursts during saccades, otherwise silent. Electrical stimulation inhibition in SNr. Inputs: association cx, oculomotor thalamus, FEF, DLPFC. Heavily innervated by DA & ACh. Firing is context-dependent. Some show anticipatory tonic firing before expected target or expected reward; selective for reward position. Reward-position-selective - 80% of cells; some even reverse their directional sensitivity. D1 blockade attenuates this reward-dependent bias. (Nakamura & Hikosaka 2004)

SC: asymmetrical reward causes L/R bias in SC, even prior to motor instruction → explains reward-related speeding.

SNr fires tonically and rapidly, also reward-modulated.


Nakahara Nakamura Hikosaka Neural Netwks 19:1027-46 2006

Extended LATER model can account for trial-by-trial variability of both pre- and post-processes

Reddi 2001: detection variability = changes in baseline, decision variability = changes in slope.

“preprocess activity”: in premotor cx, SMA, primary motor cx, cbm, FEF, SEF, parietal cx, cingulate, OFC, DRPFC, somatosensory cx, visual cx, caudate, putamen, GP, SNpr, STN, thalamus, SC. Modulated by expectation of: motor acts, WM, sensory cues, monitoring events, visuomotor transformation, sequence order, reward prediction, response conflict, free choice, predictability. Postulate: also ‘internal drive’

But now let , so instead of having s as a free parameter, it is independently chosen from a normal distribution. Substituting , we get .

This reduces to LATER when s=ms, i.e. ss=0, or , or locally where

Distance variability ss only occurs in 2 terms, and only affects latency additively in relation to t, rather than multiplicatively. Alone, it cannot account for skew. Deviation from LATER is greatest when mean latency is short.

→ can model data from 1-direction-rewarded vs all-directions-rewarded task: A single caudate neurone’s preparatory activity ↑ in the 1DR block with contralateral reward.

1DR small reward: curved data – alters both ms. and ss.




Roy Wise NRN 2004

Dopamine, learning and motivation [Theory, Reward, Learning]

Dopaminergic cells develop from mesencephalic-diencephalic junction

Nominal systems:

  • nigrostriatal (SNpc) – motor function, → caudate/putamen (=dorsal striatum);
  • mesolimbic (VTA) – motivational, → n accumbens, olfactory, septum, amygdala, hippo.
  • mesocortical (medial VTA) → medial PFC, cingulate, perirhinal cx

Ill-defined boundaries, with adjacent sources and terminations.


Dopamine is important for effectiveness of rewards: Hypotheses


  • conditioned motivation before instrumental act, (expectation) vs.
  • reinforcement of stimulus--reward & reinforcement of response--reward (after)

Ungerstedt 1971: lesion of nigrostriatal fibres ≡ lesion of lateral hypothalamus → eating/drinking deficit; lesion to mesolimbic fibres → decreased motivated forward locomotion

Wise 1986: neuroleptics: block instrumental responding for food, more than free feeding. (“attenuate the ability of food to maintain eating, far more than they attenuate the motivation to feed”)

Franklin 1979: Neuroleptics block intracranial self-stimulation in lateral hypothalamus. Showing a light that had previously conditioned been associated with reward was given → responding is temporarily reinstated not a motor effect, actually impairs rewardingness

Fouriezos & Wise 1976: On neuroleptic, reinforcement value diminishes gradually over sessions. Reinforcement is normal on drug-free days. → memory of devalued reinforcement is remembered.

Ettenberg 1985: Probabilistic reinforcement → slower to unlearn on stopping reinforcement. Training on neuroleptic → same effect. conclude haloperidol reduces reward value


Phencyclidine/NMDA antagonists – dopamine-independent reward block

Proactive reward (incentive motivation):

  • Priming effect: next trial is speeded by reward on previous trial. 1 peanut causes desire for another.
  • Short lasting only – ie = WM not LTM. Is also dopamine sensitive.
  • Incentive motivation for learnt predictors: not initially sensitive to DA blockade.
  • However, phasic DA released to predictors.
  • Giving DA slightly increases the motivational effect of the predictor.
  • Giving DA during extinction slows down unlearning

Conditioned reinforcement

Thirsty rats learn to work for a light previously paired with water

Taylor & Robbins 84: Amphetamine inj to n accumbens → enhaced responding for light. DA-selective lesion of n acc → reduced responding

Pleasure not necessary for reward / dopamine not necessary for pleasure:

  • Kelleher & Morse 1968: monkeys can work for painful stimuli (eg if previously paired with food)
  • Subjects respond at a constant rate for a dose of drug that has subjectively weaker effects over time
  • Orofacial movements for like/dislike are unaffected by dopamine lesions/blockers
  • Liking and wanting can be preconscious. Berridge & Robinson 1998: wanting = dopamine (e.g. sensitisation to drugs); liking = non-dopamine (e.g. taste reactivity)

Location: morphine is self-administered into N acc & VTA. Cocaine is self-administered in N acc, mPFC, olfactory tubercle.

LTP is seen in excitatory synapses on dopamine-containing neurones of SNc. Not seen in GABA-containing neurones of VTA. LTD seen in excitatory synapses of dopamine-containing neurones in VTA, and is blocked by D2 activation.

White & Viaud 1991: Injection of amphetamine into posteroventral caudate → ↑conditioned emotional visual response; inj venterolateral → ↑ conditioned olfactory response.

Packard 1994: Post-trial inj amphetamine into hippo → ↑consolidation of spatial but not visual task; inj caudate → visual but not spatial consolidation

Post-trial inj D1/D2 agonist into hippo → ↑retention of win-stay strategy; in caudate → consolidation of win-shift strategy

D3 inj in central vs basolateral amygdala → ↑memory consolidation in conditioned approach vs instrumental responding

  • “Information converges through glutamatergic afferents on the same medium spiny neurones that receive mesolimbic dopamine input. The glutamergic input terminates on the heads and the dopaminergic input on the shafts of dendritic spines of the N acc output neurons”.
  • speculate that “release of DA in ventral striatum, triggered by reward-associated CSs, acts primarily to energise the next response, whereas DA release in the dorsal striatum, triggered less and less by the receipt of expected reinforcers, acts primarily to stamp in the procedural memory traces” of habit learning.



Ahrens & Sahani Curr Biol 21:1-7 2011

Observers exploit stochastic models of sensory change to help judge the passage of time

Kanai...Verstraten JVis 2006: high temporal frequency stimulus of the same duration has a longer perceived duration, using reproduction of single stimulus with a key → counting?

Eagleman JVis 2004: Duration in slowed videos: observers recalibrate time to maintain physically predictable dynamics

  1. Visual cloud noise (smoothed Gaussian noise) with a given centre-frequency in its temporal spectrum. Which of 2 sequential stimuli (different frequencies) was longer (~500ms each) ? Counterbalanced staircase for subjective equality. 50ms (10%) different.
  2. Reproduction of duration of a single stimulus: double-speeded movement gives 5% lengthening, half-speed movement gives ~5% shortening.

“not limited to periodic or predictable stimuli”

But why the nonlinearity? Visual system filtering? Non-discrete sampling?

  1. mere presence of constant-frequency visual noise improves performance. Peripheral stimuli, 2 time periods in sequence, with constant central noise throughout. 2afc which is longer. dynamic noise → less variability in estimate, compared with a single static frame. [only for some subjects]. with little change in RT/lapse rate – i.e. not due to increased difficulty / distraction with static stimulus.
  2. Weber’s law holds, as predicted by Bayesian model. Simpler ‘event-counting’ model predicts sqrt law.

stochastic change model: If you know the statistical distribution of a stochastic stimulus, multiple sensory snapshots allow a Bayesian estimate of how much time has passed. This is combined with a prior from the internal clock.


Rensink, O’Reagan, JClark Psych Sci 8:5:368 1997

To see or not to see: the need for attention to perceive changes in scenes

Subjective feeling of S’s see the scene’s ‘entire structure in great detail and can immediately notice any changes in it’

Phillips 1974: array of items presented for 100-500ms, twice with short ISI, with 1 item removed or replaced on 50% of trials. Same/different 2afc. Performance poor if ISI>70ms – as though monitoring 4-5 items. → ? no transient is generated ? transient masked because stimuli are short

Bridgeman 1975: intersaccadic change blindness for all locations except saccade target. → ?saccade blurs image or masks transient.

Flicker paradigm: longer duration allows “observer to build a representation conducive to perceiving changes in a scene”. 240ms scenes with 80ms blank, order A,A,B,B,A,A etc. “to create a degree of temporal uncertainty”

  1. Regions of interest determined by verbal descriptions of other observers. Changes in areas of Central interest required 5 sec, Marginal interest required 11s, even though marginal interest changes were 20% larger in area.
  2. although ISI<<300ms iconic range, the image may need >400ms to consolidate (Potter 1976). Using A,B,A,B with 560ms stimuli (same rate of changes as before), → no improvement.
  3. visual word clue before trial, to locate change – 100% or 50% validity. → valid cues always improved performance → flicker doesn’t reduce ability to perceive image

attention required for change detection.

Rensink, in Itti, Rees, Tsotsos Neurobiol of Attention 2005

Change blindness


1)      motion vs change – change involves “transformation of an enduring structure over time” ; “motion does not involve structure”

2)      dynamic vs completed change – dynamic change maintains spatiotemporal continuity of representation

3)      change vs difference: change involves “notion of temporal transformation”; similarity is “atemporal comparison”

bottleneck: proto-objects → links → nexus of pooled information, STM.

One solution to the problem of “how to prevent properties of one object representation from being erroneously assigned to another” (binding) is “that only one object is attended at any time”

Triadic architecture: early processing → layout and gist (nonattentional); and also → object (attentional)

Change detection uses ‘volitional responses’ → “ ‘see’ must be restricted to conscious visual experience”.

  • Bridgeman 1979 describes accurate change-detection saccadic responses without consciously reportable change-perception.
  • Fernandez-Duque 2000 describes above-chance 2afc performance “without any visual experience of a change”.
  • It may be that a subset of attended properties is actually compared on a given task.


Ronald Rensink Psyche 6:09 2000

When good observers go bad: change blindness, inattentional blindness, and visual experience, commentary on Mack & Rock 1998

Mack & Rock 1998: report if horizontal or vertical arm of a cross is longer → Critical trial: small box appears nearby → did you notice it? → Then divided attention task (detect box and do judgement simultaneously). → Then ignore cross (full attention task)

Find 50% miss box initially. Also Simons & Chabris 1999 gorilla

Inattentional blindness: attending to a particular event or object, and fail to report presence of unexpected items: failure to see unattended items → first order. Suppression/selection is not needed.

Change blindess: failure to see unattended changes. → second-order information (transitions between quantities, not simple presence). Affects ‘visual experience’ but not saccades.

First vs second order: possible to dissociate the amount of ‘stimulus presence’ from the amount of ‘stimulus appearance’ by transition rate & timing.

Sensitivity to expectation: IB vanishes when Ss expect the change; CB is only attenuated.

Blindness vs amnesia: Wolfe 1999 – what is the perceptual status of unattended items? Could it be perception with failure to remember?

  • Moore & Egeth JEPHPP 1997 – ‘unseen’ items can induce grouping effects on other reported items.
  • Online vs offline reports: failure of online report → either not perceived, or can’t dual-task respond; failure of offline report → either not perceived, or not remembered
  • Direct vs indirect ‘report’: verbal/button vs priming/physiological/brain activity. Does conscious experience reliably cause direct report – i.e. can we infer no experience from a negative report?
  • Blindness to, say, gorilla, still allows for conscious experience of the colours & lines → inattentional agnosia is better
  • Independent empirical tests for: i) if a stimulus is experienced, it is attended; vs ii) if the stimulus is not attended, it is not experienced


Fred Dretske Philosophical Studies 120:1-18 2004

Change blindness

“Change blindness is more accurately described as difference blidness. It does not involve a failure to see change. The change, in fact, is carefully concealed. It is, instead, a failure to see the differences that change produces.”

McConkie & Zola 1979:

“ThE sPaCe ShUtTlE tHuNdErEd” → “tHe SpAcE sHuTtLe ThUnDeReD” -- alternating during every saccade!

“Why describe a failure to detect/perceive/notice a difference as a form of blindness?” “blindness is a failure to see visible things”

“For most values of x, S’s believing she does not see x is compatible with S seeing x. It is compatible with fully conscious perception of x. One needn’t believe in flying saucers, miracles and spies to see them. If they exist, you can seem them while beliving you are not seeing them. Skepticism about x makes one ignorant of, but not blind to, the x’s one sees” [is this nonsense?] → if you see neighbours who are spies, but you don’t think they are spies, then you have seen spies, but will not say or believe it.

Argues we can be consciously aware of objects without noticing or recognising them – cf perception of objects (no need for ability to report, e.g. that I’m aware of objects at L) vs perception of facts (requires ability to report, e.g. that there is an x at L)

Are differences like spies – can you see one without believing you are seeing one?

Stimulus detection model of perception: “a person does not see an x at location L if he sincerely reports (hence belives) seeing nothing in location L”

Object model of perception: possible for subjects to see but not notice differences; subjects “who say they see no difference are probably wrong” → differences are like objects/spies

“A friend recently shaved his moustache off. I didn’t notice...If I didn’t see the skin on his upper lip, what did this part of his face look like? What did I see between his nose and mouth? Nothing? Did his nose, then, appear to be directly above his mouth with nothing separating them? Wouldn’t I have noticed that? If I didn’t see that part of his face – his upper lip – why is that the only part of his face I didn’t see? Or didn’t I see any part of his face?” [clearly nonsense, for the same reason that ‘perceptual filling in’ arguments are nonsense. No need to invoke positive experiences in your blind spot. You belive you saw his upper lip, but you did not]

→ You can see something and not recognise it. “I didn’t see it” is wrong.

Fact model of perception: the fact that I can see the two properties of x, “does not mean that I can see, without noticing it, the difference between these states”. “Seeing a fact is (amongst other things) noticing it.”

→ you can be blind to differences but still see the objects and properties that make up the difference.

“Difference blindness is like blindness to problems, solutions, answers, and a great many other things designated by abstract nouns. The abstract noun “stands in” for a fact.” E.g. you can see the answer ‘7’ without knowing that it is the answer. [??but common sense wants me to say I can’t see the answer.]


Seeing facts: “The fact model encourages the idea that difference perception is not a visual phenomenon at all” – but actually it “is a coming to know by use of the senses” and so begins with perception and is essentially visual. “Your blindess consists of a failure to come to know by seeing, which is essentially visual, but the failure rests on the coming-to-know part, the part that occurs entirely in the head, not the by-seeing part.

[he is arguing that the object of the verb ‘to see’ describes not what our experience is like, but what our experience refers to]


Cairney Acta Psychol 39:329-40 1975

Bisensory order judgement and the prior entry hypothesis

Sternberg & Knoll 1973: prior entry of one modality when attended. Is this decision bias or channel speeding? “If attending affects only the decision mechanism, then the differences in the means of psychometric functions for order judgement should be the same for all intensities.” Converseley, if latency of one channel were reduced, then differences in means should diminish as the intensities of the two signals increase.

But: could be response bias because of instruction ‘to attend’

Expt1: range: visual 70ms before – 30ms after the auditory stimulus. Conditions:

1 – simple temporal order. 2 – easy visual discrimination task (longer of 2 lines) then temporal order. 3 – difficult visual discrimination task then temporal order. Also rate confidence of order judgement 16.

2arcsin(sqrt(p)) and bias=scale category point at which P(hit)+P(false alarm)=1.

Evidence supports does not support prior entry, rather decreased accuracy with attention to one or other streams.


Yarrow, Haggard, PBrown, Rothwell Nature 414:302-305 2001

Illusory perceptions of space and time preserve cross-saccadic perceptual continuity

Expt1: Saccades 22 or 55 degrees (=72 or 140ms). Counter at target, increasing in 1s intervals. Starts during saccade (no difference whether at start or end of saccade), but first number lasts till 400-1600ms after landing; 2AFC is the first number longer or shorter than the others?

Point of equality = when gaze had been on the target only 880ms (22 deg saccade) or 811ms (55deg saccade).

Controls: no eye movement; target rather than eye ‘saccades’ into view – 1s=1s.

Expt2: Posner cueing (50% valid endogenous) on a proportion of trials → shift of attention before saccade → chronostasis still occurs (but less) at time of saccade

Expt3: Counter stationary / moved location during the saccade but not perceived / moved and perceived. Chronostasis modulated by movement of counter, but only partially if subliminal.

Control: distractor near the counter did not reduce the effect

Conclude: duration of chronostasis = duration of saccade + 48ms, i.e. subjects extend the time they though the first target was seen backwards in time to 50ms before the saccade starts. Not attentional, but requires perceptual continuity of target.

Cf Duhamel et al, Parietal receptive fields updated 80ms before saccade.

Cf Lappe et al, Postsaccadic visual references generate presaccadic (50ms) compression of space. Cf saccadic suppression 50ms before.

Temporal extension – “fills in the perceptual gap” [but there’s no need to fill in gaps with no experience]. Refers to

  • Dennet & Kinsbourne Time and the observer BBS 1992
  • Kolers & von Grunau Shape and colour in apparent motion VisRes 1976
  • Eaglemann & Sejnowski, Motor integration and postdiction in vis awareness, Sci 2000
  • Geldard & Sherrick, Space time and touch, Sci am 1986
  • Nishida&Johnston, Influence of motion signals on perceived pos of spatial pattern n99
  • Haggard...Prinz, in Attention & Performance 19


Dennett & Kinsbourne

Subjective point of view – smeared out over the brain and over 500ms

  • Visual motion interpolation
  • Cutaneous rabbit
  • Libet’s backward referral of cortical sensation
  • Libet’s delayed moment of conscious urge/intention

These 3 premises together are not compatible with experiment:

  • Conscious perceptions are caused by events in the nervous system
  • Conscious acts cause events in the nervous system
  • Causes precede their effects


Nisbett & Wilson Psych Rev 84:3:231 1977

Telling more than we can know: verbal reports on mental processes

“such introspective access as may exist is not sufficient to produce accurate reports about the role of critical stimuli in response to questions asked ..a few seconds after the.. response [is] produced”

Opposite of Polanyi 1964: “we can know more than we can tell” e.g. skills

  1. Dissonance and attribution studies: all show introspective denial
    1. Unaware of a response: Zimbardo 1969: Repeating a shock-learning experiment with little justification showed ↑learning (& ↓GSR) than if they were given good justification for repeating.
    2. Valins & Ray 1967: Snakephobes shown snakes silently or shocks with pounding heart sounds; Ss told that the sounds were their own heart became less snakephobic (attribution of their fear response to the shock lessens fear of snakes)
    3. Unawareness of a change: questionnaire on opinion on school-busing → debate with an antagonist → requestionned → massive moderation of opinion, but without any awareness of change in opinion
  2. Learning without awareness
    1. Greenspoon 1955: free word generation. Reinforcing by nods for certain responses → increased ‘correct’ responses → questioning: unaware of reinforcement
  3. Unawareness of factor weightings used in complex judgements
  4. Subliminal perception – unaware of the stimulus
    1. Dichotic unattended tunes presented 5 times are not 2afc-old/new-recognised, but did show a familiarity effect on liking (previously unattended but unrecognised tone-tunes were preferred to new ones)
  5. Unawareness of stimuli affecting problem solving
    1. Ghiselin 1952: Poincare, Picasso, Hadamard, Whitehead, HJames → solutions appear to come abruptly without any immediate cognitive precedent.
    2. Maier 1931: subjects had to tie two hanging ropes together; solution is to set one swinging. Ss unable to do it until Maier casually swung one; 45s later they worked it out, but when questioned, 66% invented reasons like ‘it dawned on them’. Some subjects saw him twirl a weight; this was usually attributed as the useful clue, but it actually did not help.
  6. Denial of effect of other people’s presence on helpfulness behaviour (Latane & Darley 1970 with an epileptic)

Full criteria for inability to introspectively report cause of behaviour:

  • routine cognitive processes with wide range of domains;
  • Ss fully cognizant of the critical stimulus with minimal deception (unlike snake);
  • critical stimulus is verbal (so that it is always verbalisable, unlike cord-swinging);
  • minimal ego-involvement so as not to motivate denial for social reasons (unlike with the epileptic).

Failure to report an effective stimulus as effective

  1. Associative word pairs remembered. Designed to stimulate a specific response in a generate-task later → cues ↑’d generated word by 2x. → subjects asked why they chose words → no awareness → asked if word pair was responsible → 33% agreed. But for some words, 0% of Ss report influence, for others, many more reported an influence than were actually influenced. introspection = poor indication
  2. Position/order effect in shopping (strong effect 400%) → no awareness
  3. Anchoring effect: Ss have to estimate how people will on average behave. Some are shown an ‘anchor’: an exemplar of 1 person behaving. Large variation in strength of anchor effect; but on questioning, all subjects report ‘moderate’ influence of the anchor.
  4. Halo effect: manipulated the coldness/warmth of person’s character. Judge appearance, speech, mannerisms to be different (though they are the same). Ss attribute their liking of a person to the appearance/speech etc.

Reporting an ineffective stimulus as effective

  1. Reading prose with some passages deleted has no effect on the rated ‘effectiveness’ of the passage. But when shown the passages and asked how much they would have changed the effectiveness (by being added or removed, depending on the S), Ss in both conditions think it would change it a lot.
  2. Ss viewed a film in noisy conditions then rated its interest/effectiveness → rated how much the noise affected their ratings. Ss thought their ratings were lowered by the noise, but they weren’t.
  3. Testing pain threshold to increasing electric shocks. Some Ss told ‘no permanent damage’ → no effect on max shock. But when asked ‘were you affected/would you have been affected by “no permanent damage”,’ all thought it (would have) increased threshold.

Read job portfolio; 5 factors manipulated, including whether S is about to meet them. Ss rate the employee on 5 points. Then rate how they think each factor would have altered their ratings. Observer rates how much each factor would alter a rating (without seeing a portfolio). Ss and observers correlate highly; neither correlates with the actual effects.


Nisbett & Wilson argue for systematic erroneous reports. Observers and subjects draw their reports from same source: ‘a priori causal theories’. Culture/context plausibility, implicit or empirical or novel causal theories. Similar to K&T’s representativeness heuristic.


Sternberg Science 153:652 1966

High speed scanning in human memory

1-6 digits at fixation for 1.2s each → 2s gap → test digit (50% in sequence) → 2afc lever pull ‘was it in the sequence?’ → feedback light; payoff encourages rapid response and low errors → attempt to recall whole sequence in order.

Ss 98% accurate even for recall. RT is 397ms + 38ms per symbol. ‘Yes’ responses 50ms faster than ‘No’.

Expt1: Test-stimulus entropy covaries with number of items. (trial to trial change of test item)

Expt2: Control test-stimulus entropy: Told previously a set of items (1,2 or 4 items) requiring a ‘Yes’ response if the item is in the set; respond ‘No’ if not in set, or if not presented.

[no repetitions → information from other stored digits]

Slope is a measure of purely internal events (comparisons) → serial search at 25-30 symbols per second = 4 times the rate of subvocal speech for digits.

Yes & no RT is nearly the same → search is exhaustive

No change with practice

Parallel model could account for rise with N – e.g. if it terminates when the slowest process ends. But it is too small to account for the results. Hartley & David 1954 – upper bound.

Variation of the size of the negative set in Expt2 – no effect stimulus confusability cannot account for results

Variation in the size of a response-irrelevant memory load had no effect on latency retrieval process



Oberauer K JEP LMC 28:3:411-21 2002

Access to information in working memory: exploring the focus of attention

Testing the ‘concentric’ model where WM is a subset of long term memory, which is ready for use. Task: Updating of working memory locations: digits shown at 2/4/6 locations, press space, then 9 addition operations (press space after each) at ‘Active’ locations, then each location probed in random order, answer keyed in.

Conditions: blockwise alternating (Active row + Passive row, 2 Active rows); trialwise set size (3+3, 3+1, 1+3, 1+1); update-location transition type - restricted by set size and active (same cell, within-row change, switch row, within-and-between-row switch).

Results show:

  1. remembering a second passive set slows updating operations less than remembering a second active set of equal length.
  2. The setsize of an active row, but not a passive row, affects updating latencies
  3. Switching active cell yields higher RTs
  4. Switching cost increases with active setsize, but is independent of passive setsize.




Weigan, Michael and Schulz H J Sleep Res 16:4:346-53 2007

When sleep is perceived as wakefulness: an experimental study on state perception during physiological sleep.

Cites Knab and Engel 1988: spontaneous wakenings signalled by button press: many EEG arousals not perceived or not signalled, mainly in NREM, more accurate in REM. Same is true for deliberate awakenings Amrhein & Schulz 2000.

19-27% of tone-woken subjects thought they were awake, mainly in NREM sleep (50%) Mercer 2002. ? because there is missing data, and subjects use psychological data to determine whether they were awake.

Woke subject once during night during stable S2 or REM and asked:


Weintraub R Int J Philos Studies 9:1:3 2001

Logical Knowledge

Beliefs are logically imperfect: 1) not deductively closed, 2) not logically consistent

Assume: p^q is different from q^p – i.e. propositions are not just their truth values

Propositions may be sentences (Fodor 1976), mental representations (Appiah 1985) or structured intensions [sic] (Cresswell 1985)

Quine suggests beliefs are actually perfect. Uses translation as an example: rationality is a natural interpretative constraint.

But “There is nothing irrational, let alone silly, in our failure to believe all the logical consequences of what we believe... It is the agent’s intellectual potential for making inferences which constitutes the standard relative to which assessments of rationality are made... The label ‘irrational’ ... should not be attached to someone because [an] inference is beyond his logical capacity”

Quine suggests that translation must preserve truth by appeal to ‘obviousness’, but 1) does not apply to chains of truths (they are no longer obvious: sorites fallacy), and 2) in fact does not guarantee truth (e.g. error about logical truths on which there is no consensus), and 3) does not rule out implicit inconsistencies.

Prefers Grandy 1973 ‘principle of humanity’ – “interpretation should make speakers explainable”; actually assumes logical error.

Ascribing truth to a speaker’s potential beliefs is an empirical hypothesis weighed up on “predictive power, explanatory strength, simplicity etc.”, not logical perfection.

Mellor 1980: apparent logical errors are in fact problems with insight (second-order propositions) about our own beliefs (first-order). Inconsistence only occurs when “in introspecting my degrees of belief, I have made some error of measurement”.

Stalnaker 1984: logical errors are failures to correlate sentences with the ‘proposition’ (a set of possible worlds), i.e. failing to accept a sentence as true = failing to correlate words with the corresponding proposition (which the subject actually does believe).


Weintraub argues that, from these two points of view, a speaker’s contradiction results in our inability to assign beliefs to him. The beliefs must be perfect but the behaviour doesn’t immediately reflect that, so we need an account of this process before we can assign belief.


Postulates that, in terms of “non-verbal action, the structurally simpler a belief, the more causally potent it will be”. Cites Evans Newstead & Byrne (1993) Psychology of Judgement for evidence of a partial ordering of proposition complexity. Also partial ordering over inferential rules: modus ponens vs tollens, syllogism types.



Rubichi, Nicoletti, Iani, Umilta JEPHPP 23:5:1353-64 1997

The Simon effect occurs relative to the direction of an attention shift

Difficult discrimination Simon task with go-nogo cue 500ms before.


  • 25ms warning tone → 300ms gap → 100ms one of 4 letters (HNZW) below fixation → 500ms gap → 100ms discrimination target either on L or R
  • Respond L/R for square/rectangle counterbalanced across subjects, but to refrain if letter was W (catch trials).
  • Feedback on RT and accuracy, ITI 1s

Errors: Nogo responses = 0.9%, Simon errors = 12.6%.

Correct RT ~ 550ms. Reverse Simon effect = 22ms. Explained as re-orienting to fixation after target presentation, before response selected. (due to feedback, next catch letter etc)

Easier discrimination: errors – catch=3.9%, simon=4.9%, RT~400ms, Regular Simon effect = 18ms, present in early RT bins, and reversed in later RT bins.

Same effect when target duration increased to 400ms.

Expt 4: tone → 300ms gap → 100ms target → 100ms gap → 250ms catch letter presented either to the left or right of the discrimination target.

Errors: catch=33% (peripheral letters harder to discriminate), Simon=2.3%.

RT~387ms, Simon effect relative to the attention shift caused by the postcue letter = 16ms.

Explained as: the direction of the first shift of attention is immaterial, the Simon effect depends on the direction of the second shift of attention.


Giordano, McElree, Carrasco J Vision 9:3:30:

On the automaticity and flexibility of covert attention: a speed-accuracy trade-off analysis

Judge whether a 2deg size peripheral grating is tilted left or right of vertical. 8 possible locations, 4 deg eccentricity, location precue validity 12% (chance), 33%, 50%, 66% or 100%. On 50% of trials, 7 vertical distractor gratings also shown at the other locations. Target 40ms, precue was 150ms before (endogenous) or 53ms before (exogenous) to maximise attentional benefit for each condition. Target tilt and contrast was adjusted to equate conditons at 80% correct. Response tone at 40 to 2000ms after target, required response within 350ms after this tone (2% trials exceeded this).

Fit to , using a range of hierarchical models, ranging from null, to saturated (adjusted R2 + consistency across observers + evaluation of systematic residuals). Set-size affected discriminability but not processing speed. Discriminability function varies according to cue validity in the endogenous, but not the exogenous condition. With endogenous cue, the benefit of valid cueing is stronger when distractors are present. With exogenous, there is no effect of cue validity, and no interaction with distractor presence.


Mulckhuyse M, van Zuest W, Theeuwes EBR 186:2:225-235 2007

Capture of the eyes by relevant and irrelevant onsets

Speeded saccade to the one circle that turns grey, others remain red. Foreperiod 500-700ms. Error beep if SRT outside 50-300ms. Feedback on distraction every 16th trial.

Expt 1: No distractor, grey square, or red square (always in nontarget position).

Expt 2: No distractor, red open square around target, red open square around red disc (nontarget).





Eimer & Schlaghecken – button response to direction of arrow; masked prime precue arrow causes incompatibility benefit at long SOA, but compatibility effect at short SOA.


Ridderinkhof 2002: conflict detection → general suppression mechanism → delayed SRT

“Delta plots ... involves dividing the RT distributions of congruent and incongruent trials into a number of equal-size RT quantiles, subtracting mean RTs for congruent trials, quantile by quantile, from those of incongruent trials and then plotting these differences, again quantile by quantile, against the average RT of the congruent and incongruent speed quantiles.”



Dayan & Balleine Neuron 36:2:285-98 2002

Reward, Motivation, and Reinforcement Learning

Classical/pavlovian vs instrumental/operant conditioning:

= predicting reward vs optimising action. Temporal difference error does both (actor-critic model).

VTA=critic → bl amygdala, OFC; SNpc=actor → cortico-striato-thalamo-cortical loop

Explains how a light associated with food can itself act as a incentive for conditioning.

“the foundations of the relationship between Pavlovian and instrumental conditioning in the existing dopaminergic model are decimated by recent data suggesting that there are at least two independent, largely anatomically distinct, reward processes, only one of which appears to be sensitive to dopaminergic manipulations... We suggest that there are two routes to the Pavlovian incentive predictions of the critic: a plastic one, copied from the original model, and a hard-wired one, with which it competes and whose output is directly sensitive to classes of motivational manipulations.”

Actor-critic model solves the temporal credit-assignment problem


  • Pavlovian conditioned responses (CR) = reflex actions whose appropriateness is determined evolutionarily
    • Chics that have to run away from foodcup to get food (recedes at twice chic speed) – unable to learn (Hershberger 86)
    • Rat persistently approaches light that reduces food delivery (Holland 79)


Estle, Green, Myerson, Holt Memory & Cognition 34:914-28 2006

Differential effects of amount on temporal and probability discounting of gains and losses


  • as amount increases, delayed gains are discounted less steeply, whereas probabilistic gains are discounted more steeply (Green et al 1999)
  • amount affects exponent of probabilistic discounting function, but not delayed discounting function (Myerson et al 2003)

Expt1: compared temporal discounting of gains vs losses – e.g. ‘pay 100 now or 150 in 1 month’.

Expt2: compared a ‘for sure’ with a ‘possible’ gain or loss, expressed in probability e.g. 0.50; staircase of reward magnitudes to find equivalence, at a fixed probability.

Conclude: gain–loss asymmetry decreases with the amount of delayed outcome, but increases with the amount of probabilistic outcome.


Pine, Shiner, Seymour, Dolan J Neurosci 30:26:8888-96 2010

Dopamine, time, impulsivity in humans

Sequential presentation of 2 choices: e.g. 15 in 2 weeks, then 57 in 7 months. 3 sec each then 2afc. 3 sessions >1wk apart: either 30 mins after Madopar 150mg, placebo or 2 hrs after haloperidol 1.5mg (peak concs half way through testing). Subjective state VAS before and after. 10% catch trials (higher value soon vs low value later) to check concentration. Paid by lottery for one of the trials! Avg 75 per session.

Modelled data as choice function for each pair of utilities V=. Delay, value (V) and undiscounted utility (U) – orthogonalised as regressors.

  • U, V, D all independently correlated with caudate activity.
  • L-dopa increases discounting (increases preference for sooner rewards), as well as the fMRI correlation of D and V.
  • RT inversely proportional to difference in utility of the 2 choices; not influenced by drugs impulsivity is not a unitary construct
  • Subjects less alert with haloperidol
  • Striatum, insula, subgenual cingulate, lateral OFC show decreased activity with longer delays.
  • Amydgala correlates subject-wise with amount of L-dopa-induced impulsivity


Della Libera & Chelazzi Psychol Sci 17:3:222-7 2006

Visual selective attention and the effects of monetary rewards

Expt 1:

Prime: Global / local digits, 50% congruent or incongruent.

Probe: one of the levels was neutral, subjects had to respond to whichever was a number.

Both prime and probe required subjects to press digit 5/6/7/8.

400ms instruction cue ‘G/L’ → prime display → response (3s) → reward (1p/10p) 600ms → 400ms gap → probe display → response (3s) → 600ms ITI

4 hrs = 1500 trials. Random arbitrary reward at end of block – fully decoupled from performance. Just incongruent trials analysed.

Expt 2: colour cue → ‘prime’ = same/different judgement for 2 overlaid shapes vs one black shape → probe = similar display with previously ignored shape in the attended colour.

Negative priming strongly and consistently modulated by reward; positive priming not so much.


Correa, Cappuci, Nobre, Lupianez Experimental Psychol 57:2:142-8 2010

The two sides of temporal orienting: facilitating perceptual selection, disrupting response selection

Expt1: Ericsson flanker – 5 arrows < or >, 1.4 degrees size, report central arrow pressing ‘z’ or ‘x’, congruent or incongruent. 400-900ms fixation → 500ms cue ‘PRONTO’ or ‘TARDE’ → interval either 400ms or 1300ms → target display until response → 300ms audiovisual feedback on incorrect responses (or blank screen if correct) → 400ms ITI

Alternating blockwise ‘early’ or ‘late’ cues. 75% validity; 50% congruent; = 32 trials, + 6 catch trials (no central arrow, withhold response).

Result: Main effect of validity and conflict. But, Congruence → larger cueing effect, and Invalid temporal cue → less conflict. Conclude: Temporal cueing gives non-specific activation

Expt2: target 100ms: 4 locations U/D/L/R, arrows appear after delay, 3 have both up & down heads, one has just up or down. ‘z’ for up, ‘m’ for down. 32 trials blocked by time cue, 75% valid. 50% trials were stimulus-response conflict ‘Simon’ i.e. target is on the L or R; and 50% were stimulus-stimulus conflict ‘Spatial stroop’, i.e. target is above or below.

Results: Error rate 716%, Simon effect 9% errors, Spatial stroop effect 3% errors.

Significant 4 way interaction between congruency, conflict type, delay, and temporal-cue-validity. “Temporal cueing modulates conflict effect in opposite directions for Simon conflict and spatial Stroop conflict”, and “increment of Simon conflict by [valid] cuing was driven by interference in the incongruent condition rather than by facilitation in the congruent condition”. RTs mirrored the pattern of error data.

Conclude- different neural pathways for conflicting perceptual features vs conflicting responses. ? what vs where. Cue may regulate ‘phase resetting and frequency – refresh rate – increments, of neuronal oscillations in visual system’



Ridderinkhof..Nieuwenhuis Science 306:5695:443-7 2004

The role of the medial frontal cortex in cognitive control

Hypothesis: “one unified function of posterior medial frontal cortex (pMFC) is performance monitoring in relation to anticipated rewards”, signalling failure or reduced reward probability, and cause increased control by influencing LPFC.


  • SEF and rostral cingulate motor area cells – sensitive to reward and unexpected reductions in reward (Stuphorn Taylor Schall N2000). Cingulate sulcus cells respond to errors and omission of reward (Shima & Tanji Sci98) “comparing expected and actual outcomes”. Also error without feedback (Ito Stuphorn Brown Sci03)
  • fMRI dACC differential processing of unfavourable outcomes (money / abstract / pain in other pMFC areas) ‘coding of motivational value of external events’. Also rostral CMZ (=~rostral CMA) for errors without feedback.
  • ERN 250-300ms post outcome, loss>gain, quantitative and normalized to subjective expected value (mean) and range of outcomes (variance) (Holroyd Larsen Cohen psychophysiol04).
  • Response ERN as well as feedback ERN: same location, but starts at same time as error-response EMG. ? Due to phasic dopaminergic input to motor neurones RCZ [but dopamine is less for losses!]

Or Conflict (Botvinick)?

  • Early conflict in correct stroop trials: ERN-like N2 just before response
  • Late conflict in incorrect trials
  • Sustained conflict in decision under uncertainty (eg equally good choices)

Unified view (penalty response, error detection, conflict detection, and uncertainty)

  • More dorsal = pre-response conflict, ventral = error and feedback
  • Neighbouring neurons involved in different aspects
  • Conflict and uncertainty => reduced reward probability; error and penalty = loss of anticipated reward


  • mPFC activity on trial n-1 predicts LPFC activity on trial n
  • rostral cingulate motor area cells fire with reward prediction error, but only when this is followed by changes in monkey’s action selection
  • ERN amplitude to negative feedback decreases during S-R reward learning, whereas ERN amplitude to choice errors increases (Holroyd 2002 Psy Rev)


Funes, Lupianez, Humphries Cognition 114:338-47 2009

Sustained vs transient cognitive control: evidence of a behavioral dissociation

L/R hand responses to U/D direction of an arrow; arrow was above/below/left/right of fixation. 5 variables manipulated:

  • conflict type (spatial stroop – targets above or below, or simon – targets left or right of fixation),
  • congruency (of target location with arrow direction in spatial stroop, or of target location with response in the simon)
  • conflict context (63% congruent – low conflict, or 37% congruent – high conflict, in each block).
  • Previous trial congruence: congruent/incongruent
  • Previous trial conflict type (simon vs spatial stroop): same/different
  • For the spatial stroop trials, there was always 50% congruence, whereas the simon task was either 75% or 25%.

Conflict adaptation effects did not generalize, whereas conflict context effects did.


Bitsios, Szabadi IJPsychopath 52:1:87:95 2004

The fear-inhibited light-reflex: importance of the anticipation of an aversive event

Bitsios 1996: Anticipation of electric shock → pupil dilatation, reduced light reflex

Background noise. 500ms cue tone, 3s gap, then 200ms green LED at 1cm distance (=full field) 0.45 mWcm-2.

  • Before expt, subjects heard a tone & demonstration shock 1.5mA for 50ms on wrist, and were instructed that in the experiment, between the cue & light, there could be a tone 50x quiter or a shock 50x greater.
  • Alternating blocks of 3 ‘relaxation’ trials or 3 ‘anticipation’ trials. ISI 25s. Relaxation: no shocks or tones expected. Anticipation: Ss either told to expect big shock or quiet tone which they had to report afterwards. Anticipated stimulus was actually delivered on the last trial of the block.
  • VAS ratings after blocks → anxiety, alertness, discontentment (via PCA)
  • Fixation on dim red spot, Dark-adapted, No meds, Avoided etoh, coffee for 12 hrs, Tested in morning only.

Open bars: relaxation. Shaded: anticipation.

  • baseline pupil diam sensitive to anticipation of anything
  • light reflex attenuated only by shock anticipation. Effect is ‘anxiogenic’

Prev studies: diazepam → selectively prevents fear-attenuation of light-reflex but not tonic fear-dilation. Clonidine: abolishes both.

Hypothesis: 2 pathways

  • conditioned fear → amygdala →locus coeruleus → ↑symp but also LC → inhibits E-W nuc → ↓parasymp
  • anticipation of neutral event → hypothalamus → direct sympathetics


Krajbich, Armel, Rangel NN 13:1292-8 2010

Visual fixations and the computation and comparison of value in simple choice

39 fasted students chose food to eat. Tray of 70 items briefly seen before expt; then images rated using VAS -10 to +10 ‘how much do you want to eat it?’. Then choice phase: free time L/R keypress, between items rated >=0. Pairs with difference of <=5, each item used up to 6 times.

[poor sampling and filling-in of eye data]

Plotted many meaningless graphs of the trends and fitted it to a random walk model where Vt = Vt−1 + d (rleftθright) + εt with ratings rL, rR, and fit parameters: d (speed of integration), q (bias towards the fixated option), threshold, and variance of e. Alternate trials used to fit or test.


Fiorillo, Tobler, Schultz Science 299:5614:1898 2003

Discrete coding of reward probability and uncertainty by dopamine neurons

Sustained activation precedes uncertain rewards

Discusses Pearce & Hall 1980: attention and learning is proportional to uncertainty about reinforcers

Only under uncertainty do outcomes provide information.


Pearce & Hall Psych Sci 1980

A model for Pavlovian learning: Variations in the effectiveness of conditioned but not of unconditioned stimuli.

Failures of rescorla wagner:

  • US-CS pairings don’t result in learning
    • Blocking (Kamin1969): previous US:CS2 pairing blocks learning US:CS
    • Rescorla Wagner incorporates this with DwCS:US=CS x (US-SwCSi:US); i.e. US is only effective when it is not predicted, i.e. it is surprising.
    • However, Dickinson Hall & Mackintosh 1976: reducing the intensity of the US reinforcer, also causes unblocking. RW predicts only increasing the US should work! So the surprise of lower reinforcement causes learning.
    • Mackintosh 1975: conditioning is normal (not blocked) on the very first trial when the new CS is presented. There is a small amount of learning on this first trial, but none (i.e. blocking occurs) on further presentations.
    • Wagner 1978: the CS can also become ineffective: DwCS:US=d(L -vCS:context)(US-SwCSi:US), i.e. the CS association with the environment has an asymptote of L. However does not fully explain Mac75 or Dick76.
    • A surprising shock on the first compound trial potentiates conditioning on the next trial, but not the current trial. This is explained by Mackintosh 1975, using CS attenuation only.
    • Pearce: Latent inhibition: tone-shock group vs light-shock. Then tone-bigshock conditions slower in the tone-shock group. This disproves Mac75 model.
  • Inhibitory learning if CS and US are not paired
    • RW allows negative w. What does it mean?
    • But Zimmer-Hart & Rescorla 1974: A stimulus established as an inhibitor maintains its inhibitory properties in spite of being repeatedly presented in isolation.
    • Remedied by Kornorski 1948: inhibitory CS does not have negative associative strength, rather raises threshold for an excitatory CS. Correctly predicts that nonreinforced inhibitors fail to extinguish, and that inhibitors are ineffective in the absence of excitatory stimuli.
    • But Wasserman 1974: a lighted key that is negatively correlated with food → pigeons actively withdraw from it when it lights up → inhibitory learning can itself elicit behaviour
    • Kornorski 1967: posits learning of CS:notUS pairing. (can only arise if US is expected, and not delivered). Cf relaxation on omission of shock conducive = appetitive; frustration on omission of food = aversive.
  • Overexpectation
    • Kremer 1978: when 2 stimuli that have been separately paired with US, are presented in compound with the same US, there is reduced associative strength of each element.
    • Kremer 1978: a novel stimulus on the compound trials gives inhibition. Therefore a US less intense than predicted allows inhibitory learning.
  • Superconditioning: Rescorla 1971: activation of notUS memory by an inhibitory CS, suppresses US memory, increasing the associability of CS on subsequent trials.


Steiner Glaser Hawilo Berridge Neurosci Beh Rev 25:1:53-74 2001

Comparative expression of hedonic impact: affective reactions to taste by human infants and other primates

Facial expressions to sweet vs bitter taste is phylogenetically conserved – similar between great apes and humans


  • corner elevation of mouth, Duchenne smile (full smile: 1500ms zygomaticus + orbicularis oculi), upward tongue protrusion, lip smacking, finger licking.
  • corner depression (platysmus), nose wrinkle, nasolabial furrow deepening, upper lip raisin, brow furrow (corrugator), downward tongue protrusion, lip puckering, eye squinch, grimace (exposing gums), spitting.


Kelley & Berridge J Neurosci 22:9:3306-11 2002

The neuroscience of natural rewards: relevance to addictive drugs

  • Drug reward and withdrawal modulate behaviour by incentive principles, rather than aversive drives.
  • Potential DA functions: mediate hedonic pleasure? incentive salience (goals and wanting)? Learn and predict occurrence of reward? Attention? Complex sensorimotor integration? Effort? Switching between behavioural programs?
  • Reward system = “Dynamic network” between subregions of PFC, accumbens, amygdala, lat hypothalamus, vent pallidum.
  • Striatum and PFC have D1-plus-NMDA dependent potentiation → reward learning


Botvinick & Rosen Psych Res 73:835-42 2009

Anticipation of cognitive demand during decision-making

Bechara 1999: phasic GSR change just before a high-risk option is selected

Hull’s law 1943: given two actions with equal rewards, the least effortful is typically chosen. “drive for cognitive economy” in Baroody & Ginsburg 1986 for arithmetic strategy selection.

Botvinick 2007: 2 deck choice. Red or blue digits; cards to be classified according to parity (red) or magnitude (blue). One deck alternates colours, the other remains constant. Subjects biased towards constant deck (low cognitive demand).

Expt: keypress L/R deck choice, verbal yes/no response for large (red) or even (green) numbers. 2 decks: 90% task switches, or 10% task switches.

Conclude: could be anticipation of cognitive demand, or of error likelihood.

Damasio: SCR “marks a particular option-outcome pair with a negative bias”

Consistent with SCR in aversive conditioning.

Consistent with SCR reflecting a behavioural inhibition system.

ACC lesions eliminate GSR; perhaps ACC reflects ‘task engagement or energisation’, or effort cost.

Botvinick M, Rosen Z Psychological Research 73:835-42 2009

Anticipation of cognitive demand during decision making

Cites Baroody & Ginsburg 1986 “drive for cognitive economy” in arithmetic strategy, McGuire 1969 humans are “lazy organisms”, Camerer & Hogarth J Risk+Uncert 1999 economists “instinctively assume thinking is a costly activity”, Balle Sciences humaines 2002 “Law of least mental effort: mental representations”.

Botvinick 2007: 2 decks of numerals, colour of numeral indicates task (parity or magnitude judgement). Deck 1 tends to alternate between tasks, deck 2 tends to keep task constant. Preference for deck 2 (less switching less effort?)

This experiment: deck 1 90% task switching, deck2 90% task repeat. 34 blocks of 10 trial. Instructed to ‘select randomly’ between 2 decks at the start of each block, then 10s delay, then all 10 cards from same deck. GSR measured during deck selection (before the block), and during performance.

Results: behaviour: 49% choice of difficult deck. Significant RT effects of deck, task-switching, and significant interaction.


Bari Theobald...Robbins Neuropsychopharm 35:1290-1301 2010

Serotonin modulates sensitivity to reward and negative feedback in a probabilistic reversal learning task in rats

Depression as abnormal reaction to positive/negative feedback.

Citalopram is known to cause probabilistic reversal learning deficits, and increased negative feedback sensitivity in healthy volunteers (Chamberlain 2006).

Rats 2AFC nose-poke for food, ‘correct’ hole lasts until 8 consecutive responses made there. 80% valid location.

Given citalopram: low dose, high dose, 5 days, 7 days, and ‘serotinin depletion’ by intracerebroventricular 5,7-dihydroxytriptamine infusion.

Measured reversals completed, win-stay, and lose-shift.


Also RT to holes shorter after depletion; RT to collect food unchanged though (so unlikely to be reward motivation)

Conclude: reversal performance worse with depletion, or single low-dose; better with high dose or regular citalopram.

Mechanism unclear - ? different receptor types. Short term low dose 5HT increases impact of negative feedback. Long term 5HT enhances reward sensitivity. ? differential downregulation of receptor subtypes. Transient increase in lose-shift after depletion - ? functional adaptation

Deakin 1991: serotonin controls aversive motivational system. ?opponent DA


Krawczyk, Gazzaley, D’Esposito Brain Res 1141:168-77 2007

Blockwise: remember scenes & ignore faces, or vice versa, or passive viewing.

Trials: 2s possible reward shown 0/3/10 → 4s fixate → 4 sequential images for 1s each → 8-12s delay → 2s probe (face, scene or arrow) → L/R hand response: seen probe in the encoding phase? → 4s fixate → 2s feedback.

ROIs from functional localiser: view scenes/faces and press button if they repeat. Left parahippocampal (Scene-Face) Bilat fusiform (Face-Scene), Lateral PFC including bilat middle and inferior frontal gyrus.




Brown, Preece, Hulme Psych Rev 107:1:127-81 2000

Oscillator-based memory for serial order

Model ‘OSCAR’ oscillator-based associative recall:

  • several clock-like oscillators at various frequencies are linearly combined to make a learning context vector
  • The context is associated with another vector representing the item to be remembered
  • Local temporal distinctiveness (dynamic context model)
  • Bidirectional recall (context from item as well as item from context)
  • Posits arbitrarily slow-moving context oscillators
  • Represents many levels of ‘time hierarchy’ – accounting for neighbouring phoneme errors, syllable errors, within-list errors and between-list errors
  • Huge number of parameters – N, frequencies, vectors, learning and decay rates, ability to reinstantiate context at recall, vocabulary size, output interference etc

Effects modelled:

  • Nearby items switched in order more frequently than distant ones: A rapidly presented 4-item lists order-only recall, B 6 dissimilar items C 6 similar items, D memory of within-list position, E memory of list-within-trial position, F 12 item lists after learning
  • Long retention interval enhances primacy, immediate recall enhances recency

  • Serial position of items when recalled, after various delays.
  • RT and accuracy for earlier-later judgements of pairs of items, as a function of separation (Which was presented first, X or Y?)
  • Recency judgements (how many items ago was X?)

Partial reinstatement of context at recall leads to activation of recent items, only for immediate recall.

Grouping improves recall and gives small recency effects, but increases temporal ‘distance’ of order errors

Hierarchical time – e.g. remembering time of day a storm occurred, but not the day; (Nairne 1990 – Which serial position was X in its list? Which list number was X in?)

Drenowski 1980: no effects of vocabulary size.

Doesn’t account for item-length effect (evidence for rehearsal) Baddeley 1986


Wylie, Ridderinkhof... Brain 2010

Subthalamic nucleus stimulation influences expression and suppression of impulsive behaviour in Parkinson’s disease

Healthy, DBS on, DBS off.

Simon task (single coloured disc on left or right, arbitrary colour-hand mappings) errors and slowing occur for incompatible trials, similar effect in all subjects.

RT analysis: 7 bins. DBS makes you faster (60ms) and less accurate.

Note that DBS gives an upsloping delta plot: incompatibility effect is greatest for responses at 500ms.

They argue that DBS increases ‘impulsive’ behaviour. Worsens early susceptibility to response capture, but improves later controlled responding.


Marshall, Chen & Jeter Am J Psychol 102:1:39-52 1989

Retrieval influences on tests for the automaticity of the encoding of temporal order information

Hasher & Zacks 1984: automatic encoding of frequency of occurrence, temporal order spatial location, and word meaning.

Marshal Chao Horner & Lockwood 1985: subjects performed a word rhyming task, and afterwards had to make frequency estimation judgements. If subjects were informed about this beforehand they did better [→ frequency encoding is automatic?]. However if the estimation was non-explicit (forced choice more- or less-frequent), then there is no difference! → it is a “retrieval effect

Could the learning improvements in temporal order be explained as a retrieval effect?

Expt 1: serially learn visual 30-word list. Then see a shuffled 34-word list (4 new), and task is to write a serial order number for each word. As you go, also write the order in which these judgements were made.

Expt 2: showed that if order of making the judgements was fixed, the learning effect still obtains.

Conclude: “we never observe the encoded information directly.” Encoding of serial order may still be automatic.


Smyth MM, Pelky PL BJ Psychol 83:359-74 1992

Short-term retention of spatial information

WM for 3 items with secondary distraction task.

  • Experimenter touches 3 spatial locations from 9 corsi blocks at ~1s intervals
  • Asked to recall after 0/5/15s (varied trial-to-trial, subjects were told this before each trial)
  • Secondary task: 4 metal tapping plates in a square. Blockwise:

◦       simple (tapping repeatedly on one plate),

◦       spatial tapping (clockwise round square 'as quickly and accurately as possible')

◦       backwards counting (start from given number 100-999, 'as quick and accurate as poss).

No repetitions within list, no repetitions of triplet sets. [no record of accuracy on secondary task!]

Expt 1: secondary task starts after encoding, until recall

Expt 2: secondary task starts before encoding, until recall

Expt 3: interleaved encoding and maintainence.

Expt 4: eyes closed after encoding, and compared with 3-digit verbal recall.


  • backward counting affects encoding as well as maintenance. Spatial tapping does not – it only affects maintenance (when N=3).
  • “Adding spatial tapping to the maintenance of 3 digits has little effect on recall but adding backwards counting has a large effect” → spatial tapping uses different resources to maintaining 3 digits, whereas backwards counting does not.
  • “Adding either backwards couting or spatial tapping to maintenance of 3 spatial targets produces a small but significant error, which tends to increase as the recall interval increases. These costs may best be understood as involving executive processes”
  • Cf morrison's paragidm 1987: counting interferes with spatial memory at encoding but not during maintainence – how to explain difference? Maybe “storage of a spatial pattern and of a path through a pattern may be different.”
  • 'general (nonspatial) place-keeping' as an executive function – needed for spatial paths and numeric squences but not patterns
  • Navon & Gopher: “The process of organising resources may require resources itself”; if spatial path tasksrequire 'place-keeping' functions, “a triple-dissociation between articulatory, visuospatial and executive components” may not be possible.

Flexser & Tulving Psych Rev 85:3:153-71 1978

Retrieval independence in recognition and recall

Tulving & Thompson 1973: A-B pairs presented (subjects expected to recall test of B given A as a cue)

→ unexpected free association task: extralist cues, chosen to be semantic associates of B words, such that subjects frequently generated B words.

→ Shown the words they generated in free association, and had to circle those recognised as B-words.

→ then cued recall test (given A, respond B).

Result: failed to recognise many B words, though they generated them given A.

Explained as 'encoding specificity principle'.

Postman 1975: A-B presentation, then (B+others) recognition, then A-cued recall.

Same result, 'recognition failure' for subsequently recalled words.

Meta-analysis of 33 experiments, 89 conditions, P(not recognised | recalled) depends on P(recognised), but not on subject or word frequency etc.

Explained as independence of contextual cues in recognition (i.e. copy-cue, = B-word) vs recall (A-word). Model – multiple contextual features encoded in episodic trace.

p, r, s sampled rectangular distribution in (0.2-0.8). N has no effect. When there is independence of cues, then the given curve is obtained. Correlations in retrieval cues gives a strongly convex function.


Tatler & Land Proc Roy Soc 2010

Vision and the representation of the surroundings in spatial memory

1898 Erdmann & Dodge: I can't see my eyes moving in a mirror

Simons & Levin: interrupted conversation – people don't notice change in person!

Tatler: Blackout during tea-making: subjects report final or penultimate target. Good detail for that fixation but no detail for targets of previous fixations.

Transsaccadic information survives only as integrated with gist, layout, proto-object representations, sensorimotor contingencies (Rensink)

  • Change detection: even when unable to consciously report transsacadic changes, subjects have better than chance localisation of a change (Fernandez-Duque & Thornton 2000).

Dynamic scenes – limited information retained between saccades

  • Ballard Hayhoe et al 1992: block copying task → coupling between vision and action. Looking at example =1 fixation before block selection, 1 fixation before block placement. Subjects gradually learn to do it with just 1 fixation, before looking at the construction area. → retention of information only when task-relevant. Aivar 2005: saccades are launched to remembered, not actual, locations of source bricks, even when they are visible.
  • Droll & Hayhoe 2007: VR block sort, cue which block to pick-up, and where to put down, by different features. Block is refixated only when put-down-cue is initially unknown → retention of both features if subjects know in advance. But: if the unpredictable put-down-cue is same as the pick-up-cue, refixation is common → “if it not predictable that the information will be needed, that property is not retained”.

Static scenes – most information retained

Castelhano & Henderson 2005: visual memorisation task or search task. Memory tested afterwards, was good in both conditions → incidental encoding of task-irrelevant features

Montello 1993: 4 scales of space: figural, vista, environmental and geographical.

Wang & Simons 1999: Change detection between 2 viewpoints is much better if the movement is endogenous.

Land 1999: Large amplitude gaze shifts (180deg), with long blink, continuously moving until about 10deg from target, + corrective saccade 300ms later.

Burgess 2008: allocentric hippocampus/MTL, egocentric parietal ?precuneus, translations occur in retrosplenial cortex. Does the egocentric model change every time we move? Yes...


Abler, Hahlbrock et al Brain 132:9:2396 2009

At-risk for pathological gambling: imaging neural reward processing under chronic dopamine agonists

12 RLS patients on cabergoline / pramipexole / ropinirole, tested on regular drug, then off drug for 5d before testing

Monetary incentive task: 1 won with probability 0%, 25%, 50%, 75% or 100%. Coloured symbol as probability cue → delay → button press to 2 symbols, correct/incorrect → feedback.



Morrone, Ross, Burr NN 8:950-54 2005

Saccadic eye movements cause compression of time as well as space

fixation → black 1deg target L/R 15deg → 8ms green equiluminant bar top → onset asynchrony 100ms → 8ms bar bottom → saccade

Exp1: → 2s gap → comparison bars also 8ms, but separated by 8 – 100ms → verbal report which interval appeared longer, logged after experimenter verified the saccade was correctly executed.

Exp2: Varied interval between bars; 8 – 200ms. Ss reported apparent order & estimated the separation in time of 2 bars. After the response, a stimulus with the reported separation/order was presented, for subjects to confirm or revise their estimate. “Extensive training [on estimation during fixation] before the experiment”

Controls: 1) Short vertical bars near the saccade target – same effect, 2) blinks instead of saccades – no effect, 3) auditory clicks 4ms instead of bars – no effect

eccentricity colour and separation of bars “chosen to minimise sensation of motion”

Precision of distorted times follows Weber's law, Weber fraction ~0.25


psychometric function for 100ms after saccade (L) vs -70 to -30ms before (R).

(open symbols for inverted reported durations 'just for clarity')


Bernacchia, Seo, Lee, Wang NN 2011

A reservoir of time constants for memory traces in cortical neurons

'Competitive game task': fixate 500ms → 2 green discs (L&R) 500ms → fixation disappears, monkey makes saccade to one target → 500ms → ring around choice 500ms → juice.

Isolate trial-to-trial memory time constants vs epoch-within-trial periodicity. Fitted 12 epochs in a trial plus 5 trials back time constant

Time constants estimated from pairs of simultaneously recorded neurons are uncorrelated → different time constants for different cells.

Distribution of power constants is ~ t-2. Same for dACC, DLPFC, LIP.

Neural time constants don't correlate well with behaviour. Behaviour and neural time constants both vary between sessions.

Neural network model: need a wide variation in connection weights to give the breadth of time constants.

Botvinik & Plaut Psych Rev 113:2:201-33 2006

Short-term memory for serial order: a recurrent neural network model

Recurrent activation networks: do they just do 'chaining' – i.e. rely on prior state to recall next item?

  • Baddeley 1968: 6-item consonant lists with alternating confusable and nonconfusable items: accuracy of nonconfusable items = accuracy in a list of only nonconfusable items → cannot be explained by chaining.

  • Henson 1996: relative errors (adjacent items in the input list appear together but out of place at recall) are not as frequent as chaining account predicts.
  • Context-based accounts: context = list position (anderson & matessa 1997), distance from start or end (Henson 1998), oscillator states (Burgess & Hitch 1999). Unlike activation models, context is associated with items by synaptic weight.
  • Context account does not account for long-term background knowledge e.g. bigram frequency effect (Baddeley 1964). “Recall for highly probably sequences is better than for less probable ones.” - this “knowledge involves transition probabilities” & “context models eschew...item-to-item associations”.
  • Estes 1972: item representations undergo cycles of suppression and reactivation during rehearsal; subject to perturbations giving transpositions. Baddeley 1986 shows evidence against centrality of rehearsal and transposition; doesn't explain similarity effects. (Page and Norris is descendent of this)
  • Lewandowsky and Murdock 1989: TODAM. Distributed representation with chaining. Parameter to model primacy.
  • Farrell and Lewandowsky 2002: SOB. attractor network: on each stimulus, weight changes are made that give a tendency to settle into the sequential patterns. By autoassociation. Modulated by Hopfield energy decreasing through sequence. After each item is recalled, its connection strengths are eliminated. Orthogonal representations → can't account for similarity effects. Unclear how it could accommodate rehearsal.
  • Anderson & Matessa 1997: ACT-R formulation, item & position linked to a common node. Therefore it is essentially a context-based model. So doesn't explain sensitivity to domain structure.
  • Nairne 1990, Neath 2000: feature-based models. List element = feature vector encoding item identity & list position. List = vector of vectors.
  • Brown Neath & Chater 2005: SIMPLE model. Elements = points in a multidimensional similarity space, including features, position, and time.

Standard 200-cell hidden layer, time step per input or output, one unit of input is 'start-recall' and one unit of output is 'end-recall'. Train on blocks of 10,000 trials with backpropagation sum(tlog(t/a)) with rate 0.001, and the feedback output-to-hidden-layer was the 'teacher' (desired output). Bias units to prevent excess hidden layer contribution early on.

They find a superpositional code, for vectors of hidden units, where

  • items are represented independently of other items in the same list,
  • item position and identity are represented conjunctively
  • there are similarity relations of the representations of similar-item and similar-position.

The output visible on each step is only a fraction of what is stored ('output gating'). A given output unit is only driven when the hidden vector is aligned with the weight vector for that output (i.e. dot product). Elements are encoded so that on non-ouput timesteps, the item is invisible to the output layer.

Recency: Jahnke 1963. Recall is “better for item information than for order information”: Bjork & Healy 1974. Misbinding to nearby serial positions (Henson 1966 = “locality constraint”, or “transposition gradient” - 3rd figure above)

Repeats: “When there are few or no repeats in the presented lists, repetition errors are infrequent” - Henson 1996 “repetition constraint”.


Meck Brain & Cog 58:1-8 2005

Neuropsychology of timing and time perception

seconds-to-minutes: heterogeneous cognitive processes and substrate.

Regions specialised for stimulus duration: Ivry 1996/7, Hinton 2003, Friston&Frak04

Neurological and psychiatric disorders: Lustig 2003, Malapani 2003

Depression: abnormal experience, variable in bipolar, altered by lithium

Schz: poorer discrimination. Corticostriatal & corticocerebellar dysfunction

Haloperidol (Rammsayer 1999)

“millisecond time estimations are important for motor control” etc [but is this time representation at all?]

Pashler 2001: oscillator or hourglass?

Bottom up (ms) vs top-down (sec to min); top down concatenates intervals generated by bottom up.

HM underestimates durations >20s. Perbal, Pouthas, van der Linden 2000.

Rt MTL resection → impaired precision & underestimation of retrospective but not prospective duration. Lt MTL → ?improved precision + correction of underestimation in retrospective time judgements + overestimation and underproduction of prospective duration. (Drane Lee Loring Meador, 1999, Vidalaki Ho Bradshaw Szabadi 1999)

Explanation: Hippocampal lesion → increased DA transmission in striatum

Interval reproduction (interval demonstrated only in training trials at start) with post-trial feedback: ADHD poorer than normals if feedback on only 25% of trials. Nicotine improves performance. Explained as: slowing of internal clock → ADHD impairs gating of “pulses from a pacemaker into an accumulator” [explains nothing?]


Lustig & Meck Brain & Cog 58:9-16 2005

Chronic treatment with haloperidol induces deficits in working memory and feedback effects on interval timing

5 neurotic patients on chronic Halo vs 5 healthy.

Generated intervals lengthen as the feedback becomes more remote in time.

Timing and cognition: precise timing requires attention (Pang & Mcauley 2003), timing fundamental for learning and conditioning, interval timing important in serial recall and timing a signal with gaps: Buhusi 2003, Lustig 2004 Memory, Meck Church & Olton 'hippo time + mem' Beh Neurosci 1984.

Dopamine dependence of timing: Miller & Cohen 2001

Scalar Expectancy Theory: Gibbon 1977: Pacemaker-switch-integrator.

Clock is frontal/BG, D2 receptors. D2 antagonists = slow clock, overproductions, underestimations. Can't dissociate pulse emission, gating, and accumulation. Feedback allows subjects to compensate for clock-speed-changes e.g. under drugs.

Memory is cortex + hippo, ACh. Cholinergic agonists = shortening of time representation in reference memory → underproductions of target time, overestimations of current time. Gradual distortion develops that is resistant to feedback.

PD: overproduction of intervals without feedback.



Wittman, Dolan, Duzel Learn & Mem 18:296 2011

Behavioral specifications of reward-associated long-term memory enhancement in humans

Coloured word cue → category judgement → digit → hi/lo judgement → reward for correctness

Reward contingency: Expt1: cue-colour-dependent or cue-category-dependent, hi/neutral/lo reward trials. Result: effect of reward for semantic condition only

Expt2: uncertainty in reward prediction by precue – no influence in performance

Conclude: semantic identity-specific enhancement of recollection by reward association. Synaptic capture / behavioural tagging (Frey & Morris 1997) “Events that cause dopamine release should modulate memory for other items presented in the same context”


Onishi & Xavier Behavioural Processes 86:2:263-71 2011

Negative anticipatory contrast: does it involve anticipation of an impending reward? [Behaviour, Learning]

NAC: 2 rewards, given one after the other every day. First one becomes less valued under simultaneous contrast.

Explanation: ?anticpation of second reward when given first.

Williams 2002: NAC = 2 opposing processes: pavlovian association gives increased response to first reward, + comparison between first and impending rewards.

BUT: Devaluation of second reward does not reduce this effect → NAC is not due to “anterograde comparison of the first solution with the representation of the impending second solution”


Gavornik & Shouval J Comput Neurosci 30:501-13 2011

A network of spiking neurons that can represent interval timing: mean field analysis

'Reward-dependent expression': hebbian plasticity is modulated by a reward signal paired with feed-forward stimulation during training (LTP+ongoing network inhibition)


Morey, Cowan, Morey, Rouder AP+P 2010

Flexible attention allocation to visual and auditory working memory tasks: manipulating reward induces a trade-off


Tones and coloured squares, temporally overlapping WM tasks, 'whole report probes' i.e. same/different. Reward for accuracy on both tasks = 1000 points total, divided between tasks blockwise 1000,750,500,250,0. Articulatory suppression ('the the the' 2/sec), 2x2AFC.

Pashler k=S(h-f)/(1-f); Bayesian hierarchical working memory model (Morey 2010).

  1. A trade off between visual-spatial and auditory-temporal memories occurs with manipulation of rewards as well as task difficulty.
  2. Effect of payoff is not only motivation between conditions


Blumebthal, Steiner Seeland Redish Neurobiol L+M 95:3:376 2011

Effects of pharmacological manipulations of NMDA-receptors on deliberation in the multiple-T task

Rats navigate maze with T's. They initially make lots of hesitations, which decay with laps. Changing contingency increases these hesitations.

Task: must initially recognise which contingency was rewarded, but also that a contingency change had taken place.

4 trials of “vicarious trial-and-error” and 4 normal trials, at a T-junction.

After NMDA-receptor antagonist, there are fewer hesitations. Modulation of PFC glutamate release → unable detect a change in reward contingency. (cf reversal learning)


Shallice and Burgess Roberts/Robbins/Weiskranz the PFC 1998

Domain of supervisory processes and the temporal organization of behaviour

Evidence for separate subprocesses in executive function?

Hayling sentence completion 1996: very high cloze sentences (0.99), A=‘complete as quickly as possible’; B=’any word that makes no sense given the sentence frame’. PFC lesions much worse on A and B, but performance uncorrelated.

A requires only contention scheduling, B requires temporary schema (=novel strategy) – e.g. make a word beforehand. Both tasks activate L frontal operculum + R ACC.

Partialling out strategy use gives residual deficit – “presumably related to monitoring / error correction process”.

Brixton spatial anticipation 1996: predict movement of a circle on a numbered 2x5 grid. 9 simple rules – alternation, direction etc., Switches rule every 3--8 trials. ‘Number of responses never given by a control’ does not correlate with ‘number of times switching away from a rule that had been attained, without any negative reinforcement’. Both worse on frontal>posterior.

Strategy generation vs errors in selecting the appropriate processing mode (searching for schema instead of realization of schema).


Tulving 1994: encoding = Lt PFC; retrieval = Rt PFC.

Fletcher 1996: word pair cued retrieval. Orthogonally varied semantic distance and imageability of the pair. No effect of imageability. Related = less activation with increased semantic distance. Reverse for random pairs, in medial frontal and right prefrontal. ?verification proces.


Mason & Iversen Brain Res 150:135-148 1978

Reward, attention and the dorsal noradrenergic bundle

Dorsal noradrenergic bundle in rats:coeruleocortical NA.

  1. Lesion impairs acquisition of rewarded-runaway-response.But: depletions of cortical NA can still learn a wide variety of task.
  2. supports self-stimulation.
  3. Lesion produces resistance to extinction → necessary for internal inhibition / withholding of response.
  4. Increased distractibility – scan more stimuli in the environment
  • 6OHDA used to give NA-specific lesion.
  • Rats learnt to choose sucrose over water; → given LiCl (taste aversion) → 2 day gap to prevent ‘enhanced neophobia’ → day 7: repeat.
  • Then reward learning for responding in a light vs dark room → reversal learning.

water solid, sugar dotted; rapid reversal after LiCl.

. Contingency learning: treated rats slower to learn not to respond in dark

Reversal learning: treated rats much slower to extinguish light responses.

Conclude NA is important not in learning but in attention.



Von Neumann & Morgenstern 1944

Theory of games and economic behaviour

Game = rules. Play= single run of the game. Moves= abstract occasion of choice by one player ro dice (a component of a game). Choice = the specific alternative chosen on a move (a component of a play).

Preliminarity: information of the player about each of the previous choices, at the time of his move.

Poker (inverted signaling) vs bridge (direct signalling) as a consequence of nontransitivity of information between moves. Signalling is costly compared to ‘unsophisticated’ play.

Set theoretic version: set of all possible plays p. 4 sets of subsets at each move: A (the umpire’s knowledge), B(k) (the pattern of assignment for player k), C(k) (the choices available to player k), D(k) (his knowledge). A outcome function Fk(p) of the play for each player. Umpire as the random chooser.

Reduction of each game to single move: choose a strategy S which specifies your choice at each move as a function of your knowledge at that time. All players’ strategies plus the umpire’s (probailistic) strategy fully determines the play p.

  • Chance moves can be reduced to expectations: Outcome = expectation of Fk(as a function of all players’ strategies) over all the umpire’s choices.

2-player zero sum

for strategies t1 and t2, H(t1,t2) is the outcome (positive favours player 1). Player 1 wants to maximize H but only controls t1; player 2 wants to minimize it but only controls t2. Maxx(Miny(f)) < Miny(Maxx)) except at saddle point when they are equal.

Minorant game: If p1 chooses his strategy, then p2 chooses it knowing p1’s strategy: p1 can essentially predict what p2 will do, so p1 will choose to get v1=Mint1(Maxt2( H(t1,t2) ))

Majorant game: if p2 goes first, then he will choose such that v2=Mint2(Maxt1( H(t1,t2) ))

General game has value v1<v<v2 for player1. But whenever a player discovers his adversary strategy, a majorant or minorant game occurs. If v1=v2, there is no advantage to knowing the opponent’s strategy = ‘strictly determined’ game. Here, by playing appropriately, p1 can secure >= v, and player 2 can secure >= -v.

If preliminarity coincides with anteriority (each player knows all previous moves) then a 2pzs game is strictly determined, because, for a function of the form f(x,g(x)), a saddle point must exist – and by induction, for each move, p2’s choice t2 is a function of the previous move t1.


In non-strictly determined games, (scissor/paper/stone) there is an advantage of knowing opponent’s strategy. Hence probabilistic (mixed) strategies arise. Let the players choose probabilistic strategies x and h, rather than t1 and t2, where x is a vector of the probabilities of using each pure strategy t1 (on the plane Sx=1). Clearly the majorant and minorant value are such that v1<=v1’, and v2>=v2’. Moreover, using the convexity/hyperplane theorem: Let K(x,h) = S (H(t1,t2) . xt1. ht2), then Maxx(Minh(K)) = Minh(Maxx(K)), i.e. a ‘saddle point’ of K must exist over the (multidimensional) arguments x and h, and v1 = v2.

E.g. Matching pennies (2p, 2 strategies available to each, payoff matrix H).

If H(1,1)>=H(1,2)>H(2,2) and H(1,1)>=H(2,1)>H(2,2), then v=H(1,2).

If H(1,1)>H(1,2)>=H(2,2)>H(2,1), then v=H(1,2).

In these cases, either one row has all its elements greater than another row, or mut mut for columns.

Otherwise, H(1,2)<H(2,2), and game is not strictly determined, so mixed strategy: x1, x2, h1, h2 can be found as ratios. If U=H(1,1)+H(2,2)-H(1,2)-H(2,1), then

x1 * U = H(2,2)-H(2,1); x1 * U = H(1,1)-H(1,2);

h1 * U = H(2,2)-H(2,1); h2 * U = H(1,1)-H(2,1)

  • All probabilistic moves can be converted in to two nonprobabilistic ones with limited information → strategically identical.

E.g. Poker (2p): each gets a card numbered 1 to a million, higher is better. The turns: unconditionally concede, see bid & check cards, overbid by a, overbid by b). Bluffing with probability b/(a+b) protects against bluffing by the other player.

3-player zero-sum

Basic symmetrical version: each player chooses 1 other player. If p1 and p2 choose each other, p3 pays them sum c. Players free to strategise/make deals with each other before the game.

Analyse as 3 possible coalitions e.g. (1,2) vs 3. This is a 2pzs game, where one player has as strategies the pair (t1,t2), and the other just t3. since

, it is also zero sum. The value v of this game (+a/-a) is the value of the coalition. The three values a,b,c are given by

, and a+b+c>0. (if a+b+c=0, then no coalition has a raison d’etre; playing without coalition is equally valuable = inessential game)

Note that the inductive method does not work here, because at a single move, p1 cannot rely on p2 maximising his F2 on the next move; several ss (choices) may give same F2 but different F1. In 2pzs game, F1=-F2, but not in 3pzs. → a difference that is unimportant for one player might be significant for another. → one must envisage that p1 would try to induce p2 to choose a s that is even suboptimal for F2, as long as p2’s loss < p1’s gain: then p1 could compensate p2 for his loss and possibly even give up some part of the profit.

→ coalitions are necessary to optimize 3pzs game. Imputations (including compensation) will depend on all of (a,b,c), not just the one that actually happens.


Doll, Hutchison, Frank JNeursoci 31:16:6188-98 2011

Dopaminergic genes predict individual differences in susceptibility to confirmation bias

COMT polymorphism rs4680 (better prefrontal structure), DARPP-32 (associated with better reward learning) rs907094, DRD2 (associated with better punishment learning), rs6277.

Miller & Cohen “guided-activation theory”: PFC representations modulate downstream neural activation → 2 models: either A) striatum learns by reinforcement, PFC overrides, or B) PFC instructions bias striatal action selection and learning.

Expt: 2AFC symbol reward/penalty learning, “some symbols have a higher chance of being correct than others... This symbol ‘A’ will have the highest probability of being correct”, 80:20, 70:30, 60:40. Feedback “correct”, “incorrect”. Trained until above chance. The prior information was correct or incorrect.

Conclude: Val/val advantage in flexible gating of alternative hypothesis. Met/Met increases prefrontal function. DRD2 carriers more likely to dismiss negative outcomes when inconsistent with prior belief. Does not fit with obedience model, rather bias model of PFC-striatal interaction.

Evolutionary significance: poorer assessment of objective reality, but allows reaping benefits of others’ experience. ? Frontal = prior, striatum = likelihood ratio? Doll & Frank 2009 handbook of reward and decision making: prediction error model fits better than Bayesian model.


Mirpour, Ong, Bisley JNeurophysiol 104:3021-8 2010

Microstimulation of posterior parietal cortex biases the selection of eye movement goals during search

Test of Itti+Koch 2000, Koch+Ullman 1985: LIP=priority map.

Evidence so far: LIP recordings: reward sensitive, selection in search, decision accumulator. LIP outputs to FEF, SC.

Expt: monkey LIP found by memory guided saccade task: visual burst, sustained delay activity, perisaccadic burst.

Foraging task: display 5 targets ‘T’ & 5 distractors ‘+’. Arranged such that the “multiunit receptive field” encompassed 2 of the stimuli.

Stim 150ms after end of 3rd saccade; 350ms burst of 200Hz pulses.

  • Increased probability of saccade to distractor when distsractor in RF
  • Increased probability of refixating previous target
  • Effect is limited to the next saccade only. (could be decay, or non-remapping)
  • If RF is empty, stimulation may even reduce probability of a saccade into it.
  • No change in saccadic metrics with stim.


Heron, Stockdale,...Whitaker ProcRoySoc B 10.1098/rspb.2011.1131 2011

Duration channels mediate human time perception

Benefits: place coding→less susceptible to noise; interpolation→↑precise

Adaptation method using luminance blob or white noise burst, then 2IFC.

100 x Adapt: stim 40, 80,160,240,400,640,1280,2560ms, ISI 0.5-1s.

Then 2s pause, then

70 x Test: 4 x Adapt stim, then 1 fixed reference stim of opposite modality = 320ms, then 1 variable test stim (237 – 421ms, 7 levels, 30ms intervals) in adapting modality. 2IFC keypress. Repeated 10 x for each test duration.


[could they have just missed the reference stimulus? It was in a different modality. Then they’d just judge the test stimulus relative to the adaptors rather than the reference.]

[Cross modal effect?]


Rabbitt Nature 212:438 1966

Error correction time without external error signals

Previous study (Rabbitt JEP 1966) shows fast corrections after immediate “wrong” feedback.

Expt: ‘continuous performance test’ – digit, respond hi/low = L/R. ITI = 50ms; no feedback. Instruction “If you notice you made a mistake, correct it ASAP and then wait for 5-10 seconds before carrying on”. (to discriminate intentional corrections from fortuitous or responding to the next trial where the response would be the same).

Response alternations 454

Repeat response 382

Repeat response and stimulus 365

Error correction 327, are some within 40ms of error

Conclude: efficient error correction without feedback.


Rabbit QJEP-A 55:4:1081-92 2010

Consciousness is slower than you think

  • Subjects correct errors very fast (same response as they should have made)
  • Very slow at signalling an error (e.g. with a separate ‘I made an error’ response)
  • Poor memory of unsignalled/uncorrected errors
  • Instruction to ignore errors → involuntary corrections, post-error slowing persists
  • Unsignalled errors → still have post-error slowing
  • Age doesn’t impair P(correction|error) or post-error slowing. But reduces accuracy of error-signalling and error-recall

Rabbitt 1990: Normally, P(correction|error) ~ 86-96%. → “so these correction responses must sometimes be, in effect, delayed correct responses that are initiated before the error is complete”. – this may coincide with the inability to completely suppress corrections.

Parallel processes → maybe corrects are also followed by short-latency responses; but this is not seen. Consistent with SAT: later on=less likely to make erroneous corrections.

Townsend & Ashby 1983: theoretical review of error corrections.

P(error-signalling|error)=65%, RT~580-640ms. = slow and difficult.

Expt: 40 young, 40 old subjects. 4-button choice reaction time to 4 lights. (2 fingers of LH, 2 fingers of RH). Interval between response & next stimulus (ITI) = 150 / 500 / 800 / 1000ms. 3 blocks: 1) Correct your errors asap, 2) Press a 5th button if you detect error, 3) Ignore errors (verbal questions “do you remember making a mistake in last 3 trials?”: either 3 trials after an error, or 9 trials consecutively correct).

Significant interaction between age and ITI, but only in the signalling & recall measurements. Post-error slowing and direct correction are unaffected by shortening ITI. Older: need 1000ms, Younger: need 800ms.

Rabbitt & Vyas 1981: difficult line discrimination, masked. As ↑difficulty, ↑errors; but as presentation duration↑, proportion of errors corrected↑ → information accumulation continues after response.


Seung & Sompolinski PNAS 90:10749-53 1993

Simple models for reading neuronal population codes

MLE is optimal for large N neurons with independent noise → error ~= sqrt(N)

Tuning curves: (m=2, q=stimulus, r=response). Variance of ML estimate = Fisher information ~ curvature of log likelihood function = proportional to N. So typical error is N, bias is 1/N. Information in 1 neuron = = signal-noise-ratio^2. Total information =.

for the smoothed curve, max information occurs at a – where r is baseline firing. There, total.

Discriminability = . For the two hypotheses (same/different), the greater likelihood method gives error erf(d’/2) for single-interval and erf(d’/sqrt2) for two-interval. Because: is equivalent to discriminating ML estimates of q, and estimate is normally distributed for large N.

population vector’: single layer network → sums of ri → vector = the 2 quantities [cosq, sinq], which is distributed as a 2D gaussian, and the error for angle is where fn are the fourier components ~= (population averaged SNR)^2 = insensitive to the tail of the tuning curve. This is worse with narrow tuning curves since there are fewer active cells and, unlike ML, it cannot make use of the increasing gradient of the response.

Perceptron to judge which of 2 stimuli was more clockwise: take R=Swiri for each of 2 stimuli, and compare. This can only judge one angle optimally, as it is a linear function not a vector. When fully trained on angle q0, = MLE.

Double perceptron: calculate [sinq cosq] in separate perceptrons, then second layer takes sinq0 sin(q-q0) – cosq0 cos(q-q0). Simple cell r=0.01 and HWHM 14–22


Krebs, Boehler...Woldorff J Neurosci 31:26:9752-9 2011

The neural underpinnings of how reward associations can both guide and misguide attention

Stroop with hi-/low-rewarded colours. Text could be a high- or low-reward colourword, and could be congruent/incongruent.

  • (Relevant reward $>0): accumbens, rt DLPFC, rt IFG, ant insula, bilat fusiform
  • (Irrelevant reward $>0): ↑ preSMA, dlPFC, bilat fusiform
  • preSMA correlates with longer RT in trials with high-reward distractors.

Irrelevant rewards don’t generally distract from the task; they induce the prepotent response tendency.


Rabbit QJEP 29:727-43 1977

What does a man do after he makes an error? An analysis of response programming

Laming 1968: RT decreases for 3 trials before an error. E+2 and E+3 responses also slowed.

Post-error slowing >300ms stronger than just to return RT to a ‘safe’ level

Expt: Various fingers press for digit. 2AFC (5,6) vs 4AFC (3456) vs 8AFC (23456789).

After a correct, RSI=random 20 or 200ms. After an error, blockwise either 20ms or 200ms (to keep proportions balanced).


  • Double errors = 17% 12% 11% (for 2,4,8AFC respectively), expected <4%
  • Of these, 74%, 60%, 54% are involuntary correction-responses (expected 50/25/12%)
  • Even accounting for this, there are more double-errors than expected.
  • Some of the E+1 corrects, then, are also corrections.
  • But even accounting for this, there are more than expected E+1 corrects for signal-repetition than non-signal-repetition → “if an error was immediately followed by a repetition of the same signal, subjects could respond to this signal more accurately than they could respond to any other E+1 signal.”
  • For E+1 correct trials, RT fastest for signal-repetition than response-repetition (i.e. different signal that requires the same response as the erroneous one just made) or new-signal.
  • RSI 20 vs 200ms had less effect on E+1 RT for signal-repetition than on response-repetition or new-signal trials. i.e., 200ms RSI speeds responses, except when a repeated stimulus occurs after an error.

Subjects are slow on E+1 because of response competition from the involuntary error-correction.

→ E+1 corrections are not suppressed if the signal repeats → E+1 signal is processed before error-correction process completes



Ratcliff, Van Dongen PNAS 108:27:11285 2011

Diffusion model for one-choice reaction-time tasks and cognitive effects of sleep deprivation [Later]

“Classic vigilance”: wide spacing e.g. 12 events every 29 min

“psychomotor vigilance test”: several per minute

“simple RT task”: every few hundred ms.

Hazard=f(t)(1-F(t)) reveals shape of right tail.

Model: decision time: from drift rate (varies trial to trial normal distribution).

“if there were no across-trial variability in drift rate & nondecision time & drift was +ve or 0, then the RT distrib would be inverse Gaussian”.

nondecision time varies from trial: flat distribution around a mean with given width. Unlike 2-choice model, no need for variation in starting point.

m/s is an index form of the hazard function for 1-choice tasks: comparable to d’.

Expt 1: dynamic noisy stimulus, 1-choice (“respond as fast as possible when change”), vs 2-choice (“decide as fast as possible between dark and light”) .

2-choice: s constant across difficulty, m varies.

1-choice fit: constant nondrift time, constant nondrift stdev, constant s, but m/s varies with difficulty.

No correlation in parameters of 1-choice or 2-choice fit: ?“different cognitive processes involved”.

Expt2: 18 x every 2-hrly psychomotor vigilance test (= 36h sleep deprivation).

Test lasts 20 min. Every 2-10sec, counter appears on screen → speeded keypress → numeric RT feedback

Diffusion fit: m/s changes over time. But also, nondiffusion time ↑12ms & stdev ↑20ms. Correlation between test-bouts = 0.6.

Biomathematical model of fatigue: alertness = linear decline modulated by 24hr cycle.

m/s: 2 → 2.3% of trials have negative drift rate; 1 → 16% negative drift rate. <1 → hazard low right tail.

Smith 1995: Change detector + level detector → peak + sustained tail. “too many parameters”


Donkin Brown Heathcote & Wagenmakers PsychBullRev 18:61-9 2011

Diffusion versus linear ballistic accumulation: different models but the same conclusions about psychological processes? [Later]

Complains that Ratcliff 1978 is nonanalytic. Brown & Heathcote (2008) proposed Linear Ballistic Accumulator: “closed form expressions for the likelihood of an observed choice”... “even though it is relatively new...” [!]

Straightforward correspondence; questions of extra parameters like





starting point of diffusion (z)

relative thresholds for the two processes


both thresholds get further from 0 (a)

?threshold – prior/2 [no real equivalent]

Nondecision time

Good correspondence

Caution: except at the fastest speed-accuracy settings, accumulator generates larger shifts in the leading edge of the RT distribution, for changes in caution. In the diffusion model, moment-to-moment fluctuations allow very fast evidence accumulation immediately after the non-decision time.

Fitting one model to another: Caution in diffusion is reflected in all three accumulator parameters. Caution in accumulator model is reflected in diffusion’s caution and also non-decision time.


Bristow, Frith, Rees Neuroimage 27:136-45 2005

Two distinct neural effects of blinking on human visual processing

Blink Suppression vs Perceptual Continuity: latter requires active mnemonic maintaining of information across blinks → should be activated.

Blink suppression peaks 30-40ms before pupil covered. Strongest for low spatial freq.

5 conditions. Visual stim=reversing chessboard

  • ((BV+BN) – (FV+FN)) = main effect of blinking = precentral gyrus, FEF, SEF, cerebellum + cingulate gyrus adjacent to SEF, precentral sulcus, lateral fissure, posterior lateral orbital gyrus, putamen, IFG, occipital gyrus, precuneus, cuneus etc. Likely magnocellular.
  • ((FV-FN) – (BV-BN)) = blink suppression = effect of visual stim is reduced in the presence of blinking = V5/MT (lateral temporooccipital), fusiform, inf & middle temporal gyri, other occipital sulci, collateral sulcus. + superior parietal, postcentral. Nothing in V1/2/3.
  • ((BV-FV) – (BN-FN)) = visual continuity = voluntary blinks enhance the response to a visual stimulus = single region of parietooccipital fissure. Not activated by visual stim alone. ?WM
  • (BV-DV) = FEF, SEF, cerebellum (oculomotor of course) + parietooccipital fissure exactly overlapping “blink suppression” region. Same region also ↑ for blinking in the dark compared to darkenings (BN-DV)

Assumption: voluntary blinks similar to involuntary.


Ratcliff & McKoon Neural Comp 20:873-922 2008

The diffusion decision model: theory and data for two-choice decision tasks [Later]

SAT: boundary separation changes. Quality of evidence: drift rate varies.

Ratcliff et al. 1999: in a single experiment, error RT > correct RT → can move to error RT < correct RT.

But: Model takes quintiles only. Remove outliers 2-3%. Minimises chi-square (sum (observed-expected)2/expected over conditions and err/correct) using simplex. Modelling proportion of ‘contaminant’ responses (= uniform distribution between min&max RT) improves fit.

Graph of RT quantiles against accuracy. Errors on left, corrects on right (e.g. 0.1 corresponds to 0.9). Says 4 parameters: t0 for RT shift, a (top threshold), h (drift stdev), sz (range of starting point across trials). Mean drift rate determines P(corr), on the x-axis, and a determines spread at each difficulty level. [but also has spread of variability in t0]

Zero variability in start point → errors slower; Zero variability in drift rate → errors faster than corrects. Effect of variability in start point is in scale to threshold a: when criterion is ↓, trial-to-trial var of prior z is relatively ↑ → dominates var of drift rate → fast errors.

↑Spread of t0 enhances differences between conditions.

Expt: 2afc random dot motion. ex1) coherence 0-50% = difficulty = drift rate. ex2) speed or accuracy blocks = boundary a. ex3) P(left)/P(right) odds 3:1 or 1:3 ?starting z or a or both.

1 param varied per expt.

Also modelled response-signal task: “respond at time of signal”.

Inter-individuals: accuracy correlates w drift rate, mRT correlates w threshold & nondecision time. Within individuals, a correlates w t0, and with drift rate. ↑Age → ↑a but doesn’t change drift rate.

Same for lexical decision in aphasia: ↑a only.

1981: comparing letter strings in memory – degree of match → drift rate.

Smith 2004:

  • stimulus → representation in a visual STM
  • onset of information is delayed for unattended locations (attention has to move)
  • masking → buildup of information stops; after, the representation is stable
  • strength of representation ≈ stimulus duration x contrast

Nosofsky & Palmeri 1997: multidimensional categorisation → 2afc drift to 2 thresholds

Predicts neural noise [but does it predict ratio for intra-trial vs intertrial noise?]

The 2-thresholds model doesn’t fit the neural data → 2-horse



Flagel, JClark, Robinson et al. Nature 469:53 2011

A selective role for dopamine in stimulus-reward learning [Neuron, Reward, Learning]

CS stops and US starts simultaneously:

Sign-tracking = “some animals engage the CS itself, and go on to the location of food delivery only upon CS termination”. → CS is an attractive incentive stimulus – ‘wanted’

Goal-tracking = “other individuals do not approach the CS, but during its presentation engage the food location even though it is unavailable until CS termination”

Rats bred for high locomotor response to novel environment = sign-trackers; low response = goal trackers.

Baseline: High responders-to-novelty have

  1. higher sensitivity to DA agonists,
  2. increased proportion of striatal D2 receptors in the high-affinity state,
  3. higher frequency of spontaneous DA transients,
  4. higher DA response to reward before conditioning.

Cycling voltammetry in NAcc: high response-to-novelty rats showed an ↑ in DA in response to CS during learning, whereas lo-response rats did not. After acquisition, DA response to US was lower in high-responders than lo-responders.

Conclude: “CS-evoked dopamine release does not encode the strength of the reward prediction, as previously postulated, rather it encodes the attribution of incentive value to the CS.”

Flupenthixol during learning: reduced learning of sign-tracking but normal learning of goal-tracking (if tested off drug; testing on drug impairs performance for both types, since flupenthixol after acquisition also reduced performance ?nonspecific effect on activity)

Conclude: dopamine is necessary for both learning and performance of sign-tracking CR, but necessary only for the performance of goal-tracking CR. Goal-tracking learning can occur without DA.

  • But: The baseline dopamine-tone is unlikely to be causative, as outbred rats with identical baseline properties develop either into high-/low-responders after conditioning, with the characteristic DA release patterns.
  • DA attributes incentive salience to reward cues. “Individuals who attribute reward cues with incentive salience find it more difficult to resist such cues, a feature associated with reduced impulse control”.



Berdyyeva & Olson J Neurophyiol 105:5:2547-59 2011

Relation of ordinal position signals to the expectation of reward and passage of time in four areas of the macaque frontal cortex [Neurone, Memory]

Cells in PFC fire at specific moments during a sequence memory task.

Confound with 1) passage of time 2) reward expectation → use 4 tasks.

A)    serial action task: 3 sequential saccades, symbol indicates order of directions

B)    reward size anticipation task

C)    long delay task.

D)    Serial object task: 3 sequential saccades, display has 3 different objects.

independently sensitive to reward and direction

  • Often similar timecourse in delay task.
  • However, correlation between serial-rank sensitivity and time-passage sensitivity explains only 30% of variance.
  • SEF shows least correlation.
  • DLPFC shows least correlation between reward sensitivity and serial-rank sensitivity.
  • Highest correlation is between serial-object and serial-action tasks.

Fits with preSMA & dlPFC more cognitive, SMA & SEF more motor/reward.

Time and preSMA: Sakai 2002, Bengtsson 2004, Coull 2004, Knutson 2004, Lejeune 1997, Macar 2002, Nobre&OReilly 2004, Schubotz 2000

Representation of ordinal position is “widely distributed” – all areas carried equal rank signals.


Hong & Hikosaka Frontiers in Beh Neurosc 5:15:1 2011

Dopamine-mediated learning and switching in cortico-striatal circuit explain behavioural changes in reinforcement learning

Use of Frank model: “increasing (LTP) and decreasing (LTD) “forces” of opposing processes in each pathway”:

  • direct pathway: co-occurrence of pre- and post-synaptic activity, together with DA concentration above a threshold, → LTP (DA-dependent LTP), while the co-occurrence of pre- and post-synaptic activity alone → LTD (DA-independent LTD).
  • indirect pathway: co-occurrence of pre- and post-synaptic activity, together with DA concentration above a threshold, → LTD (DA-dependent LTD), while the co-occurrence of pre- and post-synaptic activity alone → LTP (DA-independent LTP).

If the indirect pathway has some LTP, it is adenosine A2a controlled, not DA.

If the direct pathway has some LTD, it is endocannabinoid CB1 controlled, not DA.

Expt: 1DR task: single target, L/R side, one side rewarded x 20-30 trials, then switch reward.

Over 10 trials, speed bias develops. After extensive experience, speed bias develops much faster.

  • Shortening of latencies to rewards is quicker than the prolonging to no-reward → faster acquisition, slower forgetting, of a motivated behaviour → direct vs indirect

  • Direct: D1-mediated LTP; Indirect D2-mediated LTD

Hood 2007: antisaccades in well medicated PD: L-DOPA→prosaccades slower, ↑accuracy. Explained as SAT, “enhanced compensatory cortical mechanism in PD”.

Why 2 pathways? Flexibility

  • 2 conflicting goals – appetitive go, and fearful nogo
  • Rat: Indirect pathway input from deep cx (motor outputs to cord) – to terminate commands, or useful in sequencing commands. Direct pathway input from intermediate cx (which projects bilaterally).


Middlebrooks & Sommer JEP:LMC 10.1037/a0021611 2010

Metacognition in monkeys during an oculomotor task

4 SOAs 16 to 67 ms = Difficulty.

Decision correct/incorrect. Bet low/high




High bet

5 drops

5s timeout

Low bet

3 drops

2 drops

No latency differences.

Phi: correlation between correct/incorrect and high/low bets = 40-75% correlation.


Conclude: metacognition. ?mediated by corollary discharge.



Lamme VF TICS 10:11:494 2006

Towards a neural stance on consciousness

  • NCC depends on how we measure consciousness – report? drawing? Choosing? Memory? (Wolfe 1999 inattentional amnesia, Landman 2003 change blindness: → recurrent processing in visual areas but not higher areas)
  • Maybe consciousness != phenomenal report, but neurally definable?
  • By stipulating which measurements comprise consciousness, we jump over it and onto the cognitive processes.
  • But it is independent: attention, oject recognition, perceptual learning without perception, semantic priming, value-decisions, can all occur without consciousness
  • Can’t say they are not conscious, on the basis of verbal report, without presupposing some processes (attention, memory, inner-language) are privileged.

Dehaene defines ‘preconscious’ representations = not in global workspace. But are they phenomenally conscious? “does not seem to be a scientifically addressable question”

Candidates: recurrent interaction, sustained activity in certain high-order neurons,

“there is no question as to the absence of conscious experience in blindsight”

Presence of recurrent activity in masking there is conscious experienced, even though no memory or attention etc.


Dehaene, Posner, Tucker Psych Sci 5:5:303 1994

Localisation of a neural system for error detection and compensation

ERN: 100ms after error EMG too short for sensory feedback → = internal monitoring.

↑Amplitude for emphasis of accuracy>speed.

Large ERN → ↓force on error key, ↑P(error correction), ↑post-error slowing.

ERN best seen averaging on response, but also seen if locked to stimulus.

Dipole model: either ACC or SMA.

Hypothesis: ERN for slips not mistakes: feedback on a difficult categorization/memory task; error feedback → broad frontal positivity P300 = surprise response. ACC=“attention for action”


Compton...Carp Psych Sci 19:7:702 2008

Error-monitoring ability predicts daily stress regulation

  • Stroop with ERN measurement. Subjects with higher “cognitive control...were less reactive to stress in daily life”.

Stroop 600ms blank, 150ms word, no feedback. Accuracy 89%, error RT 557ms, correct RT 683ms, posterror accuracy 85%, posterror RT 774ms. (same as rabbit).

Then trait neuroticism, extraversion, agreeableness questionnaire. Then 1 week x daily questionnaires on stressors & anxiety. Plotted correlation between stress & anxiety.

3 stroop scores: ERN difference, E+1-RT difference, and E+1-accuracy difference. → all predicted slope of relation:

“cognitive control moderates stress-reactivity”.


Johansson,...Olsson Science 310:5745:116-9 2005

Failure to detect mismatches between intention and outcome in a simple decision task

2afc attractiveness of faces judgements. 15 choices, 3 were manipulated so that Ss were then given the other face as though they had chosen it. Then asked to describe why they chose it. 2s/5s/inf deliberation time; similar or dissimilar face pairs.

13% of switches were detected, 27% in the inf-time low-similarity condition.

Reports of reasons: scored by independent raters. no difference in tense, content, length, laughter, confidence, emotionality, specificity. More dynamic self-commentary when manipulated. = Choice blindness

“Warns of the dangers of aligning the technical concept of intention too closely with common sense.”: confabulation as a norm.


Greenwald, McGhee Schwarz J Persnalty Soc Psy 74:6:1464-80 1998

Are a pair of target-concepts A,B are related to a pair of evaluative-attributes C,D?

50 trial blocks, reminder category names remains on screen L/R, error feedback for 300ms, ITI 100-700ms, end-of-block feedback for accuracy and RT.

1.      target-response mappings {AX BY}

2.      attribute-response mappings {CX DY}

3.      then combined {AX BY CX DY}

4.      reverse target-response {AY BX}

5.      reverse-combined task {AY BX CX DY}

Association of A-C vs A-D, and B-C vs B-D = comparison of whether the combined (3) is easier than the reverse-combined (5).

e.g. insects-flowers, pleasant-unpleasant words.

IAT discriminates better than explicit liking questionnaires for racism.


Forstmann...Brown...Ridderinkhof, Wagenmakers PNAS 105:45:17538 2008

Striatum and preSMA facilitate decision-making under time pressure

Precue: ‘Fast’, ‘Accurate’ or ‘Neutral’ letter cue 5sec. 2AFC random dot motion direction, respond during stimulus (<1500ms). Feedback: Fast → >450ms ‘too slow’; Accurate →‘correct’/‘incorrect’; Neutral → >750ms and correctness feedback.

Fit 2-horse race model, prior drawn from uniform distribution [0-A]. Considered 25 models where different parameters were constrained or varied by precue or stimulus direction. m and s vary for L/R. For each subject,

Subjects with large decrease in ‘caution ratio’ (=threshold/startrange = b/A) have larger changes in anterior striatum and preSMA.

→ decision under time pressure activates preSMA and ant striatum. (cf Lo & Wang Nat Neurosci 2006)


Forstmann...Brown...Ridderinkhof, Wagenmakers PNAS 105:45:17538 2008

Striatum and preSMA facilitate decision-making under time pressure

Precue: ‘Fast’, ‘Accurate’ or ‘Neutral’ letter cue 5sec. 2AFC random dot motion direction, respond during stimulus (<1500ms). Feedback: Fast → >450ms ‘too slow’; Accurate →‘correct’/‘incorrect’; Neutral → >750ms and correctness feedback.

Fit 2-horse race model, prior drawn from uniform distribution [0-A]. Considered 25 models where different parameters were constrained or varied by precue or stimulus direction. m and s vary for L/R. For each subject,

Subjects with large decrease in ‘caution ratio’ (=threshold/startrange = b/A) have larger changes in anterior striatum and preSMA.

→ decision under time pressure activates preSMA and ant striatum. (cf Lo & Wang Nat Neurosci 2006)


Plotinus Ennead 1 250

Individual soul as a branch of the ‘All-Soul’.

Where is seat of the affections? Is it in the body or soul or in some ‘coupling’ of the two? Affections and actions must be seated in the same place. Could the soul be interwoven in the body?

Plato: “it is absurd to say the soul weaves”. But since affectations are ‘of things’, they cannot be solely in the soul, unless there is coupling.

“Is it any explanation to say that desire is vested in a Faculty-of-desire, and all tendency is seated in the Appetitive-Faculty?... does not help towards making the affectations common to the Couplement; they might still be seated either in the Soul alone or in the body alone” – sometimes mental causes physical, sometimes vice versa.

Only solution is that “impressions are already Intelligibles, while outer sensation is a mere phantom of the other which is nearer to Authentic-Existence as being an impassive reading of Ideal-Forms”.

Evil = “we credit only the lower perception, that of the Couplement, without applying the tests of the Reasoning-Faculty”

Couplement as “the Understanding, as passing judgement upon Sense-Impressions, is as the vision of Ideal Forms, seeing them as it were with an answering sensation (i.e. with consciousness)”

Animals = ?human souls that have sinned!

Birth = “coming-into-being of that lower phase of the Soul” – the coupled part. Soul illuminates its object.

Reasoning: is an act of the Soul alone, “without movement; any movement that can be ascribed to it must be utterly distinct from all corporal movement and simply be the Soul’s own life.”

Although the Soul contains the divine, the Civic Virtues (prudence etc.) cannot be found in it. Where do they come from? (“must there needs be something to warm the source of the warmth?”) Virtue is a state, and the divine is stateless; but there is a Likeness that mediates virtue. “As speech is an echo of thought in the Soul, so thought in the Soul is an echo from...the higher sphere.”

→ Soul sees Forms, images of which it has always possessed. → “purification”

By seeing the forms we can either improve, or “for sheer shame, never veture any act which the nobler mind would disapprove”

Are there different ways for each individual to improve? E.g. the musician “must be shown that what ravished him was no other than the Harmony of the Intellectual world and the Beauty in that sphere, not some one shape of beauty but the All-Beauty”.

This kind of Dialectic (=the precious part of Philosophy) ends in “discernment of the Ideal-Forms, of the Authentic-Existence and of the First-Kinds.” It is non-propositional. It induces the virtue of Wisdom; lower virtues require Wisdom, but wisdom springs after the other lower ‘natural’ virtues.

Happiness is low, plants can have it. “Those that deny the happy life to the plants on the ground that they lack sensation are really denying it to all living things.” Why do we call reason good? Because it is useful, not for itself. But because it is in our nature, for us, the good life includes Reason.

“The sign that this state has been achieved is that the man seeks nothing else.” It does not “require freedom from pain, sickness, misfortune, disaster”. Those needs “serve towards the integrity of his being”.

The Sage must estimate death to be “better than life in the body”.

On consciousness: “When the Intellect is in upward orientation that [lower part of it] which contains [or, corresponds to] the life of the Soul, is, so to speak, flung down again and becomes like the reflection resting onthe smooth and shining surface of a mirror; in this illustration, when the mirror is in place the image appears but, though the mirror be absent or out of gear, all that would have acted and produced an image still exists; so in the case of the Soul; when there is peace in that within us which is capable of reflecting the e activities upon which it is exercised, and that in the degree in which these pass unobserved they are purer and have more effect, more vitality, and that, consequently, the Sage arrived at this state has the truer fulness of life, life not spilled out in sensation but gathered closely within itself.” images of the Rational and Intellectual-Principles these images appear. Then, side by side with the primal knowledge of the activity of the Rational and the Intellectual-Principles, we have also as it were a sense-perception of their operation.

When, on the contrary, the mirror within is shattered through some disturbance of the harmony of the body, Reason and the Intellectual-Principle act unpictured: Intellection is unattended by imagination.

In sum we may safely gather that while the Intellective-Act may be attended by the Imaging Principle, it is not to be confounded with it.

And even in our conscious life we can point to many noble activities, of mind and of hand alike, which at the time in no way compel our consciousness. A reader will often be quite unconscious when he is most intent: in a feat of courage there can be no sense either of the brave action or of the fact that all that is done conforms to the rules of courage. And so in cases beyond number.

So that it would even seem that consciousness tends to blunt th



Ainslie Psych Bull 82:4:463 1975

Specious reward: a behavioral theory of impulsiveness and impulse control

Impulsivity = choosing bad option despite knowing it is bad. Described by hyperbolic discount function.

3 reasons: 1) Socrates: subjects choose without having properly learned consequences.

2) impelled by a lower principle (devil, repetition compulsion, classical conditioning)

3) distorted valuation (Mowrer & Ullman 1945, Jevons 1871) = ‘improvidence’, ‘defect of will’, inability to imagine future goals, lower expectation of getting things in the future, waiting is itself aversive, perceived risk/uncertainty in the future, ‘perspectival’ innate property of time perception (Samuelson 1937).

Descriptions “deferred gratification”, “impulse renunciation” as Piagetian stages.

?Low positive correlations with social class, intelligence, also in adolescence: age, father-presence, nondelinquence, yea-saying personality.

[why the curve? Because we can partially consume resources]

Buss 1964: Ss chose some actual and some hypothetical rewards – no correlation.

Freud 1911: Pleasure-ego vs reality-ego → “repetition compulsion” and “death instinct” (punishment is itself rewarding, in beyond the pleasure principle). Rapoport: transition from externally imposed delay → ability to control and delay.

Blachly’s (1970) 4 characteristics of maladaptive behaviour: 1) active participation in your own victimization, 2) negativism 3) short term gain, 4) long term punishment.

Animal: learning remains strong for rewards up to 1 minute. Chimps don’t learn for delays >1hr. Logan 1965 estimates Value = amount – 0.13 t0.5 (for t in sec). Herrnstein 1967 estimates an inverse law: proportion of responses = d1/(d1+d2). Killeen (pigeons, 1970) found a t2.5 relation. [natural discounting rate – growth of capital = interest?]


Abe, Schambra...Cohen Cur Biol 21:7:557-62 2011

Reward improves long-term retention of a motor memory through induction of offline memory gains


Daigneault, Bran, Whitaker Brain & Cog 19:48-71 1992

An empirical test of two opposing theoretical models of prefrontal function

Confirmatory factor analysis between large number of patients on several prefrontal measures.

Neuropsychological model: Different functions are localised differently

1.      Planning (self-ordered pointing task, wrong entries in porteus maze test)

2.      Self-regulation after feedback/contingencies (perseverative errors in WCST & porteus maze)

3.      Maintainence of behavioural set (Interference errors in stroop, category breaks in WCST, alternation errors on Trail Making B)

4.      Spontaneity/sustained productivity (slow verbal fluency & design fluency)

5.      Spatiotemporal segmentation/organisation = Schachter’s contextual chunking (recency judgements bad but recognition spared, confusions between recall lists)

Goldman-Rakic model – that units represent SR mappings, so lesions should knock out specific mapping types

1.      Verbal representation → Manual responses (WCST category breaks and perseverations, trail-making alternations)

2.      Verbal representation → verbal response (recall and recency confusions)

3.      Visuospatial representation → Manual responses (porteous maze task, design fluency)

4.      Verbal and visual tasks → Simple responses (SOPT, relative recency), requires collaboration between verbal and visual representations

Conclude: Goldman-Rakic model fits data better

Hickey Chelazzi Theeuwes Vis cog 19:1:117-28 2011

Reward has a residual impact on target selection in visual search, but not on the suppression of distractors

When reward influences attention, does it value/devalue the target, or block/enhance the distractor? Separate this out by having variable/constant salient distractor colour, or variable/constant target colour.

Search for unique shape containing a horz/vert line. 2afc keypress for horz or vertical. One shape could be in a distracting colour. R/G/B. Either distractor is variable colour, or targets are all the same variable colour.

Random +1 or +10 reward for a correct response; -10 for error. Accuracy ~90%.

Distractor slowed response 700ms→716ms.

Conclude reward affects target selection but not distractor filtering.

Reward influences “rapid detection and localisation of target features”, but not later suppression of distractors required for target identification.


Nieuwenhuis Stins Posthuma Mem & Cog 34:6:1260-72 2006

Accounting for sequential trial effects in the flanker task: conflict adaptation or associative priming?

Eriksen with flankers 100ms before central target. Entire array remains until response, then 1s ITI. 6 conditions: congruent/incongruent x congruency changed x response target same/changed.

Conflict adaptation only for response-repetition trials. Same under time pressure, in children and elderly, with loger ITI. Response-repetition-stimulus-change trials are maximally impaired by previous congruence as opposed to incongruence.

Conclude: pure associative priming account of conflict adaptation effect.

Inconsistent with Ullsperger/Botvinik. Cf Stroop & Simon tasks: congruency effect smaller after high-conflict trials, irrespective of previous stimulus change or response change. Could be because the distractor belongs to a different stimulus dimension (word vs colour, location vs colour), but in Eriksen it is same dimension. Maybe cognitive system can only select between dimensions, not within.

Some negative priming seen – this is why some incongruent trials make subsequent performance worse. This is only evident when the stimulus set is small (cf Stroop).


Anderson, Laurent, Yantis PNAS 108:25:10367-71 2011

Value-driven attentional capture

Task: 2afc respond to orientation of line in target circle. Target identified by either being red or green. (Only one present on each trial). If red → high reward, green → low reward. Training 1008 trials. Then test task: identify line orientation in the unique shape. Critically, 33% trials had one red distractor circle, 33% had a green, 33% only other colours.

Train: Test:

Result: RT high value distractor > low value distractor > no distractor.

Distracting effect despite being nonsalient and non-goal-relevant. “clearly violate the predictions of both a salience-driven and goal-driven account of attentional capture”

Visual WM capacity correlates with lower capture.

Conclude: cf shiffrin & schneider – training on a target letter in visual search → letter captures attention even when it is no longer task relevant.

Also: “Responses were on average 66 ms slower when the target appeared in a location

formerly occupied by a high-value distractor than when it appeared in another location, confirming that high-value distractors indeed capture attention in a spatially specific manner.”


Lou,Skewes...Roepstorff JVis 11:2:15 2011

Target word 33/50/66ms, then 200ms mask, then confidence Likert 1-4 that the word was seen, then 2 words simultaneously, one is novel, one is the target word, 2AFC recognition.

12 subjects on pergolide, 12 on placebo.

Conclude: dopamine influences confidence ratings. Relates to schizophrenia.

Optimal reward-harvesting = combining confidence with expectation of reward.


Aupperle, Sullivan...Stein BBR 225:2:455-63 2011

A reverse translational approach to quantify approach-avoidance conflict in humans

Volunteers with anxiety.

Approach-avoidance conflict task: runway, move avatar towards cloud or sun. Cloud can be accompanied by 0, +2, +4 or +6 points in value. Final location represents probability 10% – 90% of an outcome in each direction. Feedback = nasty picture plus gaining points, or just nice picture. Starting location randomised between -4 to +4.

Approach avoidance conflict shows correlation with post-task questionnaire “motivation to get points”. approach behaviour correlate negatively with anxiety sensitivity index physical subscale in Males, and positively with Fun-seeking in Females.


Ross..Manoach Barton Neuroscience 2011

Human prosaccades and antisaccades under risk: effect of penalties and rewards on visual selection and the value of actions

Do humans show the Peck effect: attention drawn to locations of positive reward cue?

3 peripheral reward cues and one central no-reward cue. Manipulate congruence of cue and target location.

Sess1: Rewards in reward trials, Penalties in penalty trials.

Sess2: Reward/penalty only when (8 subject) congruent or (8subjects) incongruent.

Strong IOR observed

Conclude: reward modulates IOR magnitude. IOR ‘facilitation’ of contracue location ONLY when reward or penalty is expected. Also true for sess2 (contingent reward / penalty) → we have flexible control over the cue’s effect – unlike monkey.


Hagoort TICS 9:9:416 2005

On Broca, brain and binding: a new framework

Goal: account for fMRI language studies in terms of architecture rather than localising specific tasks.

Foundations: merge-operation from Chomsky’s minimal model; Jackendorff unification of pieces stored in a common format; Joshi’s tree-adjoining grammars, Voss & Kempen 2000 model. → ‘unification’

“memory, unification, control”. Unification occurs at syntactic, but also semantic and phonological levels.


Laeng & Endestad PNAS 109:6:2162-7 2011

Bright illusions reduce the eye’s pupil

Equiluminant images with Gaetano Kanisza figure, Kitaoka figures.

Classic account of pupillary light reflex as a “closed loop” and “impenetrable servomechanism”: challenged by information content, colour, ?expectation of brightness.

“it is known that the fovea and its immediately surrounding area can contribute more than the rest of the retinal field to the pupillary response to luminance”, Kardon Kirkali & Thompson 1991, Clarke Zhang & Gamlin 2003. → did it at fixation and got the same result.




Curtis & Lee TICS :216 2010

Beyond working memory: the role of persistent activity in decision making

1)      WM: seems to occur when stimulus needs to be remembered & vanishes after no longer needed, impersistence correlates with WM errors, magnitude increases as WM load increases, plateaus at limit of WM capacity, selectivity of firing encodes location of remembered items. But: Persistence encodes visual properties even in the absence of WM demands

2)      Anticipation of future stimuli (Rainer prospective coding for objects1999), or representation of forthcoming action metrics or goals

3)      ‘active representation’ of rules, associations, categories (wallis nature 2001)

4)      Integration: an integrator holds information in its sustained activity, but can also temporally average sensory noise

Example: information held between trials → integration of learning . this information decays gradually, unlike delay activity in WM task (3 x J Neurosci papers by Lee 2007, 2009).

         Holding previous action until outcome is known

         Holding current expectation of reward until outcome is known

         Holding expected reward function between trials (it is updated every trial)

         Eligibility trace: previous trial information necessary for solving credit assignment problem

         Previous outcome: effect often continues into ITI.

         PFC, PPC, BG all have persistent traces of previous-trial-action.

         Average reward rate needed by some RL algorithms


Olivers, Peters, Houtkamp, Roelfsma TICS 2011

Different states in visual working memory: when it guides attention and when it does not

Distinguish ‘active’ WM items (task set and focus of attention) from ‘accessory’ ones.

If active WM item is visual, then it “becomes an attentional template” and “drives” selection.

  • Search tasks require WM of “search template”. Olivers 2006 JEPHPP – dual task WM for a colour and then search → a distractor of the same colour as the WM is more distracting.
  • But Peters 2009: Ss never confuse the concurrent WM item and the search target
  • however Houtkamp 2006: if remembering 2 search targets – for immediate & later use → no interference.

Compatible with guided search & Desimone/Duncan biased competition, all items in WM compete for status of “search template” – of which there is only 1.

When search is repetitive it becomes “automated” – search template is “offloaded to other systems” (Rossi Bichot Desimone Ungerleider JNeurosci 2007) → other WM items interfere more. Olivers 2009 JEPHPP If target varies from trial to trial, distraction is less. Under these conditions, distraction does not increase with WM load.

Houtkamp & Roelfsma 2009 Psych Res: RSVP disjunction search: ‘were any red or green items present?’ vs ‘was a red item present?’. Accuracy analysed with signal detection → number of templates = 1.

An item’s cellular representation becomes completely different (‘orthogonal’) when it is the second of two items in memory (2 x Warden & Miller papers). When it is relevant again, it goes back to state 1. ?? only one of these can guide attention

Active visual WM representations guide attention ; inactive WM items do not, and have different neuronal encodings.



Rypma...Gabrieli Neuroimage 9:216-26 1999

Load-dependent roles of frontal brain regions in the maintenance of working memory

Sternberg task, 1, 3 or 6 uppercase letters simultaneously presented for 1.5s, 5 sec blank retention interval, then single probe lowercase letter; single button response if the item was in the set.

(Load 3 – Load 1): caudal Lt IFG

(Load 6 – Load 1): bilateral DLPFC esp MFG + SFG, caudate

Interpretation: left IFG = “phonological slave process in WM” = broca (BA44) = a “limited-capacity domain-specific buffer”

Bilat DLPFC: usually involved in manipulation in WM; but here just correlates with capactity. “Strategic processes may need to be employed to maintain” >3 items in WM, i.e. manipulation using cognitive operations similar to those in more complex tasks e.g. Ravens.

Baddeley & Hitch 1974: Prose comprehension is impaired for WM >3.

Tasks of executive: monitoring WM content (self-ordered tasks), updating it (n-back), coordinating slave processes (dual-task), allocation of attention among multiple stimulus domains (divided attention), planning (reasoning task).

Could also be strategy-shifting: several strategies for WM of letters; temporal, spatial, associative etc.

Organisation of PFC theories:

6.      Goldman-Rakic 1995 domain-specific VL=nonspatial DL=spatial

7.      Jonides 1993: spatial Rt, nonspatial Lt

8.      Owen 1996, Petrides 1996: Process-specific: VL=maintenance, DL=manipulation

Findings suggest domain specificity if low demand, process-specificity if high-demand.


Desposito, Postle, Jonides, ESmith PNAS 96:7514-9 1999

The neural substrate and temporal dynamics of interference effects in working memory as revealed by event-related functional MRI


Godijn & Theeuwes Psych Res 66:234-6 2002

Oculomotor capture and inhibition of return: evidence for an oculomotor suppression account of IOR

Colcombe 2000: Latencies of correct saccades are slowed by 20ms when a distractor onset is present. Can be explained by:

Folk & Remington 1998: filtering cost. But this is assumed to be nonspatial – there is no actual shift of attention to the distractor when a correct saccade is made. → does not explain capture.

Rather, initiation of saccade programming – competes with target.

Posner & Cohen 1984: IOR= uninformative peripheral onset ‘to be ignored’, SOA>200ms, slowing of saccades to that location.

Theory of Klein & Taylor 1994, 1998: saccade programming without a saccade can give IOR

Expt: distractor onset 50ms before target. After 800ms, second target at either onset or non-onset location.

Result: 36% capture. Stopping near distractor ~100ms. S2: 10ms slowing if T2 = onset.

Saccades curved away from onset.

Hypothesis: exogenous → LIP + extrastriate + superficial SC → intermediate SC. Endogenous → LIP + SEF + DLPFC + FEF → intermediate SC

Model: target and distractor are mutually inhibitory (lat inh). PFC suppresses the distractor & fixation pos selectively.


Tversky Erkenntnis 9:2:163-73 1975

Critique of expected utility

1. Negatively accelerated utility curve cannot explain:

A=(1000,1/2,0) vs B=(400)

C=(1000,1/10,0) vs D=(400,1/5,0)

Most subjects prefer B>A, and C>D. but this entails U(400)>1/2*U(1000), and 1/10*U(1000)>1/5*U(400), so contradiction.

2. Also consider framing:

Game round 1: P(ending game)=4/5, otherwise continue

Game round 2: choose E=(1000,1/2,0) vs F(400)

Most subjects choose F>E, although in terms of actual probabilities, the total gambles are E=C, F=D. So the effect that causes C>D is not calculated when multiplying the probabilities out separately.

= risk aversion + “positive certainty effect”– preference for positive outcomes that are certain.

3. Mirror image effect:

A'=(–1000,1/2,0) vs B'=(–400)

C'=(–1000,1/10,0) vs D'=(–400,1/5,0)

Most subjects prefer A'>B', but D'>C'.

= risk-seeking + “negative certainty effect” – aversion for sure losses.

4. Allais

A=(1million) vs B=(1million x 0.89 or 5million x 0.10 or 0 x 0.01)

C=(1million x 0.11 or 0x0.89) vs D=(5million x 0.10 or 0 x 0.90)

Most prefer A>B, but D>C. Violates independence assumption: “changes in outcome that are common to both options should not change choice”. Explainable by certainty effects.

Some people change their choice when EU is pointed out; others don’t.

Allais: not only is it natural, but also perfectly rational

Raiffa 1968 & Savage 1954: utility theory has normative force. To choose A>B & D>C is mutually inconsistent.

Tversky: B is dragged down by the ‘missing the prospect of getting 1million for sure’. Not just expected money – a broader interpretation of the consequences is required (including psychological factors)

In the absence of constraints on how to interpret outcome, EU theory can always be interpreted as to satisfy the axioms and is therefore “empty from both descriptive and normative standpoints”.

= decision theory can say how to act in light of your values, but not what values to have.

A comprehensive theory is likely to be “explicative, or even therapeutic, rather than normative in nature”.


Frank, Loughry & O’Reilly Cog Aff Beh Neurosci 1:2:137-60 2001

Interactions between frontal cortex and basal ganglia in working memory: a computational model

Brown & Marsden 1990 – BG & frontal lesions produce similar impairments

“WM andcognitive control can be seen as two different manifestations of” actively maintained information. 2 psychological constructs, subserved by a common mechanism.

They say, to trade off robustness with rapid updating, selective updating is characteristic of WM → gating.

O’Reilly 2000: Dopamine strengthens other inputs to PFC. ↑ for expected reward. But: nonselective.

Chevalier & Deniau 1990: striatal activation enables, but does not directly cause, motor effects. Disinhibits / gates the thalamus.

Alexander 1986: there are 5 basal ganglia-thalamocortical loops [?*]

  • Not enough thalamic neurons to support all representation by thalamocortial oscillation. Also, any gating at thalamus would destroy the memory. So the reverberation is intracortical.
  • Simple recurrence: cortex involved in WM cannot (even transiently) be driven by any sensory input, without disrupting WM.
  • But Miller 1996 sees transients during WM tasks: cells recover their memory-based firing after representing transient stimuli.

Lewis & O’Donnell 2000: PFC neurones have bistable resting potential - up/down state – might solve this problem. Robustness to transients + ability to ‘open the gate from the inside’ – gating is controlled by the PFC neurones themselves.

PFC neurones should themselves learn what to gate → trial-and-error gating

  • Striatum 111million, GPi 160,000, SNR 160,000 → information transfer from BG to Cx is not ‘content’ but ‘when to update’ → lesions give deficits of initiation but not execution.
  • Striatum encodes all conjunctions of possibly relevant cortical states – including task context, attentional set, as well as stimulus history.
  • Frontal cx is also conjunctive, but for maintaining appropriately contextualised information.

“Gating regions” ~= cortical stripes 0.2 x 0.4 x 2-4mm, each can be separately updated by the BG. Frontal cx = all cx; ~= 20,000 stripes.

Striatal neurones “general silence” = high threshold. Regulated by tonic DA. Phasic DA in shaping the timing.

WM and motor control: continuum; future action plans. Updating WM = initiating action.

3 stripes per layer: Lt=stimulus that triggers an action; Mid=inner loop – active WM item; Rt = active task set.

“PFC_Maint” = cortical layers 2-3 and 5-6. “PFC_Gate” = layer 4.

Stim → PFCMaint transient → (if important) striatum as conjunction of stimulus, current WM content, other context → inhibits GPi → activates thalamus → enables corresponding strip of PFCGate to be activated by PFCMaint representation – which selects the specific gate unit. → controls switchable ion channels in corresponding strip of PFCMaint: “store your current activation state”.

Old models:

Berns & Sejnowski 1996: BG neurones inhibit each others’ output at the STN, to give ‘selection’ of an action.

Amos 2000: WCST; striatal units = match detectors between stimulus and ‘target card’ → thalamus disinhibition = response. Frontal attentional signals modulate BG.

Beiser & Houk 1998: BG disinhibits recurrent corticothalamic WM loops → retention of sequences.


Hodgson, Mort...Kennard Neuropsychologia 40:1891 2002

Orbitofrontal cortex mediates inhibition of return

1 patient bilat surgical OFC. Neuropsychology:

decreased Corsi.

Cantab: foraging SWM task for within-trial and between-trial errors: both impaired.

Tower of London = Normal. IGT impaired.

Intradimensional/extradimensional shift in sorting cards = Normal.

2AFC recognition of which abstract patterns was previously shown = Normal.

But recognition memory for which of 5 squares is presented at a previously occupied position (swm) = impaired.

IOR Experiment: fix 800ms → central colour cue → saccade L or R → happy/sad face + tone at chosen location. “rule will reverse at several points during test”. 200 trials, 14 rule reversals.

Patient: reduced post-error slowing, reduced IOR after error – latency slowing for saccades that need to go to the same location that was previously penalised.

In a visual search task, OFC patients make many more refixations.

Conclude “deficit in reward-dependent IOR”. [but could be due to tendency to emit an error correction (Rabbitt), or difficulty updating one half of the rule]

Link & Heath Psychometrika 40:1:77 1975

A sequential theory of psychological discrimination

Antecedents to random walk: Wald 1947 sequential probability ratio test for statistical analysis: identity for constrained random sums → Stone 1960 ‘random walk model’,

Kullback 1959 defined ‘information statistic’ → Laming 1968:

Both approaches use increment=LogLR, and both predict P(correctRT=t|correct) = P(errorRT|error).

Here: generalisation: subjective referent => starting point is somewhere between the two boundaries.

Laming 68: starting variability → fast errors.

Assume no overshoot of bounds = mean and variance of step << bounds

MGF E(exp(-qd)) walks along the ‘psychological difference continuum’. Wald’s identity: . If the MGF is symmetrical about q0, then error and correct RT distributions are equal. Wald shows that where the roots r1(z) and r2(z) are the values of q where MGF=z. With a bound z=1, , and RT=EA = , where the diffusion rate is found by and the asymmetry is . The overall RT distribution is



Seo & Lee J Neurosci 27:31:8366 2007

Temporal filtering of reward signals in the dosal anterior cingulate cortex during a mixed strategy game

Monkeys played matching pennies with saccades. (2pzs game).

Computer: exploited statistical biases in monkey’s choices.

  • Estimated P(choose left | choice sequence in previous n trials) n=0,1,2,3,4
  • Estimated P(choose left | choices and prewards in previous n trials) n=1,2,3,4

         Of these 9 estimates, computer chooses the one with the greatest deviation from 0.5 that was statistically significant (binomial p<0.05), and selects left target with that probability.

“to maximise total reward, therefore, the animal needed to choose both targets equally often and make its choice independently from previous choices and their outcomes”.

Dorsal bank of cingulate sulcus 24c, ventral to SEF.

Behaviour modelled with TD learning: value of each target varies from trial to trial at rate a, depending on reward prediction error , and choice is noisy based on softmax of value difference ; b is ‘inverse temperature’.

Monkeys used win-stay-lose-switch. (using BIC).

GLM of firing rate in each window, regressors:

1)      previous monkey choices, previous computer choices, previous outcomes, for up to 3 trials back = 13 regressors. → k-means cluster analysis of regression coefficients.

2)      Interaction terms of reward on trial n, n-1, n-2

3)      Current trial net value, difference in value, and reward prediction error

3-way interaction shown between previous rewards: “when a neuron displayed a significant interaction between the reward in the current trial and that in the previous trial, the signals related to the reward in the previous trial were conveyed more reliably by the neurons that increased their activity according to the outcome of the current trial. This was true regardless of whether the reward in the previous trial influenced the activity of a neuron in the same direction as the reward in the current trial or not.”

→ majority encoded reward

→ modulated according to reward history; most cases = linear function of successive rewards

→ some signals related to previous choices – but fewer than in DLPFC

→ few neurones encoded relative value of two choices → ?not involved in choosing

In some tasks, ACC encodes conjuction of choice and expected or actual consequence. Not in this task, ?because previous reward only weakly influences next choice.

Reward-related activity is sustained across trials, unlike phasic DA. → ? dACC calculates the RPE seen in DA neurones?

Sensitivity to overall reward rate → “hedonic reference point”


Bryden, Johnson...Roesch J Neurosci 31:50:18266-74 2011

Attention for learning signals in anterior cingulate cortex

Pearce & Hall: unsigned prediction error needed for alerting to changes = attention

Basolateral amygdala signals these, but only at the time of unexpected reward, not after → can’t be used for attending/learning

2 fluid wells. Start of trial: light goes on, one of 3 odors (cue) presented. After smelling port, rat has 3sec to respond in 1 of the 2 wells. 3 Cues in pseudorandom order: reward Lt, reward Rt, or both (free choice).

Then: 4 blocks. New forced choice cues introduced. manipulate delay & reward magnitude:

Rats slower and less accurate for the long-delay and small-magnitude well.

ACC sensitive to expected reward magnitude but not delay, despite the preference. (cf Rushworth 2007 and Rudebeck 2006)

Compare early and late trials in each block → sensitivity to breach of expectaction

Conclude: “Activity was high during trials after violations of reward expectations, regardless of valence. Activity in ACC also detected errors of commission and reward prediction, and signaled when the expected reward was to be large.”


Bartlett,... Logothetis, Hoffman J Neurosci 31:50:18423-32 2011

Saccades during object viewing modulate oscillatory phase in the superior temporal sulcus

In V1, saccades produce resetting of oscillations to a phase associated with high firing rates.

IT and upper-bank of STS are thought to process independently of fixation-location

Monkeys fixate 500ms, then view face or nonface 500ms, then juice. Monkey allowed to make small saccades to edge of fixation window. Fixations, LFP and single units recorded from upper bank STS.

Small spike at 0ms shows fixation-related spiking. Calculated ‘phase concentration’ at each time point in each frequency band → LFP is locked to fixation.

First image-evoked response = 50-100ms after image.

→ Separated out effect of image onset from fixation time with oblique plot. “band-limited residual phase concentration is greatest when fixation and image onset coincide... suggesting a supra-additive effect”. – “as occurs during active perception”.


  • “consistent with recent reports on the opposing relationships of alpha band activity to sensory responsiveness in early versus late cortical areas (Bolimunta et al 2008/2011”
  • consistent with “visual processing is augmented following saccades”
  • “the experimental convention of delaying stimulus onsets by hundreds of milliseconds relative the end of saccades...micharacterises the function of the visual system during active vision”.

Spruston NRN 9:206-21 2008

Pyramidal neurons: dendritic structure and synaptic integration

Location: neocx layer II/III, layer V; CA3, CA1, subiculum

Domains’ have different inputs:

  • Basal (proximal) dendrites: from layer IV cells & local circuits.
  • Tufts (distal) dendrites: from thalamic and other cortical areas

Basal dendrites – greater & denser in PFC, greater in humans → “do neurones with more dendrites fire more? If not how is firing maintained at the same rate” [why would the dendrites all be excitatory? They may not be active at the same time? They may not literally summate?]

“Pyramidal neurones might be designed to respond to coincident input to the tuft and the more proximal dendritic domains. Alternatively input at the tuft might control responsiveness to more proximal inputs.” [why not the other way round?] = coincidence detection vs. gating

HC: (“functional significance of this remains mysterious”)

  • Entorhinal cx → perforant path → CA1 distal tuft
  • Nearby CA3 cells → Schaffer collaterals → CA1 proximal dendrites
  • Far away CA3 cells → CA1 distal tufts
  • Dentate gyrus → mossy fibres → CA3 pyramidal cells


30000 excitatory/cell. Every spine is contacted by at least one synapse. 10-30% ‘perforated’; ?majority of other synapses are small/weak or silent. “Silent”: may affect local integration but not detectable in the soma. “several tens of synapses must be activated to produce a single AP”

Synapses distant from soma & on narrow dendrites & many branch points:

  • ‘synaptic scaling’ = increase conductance → normalizes influence on soma. Found in CA1 neurones, but not layer V.
  • Dendritic voltage-gated channels / dendritic spikes
  • Hyperpolarisation-activated cation channels → small depolarisation after EPSP → limits distance effect

Variation between synaptic spines: in receptor concentration, proximalness, plasticity.

GABAergic Inhibitory Interneurones:

Basket cells → perisomatic region = global inhibition

HC OLM interneurons / neocortical Martinotti cells → apical tuft = selective

feedforward: interneurone activated by same inputs as cell itself → shortens temporal window of summation

feedback: activation of interneurone by the pyramidal cell itself → limits sustained firing.

This can be hi-pass = “onset transient” (target soma) or lo-pass “late-persistent” (target distal)

CA1. Red=interneuron axon, Blue=interneuron dendrite. Layers = stratum lacunosum-moleculare, radiatum, pyramidale, oriens. [? Gives resistance to rapid changes in local environment & sensitivity to only rapid changes in distal inputs]

Gabaergic chandelier cells → pyramidal axons → excitatory due to Cl- reversal potential → ??timing of rhythmic firing

Functional diversity

  • Somatic current→ regular spiking with frequency-adaptation OR intrinsic burst firing (due to afterdepolarisation, influenced by prior activity & modulatory neurotransmitters)
  • Dendritic VGK+ channels – uniform or somatodendritic gradient → different EPSP duration
  • Backpropagation of AP depends on VGNa+ channels which can inactivate → in CA1 & layer V there is activity-dependent reduction of backprop, but not in layer II/III.
  • In layer V, backpropagation-activated Ca2+ channels can allow previous APs to sum with tuft inputs when basal dendrites are active → bursting.
  • Because backpropagation does not reach apical dendrites, pairing firing with synaptic activity does not induce LTP there in CA1 cells. In layer V, dendrites can be depolarised to amplify this backprop.

LTP only occurs with high freq pairings, or with post-synaptic bursts.

Some forms of LTD do not require AP firing.


DA, 5HT, NA, Ach → synaptic strengths, base firing rate, firing mode, dendritic excitability (integration gains and potentiation), up/down states, intrinsic oscillations.

mACHR suppresses CA3→CA1 transmission at proximal apical dendrites only. → “different types of presynaptic information might be weighted differently during different behavioural states”

mACHR in entorhinal layer V → non-specific Ca2+ current → persistent firing

distal basal dendrites require BDNF for potentiation; proximal dendrites don’t.

Final puzzle: remapping of CA1 cells during exploring novel maze: is the change due to 1) plasticity of excitatory or inhibitory synapses, 2) local dendritic excitatability changes, 3) neuromodulatory input.

Kuhl, Williams...Lindblom Science 255:606 1992

Linguistic experience alters phonetic perception in infants by 6 months of age

Prototypes function like perceptual magnets. Linguistic experience shrinks the perceptual distance around a prototype.

Trained to head-turn when a repeated vowel changes, rewarded by activation of toy bear.

6-month infants have magnet effect only for native-language prototypes.

Language-specific perception does not depend on emergence of contrastive phonology or understanding of word-meaning. Before uttering meaningful words.

May assist in organising speech sounds. In place before word meaning is acquired.

Phonetic prototypes are therefore fundamental building blocks.

Petitto in Cambridge companion to Chomsky, McGilvray

How the brain begets language

  1. Monkeys have no language. Taught “Nim Chimpsky” sign language.
    1. No syntax: Stringing words together: syntax matrix of 2. After that, ‘word salad’ of most frequent 5 words.
    2. No morphology: cannot modify any words by morphemes.
    3. No sign phonology: Variable production of forms – unpatterned and random errors
    4. No abstraction: Cannot acquire ‘fruit’ as a generic
    5. No denotation – i.e. No scope for kind-concepts. → Cannot use ‘apple’ to refer to an object. Errors: global association, e.g.: using ‘apple’ for the location, action, tool etc. that was associated with the apple.

2-yr old does all this perfectly, rapidly, & without being taught.

  1. Deaf babies hand-babble. Also hearing babies exposed only to sign. = 2.5–3 Hz hand open/close, normal baby movements are 1Hz.
  2. Monolingual, Bilingual babies, ASL, LSQ, and bi-signlingual babies all have same babbling/language milestones. 6/12 distributional patterns influence perception.
  3. Deaf signers activate planum temporale when viewing ASL phonetic-syllabic units, and LIFC when viewing words. → two hierarchical levels of Chomsky’s language acquisition device.
    “due to some yet unknown factor, PT tissue in humans has a sensitivity to certain specific patterns found only in natural languages” == “specialisation for highly specific, maximally contrasting rhythmical patterns”, but not music.

“Adaptive phonological differentiation” guides attention to find salient aspects of input: rhythmic, prosodic, [ordered] elementary units, bundles of ~1.2 – 1.5 seconds.

Capa, Bustin, Cleermans, Hansenne Exp Psychol 58:5:370-5 2011

Conscious and unconscious reward cues can affect a critical component of executive control: (Un)conscious updating?

“it is thought that consciousness is required for executive control.”

“unconscious stimuli...are assumed not to be able to be the source of executive control”.

Bijleveld, Custers & Aarts 2009: supra/subliminal reward cues → pupil dilatation for difficulty was greater with high reward cue. no effect for easy trials.

Pessoa 2009: conscious reward cues improve updating.

Salthouse Babcock and Shaw 1991: updating task

Expt: N=28; 60 trials. Premask (100ms), reward stimulus (70ms/300ms), post-mask (100ms), then 5 digits in a row (7s), then 6 add/subtract ops -2 to +2 each at one location (each 3.5s). Response for all 5 numbers on keypad; All correct → “win/lose” and cumulative earnings.

Then: same 60 cues shown, perceptual discrimination 4AFC “seen 1euro”, “seen 5p”, “guess 1euro”, “guess 5p”, unlimited time.

  • Updating was better with subliminal cues! (why this is “is open to argument”)
  • Main effect of reward, no interaction with cue visibility.
  • Second analysis using %correct; Updating was improved by high-reward in the subliminal but not the subliminal condition.
  • RT slightly faster for high-reward (3.5 vs 3.67 s), but only with subliminal cue
  • SAT? covariate 2-way ANOVA within subjects: no association of RT & accuracy










[Very strange numbers in this table] [calculated visibility for both “seen” and “guess” together] d’=0.09

?conscious processing of rewards interferes with maintenance. [choking?]

Difficult task (accuracy ~35%): unconscious prime → uncertainty about reward → “attraction” to that condition → better performance [??]

Baines, Ruz...Nobre Neuropsychologia 49:2489-97 2011

Modulation of neural activity by motivational and spatial biases

ERP with independent reward and Posner cueing.

  • Much research on “role of motivational factors upon motor processing and decision making (Rushworth & Behrens 2008)” but little investigation on influence “upon perceptual and cognitive analysis”.
  • Separating reward from attention: Engelman & Pessoa 2009, 2007, Padmala & Pessoa 2010, Pessoa 2009.
  • Does reward “act through the attentional system (ACC/PPC)” or “direct perceptual modulation by reward and motivation” (amygdala/OFC→extrastriate).

Perceptual effects = P1 and N1; cognitive effects = P300; motor selection = LRP.

N=14, pay 32

Circles as placeholders always present at L/R target locations. Symbolic compound cue (red/green) (+/) → (Rewarded/unrewarded) (L/R) counterbalanced across subjects. Cue validity 25% (100ms), SOA (800-1200ms), grating 12 degrees off-vertical (33ms), then L/R index finger response for cw/ccw. Compound symbolic feedback: (triangle/square) (filled/hollow) → (incorrect/correct) (too slow?) counterbalanced. ITI 1.8-2.8s.

Accuracy 94% valid, 92% invalid (different, p=0.05). No effect of reward stake p=0.14, no interaction. RT independently speeded by reward and validity.

[?simon effect: cue direction/target side/response side ]


  • RT, rather than accuracy, was more critical in optimising reward, so that is where motivation had its effect. Not ceiling effect, because cue validity did affect accuracy.
  • Reward speeds N1 latency → not arousal, as arousal typically affects later stages.
  • Valid cue increases amplitude of N1, and speeds and amplifies P1. could be ceiling.
  • Reward did not affect P1 – cf Hickey: P1 enhanced by motivation by stimulus colour. In current study, reward was not stimulus-feature-associated; rather it engenders “greater degree of preparatory processes” – separate from attentional processes.
  • Independent effect of motivation and attentional expectation on P1, N1 and P31.
  • Motivation P31 = contextual updating =? Encoding into WM, integration with history/goals, updating stimulus-outcome representations. Could also be arousal. But: “It may be the case that only motivation...can change arousal” (Sabatinelli et al 2007)
  • But attention validity ↓ P31, = unexpectedness/novelty → independent effects
  • Interaction in late P32 and late portion of LRP: Valid rewarded is ↑↑. If reward expected, → the effect of validity is to ↑ P32, even though it ↓P31.
  • Interaction in LRP: Reward boosts motor activity when target occurs at the expected location.
  • Increased LRP for invalid targets, on non-reward trials → more rapid resetting of the attention system following low reward.

“late integration” of reward and attention information. → “coherent stimulus-outcome representation” [‘independent’ main effect does not mean they are ‘independent’ processes – could be same process!]

Padmala & Pessoa Neuropsychologia 40:558-65 2010

Interactions between cognition and motivation during response inhibition

fMRI stop signal task with rewarded go-trials → longer stopping-time: impaired by reward.

Expt: N=35, 8 blocks of 150 trials, alternating rewarded and unrewarded blocks.

1 of 2 shapes shown 1s → 2AFC identification index/middle finger of Rt hand. “respond as soon as possible; sometimes it may not be possible to stop” <1s. 16% Stop signal = tone 300ms; 1up-1down-50ms staircase.

“In rewarded blocks, the computer will choose 5 go trials, and each correct response will add $1. There is no reward associated with stop trials. The amount awarded will be displayed after each reward block”.




Go RT (ms)

487.1 18.6

484.0 19.12

Inhibition rate (%)

50.5 0.6

9.1 0.8

SSD (ms)

294.5 25.8

270.7 26.1*

SSRT (ms)

192.6 11.0

213.2 10.9*

unsucc RT (ms)

461.0 17.4

471.7 17.7

Go error rate (%)

2.3 0.3

1.5 0.2*

Logan & Cowan 1984: SSRT = Mean correct go RT – average SSD

Go time unchanged by reward, but SSRT ↑ by 20ms. [but how can we explain ↑error RT with reward, but no change in correct RT & ↓ error rate?]

(Successful stop – failure to inhibit) = large parietal, SFG, MFG, M1, IFG, caudate, putamen

(failure – success) = L ant insula, ACC, PCC

(successful – failure) x (reward – noreward): reduced effect of (success-failure) in rIFG & M1

[you’re bound to go more if the go is rewarded and stop is not. ]

Conclude: these areas may integrate motivation and inhibition.

Rewarding go trials is similar to ADHD.


Warden & Aylesworth J Compar Psychol 7:2:117-27 1927

Punishment has a higher incentive value than reward.

Rats learned to discriminate a bright and dim light, to choose one of two rooms. 3 groups: food for correct, or shock for incorrect, or both. Rats given 5min to react; failure to react is classed as incorrect.

Result: rats always responded (never failed to make a response) in the reward condition. However they took longer to learn than the punishment-only group, where they often did not approach either room.

Punishment → aversion to any response

[But: means the reward wasn’t good enough or the shock was too strong?]

Kahneman & Tversky Econometrica 47:2:263-92 1979

Prospect theory: an analysis of decision under risk

A prospect P={xi,pi} is a contract that yields outcome xi with probability pi, where Σp=1.

Critique of expected utility, which assumes

  1. Expectation U(P) = Σpiu(xi),
  2. Asset Integration so winnings become part of your assets: prospect P acceptable iff U({w+xipi})>u(w)
  3. Risk aversion: d2u/dx2 < 0

Considers several problems

  1. Allais paradox – breaches substitution (independence) axiom; can be breached by non-monetary gambles, so people can switch preferences in this: {50% A, 100% B} vs {5% A, 10% B}. Manner of violation: “If (y, pq) is equivalent to (x,p), then (y,pqr) is preferred to (x,pr).”
  2. Reflection Effect – If the preference for certainty (smaller variance) were used to explain the Allais effect, it should not be reversed for penalties, i.e. subjects should prefer a certain smaller loss. However they do not. Certainty increases the aversiveness of loss.
  3. Aversion to Probabilistic Insurance – normal insurance is just worth its cost; probabilistic insurance is 50% price, but if the accident happens on an even numbered day, you get your money back and no insurance, on an odd day you pay the whole premium and the insurance covers the losses. Only 20% of people take the insurance. “a burglar probabilistic insurance”. → goes against concave utility theory, because if y is the price of the EU0 insurance, and r is the probability of probab-insurance, then “one reduces the probability of losing x from p to (1-r)p” which is worth at least > ry.
  4. Isolation effect – different decompositions of a gamble lead to different preferences. Violates supposition that “choices between prospects are determined solely by the probabilities of final states”.


Prospect theory = 2-stages, 1) Editing, 2) Evaluation.

Coding – gains and losses defined relative to some reference point

Combination – of any probabilities associated with identical outcomes

Segregation – decomposing of common riskless components of a prospect

Cancellation – discarding components that are shared by the offered prospects

Simplification – rounding probabilities up or down, discarding extremely unlikely outcomes

Dominance – prospects are scanned for fully dominated alternatives, which are rejected

Then: for regular prospects, i.e. (p+q)<1 or x>0>y or x<0<y, then V=.

For strictly +ve or –ve prospects i.e. p+q=1 and (x>y>0 or x<y<0), V=

Where π gives overweighting of rare events, and π(p)+π(1–p)<1. Distinguish overweighting from overestimation of rare probabilities.

e.g. People would pay more to reduce the number of bullets in Russian roulette from 1 to 0, than to reduce from 4 to 3 – despite the utility of money is presumably less in the 3/4 case.


Wunderlich, Rangel, O’Doherty PNAS 107:34:15005-10 2010

Economic choices can be made using only stimulus values

  • Comparing action-values: value signals in caudate, SMA, action-related-value signals in LIP. Supported by Glimcher/Dorris, Shadlen/Newsome accounts. Cf Thorndike law of effect. As a general law of behaviour – i.e. reinforcement learning.
  • Comparing goods: abstract goods-based-value in OFC; lesions to OFC→impaired stimulus-reward learning, lesions to ACC→ impaired action-reward associations (Rudebeck 2008).

fMRI 2-armed bandit for money; 2 of 3 possible shapes shown on each trial. Each shape’s P(reward) drifted over 50 trials. Strange response mapping: prefer shape above→saccade to right; prefer shape below→rt hand button press.

50% trials: shapes shown initially side-by-side 3-5sec before moving to the response positions.

Model: reinforcement learning, 2 parameters – learn rate and softmax noise.

  • RTs faster when stimulus choice was visible before action choice (800ms vs 1100ms).
  • In SC trials, correlation with value of chosen stimulus = vmPFC, but only before action was revealed.
  • GLM with orthogonalised chosen and nonchosen stimulus values. → vmPFC
  • Effect of average of two shown stimulus values → vmPFC
  • Effect of value of the action made (eye or hand movement) = none

Conclude: → “the brain is capable of computing a decision purely in goods space when action pairings are not available”.

In the stimulus-choice condition, stimulus-value signals are only transiently represented. Region of average-value-of-stimuli is adjacent to region of value-of-currently-chosen stimulus. Don’t have enough temporal resolution to see what order.


Glimcher, Dorris, Bayer Games Econ Behav 52:213-56 2005

Physiological utility theory and the neuroeconomics of choice

  • Bounded rationality: Decisions based on EU occur only under some conditions. These models “have little or no predictive power outside of their bounded domains”.
  • Modified utility theories: “fail to be parsimonious and often appear ad hoc or underconstrained.”
  • Neuroeconomics: to “reconcile prescriptive and descriptive economics by producing a highly predictive and parsimonious model”.

Dorris & Glimcher 2004: “inspection game”: (kreps 1990). 2x2 payoff matrix. Employee chooses to work at cost C, or shirk. The inspector chooses to inspect at a cost I, and he pays the employee wage W, which is less than the value of the work to the employer, V. At the Nash equilibrium, EU(shirking) = EU(working), and EU(inspecting) = EU(not inspecting). Therefore, P(Inspect)=C/W, and P(Shirk)=I/W.

Expt: varied I to get different P(shirk).

Human vs human with no instructions except “make as much money as poss”, unlabelled buttons. Result: overshirking with lowest I.

Human employee vs computer employer: algorithm tracked estimated P(shirk) and P(repeat last response) with λ=0.1 → EU(inspecting). Added an exploration bonus → computer choice softmax. Result: same as human-human.

Monkey employee: fixation, then two targets, red shirk and green work, then fixation stop blinked, then respond within 750ms. Result: same as humans.

  • LIP neurones encode expected value in simple tasks e.g. reward-probability cue
  • but no variation during game “when the animal is engaged in a strategic conflict the firing rate associated with this same movement is fairly constant at an intermediate level.” LIP encodes the desirability / physiological EU of movements.
  • Doubling the absolute rewards gave no change in firing rate → encode relative EU.
  • Humans and monkeys adopt almost perfectly stochastic strategies. Tolhurst 1981: all variance in cortical firing rate = 1.07 mean firing rate. ??poisson noise? Mainen & Sejnowski 1995: AP is deterministically related to membrane potential; but Stevens 1994: synaptic transmission appears to be irreducibly stochastic. Shadlen 1996: uncorrelated noise means the mean rate of firing can be accurately extracted; if neurones are highly correlated, then the noise does not cancel out.
  • Source of EV signal: dopaminergic neurones fire ~3Hz at rest; reward prediction error. Schultz 1997: SNpc receives expected reward and actual reward signal, & subtracts. Bayer & Glimcher: DA firing = current reward – Σ(previous rewardt-i e-i)
  • Herrnstein’s (1961) two-alternative lottery: once a target was armed, it remained armed until it was chosen. Staddon 1980: matching the probability of responding to the probability of arming is nearly an optimal solution.

“one might be able to propose a single identified mechanism for both [stochastic and rational] behaviour”.

Rejects dualism of “rational and irrational decision making”. “First, there is no neurobiological evidence that emotional and non-emotional systems are fully distinct in the architecture of the primate brain” [weak statement, and primates don’t have concepts] “Second, there is no evidence that rational and irrational behavior are the product of two distinct brain systems, one of which is uniquely rational and one of which is uniquely irrational.” [but we have pretty good reasons to believe there is a strong distinction in computational terms – no concepts mean different decisions.- do primates overmatch?]

1898 Veblen “why is economics not an evolutionary science?” → Ultimately economics is a biological science.

Zhang & Barash Nature 408:971 2000

Neuronal switching of sensorimotor transformations for antisaccades

Mechanism for vector inversion: could be

  • Receptive field remapping: a critical set of cells with two alternate receptive fields – pro and anti.
  • Visuomotor switching: switching of the connections between visual and motor cells.

Prosaccade or antisaccade signalled by colour of the stimulus. Stimulus positioned so that either it or its antisaccade direction falls in the RF: → prosaccades are either V+M+ or V-M-, antisaccades are either V+M- or V-M+.

Calculated ‘differential activity’, mapping the activity onto the range between visual and motor

Visual neuron, Motor neurone, and a cell with ‘paradoxical’ activity

Paradoxical cell has a visual-like (short-lived) time course, but becomes motor. Visual cells have latency 60ms, paradoxical activity latency ~110ms.

Conclude: Shows the existence of visual switching. May be remapped visual response, or feedback from higher areas.

Could be related to perisaccadic remapping seen in LIP (Duhamel 1992), and acquisition of auditory response fields in LIP.


Stuphorn, Taylor, Schall Nature 404:857 2000

Perfornance monitoring by the supplementary eye field

Saccadic stop signal task (countermanding) in monkey. Reward = juice+sound, or just sound.

Modulation in cancelled trials = difference in firing between cancelled and latency-matched go trials. 3 groups of cells:

  • 15% specifically modulated by failure to cancel (different even when matched for visual stimulation, timing of responses etc). May correlate with medial frontal error-related ERP potentials
  • 13% specifically modulated by successful cancelling planned movements. (occurs after the stop-signal RT, so cannot contribute to cancelling the movement). Magnitude does not reflect SSD, but correlates with inter-session variability in stopping performance (i.e. an internal variable) → measure of conflict
  • 22% active before and during the delivery of reinforcement – independent of whether it was a stop or go trial. Occurred for both primary and secondary reinforcer. = “functional complement of the putative error-related neurons”: expectation and receipt of reinforcement.

Stimulation of SEF inhibits activity in FEF (Sadeghpur et al., 1998) → Suppression of reflexive behaviour

“Neurons in the SEF may signal the production of an error, the anticipation of reinforcement, or the presence of processing conflict”... → SEF as a “node in the brain’s supervisory control system”


Brown Psych Bull 129:3:394-413 2003

A review of the dj→ vu experience [Theory]

Def: Titchener 1928: “paramnesia or wrong recognition: a definite feeling that all this has happened before, inspite of the fact and knowledge that the experience is novel”.

Woodworth 1940: “a weird feeling that one has been through all this before, as if time had slipped a cog and were now repeating itself. The illusion of having been there before”

  • Incidence 30 to 100%. M=F, higher socioeconomic class & education, travel. Evening>morning; in company of others, fatigue, or following unpleasant/confusing mental activity, physical exertion, hangovers.
  • Often visual trigger; lasts few seconds, primary psychological reaction is surprise, time sense slowing down, mild stress, but no changes in thinking or emotion.

Pathology: HSE; HI w LOC, Drugs: Amantadine+phenylpropanolamine, tolune inhalation, benzo withdrawal

  • Schiz: may have long dj→ vu (hours), assoc with depersonalization, but qualitatively different from nonclinical dj→ vu → different entity
  • TLE: dj→ vu aura present in 1% to 80%. minutes rather than seconds. Mullan & Penfield 1959, surface stim of cx → dj→ vu in 5% of TLE. Nonrepeatable – occurs at different cortical locations in each session, and also work in the nondiseased hemisphere. Memory often clouded by subsequent sz.

Psychodynamic vs scientific explanations

Correlation with dream memory. Zuger 1966: dj→ vu could be “a dream state intruding into waking consciousness”

  • Dual process theories: 2 cognitive processes normally operating in synchrony become uncoordinated. E.g. familiarity and retrieval (also accounts for jamais vu), encoding and retrieval (simultaneous recording and playback), perception and memory, Hughlings Jackson’s 2 forms of consciousness – extero and interoceptive become simultaneously active.
  • Neurological: epileptic hippocampal outflow generates familiary, Geschwind ‘general right hemisphere dysfunction’, delay in perceptual transmission to high-order areas (but note perceptual fluency: faster processing is usually interpreted as experiencing earlier), Weiskrantz 1986: 2-pathways = disrupted integration gives 2 sequential experiences, delay to dominant hemisphere.
  • Memory: some familiar aspect but source of familiarity is forgotten. Conflict in source monitoring processes, the cognitive processing (but not content) was experienced before, single-element familiarity without recollection gives familiarity of the whole scene (=restricted paraamnesia), single-element emotional association, gestalt familiarity
  • Attentional: 1 perception under distraction followed by another perception under full attention. Perceptual fluency (unconscious preview benefit as model), Inattentional blindness (including ‘time gap’ experience)


Branco, Clark, Hausser Science 329:1671 2010

Dendritic discrimination of temporal input sequences in cortical neurons [Neuron, Time]

“two photon glutamate uncaging” at 8-10 spines on signle basal and apical oblique dendritic branches. Control order of synaptic stimulus, record soma response.


  • Directionality of input: inwards >> outwards by 2.8mV, even when stimulating only 3 inputs → spike probability increased by 38%.
  • Optimal direction selectivity at 2.6 um/ms. Different sequences → 2.3mV variability → likelihood of discriminating any 2 sequences = 40%.
  • EPSP nonlinearity and direction-sensitivity both abolished by NMDA blocker.
  • Found in Layer V pyramidal, and hippocampal dentate gyrus granule cells.

Mechanism: (1) gradient of input impedance along dendrite; EPSPs travel faster in distal branches. (2) Highly nonlinear voltage dependence of NMDAR conductance.

Sigman & Dehaene PLoS Biology 3(2):e37 2005

  • Visual number & auditory tone, each for 150ms, SOA=0 to 1025ms.
  • Dual task. Number task = 2afc number > 45. Tone task = 2afc was tone high (880Hz) or low (440Hz) – Constant and simple.
  • Respond RH for number task, LH for tone task. ‘respond to both as fast and accurate as pos’
  • Manipulated the number task: ‘P’ = words or digits, ‘C’ = far or close from 45, ‘M’ = double tap or single tap response slower or longer RTs, additive.
  • Number task was either task 1 or task 2 in different subjects → change P/C/M of task 1 or 2.

[double tap only for numbers → increases difficulty of tone task too!]

Only ‘both correct’ trials analysed = 83%.

Model: fixed P and M duration, noisy integrator C.

Result: motor manipulation and word-form manipulation only affect the fixed delay, whereas number-distance manipulation affects the stochastic integration time.

P+M=180ms, C=550ms → C-component represents ~70% of total RT.

Also, the first task always starts later after its stimulus than the second task after its stimulus (independent of Δ). Explanations: temporal attention? Preparation delay?

Conclude: motor planning occurs independently of word-form analysis. ‘Central bottleneck’ [actually what it shows is that the number magnitude judgement is the source of variation in RT].


Sourani, Eitan, Goelman Nature preprint 2012

Injection of 6-OHDA into rats → model of PD; treat with apomorphine, or lesion to habenula.

Background: OHDA rats have ↑neuronal projections from GPi to lateral habenula, and from lateral habenula to dorsal raphe nuclei. PD→ enhanced GPi activity→ excitatory connections → ↑habenula excitability → inhibitory projections → downregulation of serotonin system

Hypothesis: habenula mediates coupling of dopamine to serotonin; in PD, low dopamine causes low serotonin.


  1. Manganese-enhanced MRI inject Mn into left raphe interpositus nucleus: Demonstrated reduced connectivity and excitability of raphe. Particularly habenula, dentate gyrus, thalamus, hypothalamus.
  2. Apomorphine: reduced glucose utilisation in lateral habenula. Forced-swim and novelty-suppressed-feeding tests = index of depression (time and thigmotaxis). ‘PD’ vs ‘PD-apo’ vs ‘sham’ rats. Apomorphine normalised behaviour.
  3. Habenula electrical lesions (1mA x 15s): did not affect sham mice, but helped PD mice.

Conclude: this is the mechanism whereby dopamine replacement improves depression in PD. ? DBS



Anderson, Laurent & Yantis PNAS 108:25:10367-71 2011

Value-drive attentional capture [Behaviour, Reward]

Training = 1000 trials with Fb: Search for red or green amongst other nontarget colours. Response = discriminate orientation of bar within. Feedback: Target was red → 80% 5p, 20% 1p; Target was green → 20% 5p, 80% 1p.

Test = 480 trials without Fb: target is the unique shape, “colour is now irrelevant”. Target was never red or green, but 25% had 1 distractor red, 25% 1 distractor green, 50% had all distractors other colours.

  • ↑RT for trials with previously high-valued colour distractors.
  • Even the fastest 25% of RTs in the high-value distractor condition were slower than those in the distractor-absent condition.
  • Change-detection performance and BIS correlate positively with value-driven capture.
  • Expt2: removed all trial-by-trial fb for training → abolished effect of distractor colour in test phase. → not a generic effect of previous search targets, rather, genuine effect of reward.
  • Expt3: 250 training and 240 test trials, and subjects returned for a second test session after 4-21 days → robust retention of effect.
  • RT in trials without a distractor, which were preceded by a trial with a high-value distractor. Responses were 66ms slower when the target appeared in a location formerly occumied by a high-value distractor than when it appeared elsewhere. → spatially specific capture of attention

Results “violate the prediction of both a salience-driven and goal-driven account of attentional capture”: distractor is neither salient nor goal relevant.


Fecteau & Munoz NRN 4:1 2003

Exploring the consequences of the previous trial [Theory, Attention]

Explanations of previous trial effects: priming, procedural learning, maintaining or switching task set, attention, guessing strategies (Soetens JEPHPP 1998), competing motor programs.

FEF and SC correlates of previous trial effects:

  1. 2afc: speeding for targets appearing at the same location as before (in monkey). Repetition advantage = pre-target activity predicts RT; ?allows threshold to be reached sooner. But: Humans have opposite effect: alternation advantage. ?because monkeys are practiced.
  2. oddball localisation: speeding when target colour remains same = priming of popout, = earlier time of ‘selection’ (time that cell responds differently to a target vs distractor); competing motor plans = fast error corrections / curved saccades.

    neurone with target in its RF has elevated activity even when saccade is erroneous [different for uncorrected errors? what about the cell encoding distractor location?] ... “the salience of the distractors is accentuated” → explains both phenomena
  3. oddball localisation – repetition of location of target → slowed responses.
  4. exogenous Posner cueing with SOA>200ms = IOR = cue elevates baseline firing before target, but attenuates phasic visual signal from target → conflict → slowing.
    RT correlates with amplitude of target response, not with pre-target baseline firing rate.

Funny IOR facts: saccadic vs manual → different SOA needed to produce IOR (Briand Larrison & SerenoA 2000) & different reference frame (Tipper Jordan & Weaver 1999).

SOA changes if task is to discriminate a target feature vs just detect target onset.

Commonality: IOR = diminished salience of target. → concept of salience map for previous trial effects.

How to explain priming vs negative priming? “Perhaps the parameters of the task used to explore previous events change the way the same signal is translated in the salience maps”


Padmala & Pessoa JCN 23:11:3423 2011

Reward reduces conflict by enhancing attentional control and biasing visual cortical processing [Reward, fMRI]

Expt: reward cue 20 or 0. Image of house or building, with word superimposed “HOUSE” / ”BLDNG” / “XXXX”. Respond to image only; word is irrelevant: congruent / incongruent / neutral. Rt hand button-press 2afc.

Feedback = winnings and running total. Time limit = 800ms. Slow and error trials →0.

fMRI functional localiser with 1-back task for images and words.

BAS: “behavioural activation scale” Carver & White 1994 – “strongest predictor of positive affective responses to reward”.

Result:  ↓RT for congruent>neutral>incongruent, and for reward>noReward, and interaction.

“motivation enhanced attentional filtering, reducing the influence of the task-irrelevant word”

response to reward cue. Interaction only mPFC.

Mediation analysis: correlation between IPS & mPFC activity “was significantly reduced once the contribution of Lt Fusiform was taken into account” → word perception was filtered out more in reward trials. Same for IPS & accumbens.


Goldstein, Barnett...Roesch JCN 32:6:2027-36 2012

Ventral striatum encodes past and predicted value independent of motor contingencies [Neurone, Reward]

Previous studies: “predicted reward and motor output signals have been intertwined in a way that makes it difficulty to dissociate encoding of value from the direction and speed of action initiation” (confound by RT and action contingency)

Expt: Odour precue signals large or small reward, delay 250-500ms, then L/R light cue, response, then 500-1000ms delay, then reward. Extensive training.

Behaviour: Slower RT and ↑accuracy for high-reward.

Record 488 ventral striatum neurones. 229 (47%) increased after odour. Of these 33 increased for big reward, 9 increased for small reward [opponent coding of predicted reward?]. Before odor, these cells encoded previous trial reward size.

[post-hoc and confounded] comparison of predictability: they didn’t allow sequences of 3 Hi or Lo trials in a row, so analysed next trial which was predictable. – shows that pre-trial activity reflects previous trial not prediction about current trial.

Activity correlated +ve with RT, only for big-reward condition. [only just]


  1. striatum encodes value independent of motor contingencies, and previous trial reward.
  2. speed-accuracy-tradeoff → “predictive reward signals [don’t] just energise specific actions but signal value in a way that might be used to slow reaction times to improve task performance when more is at stake”.

[but does not dissociate from effort or motivation]

Previous studies: ventral striatum does not discriminate between odours that predict the same reward; encodes how much effort is required for a reward; OFC does all these too.


Jacobs, Lega, Anderson JCN 24:3:553-63 2012

Explaining how brain stimulation can evoke memories [Disease, Memory]

Epileptic patient ECoG. Stimulation of 1 electrode → elicits high-school memories. Recording from same electrode during visualisation of old memories → lower high-gamma activity.

Questions: high-school vs non-high-school, and person vs non-person.

GLM to identify phase-amplitude coupling between phase of low-freq oscillations and amplitude of high-freq oscillations. Circular-linear regression. Effect of personhood on phase-amplitude coupling.

No difference seen for ERP.


Selen, Shadlen, Wolpert JNeurosci 32:7:2276-86 2012

Deliberation in the motor system: reflex gains track evolving evidence leading to a decision

Subjects hold lever at centre against a 4N elbow-extending force, then respond in direction of RDK motion. RDK coherence 0 to 50% for 100-1500ms nonageing, immediately followed by 10mm perturbation of hand → response → feedback. EMG measurement of tendon reflex after perturbation.

Analysis: estimate DV for each trial based on

Conclude: Motor control systems are continuously informed about decision variables


Mullally, Intraub, Maguire Curr Biol 22:4:261 2012

Attenuated boundary extension produces a paradoxical memory advantage in amnesic patients

Boundary extension: memory of a scene extends beyond the actual boundaries – childrens drawings etc.

Distortions seen in controls but not hippocampal amnesics:

Scene construction from a verbal description showed patients were worse

Describing a scene from memory was normal detail except for omitting spatial references from descriptions of what was likely to be beyond the view, and lower vividness ratings. Patients could not visualise anything but provided rich semantic/associative information.

RSVP 2 images 250ms each 250ms ISI: judgement “was the second picture zoomed in closer or out bigger than the first”: controls show 61% zoom out, patients 31%

Drawing scene from memory: controls distortion to 61% of original size, patients to 63%

Haptic scene presentation → manual demonstration of size: size increased to 113% in controls, 95% in patients


So & Stuphorn J Neurosci 32:9:2950-63 2012

Supplementary eye field encodes reward prediction error

Oculomotor gambling task: saccade to peripheral coloured squares, 7 colours = 1-7 units of water. Gamble = two colours, probability as proportion.

choice vs no-choice trials. Choice = certain vs uncertain reward, uncertain = hi-valued (4 vs 7u) or low valued (1 vs 4u), probability of higher reward = 10%, 50% or 75% → 6 gamble types. Each gamble paired with 4 certain options 1/2/3/4 or 4/5/6/7.

→ measurement of risk equivalence value: choosing gamble gets less as the sure-value increases.

Spike density function: convolve spikes with (1-e^(-t/1ms)) e^(-t/20ms) = EPSPs

Regression of activity based on subsequent choice: 1) subjective value of chosen item, 2) choose-gamble/sure, 3) value-contingent-on-choosing-gamble/sure.

Some cells represent the gamble-value exclusively, others sure-value exclusively.


Isham & Geng Consc & Cogn 2012

Rewarding performance feedback alters reported time of action

Random rewards in Simple reaction time task. Clockface 3s. Then feedback win/lose/nil 33%, random. Then estimate of action time.

Expt 1) subjects told “reward for fast responses, penalty for slow”. Feedback win/lose on each trial → strong effect on perceived time of action

Expt 2) feedback was “fast” or “slow” or “” i.e. computer does not tell you → No effect

Expt 3) subjects told “press button at any time of your choosing”.. “your winnings are based on a random algorithm” → no effect

Conclude: Combination of reward feedback and cognitive interpretation is required. Reward alone does not influence perceived action time, nor does timing feedback alone.


Mazaheri, DiQuattro, Bengson, Geng PLoS One 6:2:e16243 2011

Pre-stimiulus activity predicts the winner of top-down vs bottom-up attentional selection [EEG, Attention]

Target = T or inverted T, distractor 90 rotated Ts. Fixation 1500-2000ms, target is grey. 33% alone, 33% nonsalient distractor (grey), 33% salient distractor (white). 600ms display, then response 2afc button ‘upright’ or ‘inverted’. 140 trials per condition. [Subjects were essentially free viewing!]

Behaviour: proportion of first saccade that went to distractor. When equisalient, =46% to 53%. With salient distractor, ↑heterogeneous, 13% to 85%. Trials where first saccade was distracted → longer manual RT (obviously). Capture trials = shorter saccadic latency by 40ms.

ERP & spectrum before capture vs noncapture trials: high frontal alpha, and low stimulus-locked posterior alpha. “Attenuated N1 response to the stimulus array” on capture trials


Van Maanen, Grasman...Wagenmakers Frontiers Decsn Nsc 2012

Pieron’s law and optimal behaviour in perceptual decision-making

Mean RT = αI+γ. Detection time is simply αI

Pieron 1914: button press for detection of brightness, tone, taste.

Jaskowski & Sobieralska 2004: Go-nogo pieron

Palmer et al 2005: 2AFC pieron

Stafford et al 2011: Stroop RT depends on luminance of the color dimension

Pieron as special case of ease of discriminability.

Expt1: RDK motion at an angle θ, 2AFC discrimination between 2 axes displayed at θ and θd. Trialwise manipulate 7 difficulty levels: similar (d~11.5 deg) vs different angles (d~90 deg). 2 axes shown for 500ms before motion; Motion shown until response. Feedback correct/incorrect.

Compare pieron with exponential (αed), Nelder & Mead optimisation→BIC. Evidence ratio as e-(BIC1-BIC2) (Wagenmakers & Farrell 2004) = “how many times more likely the data are to have occurred under model 1 than under model 2”. ~= 0.2 to 0.67

Unfortunately plain pieron or exponential fits give γ<0 or γ>800ms → implausible. So:

Linear ballistic accumulator fit to estimate nondecision time from RT quintiles. Nondecision time ~=300-500ms [!]. rate of rise = 0.6-1.3 depends on discrimination distance. Threshold = 0.3-0.9, sigma=0.4-0.6. [error rate 33% on hard condition, but sounds like they didn’t even model the error response?]


Observer “needs to decide whether a certain amount of evidence for a particular response alternative outweighs the evidence for other response alternatives”.

“Optimal”: observer has “minimum time required for a particular error rate”.

For equal priors, wait until P(Hi|D) = P(D|Hi)/ΣP(D|Hj) > Threshold of evidence.

Reduces to sequential probability ratio test (Wald 1947)

Assum each evidence sample is Gaussian about the actual motion angle; σ depends on individual’s perceptual ability. P(x)=φ((x-μi)/ σ).

Evidence for an alternative after one sample: log[φ((x-μ1)/σ)/ φ((x-μ2)/σ)]

“i.e. proportional to (μ1- μ2)2.” [this isn’t completely right, but OK for both case if x=μ1 and x=μ2, which might come out in the wash when expectation of evidence is considered; E[x]= μ1 or μ2]

Stafford & Gurney 2004: decision times decrease with discriminability:

, so

[?is equivalent to rate of rise = distance / stdev = d prime?].




Mesulam Phil Trans Roy Soc B 1325:46 1999

Spatial attention and neglect: parietal, frontal and cingulate contributions to the mental representation and attentional targeting of salient extrapersonal events

Bisection of short lines <5cm is reversed – leftward bias? Inconsistent.

Randomly distributed items much worse to detect than grid of items

Popout preserved on left. Cancellation with erasing targets is much better.


  • Blindfolded manual search with either hand is worse on left space → motor programs organised by target space rather than muscle group. (Weintraub & Mesulam 1987)
  • Improvement for moving to a left-side target if the movement required is towards the right (Driver & Mattingley 1998).
  • Search exposing items in a central slit by moving sheet below slit: more targets detected on right side → hypokinesia to left is not a determinant.
  • Anecdotal: monetary reward improves cancellation (Mesulam 1985), hunger improves finding left-sided food.
  • A simultaneously presented extinguished left-sided stimulus speeds response to a right-sided stimulus.
  • Covert semantic priming +.


  • Turning head/eyes to left → increase in target detection to left of body.
  • Turning head/eyes to right → increase in target detection for same retinal location.
  • Left caloric stim → improved imagination of left side of a map.
  • Bias reduced in supine position
  • Patient lying on left side → world-based left is ignored. Model tower inclined 45deg to right → still omit canonical left side of tower.

Relative: wherever 2 lights or 2 touches, leftmost is ignored. Left halves of chimeric faces. Object-based? UK map omits west coast of Wales but draws Ireland.

Kanisza amodal completion → reduced extinction. Connected daisies = omit left daisy; unconnected daisies = omit left side of each daisy. Omit left side letters of nonwords but not of words. Able to see left of chimeric face when right half is degraded.

Greater neglect in inferior quadrant. Vuilleumier some patients have specific near-space or far-space neglect.

Why? Left hemisphere ERP/activation after Rt stimulus; Right hemisphere ERP/activation after either L or R stimulus. Retrieval of spatial information = Rt parietal activation.

Monkey: posterior parietal (IPL+LIP), premotor, FEF, SEF, SC.

STS, cingulate, VL & MD thalamus, striatum, raphe nucleus, LC, VTA/SN, nuc basalis

insula, OFC, parahippocampal.

= “overlap with other networks: eye movements, WM and temporal expectation”


Wright, Burns, Geffen & Geffen Neuropsychologia 28:2:151-9 1990

Covert orientation of visual attention in parkinson’s disease: an impairment in the maintainance of attention

25 PD, all but 2 on dopamine/agonist. Posner: Central cue L-arrow/R-arrow/cross, 80% validity, SOA 1.1sec, cue was not extinguished [overlap task]. Feedback = RT in ms. ANOVA subject, age as covariate, validity (valid invalid neutral), L/R.

PD patients slower overall, but smaller cost of invalid cueuing.

Effect of block: both groups get less cost of invalid cueing over time.

No LR asymmetry.

in PD, the generator of Nd2 is smaller.

  • PD patients “disengaged from attended locations more readily than controls”.
  • “L&R parietal lobes are asymmetrical for the disengage operation”.
  • “No impairment in attention when task demands are minimal, [but] in tasks requiring more substantial attentional demands... deficits in attention are evident”.

Conclude Effect of Bromocriptine on Nd2 → “dopamine is involved in selective attention”. Nd2 uncorrelated with motor symptoms → “cannot be nigrostriatal/caudate/putamen complex. So probably it is the frontal dopamine system”.

Contention Scheduling = striatal/thalamic system.

Prev: droperidol and clonidine reduce cost of invalid cueing


Stam Visser...Gielen Brain 116:1139-58 1993

Disturbed frontal regulation of attention in Parkinson’s disease

33PD divided by neuropsych into high or low ‘frontal’ score.

ERP during dichotic listening: 20% target 75ms tones, distractor 25ms tones. Ignore all tones in unattended ear. Blocks of 150.

‘Nd’ = ‘processing negativity’ = distractor ERP when stream attended – when stream unattended.

ERP to target, distractor, and processing negativity.

Bromocriptine dose correlates with P3 latency.

“low frontal” patients have smaller Nd2 (late Nd).


Prev: N1 longer latency, P3 longer latency


Horstink...CoolsAR JNNP 53:685-90 1990

Bimanual simultaneous motor performance and impaired ability to shift attention in Parkinson’s disease

18 iPD, 19 age and intelligence-matched controls. Ldopa 0.25 to 6 yrs, 300-3500mg/d

1. Squeeze bulb with nondominant hand “repeatedly and as quickly as possible by fully opening and closing hand”

2. Triangle-drawing with dominant hand: “connect without interruption and as quickly as possible” the corners of a 9cm triangle.

3. Simultaneous Sq + Tr

4. Simultaneous nondominant Sq and dominant-hand writing ‘eeeee’


  • Amplitude of squeezes – decrement with concomitant dominant arm task, greater in PD than controls. → PD = “diminished capacity to shift their attention to squeezing task”
  • Amplitude of triangle – not affected by concomitant squeezing. → squeeze task “lacks imperative external cues or guidance to prompt attention and divert it from” triangle task.
  • Frequency ratio LH:RH in bimanual tasks was same for PD and controls. = “Task-specific invariant interlimb linkage”
  • Frequency of squeezes increases with simultaneous eeee, decreases with simultaneous triangles. = PD construct a “bilaterally generalised time-programme” like controls.
  • Frequency & amplitude are correlated for single-handed tasks, but do not correlate for bimanual task. → “frequency corresponds to time-sharing”

Decrement “not due to a decreased ability to share time between the two tasks”, rather, inability “to shift their attention sufficiently from one of the simultaneous tasks to the other”.


Owen, Roberts, Hodges...Robbins Brain 116:1159-75 1993

Contrasting mechanisms of impaired attentional set-shifting in patients with frontal lobe damage or Parkinson’s disease

18 frontal = 11 Rt lobotomies, 4 frontal epileptics, 2 ACOMs, 3 Rt meningiomas; 5 left lobotomies.

26 early PD (pre med), 23 on LDopa. Exclusion dementia. 25 age & NART-matched controls.

Age as covariate in ANOVA.

CANTAB visual discrimination learning procedure. Simple discrimination → reversal, compound discrimination → reversal, intradimensional shift → reversal, extradimensional shift → reversal


Lee Wild Hollnagel Grafman Neuropsychologia 37:595-504 1999

Selective visual attention in patients with frontal lobe lesions or Parkinson’s disease [Lesion, Attention]

Eriksen flanker, letter ‘S’ or ‘H’ above fixation, press corresponding key. Unflanked, or Flanked by letter that was either same as target ‘compat’, the other nonpresented target ‘incomp’, or other ‘neutral’. Feedback incl RT and overall accuracy. Vary: spacing = selectivity.

15 frontal patients: Vietnam veterans, incl 2 subdurals!

Conclude: slower & less accurate, depending on lesion volume. However not specific to spatial distance of distractor. ~= null result!


Rowe, Stephan, Friston, Frak, Lees, Passingham Brain 125:2:276 2001

Attention to action in Parkinson’s disease: impaired effective connectivity among frontal cortical regions [fMRI, Lesion, Attention]

“MOVE”: paced sequential right finger movements 1,2,3,4, with tone / visual cue.

“SEARCH”: sequential central letters every 3s, conjunction – red ‘R’, respond later.

Also REST, and DUAL task = move + search.

“ATTEND” = MOVE + “think about the next move”

During motor task, compare effect of concurrent search (SEARCH-REST) and effect of attending to the movement.

Conclude: “attention to action associated with increased coupling between prefrontal cortex and medial & lateral premotor regions in healthy adults”, unlike with concomitant visual search. PD → “functional disconnection of the SMA and PMC from prefrontal influences”.


Cormack Gray Ballard Tovee Int J Ger Psychiat 19:763 2004

A failure of ‘Pop-out’ in visual search tasks in dementia with Lewy Bodies as compared to Alzheimer’s and Parkinson’s disease

15 DLB, 18 AD, 21 PD.

Simple vs conjunction red/green square/circle. 50% present/absent. Set size 2,8,16. Display duration 200/400/800ms.

Compared with basic choice RT to visual stim.

‘Serial search’ ‘Parallel search’

DLB don’t have progressive increase in RT with set size. Much worse overall.

AD, PD have same serial-search pattern as controls.

However DLB did not do worse on CAMCOG. Also no correlation with fluctuation of performance over time.

Parasuraman 1995: AD slower serial search. → attention shifting problems in AD.

DLB more occipital pathology.

Troscianko and Calvert 1993, Weinstein 1997: PD use serial search to solve popout.

???Weinstein A, Troscianko T, Calvert J. 1997. Impaired visual search mechanisms in Parkinson’s Disease (PD): a psychophysical and event-related potentials study. J Psychophysiol 11:33–47.


Lieb Bruckner Bach... Brain 122:303 1999

Impairment in preattentive visual processing in patients with Parkinson’s disease [Attention, Lesion]

Visual search – preattentive or attentive. Threshold found with ‘PEST’ algorithm – equates difficulty too.

Conclude: not popout or serial search problem, rather “impaired processing of orientation differences”. Suggest that “retina but also striate and extrastriate visual cortext are affected”.


Hartikainen, Ogawa, Knight J Clin Exp Npsychol 2012

Orbitofrontal cortex biases attention to emotional events

7 head injury bilat OFC lesions. 2AFC identification of 150ms targets on L or R side. 350ms before each target, flashed a picture centrally. Picture is 33% pleasant/unpleasant/neutral. 18% no picture, and 27% of pictures not followed by a target. ITI 1.5s

Conclude: reduced attentional allocation to affective distractors in orbitofrontal patients. OFC lesion had opposite effects on involuntary (distractor N2-P3a) and voluntary (target N2-P3b) attentional markers.

affective stimuli don’t compete as much. enhanced voluntary allocation to targets.


Preuschoff, Hart, Einhuser Front Decis Neurosci 5:115 2011

Pupil dilation signals surprise: evidence for noradrenaline’s role in decision making [Reward, Behaviour]

Auditory gambling task: Bet on first or second number higher (button press) → 5 sec → spoken number between 1 & 10 → 5 sec → second spoken number 1-10 → 5 sec → report whether won or lost (button press). Outcome: win or lose $1, or lose $0.25 if mistake on reporting win/lose.

Result: 97% accurate on outcome report. Bet “Second card higher” ~53%.

Pupil normalisation (P-P0)/P0. Decicion variables:

  • pWin(after first card), EV(after first card), RPE(after first card),
  • Risk(after first card)=E[actual exp rew – (expected exp rew) after 1st] = Uncertainty,
  • (risk prediction error after card 1) = (actual rpe – expected rpe after 1st) = Surprise,
  • pWin(after second card)=0/1, EV(after second card)=+1/-1, RPE(after second card),
  • (risk prediction error after 2) = (actual rpe – expected rpe after 2nd).

Conclude: risk prediction error correlates best with pupil dilation.

Noradrenaline plays a similar role for uncertainty as dopamine does for reward – i.e. encoding of error signals.


Nanhoe-Mahabier et al Movt Disorders 27:4:574-8 2012

The possible price of auditory cueing: influence on obstacle avoidance in Parkinson’s disease [Lesion, Attention]

19 PD pts Off. Motion analysis during walking on treadmill. 2x2: with and without auditory metronome (10% less than preferred cadence in baseline), with and without concurrent obstacle avoidance. Obstacle = 400x300x15mm block “suddenly dropped on the treadmill” on patient’s most-affected side.

  • Auditory cueing improved stride length, stride time, and cadence.
  • Patients with freezing of gait (30%) – no different at baseline, no interaction w cueing.
  • Patients with freezing of gait – worse at obstacle-avoidance.
  • Object avoidance did not worsen synchronization to metronome.
  • Metronome improved gait when obstacles present, but no interaction.

Conclude: Metronome improves gait in PD, with no negative effect on concurrent obstacle avoidance.


O’Shea, Morris, Iansek Phys Ther 82:9:888-97 2002

Dual task interference during gait in people with Parkinson’s disease: effects of motor versus cognitive secondary tasks

PD On. Walk, walk+count backward in 3s, walk+“transfer as many coins from one pocket to another using the dominant hand”. Both tasks impaired walking. PD cadence also deteriorates with secondary task. Both tasks have same effect.

Conclude: Type of secondary task is not a determinant of interference with gait.


Allman & Meck Brain 135:3:656-77 2011

Pathophysiological distortions in time perception and timed performance

  1. Prospective timing: know in advance we need to time an event, we can measure its duration.
  2. Retrospective timing: we didn’t know at the time, but afterwards we are asked to estimate
  3. Temporal Perspective: orientation, + sense of past present and future.

Animals credited with prospective only.

         DA agonists produce horizontal leftward shifts in psychophysical functions relating probability of response to signal duration → tendency to classify shorter stimuli as long, tendency to under-reproduce intervals. → equivalent to increase in clock speed

         Cholinergic effect (anticholinesterases) → proportional leftward shifts but only in temporal memory. Failure to renormalize with training+feedback if drug is continued, unlike DA effect.

         Malapani 1998, 2002: PD. When durations distributed around single value, Off patients overproduce 21s. But, when two modes present, over-reproduction of 8s and underproduction of 21s. = “migration effect”. Not correctible with feedback. Correlates with akinesia. Only present >2s. (? due to reference memory coupling? Tuning of neural accumulators? More likely attentional set-shifting, as there is nonscalar variability.)

         Pastor 1992: PD verbally underestimate duration, and overreproduce durations, even when counting aloud. Reduced accuracy in ordinality-comparison <2s (could be related to ↓clock speed). But psychophysically, no difference in temporal bisection, generalisation gradient, verbal estimation.

         PD violate scalarity (30-120s) both On and Off → requires basal ganglia integrity

         Many studies show no effect of DA meds on interval reproduction or perception

         Modality effect: auditory signals judged longer than visual signals of same duration

         Schizophrenia interval classification task (two anchors): kink in psychometric function; patients reverse-classify near the geometric mean of the two anchors. Controls also show this if ratio of short to long anchors is small. → ?difficulty effect

         Schizophrenics have lower sensitivity for duration discrimination. underestimate durations in 400:800 ms classification, but not for 1:2 seconds.

         Autism (?abnormalitites of PFC, BG, cerebellum, DA, 5HT) – temporal bisection → shorter point of subjective equality. Correlates with language tests and WM. Tendency to truncate/shorten longer durations.

         Wallace & Happe 2008: Autistic children may overestimate and underproduce durations. Overestimation greatest for short durations. No difference in accuracy of estimation / reproduction / production. Mosconi 2010: unaffected relatives also have oculomotor timing abnormalities.

         ADHD: larger duration-discrimination threshold (both aud and vis). Underproduction. Decreased accuracy of judgements and reproductions. ? attentional

         Nicotine improves ADHD on 7 and 17sec peak-interval procedures. Reduces the rightward horizontal shifts (attributed to deficit of attention, which can be corrected by feedback)

Conclude: “Striatal beat frequency model”

DA fron VTA resets cortical neurones and acts as “start gun”.

DA from SNpc resets the weights of synaptic connections in dorsal striatum

Medium spiny neurones detect coincidence of cortical oscillations (and can learn sensitivity to different frequency combinations[!]). Striatal output to thalamus: direct D1, indirect D2 == start/pause/reset.


Verreyt, Nys, Santens, Vingerhoets Neuropsychol Rev 21:405-424 2011

Cognitive differences between patients with left-sided and right-sided Parkinson’s disease. A review

Asymmetry: 85% at presentation (Rt 46% Lt 39%). Pathognomonic. Related to shorter disease duration, younger age at symptom onset, positive family history of other movt disorders.

Reviews of generalised cognitive problems: Elgh 2009, Williams-Gray 2009.

Classification: initial vs current lateralisation, hx vs exam, subjective vs standardised scale.

11 studies of mental rotation/imagery: 6 L=R, 4 Lt worse.

12 studies of block design, only 1 revealed asymmetry. No asymmetry also on picture completion, object assembly.

Object recognition: 1 study = Lt worse for global, Rt worse for local detection. No other diff

Object Naming: 11 studies, 5 nodiff, 1 Lt worse, 4 Rt worse. → probable Lt = language

Visuospatial memory: 13 studies eg Rey figure – 5 show L worse, 1 shows Rt worse.

Wechsler memory scale (verbal): Rt worse in 1 study. Many different tests –california verbal learning test, auditory VLT, remote memory, supra-span test....

WCST: 6 studies, no lateralisation. Trail-making test no difference. ID/ED and other switching tasks = conflicting results for laterality.

Controlled oral word-association test (cowat): no difference in 11/13.

Stroop: 4 studies. 1/4 show Lt worse.

Sustained attention: 1 study, no asymmetry. WAIS-R digit span = “complex attention and working memory”, 11/14 no difference, 2 Rt worse, 1 Lt worse. Also, counting by threes, serial digit learning, months backwards, corsi: no difference.

6 studies: Posner quadrant task, initial visual exploration, balloons test, line bisection, cancellation. 4/10 show Lt PD → small right-sided attention bias. 1 study shows both Lt and Rt PD have a leftward bias.

Conclude: Lt sided motor symptoms → worse visuospatial orientation and mental imagery

Rt PD patients poor at language-related tasks. Memory: performance depends on type of stimulus. Little asymmetry in executive fuction.


Williams-Gray, Foltynie, Robbins, Barker Brain 132:11:2958-69 2011

The distinct cognitive syndromes of Parkinson’s disease: 5 year follow-up of the CamPaIGN cohort

127 PD patients, COMT & MAPT polymorphisms, MMSE, NART, fluency, Cantab.

Semantic fluencty, pentagon copying and MAPT H1/H1 are strong predictors of cognitive decline.

Tower of London deterioration varies dependent on COMT, but not predictive of dementia.

Conclude: “early deficits on frontostriatally based tasks are not related to subsequent dementia risk”, whereas temporoparietal tasks are.

→ “dementing process has a non-dopaminergic aetiology”

So cognitive deficits = either anterior (executive) benign, or posterior (spatio-semantic) heralding dementia.


Elgh, Domellf...Forsgren EJN 16:12:1278-84 2009

Cognitive function in early Parkinson’s disease: a population-based study

88 nondemented PD at diagnosis. Free and cued selective reminding test. WMS Logical memory & paired associateive learning, brief visuospatial memory test, digit span, electronic tapping test, trailmaking, controlled oral word association, boston naming, Benton judgement of line orientation, WCST.

70% normal in all tests! Worst: trailmaking, episodic memory, WCST and word association test. But tapping test is globally impaired, so trailmaking is confounded by movement.

Deficit correlates with expressionlessness, shorter duration of disease, speech problems, bradykinesia, rigidity, but not tremor/axial/depression/age.

After adjustment for psychomotor function/age, significant differences for all domais except visuospatial and WCST.

Dopamiergic cell loss: Tremor = medial SN, akinetic/rigid = ventrolateral SN (Jellinger 1999).


Hangasi et al. Movt disorders 26:10: 2011

The effects of rasagiline on cognitive deficits in Parkinson’s disease patients without dementia: a randomized, double-blind, placebo-controlled, multicenter study

MAO-B inhibitor, increases frontostriatal transmission, potentiates nigrostriatal system.

  • Rasagiline improved backward digit span but not forward, and improved the “digit ordering test”: Just like WMS digit span, but “repeat these digits in ascending order” (Werheid...von Cramon 2001).
  • Rasagiline had no effect on trail-making performance, but did improve Stroop spontaneous corrections.


Farooqui, Bhutani, ... Murthy EBR 208:1 2011

Impaired conflict monitoring in Parkinson’s disease patients during an oculomotor redirect task

Double-step / countermanding-like task. Visually guided vs memory guided task.

60% no-step trials: red.

40% step trials: Target-step-delay = 50-200ms. “saccade to the later appearing green target”

post-conflict = nostep after correct step trial

Cumulative Weibull fit with pedestal at either end =.

Result: Post-error and post-conflict-trial slowing is slightly less.

PD significantly poorer in error correction.

In both visually- and memory-guided tasks, controls showed greater extent of error-correction at at shorter target-step-delays than long (93% vs 84% vis, 62% vs 42% mem).

PD patients showed no difference in error-correction between short and long delays.

Memory guided task does not show post-error slowing.

Discussion: “coactivation of movement and fixation neurons can instantiate a form of response conflict”. However in error-step trials, “activation of STOP process may not be maximal, resulting in lesser conflicts between the GO and STOP processes”

Could it be just more reflexive saccades? No because a) nostep RTs not different b) memory condition is not reflexive. c) if anything, correct nostep RT is slower in PD.

= poor conflict recognition or poor modulation of decision parameters to change performance?

ERN is attenuated in PD (Willemssen et al 2009, Falkenstein et al 2001, Stemmer et al 2007, Ito and Kitagawa 2006, but see Holroyd et al 2002).


Dissociation between errors and conflicts has been reported (Swick and Turken 2002).


Godefroy, Spagnolo, Roussel, Boucart Cerbrovas Dis 29:5:508-14 2010

Stroke and action slowing: mechanisms, determinants and prognosis value

Slowing on finger-tapping, visual inspection time, simple RT, but not choice RT.

Visual inspection time correlates with Right IPL; finger tapping correlates with Lt middle frontal gyrus + lenticular nucleus; SRT = right lenticular nucleus and posterior fossa.

Conclude: Action slowing in stroke is due to perceptual and motor processes.


Muller-Plath, Ott, Pollmann JoCN 22:7:1399-1424 2010

Deficits in subprocesses of visual feature search after frontal, parietal and temporal brain lesions – a modeling approach

28 fronto/temporo/parietal lesions. ‘STRAVIS’ – strategies of visual search. Decomposes RT: estimates size of attentional focus, dwell time of attention, movement time, and fixed delay.

Task: 4 / 6 / 8 gratings in an semicircle in left/right visual field. Target individuated by different spatial frequency. 66% target present. Vary TDsimilarity. Display remains, Free eye movements.

Search task: any of the gratings could be the target. ‘Comparison task’: only the top item could be the target. (included to help fitting model!).

Model: include only Target+ trials with correct response. 12 trial types: (search4 search6 search8 compare6) x (3 similarities). Patients coded for involvement of 7 areas: DLPFC, FEF, ant insula, IPL/TPJ, temporal pole, SPL, parieto-occipital/precuneus/cuneus. Linear model for each dependent measure: fit Error rate, mRT, RT slope.

→ 7 Binary predictors + 3 attentional parameters (focus sz, dwell, movt time) per subject.

Result: Error rate: Lesions account for = 27% of variance. Temporal pole = 21%.

RT: lesions = 56% of variance, Ant insula + SPL + parieto-occipital.

RT slope: 37% variance explained; ant insula, temporal pole.

Focus size: DLPFC reduces focus size by 0.7 items. Lesions to temporal pole & IPL enlarge it by 1 item.

Dwell time (~390ms): SPL lesions → +255msec; ant insula → +98msec.

Movt time (~51ms): FEF +62msec, SPL -35msec.


Sapir, Sorokoer, Berger, Henik NNeurosci 2:1053-4 1999

Inhibition of return in spatial attention: direct evidence for collicular generation [Lesion]

1 pt with Dorsal midbrain haemorrhage

top = ipsilesional, bottom = contralesional


Akshoomsoff & Courchesne JoCN 6:4:388 1994

ERP evidence for a shifting attention deficit in patients with damage to the cerebellum [Lesion]

Shift-attention vs focus-attention task. Shift: simultaneous visual and auditory stream, a rare target signals to change stream. Focus: long periods (minutes) in one modality. Also within modality (circles vs squares or colour). 5 patients (children) 9 controls. 2 had hydrocephalus, 2 had posterior fossa irradiation.

P3b amplitude reduced in patients. “reflects impaired rapidly activate and deactivate attention”.


Vaina & Cowey Proc roy soc B 263:1374:1225 1996

Impairment of the perception of second order but not first order motion in a patient with unilateral focal brain damage

Focal stroke near MT. mild Rt hemi, “felt disturbed by visually cluttered moving scenes and by auditory noise”. VF & CSF NAD. Initially he had an elevated threshold for coherence of random dot motion, but normalised at 2 months.

Second order: flickering bars, switching gabors,

Conclude: multiple cortical areas? Double pathway? Lesion was in MT? inconclusive.


Freed Corkin Growdon & Nissen Neuropsychologia 27:3:325:339 1989

Selective attention in alzheimer’s disease: characterising cognitive subgroups of patients

Picture recognition (DMTS). “Rebound”: a subgroup of patients show memory impairment at 24hrs but improvement at 72 hrs.

Posner-like task in 20 AD pts – “test of attentional focusing”. Central precue 33% arrow Rt, 33% arrow left, 33% double-arrow. Validity 80%; target ‘x’ on one side, press single response key when any target appears.

Some impairment.


Godefroy and Rousseaux Brain & cogn 30:155-74 1996

Divided and focused attention in patients with lesion of the prefrontal cortex

16 ACOM subarachnoids, 5 excluded for hydrocephalus, retrorolandic damage, weakness, drugs. 3 had posterior WM damage. Mean age 55.

Battery: MMSE, digit span, modified card sort.

“Attention disorders: distractibility ana aspontaneity” – observers’ rating scale 1 to 5.

Stimuli every 1/4/7/10 seconds: tones and white squares.

Expt 1: Simple manual RT, either unimodal or successive random bimodal (generating trial pairs which switch-modality or same-modality).

Patient’s crossmodal slowing was greater than control when preceded by a short ISI.

→ PFC = “deficit in divided attention between perceptual channels”; longer time to shift attention?

Expt2: Go-nogo; either auditory=go visual=stop, or vice versa. Compared Go-RT with bimodal RT. Exclude post-commission-error trials.

Patients made more commission errors (but not omission; interaction+) → “suggests a deficit of focused attention”.

Patients “remained sensitive to the crossmodal retardation effect in the Go-nogo task, contrary to controls”. → deficit of focused attention. SDT analysis: persistence of crossmodal retardation is predicted by commission errors, not age/unimodal RT/divided attention/omission/d prime. = inability to reject irrelevant information.

Lesion location: divided attention depends on LtDLPFC, MPFC, caudate, age

Focused attention depends on Lt caudate, medial PFC, age.


Sereno, Briand, Amador & Szapiel JCEN 2006

Thiamine SC lesion: 2x2 saccadic tasks - gap (187ms) /overlap (fixation present throughout), pro/anti, 40 trials of each.

Nonpredictive spatial precue SOA 80, 133, 1000ms.

Presence of multi-step movements. Fewer for antisaccades than prosaccades. → stoppages only for reflexive saccades?

Absence of gap effect (speeding in gap task compared to overlap).

Significant increase in errors for Gap Antisaccade task. Deficit of IOR in posner task.

On day 2, gap effect recovered, but still no IOR → dissociation?


Rueckert & Grafman Neuropsychologia 34:10:953-63 1996

Sustained attention deficits in patients with right frontal lesions

Previous studies: Corpus callosum is important in sustaining attention

21 Left and right frontal patients – Vietnam penetrating HI. 3 had some parietal damage, 5 had some temporal damage., 9 had some bilateral damage. All included some DLPFC. All but 4 included OFC.

vigilance task’: Simple RT rectangular and nonageing foreperiods (2-18sec).

‘X-CPT’: ISI of either 1s or 2s (different blocks, counterbalanced across subjects). 1 target approx every 9.5sec. Also nonageing.

Story task “when does the protagonist pay?” – sequential sentences on screen.

right frontal lesions: slower to respond, miss more targets than controls, “deficit in sustaining attention over time”. Slower during second half of block, but decrement was much more pronounced with rectangular foreperiod. Speed of presentation didn’t affect slowing over time.

no diff on WCST, Tower of Hanoi.

Effect of lesion location: some weak evidence that callosal lesions show greater deficits in sustaining attention. ACC: slower response on story task, nil else. BG: 4 pts, all slower on all 3 tasks.

Left frontal: worse on all tasks, subtly. Rt frontal paradoxically worse that bilateral patients.

[many good refs in discussion]


Fletcher & Sharpe Annals Neurol 20:464-71 2004

Saccadic eye movement dysfunction in Alzheimer’s disease [Lesion]

13 AD patients vs 11 controls, prosaccades and antisaccades.

Longer latency for unpredictable timed prosaccades.

Hypometria & lower accuracy. Failure of antisaccades: 74%; “22% of trials elicited no response”.

“Fixation stability: large-amplitude intrusions, up to 20 degrees, occurred in 8 patients. After intervals of approximately 200 to 600ms, a refixation saccade corrected the retinal error. These large saccadic intrusions caused impersistence of maintaining gaze. They were usually directed prematurely towards the next target step and often appeared to be anticipatory.”

~5/min in 8 patients, >10/min in 5 of them → “chaotic disruption of fixation”.

3/11 normals had IoG, usu <2/min, but one 78yo had 9/min.

Square-wave jerks = horizontal saccadic intrusions of 0.5 to 3 deg → refixation: common but no effect of AD. cf Macrosquarewave jerks <200ms intervals.

? related to “inability to sustain motor activity (motor impersistence), a recognised feature of dementia”

Previously reported in 60% of cognitive dysfunction, ?correlated with intellectual decline

Huntington’s (Levin Jones & Stark Biol Psych 1982), Schizophrenia (Lynch & McLaren, Son Neur Abstr 1984).

Conclude: degeneration of parietal fixation neurones?


Alexander, Stuss, Shallice, Picton, Gillingham Neurology 65:572-9 2005

Impaired concentration due to frontal lobe damage from two distinct lesion sites

43 chronic frontal lesions: infarcts, bleeds, trauma, operated tumours.

Classified as: Left lateral, Right lateral, Inferomedial, Superomedial

Line bisection and extinction, NART, forward and backward digit span, Token Test, Boston Naming test, Judgement of line orientation, BDI.

5-choice serial RT, light → “respond quickly and minimise errors”. Pressing button extinguishes light, next after 200ms. Order: random excluding repeats of last position.

Result: RT: right medial and superior frontal were 33% slower. = cingulated, paracingulate, preSMA/SMA, corpus callosum.

Time on task: 5 blocks of 100 trials. No effect of fatigue on RT/errors.

Errors: in 1st block, higher for left-lateral patients. Not due to speeding. difficulty “task-setting”

Conclude: continuous performance != vigilance tasks: frequency of response.

2 processes needed:

  • maintaining a fast rate of response = “energising, establishing a high probability or low threshold for responding” Rt medial/superior frontal
  • setting the stimulus-response relationship = contention scheduling. Lt DLPFC


Rafal Posner Friedman Brain 1988

Orienting of visual attention in progressive supranuclear palsy [Lesion, Attention]

8 PSP vs 8 iPD. Exogenous (brightening of peripheral box) 50% valid vs endogenous (arrow) 80% valid, CTI = 50ms, 150, 350, 550ms. Press button when target detected.

Exogenous: Horizontal = early effect of validity, Vertical = only delayed effect of validity.

Endogenous: vertical has less effect of validity than horizontal

Conclude: “PSP subjects are impaired in orienting attention in the vertical plane in response to both exogenous signals and under endogenous control, but this impairment may be relatively more severe for exogenous than for endogenous orienting”


Poliakoff, Boyle, Moore...Spence Brain 126:2081 2003

Orienting of attention and Parkinson’s disease: tactile inhibition of return and response inhibition [Lesion, Attention, Behaviour]

Excellent review of Posner in PD.

McDowell & Harris 1997: doorways = capture of attention. Beneficial visual cues Morris et al 1996.

Negative priming: Downes et al 1991, Filoteo & Rilling 2002, Wylie & Stout 2002. Stroop: Henik et al 1993. PD faster than control at responding to previously ignored attributes. Briand et al 1999: increased oculomotor reflexive orienting? Antisaccades in PD: Crevits & De Ridder 1997, Armstrong et al 2002, Kinstone et al 2002).

Posner: Early studies of cueing in PD: no deficit. Rafal et al 1984: facilitation at short SOA <600ms, no effect of meds. Bennett 1995, Filoteo 2002: no deficit. Briand 2001: saccadic posner with SOA 67, 133, 1000: no difference between PD OFF and control, but IOR increases with PD disease stage. Klingstone 2002: IOR is normal magnitude at 288ms.

Reduced cue effect: Yamada 1990: no cueing effect with visual stimuli; Bradshaw 1993: reduced cost of uncued targets.

Reduced IOR: Filoteo 1997: reduced IOR at 1000ms. Yamaguchi & Koboyashi 1998: no decrease in facilitation at long SOAs. Pollux 2001: reduced IOR at 600ms.

Why was IOR apparently reduced? Cue-target vs target-target tasks? Absence of central reorienting cue? effect of prolonged cue? Endogenous component if predictive?

Expt: 24 iPD, 15 on combination, 17 on agonist, 4 on anticholinergics. 11 couldn’t do it: 6 dyskinetic, 2 tremor, 2 anxiety.

Central flash 100ms → 400 or 800ms delay →lateral vibrotactile precue 50ms → 1000ms delay → central flash 100ms → 400 or 800ms delay → lateral vibrotactile target 100ms. Therefore, SOA = 1400 or 1800ms. “cue-target task”: respond only to second stimulus, “target-target task”: respond to both. (control for motor inhibition effect). Response = right foot tap.

Controls exhibit IOR in both tasks. PD exhibit IOR only in the cue-target task but not target-target task. Magnitude smaller in PD.

Conclude: either 1) PD has no IOR but effect in cue-target task is entirely attributable to response inhibition or 2) PD have IOR but effect in target-target task is smaller because double-motor-response interferes with response to second target. Argue that 1) is more plausible

Why is IOR reduced in PD? Reduced inhibitory control (Filoteo 1997) vs hyperreflexivity (Jackson & Houghton 1995).

Mechanism of reduced IOR in PD:

1)      reduced inhibition of SC by BG.

2)      frontal/parietal output to BG reduced.

3)      impaired integration of sensory/context/expectation information within BG

4)      ? related to short-latency phasic DA for exogenous cues.

5)      Impaired gating by BG of input to motor regions → hyperreflexivity


Tipper, Rafal...Egly,Bruce JEP:HPP 23:5:1522-32 1997

Object-based facilitation and inhibition from visual orienting in the human split brain

3 collinear objects, rotated 90 degrees so they either cross the midline or horz meridian, or 180deg so that object-based and location-based effects compete.

Peripheral item 100ms flicker → 200ms → central item 86ms flicker → rotation during SOA 90=500ms, 180=842ms → 67% probe in one of the two peripheral squares → speeded button-press if probe.

After 180deg rot, the cue gives “object-based IOR” at the object’s new location, and “location-based IOR” at the old location. Variable in healthy controls: some have small object based IOR, or small location-based IOR.

90: 2 split brain patients showed object-based IOR when objects moved within a hemifield, but not between.

180: patients showed shorter RTs for cued object’s new location, more than its old location. Could be due to either 1) facilitation of object after crossing midline, or 2) location-based IOR.

Conclude: cortex necessary for object-based IOR?


Alivisatos, MilnerB Neuropsychologia 27:4:495-503 1989

Effects of frontal or temporal lobectomy on the use of advance information in a choice reaction time task [Lesion, Attention]

Epilepsy/tumour surgery. 7 left frontal, 14 right frontal, 24 left temporal, 24 right temporal. WAIS>80

Target shape discrimination at 1 of 4 locations. Precue 3s before target, with arrow or word.

Errors 2.2-2.6% (NC=1.8%).

Controls: positive correleation between neutral RT and size of invalid-valid effect. Interpreted as inability to use the warning / ‘get-ready’ signal.


Knight Hillyard Woods Neville EEG & Clin Nphys 52:571-82 1981

The effects of frontal cortex lesions on event-related potentials during auditory selective attention [EEG, Lesion]

Dichotic listening in 13 bilateral and 10 unilateral frontal patients, with ERP.

Frontal patients did worse. Attention-related negativity reduced bilaterally in L-frontal patients. Reduced on left side only in Rt-frontal patients.


Eslinger & Damasio Semin Neurol 4:3:385 1984

Behavioural disturbances associated with rupture of anterior communicating artery aneurysms [Lesion]

3 patients. 1) ventromedial haematoma: confabulatory anterograde and retrograde amnesia – spatiotemporally misplaced recollections. Happy and unconcerned about his deficits. Swearing, disinhibition, labile mood. 2) basal forebrain and right mesial frontal: long and short term confabulatory amnesia, anterograde>retrograde, with intact WM. 3) midline basal forebrain and bilateral symmetric ACC. Retrograde amnesia up to 10 years, reliving moments of past. “acts out on the basis of her false recollections and fabriations” amnesia + deficit in monitoring appropriateness of behaviour. Flat mood, unconcerned about deficits, paranoia.


Chao & Knight Neuroreport 6:1605 1995

Human prefrontal lesions increase distractibility to irrelevant sensory inputs [Lesion, Attention]

7 DLPFC strokes, 5 temporal lobe strokes, 5 hippocampal strokes.

Auditory WM change detection task. Blocks without or with ‘distractors’: tones in retention interval.

50ms warning tone → 1.2s gap → Environmental sound 700ms → 4 to 12 sec delay → environmental sound. 62% trials no-change. Distractor condition: env sound → 1sec → repeated [100ms tone + 400ms gap] → 1sec → env sound.


Conclude DLPFC critical for gating of irrelevant sensory inputs. Hippocampal deficit only at longer delays. “chronic leakage of irrelevant sensory inputs”.


Koski, Paus, Petrides Neuropsychologia 36:12:1363-71 1998

Directed attention after unilateral frontal excisions in humans [Lesion, Attention]

Surgical lesions: 6 left PFC, 11 rt PFC, 16 left ant temporal cortex, 12 rt ant temporal cortex, 10 left transcorical selective amygdalo-hippocampectomy, 10 right selectives. (=17 frontal, 48 temporal!)

50ms warning tone → 1s → cue (50% of trials = 100% informative arrow, or neutral ‘+’) → SOA 200-3300ms → peripheral target → simple response time (1 key).

No evidence for increased anticipation / fixation breaks...

  • Frontal: Mild impairment in using advance spatial cues to speed response to the target.
  • Cueing effect smallest at 200 and 800 ms ?why.
  • But impairment equally strong at all intervals.


Alivisatos Neuropsychologia 30:2:145-59 1992

The role of the frontal cortex in the use of advance information in a mental rotation paradigm [Lesion, Executive]

9 Lt frontal, 12 Rt frontal, 22 Lt temporal, 23 Rt temporal patients.

Is a letter correct or mirrored? At various rotations. Blockwise uninformed or informed: precued 1500ms with the unreflected (but rotated) letter.

Errors<8%. Conclude “activation of preparatory processes”, “efficient and flexible organisation of behaviour”.


Pierrot-Deseilligny, Rivaud, Gaymard, Agid Brain 114:1473-85 1991

Cortical control of reflexive visually guided saccades [Lesion, Saccade]

5 Lt PPC (angular gyrus), 5 Rt PPC (dl, area 46), 10 Rt PFC, 6 Lt PFC, 4 Rt FEF, 6 Lt FEF, 4 Rt SMA, 5 Lt SMA.

Prosaccade task & antisaccade task (blocked) with 200ms gap.

Conclude: FEF is not crucial for exogenous saccades. Is required for inhibiting them.

PPC angular gyrus (particularly Rt) is required – equivalent to monkey LIP, projects onto colliculus. Not a disorder of disengagement because of gap [?]


Smith & Milner Neuropsychologia 22:6:697-705 1984

Differential effects of frontal-lobe lesions on cognitive estimation and spatial memory

7 left frontal and 12 right frontal patients after epilepsy surgery. Also hippocampal patients.

Aim: Incidental learning of locations of objects in an array. Task – shown toy models, estimate price of the ‘real item’.

Hippocampal impaired at a long delay.


Rodriguez-Ferreorp, Cuetos...Ribacoba Movt Disorders 25:13:2136-41 2010

Cognitive impairment in Parkinson’s disease without dementia [Lesion, Attention]

50 patients. Basic neuropsychological testing (less detailed than queen square book), but with added “visual search” – pick out ‘3’ from a grid of 240 digits, and added semantic association (2afc which picture is associated with this one?).

Age affects everyone. MANCOVA with age and MMSE factored out. Visual search and semantic association predicted PD independently of those. Discriminant analyses with all tests: PPV 76% NPV 86%.


Fimm, Heber...Fromm Noth Movt Disorders 24:11:1631-20 2009

Deep brain stimulation of the subthalamic nucleus improves intrinsic alertness in Parkinson’s disease [Lesion, Attention]

13 patients ~60yo 1y post DBS, bilat sensorimotor STN & zona incerta, fields of Forel and pallidofugal fibers. LED=538mg (reduced by 56%). DBS 2.85V single monopolar, 64us @140Hz.

Turning OFF: motor examination score worsens between 0 and 30 minutes.

Battery: Word fluency, Stroop, WAIS, digit spans + corsi, auditory verbal learning, conditional associative learning test, trail-making, timed finger-tapping.

“alertness”: simple RT to visual ‘X’. blocks of uncued and tone-precued 600-1500ms.

“visual scanning”: for landolt Cs in 1 of 4 orientations, 25 item grid, “instructed to use a fixed strategy, to search line-by-line left to right”.

Simple uncued RT improves on DBS. Cued RT no change with DBS. This simple RT improvement is uncorrelated with motor score improvement. Also uncorrelated with timed finger-tapping.

Conclude: “intrinsic alertness” improved by DBS, “phasic alertness” is not.


Van Asselen, Almeida, Andre...Castelo-Branco Neuropsychologia 47:5:1269-73 2009

The role of the basal ganglia in implicit contextual learning: a study of Parkinson’s disease [Lesion, Learning]

Visual search task T among Ls, is the T rotated left or right? repeated spatial context information: 16 blocks, each block = 12 new configurations, and 12 configurations that were each repeated once in each block. In the latter, T & L positions were constant, but target orientation was kept random.

Builds over 5 trials, remains > 1 week, unaware (as measured by recognition task), dependent on medial temporal lobe.

Conclude: PD deficit in implicit learning of attending spatially. “contextual cueing”.


Charnov J Evol Biology 3:1:139-44 1990

On evolution of age of maturity and the adult lifespan [Theory]

How does Age of Maturity (= α) evolve?

Mean children born in life = R0 = fitness

= (probability of surviving to α) x (‘value’ of being age α)

= (infant mortaility proportion) x ()

φ is the adult mortality rate. φ, V and IMR must be functions of population density.

For a valid fitness measure, the population must be stationary, i.e. E[R0]=1, i.e. logA= φ(α)–logV(α).

An organism can change its α. So maximise R0 with respect to α.

If φ(α)=Zα, i.e. constant adulty mortality over time and , then ↑Z gives ↓α.

Value V = (fecundity)/(yearly adult mortality) = b/q. If V changes nonuniformly with respect to age has no effect on optimal α.

Conclude: the effects of population density might not affect the optimal maturity age.


Peccei Ethology & Sociobiol 16:5:425 1995

The origin and evolution of menopause: the altriciality-lifespan hypothesis [Theory]

Menopause offers fitness benfits by trade-off between maternal investment in existing progeny and continued reproduction. Results in increases in maternal care → improved offspring survival and fertility.


Promislow, D J Theoretical Biol 170:3:291-300 1994

DNA repair and the evolution of longevity: a critical analysis

Previously thought ↑DNA repair rate → ↑ lifespan, or need for longevity improves DNA repair. Repair rate = Aek.mass.

Claim: DNA repair and lifespan both correlate with size.

Argues that larger organism, with same lifespan, requires higher DNA repair rate: a single mutation can be fatal, so ↑cells → ↑repair rate.

Also size promotes longevity from ↓growth, ↓predation, starvation, immune systems..

Cannot distinguish from correlation, whether increased size is a causal intermediate, in either direction, or if it is a true confounding variable (i.e. causes both longevity and DNA repair).

Also, could be any other variable! body size correlates with brain size, BP, pancreas mass, population size.


Zhou et al Neurosci Lett 509:1:50-5 2012

Selective attention deficits in early and moderate stage Parkinson’s disease

44 iPD. MMSE verbal fluency, digit span.

attentional network test”: precue 400ms SOA: no cue, central / double precue (spatially uninformative warning), spatial exogenous cue (100% valid). Then target = L/R arrow flanked to left & right by congruent / neutral / incongruent arrows. Target is above or below fixation, random. Respond with L/R hand. Target visible until response.

6 trials of each of 48 conditions: 4 cue types x 2 target location x 2 target direction x 3 congruencies.

‘alerting’: RT(no cue) – RT(double cue) = benefit of warning signal

‘orienting’: RT(centre cue) – RT(spatial cue) = benefit of exogenous 100% valid spatial cue = ability to utilise information from brief prior cue

‘executive’: RT(incongruent) – RT(congruent) = incongruence cost of simultaneous flankers = ability to select out the middle item

Full-size image


Result: patients have bigger benefit from spatially informative precues. No difference in warning effect or congruence effect.


Price Filoteo Maddox

PD deficits in rule generation, maintainance, shifting, and selection.

Medication worsens generation, particularly in early PD.

Rule maintainence in WM is not impaired, except when interference from other previous rules is high (‘requiring greater selective attention to the current rule’...) – which may be drug-related.

Shifting difficulty (perseveration) may be critically impaired when extradimensional shifts are required to dimensions that were completely irrelevant (Slabosz et al. 2006). Medication seems to help with shifting – possibly by increasing distractibility, but they still have problems selecting the new rule if it has no history of reinforcement.

Selection of new rule is impaired by 1) enhanced latent inhibition when Off, and 2) may ineffectively use feedback when On or overmedicated.

Conclude striatal dopamine is necessary for rule shifting, but off patients are less distractible..


Briand, Hening, Poizner, SerenoA Neuropsychologia 39:11:1240-9 2001

Automatic orienting of visuospatial attention in Parkinson’s disease

Speeded saccades to a target, unpredictive (50% valid/invalid) exogenous cues at 67 / 133 / 1000ms.

Full-size image Full-size imageC=cued, U=uncued.

  • “Patients faster than controls in terms of initial facilitation following reflexive cues”
  • Cueing effect negatively correlates with UPDRS.
  • More advanced patients have greater IOR.


Troche, Trenkwalder...Rammsayer Neuropsychologia 2006

Unimpaired negative but enhanced positive priming in Parkinson’s disease: evidence from an identity and a location priming task

Identity priming: 4AFC visual digits Erikson flanker. Two fingers of each hand. ‘Prime’ was essentially an extra trial before each trial; identical, control, or target→distractor, or distractor→target.

Location priming: 4AFC locations arranged trapezoidally on screen mapped to fingers. 1 target, 1 distractor, 2 blanks. Prime (like another trial, requiring a response) could be identical, all new, or distractor→target (‘ignored repetition’), or target→distractor.

Full-size image Full-size image Full-size image

location NP location pos priming identity priming

  • PD → ↑Identity and location positive priming
  • Controls and PD both have significant location NP, but no identity NP.
  • Identity NP proportional to severity of disease.


Stout, Wylie, Simone, Siemers 2000

Influence of competing distractors on response selection in Huntington’s disease and Parkinson’s disease

Touchscreen, Shape matching. Central white cue shape (1 of 4), plus 1 or 2 coloured shapes at any of four corners of screen. 1300ms ITI. Prime-probe pairs of trials: colour / shape / location of prime could match / notmatch the probe.

LSC = Location / shape / colour.

  • PD showed NP in location*shape and location*shape*colour priming.
  • HD had no significant NP.


Filoteo & Maddox Neuropsychology 13:2:206-22 1999

Quantitative modelling of visual attention processes in patients with Parkinson’s disease: effects of stimulus integrality on selective attention and dimensional integration

Stimului with two dimensions: 2 perpendicular lines’ lengths, or an oriented line (angle and length). Categorisation task: 2 categories, each Gaussian for each dimension.

  • ‘Selective attention’ condition: only 1 dimension is relevant for categorisation.
  • Dimensional integration’ condition: category depends on a linear combination of the two dimensions.
  • ‘Baseline’: stimuli with only one dimension

Optimal performance = 97% (Gaussian overlap)

PD show increased noise in criterial decision, whether it be selective attention to one feature or dimension-integration of two features.




Kirkwood and Rose Phil trans roy soc B 332:15 1991

Evolution of senescence: late survival sacrificed for reproduction

“Antagonistic pleiotropy” (population genetics): alleles with early beneficial effects also have later deleterious effects

”Disposable soma” (physiological ecology): reproduction is costly, so less effort available for maintainence of somatic tissues.

Semelparity vs iteroparity. Cole’s paradox: for infinite lifespans, the cost of moving from semelparity to iteroparity is equivalent to an increase in litter size of 1. (i.e., parent dies!)

Verhulst model: where K is “carrying capacity” of environment.




Botha & Carr Parki’ism & Rel Dis in press 2012

Attention and visual dysfunction in Parkinson’s disease [Lesion, Theory]

Review of dopamine in attention.

Retina: dopaminergic modulation (Cardoso et al Mov Dis 2010)

LGN: D1 and D2 receptors, independsnt in mango/parvo/konio.

V1: DA innervation of lamina VI. May alter centre-surround / spatiotemporal tuning

PD deficits: feature integration, inability to resolve visual ambiguities.

Visual hallucination = impaired recognition/integration + suboptimal WM → “disturbance of external vs internal representations, or between the dorsal attention network and the default mode and ventral attention networks”.

DA as 1) biasing selection in ventral stream, 2) gating entry to consciousness & WM.


Sampaio, Bobrowicz ...Castelo-Branco Neuropsychologia 49:1:34-42 2011

Specific impairment of visual spatial covert attention mechanisms in Parkinson’s disease

Staircase measurement of contrast sensitivity threshold [PD, Attention]

4 quadrants. 100ms SOA exogenous precue 50% central/peripheral. Target also central / peripheral. Cue is 33% valid / neutral / invalid.




a)      11% patients could not fixate → excluded

b)      PD flattening of spatial asymmetries

c)      PD did not benefit from cue.


Kingson, Klein...Maxner J Clin Exp Npsychol 24:7:951-67 2002

Orienting attention in aging and Parkinson’s disease: distinguishing modes of control

“Some investigators have reported that PD does not produce any corresponding dysfunction in attentional orienting (Lee, Wild, Hollnagel, Graffman 1999; Rafal Posner Walker & Friedrich 1984), whereas others have reported that attentional dysfunctions do indeed occur (Hodgson, Oittrich Henderson & Kennard 1999; O’sullivan shaunak hawken Crawford & Kennard 1997).”

Lieb et al 1999: Evidence of PD deficit in reflexive but not volitional attention

Shaunak et al 1999: evidence of PD deficit in volitional but not reflexive

Briand et al 1999: evidence of PD deficit in inhibiting reflexive and generating volitional

→ Disinguish overt/covert from reflexive/voluntary.

Expt1: Central cue 80% predictive: 40% L 40% R 20% neutral (50/50); → SOA 288 or 576ms → peripheral target 80%, catch trials 20% → button press if target appears. “volitional covert”.

“PD orient covert volitional attention in a healthy and normal manner”

Expt2: Exogenous precue uninformative L/R/mid → 72ms or 288ms → target 80% or catch 20% → button press if target. “reflexive covert”.

“reflexive orienting at the cued location was the same as control” incl IOR at 288ms.

Expt3: central arrow cue → saccade to direction of arrow “volitional overt”

Hypometric but normal latency.

Expt4: Peripheral go cue → saccade to target. 50% of trials had 200ms gap (fixation offset) before target.

PD faster than controls for both no-gap and gap conditions.

Expt5: Peripheral go cue → antisaccade to other location. Also had gap on 50% trials.

PD same as controls

Conclude: PD faster reflexive orienting. Requirement of antisaccades eliminates this.


Rafal Posner Waler Friedrich Brain 107:1083-94 1984

Cognition and the basal ganglia: separating mental and motor components of performance in Parkinson’s disease

10 PD patients On and Off.

Expt1: ‘scanning of STM’: Sternberg paradigm, set size 1/2/4/6. No difference in error rate.

Expt2: covert orienting: peripheral cue 80% valid → SOAs 50, 150, 550, 1000ms. Cue lasted 300ms[!] → target (remains on) → single keypress whenever target seen. [no catch trials]

Expt3: Manual choice RT: peripheral precue 80% valid → SOA 50, 200, 350, 500ms → target → move heavy lever to side of target.

Filled = Off, open = On. Valid=circle, invalid=sqr.

Result: Although motor effects are slower, mental preparation time and exogenous cueing time is not longer.

Choice RT prolonged more than simple RT.

→ evidence for bradykinesia but no bradyphrenia


Bočkova, chldek, jurk et al J Neural Transm 118:8:1235 2011

Involvement of the subthalamic nucleus and globus pallidus internus in attention [PD, Neuron, Attention]

7 iPD with STN DBS. Recording from the 4 STN electrodes!

Visual oddball paradigm, 3 types of stimulus: frequent nontarget 70%, rare target 15%, rare nontarget 15%. Button press to target.

Late positive potential 500ms linked to target processing

Distractor – early short positive peak 200ms

Simultaneous early distractor-specific potential in GPi and STN

Event-related (de)synchronisation.

Not seen in surface EEG.


Hlbig, Assuras...Olanow Movt Dis 26:9:1677-83 2011

Differential role of dopamine in emotional attention and memory: evidence from Parkinson’s disease [PD, Attention]

15 PD On & Off. Viewed pictures (emotional valenence +/0/- x arousal hi/lo, IAPS).

Rating of experienced valence and arousal. Delay 10 min, then 50% old/new judgement.

Recognition accuracy & RT.

PD Off: Reversed valence selectivity for emotional memory.

Dopamiergic treatment blunts the arousal effect on reaction time: No longer get the normal slowing for high-arousal pictures.


Solis-Vivanco, Ricardo-Garcelli... Neurosci Lett 495:2:144 2011

Involuntary attention impairment in early Parkinson’s disease: an event-related potential study

42 PD

90% 1000Hz, 10% 900Hz deviant. 50% 200ms 50% 400ms. Task: judge duration, 2AFC buttons for 200/400ms.

“Distraction potential” = (all deviant – all frequent)

“mismatch negativity” = largest negative wave within 100-250ms after tone: not reduced.

P3a = largest positive wave between 250 - 400ms: lower amplitude. Normally corresponds to the automatic orientation of attention.

Re-orientation negativity = maximal negative wave between 400-700ms : enhanced by medication.



Baddeley, Della Salla, Papagno, Spinnler Neuropsychology 11:2:187 1997

Dual-task performance in dysexecutive and nondysexecutive patients with a frontal lesion

32 frontal patients, clinically classified as dysexecutive or not. 24 with agreement of 2 raters.

7/12 dysexecutives had apathy / disinhibition.

Single/dual task WM: digit span +/- visuomotor ‘tracking’ = crossing out a chain of boxes laid in a path. Span measured first, and the experimental task used the maximum they could manage.

Word generation on one letter, and modified WCST.

Argues that dual task score is best predictor of clinical dysexecutive syndrome.

Not WCST or verbal fluency.


Cowey & Greene Memory 4:1:19-30 1996

The hippocampus: a “working memory” structure? The effect of hippocampal sclerosis on working memory

Hippocampal sclerosis vs frontal lesion (epilepsy cohorts).

Letter span single task, letter span + visuomotor tracking (cancelling a string of boxes on paper), visuomotor task alone.

Letter span was measured first, and the experimental version used the maximum length they can manage.

Result: Only frontal patients show dual task cost. No difference L/R.


Dalrymple-Alford, Kalders, Watson JNNP 57:360-7 1994

A central executive deficit in patients with Parkinson’s disease [PD, Executive]

8 iPD, Digit span +/- concurrent tracking –cursor on screen with a joystick, moving in lissajous figure. Cursor size adjusted to keep error rate constant.

“patients and controls had similar digit spans”

Tracking error “was greater in the patient group, but this difference did not approach significance”

Dual task cost did not correlate with disease severity. Age was correlated with difficulty of tracking, education was correlated with WCST, digit span correlated with word fluency.


Brown & Marsden Brain 111:2:323-45 1988

Internal versus external cues and the control of attention in Parkinson’s disease

Cued stroop, uncued stroop, control task.

Stroop – alternating 10 trials of Ink vs Word. Cued: reminded on each trial ‘INK’ or ‘WORD’ for 1s. Uncued: just ‘READY’ before each trial.

Control task: either words written in white, or squares in colour. (no conflict).

Control = open, PD = filled. Sqr = uncued, circle = cued.

PD impaired on WAIS block design and WCST. “Impaired attention” in PASAT. Unimpaired on WAIS vocabulary.

Slower than controls in all RT tasks. Specific impairment on uncued Stroop.

PD more errors than controls on Stroop uncued, but not specifically on alternation trials. No effect of time on task. PASAT (paced auditory serial addition task) attention not correlated with Stroop.


Davidson, Gao, Mason & Winocur JNNP 30:1:18-32 2008

Verbal fluency, trail making, and Wisconsin card sorting test performance following right frontal lobe tumour resection

Good Review of literature on Verbal fluency, trail making, and WCST.

Verbal fluency more affected by L frontal. Stuss 1998: Rt produces greater deficits in category than letter fluency. Inf medial PFC no effect. Troyer et al 1997: clusters (of semantically or phonologically related words), then switching (within the domain or to new domain of similarity); DLPFC switching deficit.

Trail-making: Stuss 2001, Demakis 2004: DLPFC but not lateralised, and not OFC. Stuss 2001 suggests B requires inhibition etc.; Demakis 2004: actually Frontals more impaired at A than B; but A is always administered first.

WCST: infer the current sorting rule by ‘correct’/‘incorrect’ cues. Categories achieved and perseverative errors. Milner 1963: DLPFC patients set-shift 2-3 times then perseverate. Not seen in OFC/parietal/temporal. Stuss 2000 and Demakis 2003 agree. Stuss 2000: DLPFC or superior medial damage in either hemisphere, but not OFC. OFC patients lose set before completing a category.

Study: 20 Rt frontal patients. Classed with % damage to Premotor, Anterior, Inferior, Dorsolateral, Broca, Cingulate.

But could not find any regional specificity. Trailmaking A&B same.

Chemotherapy: Wefel, Lenzi, Theriault, Davis, & Meyers, 2004; for reviews, see Anderson-Hanley, Sherman, Riggs, Agocha, & Compas, 2003; Minisini et al., 2004; Tannock, Ahles, Ganz, & van Dam, 2004; Taphoorn & Klein, 2004.


Roca, Parr, Thompson...Duncan Brain 133:1:234-47 2010

Executive function and fluid intelligence after frontal lobe lesions

42 frontal patients (tumour, vascular).

Culture-fair gi, WCST, verbal fluency

Then ‘ineco’ frontal screen:

  • Luria fist/edge/palm. 3 times with experimenter, then 6 times alone.
  • Interference task: “I tap once, you tap twice”. 1-1-1-2-2-2-1-1-2-1-2-2-2-1-1-2
  • Go-nogo: after interference task. Same but 2=nogo.
  • Backwards digit span, Months backwards, reverse-Corsi 4 blocks N=2,3,4,5
  • Proverbs x 3, explain + example
  • Hayling (high cloze sentences, complete ‘correctly’ x 3, then irrelevant word x 3)

Hotel task: sample 5 activities in 15 minutes, plus 2 self-timed events.

IGT: 2 risky decks, 2 conservative.

Faux pas: written and spoken 1-paragraph story, x 20. 50% had a faux pas. Judge “whether something inappropriate was said, and if so, why it was inappropriate”. Story repeated if they failed a memory question.

Mind-in-the-eyes: photograph of eyes, 2AFC what are they thinking/feeling?

Divided L/R lateral, inf medial, sup medial. No effect of location!

Conclude: WCST, verbal fluency, IGT → completely explained by g

After removing g, then go-nogo, proverbs, hayling, hotel, faux-pas remain significant. But no specific area.

Torralva 2007: IGT is ventromedial. Roca doesn’t find this. [very long discussion]


Duhamel, Goldberg...Grafman Brain 115:1387-1402 1992

Saccadic dysmetria in a patient with a right frontoparietal lesion: the importance of corollary discharge for accurate spatial behaviour

Double-step task: shows that retinotopic coding is inadequate to explain saccades

Present two targets A&B before the first saccade. Task – to look to A then B in turn.

Single patient, right frontoparietal lesion, left homonomous hemianopia.

Simple prosaccades: Slower + smaller saccades to left hemifield.

Task: fix offset & target A 100ms → target B 80ms → blank. “both saccades visually triggered but not visually guided”.

“marked deficit in compensating for prior contralesionally directed saccades” in a sequence.

→ frontoparietal cortex critical for performing spatial computations involving retinal information and information corollary to eye movements.


Georgiou-Karistianis, Farrow et al Cogn Beh Neurol 25:1:1-6 2012

Deficits in selective attention in symptomatic Huntington disease: assessment using an attentional blink paradigm [Lesion, Attention]

Standard RSVP. HD had poorer T1 accuracy. T2: larger AB but not longer.

HD had fewer post-target intrusion errors, and more random errors.


Vardy, Bradshaw, Iansek JCEN 25:3:361-75 2003

Dual target identification and the attentional blink in Parkinson’s disease

13 iPD. Modification of standard AB: “baseline measure [T1 detection only] is only useful I so far as participants can successfully ignore the first target”. Also PD patients “have difficulties in extradimensional shifts”, so no point in having 2 types of target.

→ 2 red target letters, 10Hz

No significant impairment. PD ↑pretarget intrusions; ? “getting stuck with the old information” / “stuck-in-set difficulties”.


Richer, Decary, Lapierre et al., Brain & Cogn 21:203-11 1993

Target detection deficits in frontal lobectomy

5 Rt, 2 Lt frontal lobectomies, compared with 10 temporal lobectomies. (all epileptic)

d2 test (Brickenkamp 1966) = character cancellation, cancel target letter in row of 40 letters; targets ~45-55%. 3 conditions: 1 target letter + 1 distractor type, 1 target + 3 distractors, 3 targets + 5 distractors.

Stroop with both no-conflict controls.

Ancova factoring out performance on (1+1) condition: frontal much worse at C than B. → multiple target detection requires frontal.

Frontal also slower at Stroop when no-conflict speed is factored out.


Goel & Grafman Neuropsychologia 33:5:623-42 1995

Are the frontal lobes implicated in “planning” functions? Interpreting data from the tower of Hanoi

“tower of H is best explained by postulating a goal-subgoal conflict resolution difficulty.” – not “planning or look-ahead abilities”. But consistent with problems in WCST, Stroop, antisaccade, A-not-B, delayed-alternation.

“the planning issue is independent of the goal-subgoal conflict issue”: planning = task where moves cannot be reversed.

tower of H is quite different to tower of L: ToL is three discs, and pegs can accommodate 1,2 and 3 discs respectively, and any stacking order possible.

9 frontal patients. Also did WAIS-R, WMS-R.

Task: 5 discs different sizes, 3 pegs, aim to stack all on middle peg. Keypad 1,2 moves disc from 1 to 2. 9 starting configurations of varying difficulty.

Dependent measures: number of moves, number of illegal moves, reversals, moves remaining to goal (if incomplete), speed.

Patients worse, but not correlated with WAIS-R or WMS-R. Moderate correlation with WCST.

Simon & Newell, strategies = ‘goal recursion’, two perceptual strategies, and ‘move-pattern’ strategy... Patients and controls use the same strategies – i.e. a perceptual strategy.

“When patients do solve the problems, they do so as well as controls”

Why do they fail? When subgoal stack requires a counterintuitive ‘backward move’ for successful completion. Interaction between patient vs control and performance subgroup.

Cf inhibition of prepotent responses: = conflict between satisfying global goal (prepotent) and local goal (appropriate).


Howes & Boller Brain 98:317-332 1975

Simple reaction time: evidence for focal impairment from lesions of the right hemisphere

29 Lt and 20 Rt hemisphere lesions. Controls / dominant / nondominant hemisphere.

100 clicks spaced by 4,5,7,10,15s. Simple RT with ipsilesional hand.

RT nondom = 635ms, dom = 334ms, control = 215ms. True at 5th and 95th percentile.

Time on task: nondom had no improvement.

Longer ISI: controls fastest for intermediate ISI. Nondom: get faster as ISI increases.

Nondom: much greater inter-individual variability; no correlation with lesion size


Godefroy, Cabaret, Rousseaux Neuropsychologia 32:8:983-90 1994

Vigilance and effects of fatigability, practice and motivation in simple reaction time tests in patients with lesion of the frontal lobe

11 ACOMs, tested in 6 sessions, unimodal and bimodal vigilance tasks

5, 10, 15second ISI, targets and distractors, lasts 25 min. Right hand key press to targets.

Vigilance – patients ~90ms slower, but no decrement through the task.

Fatiguability – when coming to a task at the end of a session: ~20ms for unimodal, ~70ms for bimodal (~=twice normal)

→ significant fatiguability, mild practice effect, slight motivation effect, and no change in vigilance over time. But – no interactions with patients!


Shallice Stuss Alexander Picton Derkzen Cortex 44:7:794-805 2008

The multiple dimensions of sustained attention [Lesion, Attention]

43 focal frontals = 11 Lt lateral, 6 Rt lateral, 15 inferomedial, 11 superomedial. (see 2005)

NART, forward/backward span, token test, boston naming, BDI

Estimate number of auditory stimuli on each trial: 8-22 brief tones, interval 230-280ms (fast) or 2500-3500ms (slow). Instruction “there will be between 5 and 25 stimuli”.

slo~0.3Hz, fast~3Hz

Conclude: Slow sustained counting judgements were impaired following lesions to superomedial PFC. Fast sustained counting is impaired by both superomedial and Rt Lateral patients. Rt lateral worse with longer trains.

patients with SM lesions underestimate rapidly presented trains of stimuli


Correani, Humphreys Cogn neuropsychol 28:5:363-85 2011

An impaired attentional dwell time after parietal and frontal lesions related to impaired selective attention not unilateral neglect

11 parietal (some aphasic, most with extinction – 3 are left hemisphere!) + 9 frontal (mixed, including temporal lobes in 4 cases!) [awful cohort].

Attentional blink paradigm, ISI 50, 150, 450, 1350ms. Expt2 with mask, so equated duration of exposure to stimuli at different rates.


Expt3: report colour of T1, and shape for T2 (or vice versa).

Correlates with Birmingham cognitive screen attention score: Selective attention = respond to 3 words but not to other 3. Sustained attention = same task but performance decrement across blocks.

Patients generally impaired. Could be due to filtering T1 even when T2 had to be reported. → “poor target selection”.

No difference between frontal / parietal.


AlvarezJ & Emory Npsychol Rev 16:1:17 2006

Executive function and the frontal lobes: a meta-analytic review

WCST: 10/16 studies suggest that frontal worse than nonfrontal. 2 find no frontal deficit. 4 no difference of frontal to diffuse/BG lesions. 5 failed to find differences between frontal and nonfrontal. – “WCST is a sensitive but not specific marker of frontal lobe damage”




Uretzky & Gilboa JoCN 22:12:2745 2010

Knowing your lines but missing your cue: rostral prefrontal lesions impair prospective memory cue detection, but not action-intention superiority

Small right frontopolar lesion.

Expt1: Word/nonword 2afc, done after memorising instructions for two practical tasks, until they could recite it, then being told which one they would have to perform later. Also tested after having done the task (in separate session). Words = 50% nonword, 16% new word, 16% words from the to-be-performed task, 16% words from the not-to-be-performed task.

Took more repeats to retain the task, could not remember which of the two sequences needed to be performed.

Patient shows normal facilitation for to-be-performed words, and inhibition of the not-to-be-performed words.

Expt2: Semantic task, 2 words shown, categorically similar or not? 2AFC. Simultaneous dual task: monitor for a target – either one of the words (‘focal’ as it is the focus of attention in task 1) or a syllable within one of the words (‘nonfocal’), in different blocks. 4 targets in 120 items = 3%.

Patient detected only 1 of 4 focal targets (RT=12sec), and 0 of 4 nonfocal targets. Controls detect 90% with RT ~2sec.

Patient did not show the normal RT cost in nonfocal-monitoring blocks over focal-monitoring blocks.

Conclude: “neuropsychological dissociation between preserved privileged representation of prospective intentions and impaired detection of cues that support the opportune recovery of prospective memory”.

Stats: Friedman nonparametric ANOVA, and Crawford & Garthwaite 2002 modified t-test with sqrt((n-1)/n) correction , and revised standardised difference test.


Ibaez Gleichgerrcht Manes Brain struct func 213:397-410 2010

Clinical effects of insular damage in humans

Intraoperative stimulation of Rt insula = tachycardia + pressor; Lt insula = bradycardia + depressor (more often) (Oppenheimer 1992). In animals: similar but rostrocaudal asymmetries.

5 papers report gustatory deficit after insula damage ? ipsilateral tongue. 1 report of decreased olfaction – contralateral unpleasant odours less strong.

Bamiou 2006: reduced auditory temporal resolution and sequencing. Duration pattern judgements;

Brocas aphasia from Lt insula (5 studies) ? speech region ? disconnection. PET shows activity in verbal WM. Left insula patients have poorer immediate and delayed verbal memory in WMS.

Animal granular insula is somatosensory. Human: 3 cases of tactile agnosia (but included PPC). 1 case - Somatic hallucinations / somatoparaphrenia. 3 cases Rt insula = hemisensory deficit, hyperaesthesia, graphaesthesia, stereognosis, body awareness.

“Posterior insula is part of the human vestibular cortex”

“pain matrix” = somatosensory cx, insula, ACC, thalamus, PFC.possible hypo or hyperalgesia in lesions.

Karnath & Dietrich 2006: Rt insula important for neglect. Anteroventral = limbic (amygdala olfactory/gustatory, visceral inputs) lesions disrupt arousal / attention / activation.

Animals stop learning taste aversion after insula lesion. 1 pt bilateral insula lesion → inability to recognise facial expression of disgust. 1 pt left “posterior part of anterior insula” → reduced experience of disgust.

Anergia, underactivity, tiredness, but not depression. ? similar to bilateral ACC ablation. But many studies show reduced insular activity in depression.

Rat lesion reduces craving and malaise in addiction. Posterior insula lesion patients much better at quitting smoking! (Naqvi 2007).

Helmuth, Mayr & Daum Neuropsychologia 38:11:1443-51 2000

Sequence learning in Parkinson’s disease: a comparison of spatial-attention and number-response sequences

SRT-like task: Digits 1-4 appear in 4 locations around fixation. Respond to identity of digit.

Types of sequence: repeating locations, repeating digits.

Block 1: both repeating, 9+10: identity repeating only, 13+14: location repeating only.

Repeat afterwards with locations widely spaced.

Patients are slower when the spatial locations are randomised, but controls are slower when the identities (motor mappings) are randomised. → Patients learned locations better than controls (when locations close together). & Patients were impaired at either stimulus-identity or motor-pattern learning.


Schutz, Trommhauser, Gegenfurtner PNAS 109:19:7547-52 2012

Dynamic integration of information about salience and value for saccadic eye movements

Task: saccade to a visual target = gradient dark – light on a grey bg.

Vary the contrast of the 2 endpoints.

Baseline = no reward; saccades shift towards highest-contrast end.

Control = only one half of gradient shown;

Reward only = dark polarity for 2p;

Reward+Penalty = dark +2p, light -2p – if the high-contrast region is penalised, saccades shift away from high-contrast end.

Results: RT-vs-deviation curve for Varying contrast (shade of green): A: no-reward = constant effect of contrast over time. B: reward for black – effect of reward increases over 250ms, diverting away from hi-contrast. C: reward black + penalty white: early abolition of contrast effect.

model’: weight of value is a function of time – cumul gaussian(t). Gives optimal direction that maximises EV, given each observer’s variability.

Corrective saccades latency 180ms, influenced by direction of reward/penalty; fewer corrections in no-reward condition. “once the value map is available, it is used automatically, even though secondary saccades did not lead to rewards”.


Hosokawa, Watanabe JNeurosci 32:22:7662-71 2012

Prefrontal neurons represent winning and losing during competitive video shooting games between monkeys

Shoot another turret by moving a joystick in the right direction. Can shoot repeatedly. Sound at end of trial plus visual signal of which side wins.

4 conditions playing against: nobody, computer turret, nobody with with another monkey present watching, another real monkey.

Matched difficulty so that always 50% win (e.g. speed of bullet)

DLPFC: cells selective for win/lose

Cells selective for competitive vs noncompetitive

Selective for presence of another monkey

When competitive, cells selective for opponent=monkey or computer.

No difference in EOM. Improved accuracy when opponent is monkey, and when competitive.

Win-competitive and lose-competitive activity – may reflect “expecting/obtaining a reward whose value may have been enhanced in the competitive condition”.

Some cells had competition-enhanced activity irrespective of win/lose - ? encode motivational context information.


Wessel, Daneilmeier.. Ullsperger J Neurosci 32:22:7528-37 2012

Surprise and error: common neuronal architecture for the processing of errors and novelty

Erikson flanker task with letters. Separate with EEG and fMRI.

Task: Constant response mapping H or Z == left, S or X == right.

i.e. flanker could be same/different letter, and different letter could be compatible / incompatible response.

10ms after response → symbol which could be standard, rare target (instructed, appears only 3 times in exp), or completely novel. Instructed target = press a third response button. The novel stimuli were matched to the Ss error rate, separately for congruent/incongruent.

Behaviour: 13% errors, post-error slowing ~30ms, post-novelty slowing ~30ms. Same for current-compatible / incompatible, and previous-compatible / incompatible.

ICA of EEG – which component best explains error-related activity? Can this IC also explain novelty-driven deflections?

fMRI GLM for voxels affected by errors vs novel events.

Dipole suggests mPFC.

(error > correct) = anterior middle cingulate, preSMA, inferior frontal junction, ant insula

(novel > standard) = preSMA, inf frontal junction, ant insula, hippocampus, parahipp, amygdalohippocampal area, visual cx, sup parietal cx.

(standard > novel) = 0.

(Error & novel) = ant middle cingulate, preSMA lateral PFC, subcortical nuclei, TPJ, STS.

(error>novel) = ant medial cing, preSMA, ant insula.

(novel>error) = OFC, medial temporal lobe, visual cx.

Conclude: error-monitoring, reinforcement-learning etc not sufficient to explain findings.


Ho, van Maanen...Wagenmakers, Serences J Neurosci 32:23:7992-8003 2012

The optimality of sensory processing during the speed-accuracy tradeoff

Task: 6Hz grating, either constant or alternating (5 deg offset) orientation, visible 3s. Respond ‘match’ or ‘mismatch’; RT any time after stim onset. Emphasise accuracy (correct=+10, error =-10) or speed (correct=+10, error=0, too slow=-10). [encourages blind guessing!]



Response time (ms)

Speed (match)

71.37 2.71

947.7 64.2

Speed (mismatch)

61.08 3.08

896.8 49.4

Accuracy (match)

90.00 1.78

1697.8 131.2

Accuracy (mismatch)

82.05 3.28

1443.9 92.9

Speed (match plus mismatch)

66.23 2.24

922.4 40.1

Accuracy (match plus mismatch)

86.02 2.00

1570.8 82.6

Match (speed plus accuracy)

80.69 2.40

1322.8 101.7

Mismatch (speed plus accuracy)

71.57 2.99

1170.4 73.7

Model: 8 LBA models: vary ‘A’ (flatly distributed prior), b (threshold), v (mean rate). Kept σconstant. “to avoid having the model results overly biased by contaminant processes such as guessing... mixture process with 2% guessing (random responses with uniform RT over observed range)” [awful! Ratcliffe and Tuerlinckx 2002] → v, A and b all vary with SAT (BIC x 1010). Trial-to-trial estimate of drift rate and starting point by MLE:

grey = AE, black = SE

Forward encoding model: Retinotopic map localiser. Estimate basis functions for each orientation feature in 5-deg. Train weights on 8 out of 10 trials

“channel responses estimated on each trial (in C2) are constrained by the estimated weights assigned to each voxel and by the vector of responses observed across all voxels on that trial in the test set”. “unique channel response estimates can be obtained on a trial-by-trial basis as long as the number of voxels is greater than the number of channels”.

Result: When accuracy is emphasised, channel responses are amplified around 50 degrees away from the stimulus, on correct trials compared to error trials. Not found in speed-emphasis.

Correct/incorrect logistic regressors best predicted by channels at 60 deg from stim.

If Trial-by-trial rate-of-rise is used as a regressor, channels around 40 deg correlate with v.


  1. “correlation between activation levels in off-target channels and the rate of sensory evidence accumulation”
  2. “instruction-dependent change in the reliance of decisions on off-target channelse in V1”.
  3. “SE tuning functions resemble those in correct-AE trials” → “poos performance on SE trials = failure to optimally rely on informative populations”.


Ballard, Aarsland...Brown Tovee Neurology 59:11:1714-20 2002

Fluctuations in attention: PD dementia vs DLB with parkinsonism

50 pathents with DLB, 50 PD with dementia, 50 nondemented PD, 80 alzheimers.

Cognitive drug research battery: SRT, CRT, digit vigilance (RSVP with single target).

groups: DLB, PDD, PD, AD, control.

DLB and PDD were similar: slower, greater fluctuations, worse vigilance.

But nondemented PD are not different to AD in vigilance.

DLB greatest fluctuation in CRT.

DLB without PD were much better.


Fan Rossi & Yin J Neurosci 32:16:5534-48 2012

Mechanisms of action selection and timing in substantia nigra

Task – rats press lever for at least a certain duration. Trained on progressively longer intervals – 20ms, 400, 800, 1600ms.

Behaviour: faster to start next trial after a failure.

DAergic neurones identified by quinpirole suppression (D2-autoreceptor agonist) and PCA.

Both GABA and DA neurones can be either action-on and action off.

When the press was short, animal rarely enters the food cup = “low confidence rating” in getting the reward [but could have heard the pellet fall!]. Cells that fire more after longer presses and less after shorter presses. But sensitive to relative duration compared to criterion can explain whether the animal enters food well or not [but could be just motivation etc]

Conclude: most neurones that increase or decrease for movement are GABA, but DA do it too. Action-off inputs from striatonigral ‘direct pathway’. Action-on inputs = either disinhibition by GPe (but inconsistent with high baseline firing of GABA neurones), or glutamatergic excitation from STN. Unsure of significance: SN GABA output disinhibits thalamus and motor midbrain/brainstem - ?antagonistic muscles.


Roesch, Bryden, Cerri...Schoenbaum J Neurosci 32:16:5525-33 2012

Willingness to wait and altered encoding of time-discounted reward in the orbitofrontal cortex with normal ageing

Aged rats exhibit deifits in reversal and delayed-nonmatching tasks that resemble OFC damage: Eichenbaum 1992.

Rats age 4/12 or 24/12 trained on ‘temporal discounting’: 3 odor cues instruct goL, goR, or free choice (33%). Manipulate: L/R gave 2 different delays with fixed reward, or two different rewards with fixed delays. Delay blocks were staircased to find 20% choice preference. Reward blocks were 1 vs 2 drops.

Old rats show less delay discounting, but same reward preference shifts.

Young rats have more OFC neurones that fire highly for rewards, but same number that fire highly for the odor cue.

No difference in proportion of reward-responsive neurones that differentiated delays.

But young rats had over-representation of immediate rewards over delayed rewards, not seen in aged.

no difference in discriminating delays (forced choice trials), but difference in choosing delay.

One interpretation could be that ageing preserves reward expectancies but disrupts learning.


Ghahramani...Aron, Poldrack, London J Neurosci 32:21:7316-24 2012

Striatal dopamine D2/D3 receptors mediate response inhibition and related activity in frontostriatal neural circuitry in humans

Fallypride PET (D2/3 ligand) and fMRI during stop-signal task.

Prev: Congdon 2008, Hamidovic 2009: DA receptor polymorphisms influence response inhibition. Stimulant abusers have impaired inhibition (Monterosso 2005, Colzato 2007) and low D2R availability (Volkow 2001, Lee 2009). Meta-analyses of fMRI SST = Congdon 2010, Swick 2011.

Aron 2007: stop rtIFC/preSMA ‘hyperdirect’ projection to STN

Result: D2/D3 receptor availability in caudate and putamen correlate negatively with SSRT. Accumbens does not.

fMRI successful stop vs go activates rIFG, preSMA, ACC, insula. “clusters corresponding to the negative correlation between Stop vs Go and SSRT” = rt caudate, putamen, SFG, ltOFC.


Buschmann & Miller Science 315:5820:1860-2 2007

Top-down versus bottom-up control of attention in the prefrontal and posterior parietal cortices

Recording from monkey LIP, FEF and lateral PFC. Saccade to a target item whose identity was indicated at the start of the trial, 3 distractors could either be very different & uniform (=popout) or similar & varied (=serial search).

For each cell, for each condition (ser vs popout) and timepoint in the trials, mutual information about target location was calculated.

Neurones found the target later in serial search than in popout. Lateral PFC and FEF are first for serial. Whereas LIP is earliest in popout.

LFP coherence between frontal and LIP: increases both mid- (22-34Hz) and high-freq (35-55Hz)in the perisaccadic period; but top-down → ↑mid-frequency, bottom-up → ↑hi-frequency.

Conclude: “localised synchrony... may help resolve competition for attentional selection”

“lower frequency bands more robust to spike timing delays” → couple distant areas → broadcast of top-down signals.

“brain emphasises coherence at different frequency bands” to modulate connections and engage the best network for the task.


Theeuwes P&P 49:1:83-90 1991

Exogenous and endogenous control of attention: the effect of visual onsets and offsets

100% valid endogenous central arrow precue, at -600, -300, +200ms relative to target.

Targets are offsets. Manipulation: peripheral marker can appear or offset near one digit, at -160, -80, 0, +80ms after target.

Conclude: precueing abolishes distracting effect of onset at nontarget location.

The extent to which onsets capture attention is under the control of intentions.

Abrupt onsets in an endogenously attended region do affect attention, albeit briefly.

Same experiment for offsets: offsets capture attention when attention is unfocused.

Offsets do not produce interference when in the attended region.

→ capture by onsets or offsets are not ‘automatic’.


Watson & Humphries P&P 57:5:583-97 1995

Attention capture by contour onsets and offsets: no special role for onsets

1, 4, 8 or 16 items in grid, conjunction: Report present/absent (50%) LH/RH.

Ex1a) blue H among green Hs and blue As, or blue A among green As and blue Hs.

Ex1b) blue H target is defined by removing a segment from an A (after 100ms)

Ex1b) blue A target is defined by onset of a segment from an H (after 100ms)

Find: both onsets and offsets have equal effect on improving detection




Glosser & Goodglass JCEN 12:4:485-501 1990

Disorders in executive control functions among aphasic and other brain-damaged patients


Roesch & Bryden Frontiers in Decision Neuroscience 2012

Impact of size and delay on neural activity in the rat limbic corticostriatal system



Weber & Fischer Vis Res 34:14:1883 1994

Differential effects of non-target stimuli on the occurrence of express saccades in man


Philips AG & Fibiger HC Behav Pharm 1:269 1990

Role of reward and enhancement of conditioned reward in persistence of responding for cocaine

Rats will self-administer cocaine directly into the nucleus accumbens. Nucleus accumbens lesions attenuate intravenous self-administration of cocaine.


Frederick, Loewenstein, O’Donoghue J Econ Lit 40:351-401 2001

Time discounting and time preference: a critical review

Hyperbolic discounting functions always fit behaviour better than exponential functions



Solomon & Corbit Psychological Rev 81:119-145 1974

An opponent-process theory of motivation: 1. temporal dynamics of affect

Model of motivation – series of rewards gives an envelope of habituation and rebound, modelled as sum of fast primary and slow opponent reactions to reward.


Wise RA & Hoffman DC Synapse 10:3:247-63 1992

Localization of drug reward mechanisms by intracranial injections

VTA, nucleus accumbens, lateral hypothalamus, PAG, hippocampus previously suggested as target of opiate reward. VTA and accumbens are supported by good evidence. Accumbens and frontal cortex may govern stimulant reward, but amphetamine and cocaine operate differently. Each site is either dopamine cell-rich or terminal-rich.


Kivetz R Marketing science 22:4:477 2003

The effects of effort and intrinsic motivation on risky choice

Shift of utility function with effort required


Samejima, Ueda, Doya, Kimura Science 310:1337-40 2005

Representation of action-specific reward values in the striatum

Striatal cells encode expected value of potential actions, and the difference in expected values of potential actions.


Cohen MX, Ranganath C J Neurosci 27:371-8 2007

Reinforcement learning signals predict future decisions

Exploration: expected information weighed up against expected immediate cost of exploring


Schultz W Nature Rev Neurosci 1:199 2000


Schultz W Ann Rev Psychol 57:87-115 2006

Behavioral theories and the neurophysiology of reward


Kluver H & Bucy PC Arch Neurol Psychiat 42:979-1000 1930

Preliminary analysis of functions of the temporal lobes in monkeys

Terminology: increased tameness, psychic blindness, compulsive oral behaviour, hypermetamorphosis, dietary changes, hypersexuality.

Weiskrantz (1956) suggests they all share the common feature – dissociation between sensory and affective qualities of a stimulus.


Rolls, Hornak, Wade, McGrath JNNP 57:1518-24 1994

Emotion-related learning in patients with social and emotional changes associated with frontal lobe damage


Funahashi, Bruce CJ, Goldman-Rakic PS J Neurophysiol 61:331-49 1989

Mnemonic coding of visual space in the monkey’s dorsolateral prefrontal cortex


Murray, O’Doherty, Schoenbaum J Neurosci 27:8166-9 2007

What we know and do not know about the functions of the orbitofrontal cortex after 20 years of cross-species study


Niv Y & Schoenbaum G TICS 12:265-72 2008

Dialogues on prediction errors


Ostlund SB, Balleine BW J Neurosci 27:4819-4825 2007

Orbitofrontal cortex mediates outcome encoding in Pavlovian but not instrumental conditioning

Scalar representation of utility – not correct, as reward prediction error is specific to reward type.


Tanaka SC, Balleine, O’Doherty J Neurosci 28:6750-5 2008

Calculating consequences: brain systems that encode the causal effects of actions

Contingency-learning for appreciation of causality: contingency signal on fMRI is conflated with that of reward expectation.



Correlate fMRI with reward expection of individual options, net value of options available, prediction error.


Kim S, Hwang J, Lee D Neuron 59:161-72 2008

Prefrontal coding of temporally discounted values during intertemporal choice

Macaques trained to choose between two spatial locations with different rewards at different delays. Recording from DLPFC cells, plot regression coefficient between activity and value expectation. Correlation with expected value of action even before correlation with eventual decision.


Wittmann, Daw, Seymour, Dolan Neuron 58:976-73 2008

Striatal activity underlies novelty-based choice in humans

Ventral striatum activity = reward prediction error + novelty bonus


Kihlstrom JF Science 237:4821:1445 1987

The cognitive unconscious

Automatisation of a conscious process into an unconscious one.

Subliminal perception, implicit memory, hypnosis: events can affect mental functions even though they cannot be perceived or remembered.

→ tripartite division: truly unconscious mental processes, operating on preconscious or subconscious knowledge structures.


Klebaur, Phillips, Kelly, Bardo Exp Clin Psychophar 9:372-9 2001

Exposure to novel environmental stimuli decreases amphetamine self-administration in rats


Reed P, Mitchell C, Nokes T Anim Learn Behav 24:38-45 1996

Intrinsic reinforcing properties of putatively neutral stimuli in an instrumental two-lever discrimination task

Novelty is a reinforcer (US)


Tobler, O’Doherty, Dolan, Schultz J Neurophysiol 97:1621-32 2007

Reward value coding distinct from risk attitude-related uncertainty coding in human reward systems

OFC signals risk, striatum signals expected value


Schultz J Neurophysiol 80:1-27 1998

Predictive reward signal of dopamine neurons


Schall Curr Biol 15:R9-11 2005

Decision making



Popescu, Popa, Pare NN 2009

Coherent gamma oscillations couple the amygdala and striatum during learning

LFP recording in cats performing appetitive learning. Learning causes increased gamma coherence between BL amygdala and ventral putamen


Pesaran, Nelson, Andersen RA Nature 453:406-9 2008

Free choice activates a decision circuit between frontal and parietal cortex


Mnte, Heldmann...Sturm, Heinze Front Hum Neurosc 1:11 2007

Nucleus accumbens is involved in human action monitoring: evidence from invasive electrophysiological recordings


Cohen, Axmacher, Lenartz, Elger, Sturm JCN 21:875-89 2009

Good vibrations: cross-frequency coupling in the human nucleus accumbens during reward processing


Cohen, Axmacher, Lenartz, Elger, Sturm J Neurosci 29:7591-8 2009

Nuclei accumbens phase synchrony predicts decision-making reversals following negative feedback


Heinze, Heldmann, Voges...Sturm, Munte Front Hum Neurosc 2009

Counteracting incentive sensitization in severe alcohol dependence using deep brain stimulation of the nucleus accumbens: clinical and basic science aspects


Kuhnen & Knutson Neuron 47:763-70 2005

The neural basis of financial risk-taking


Elliot, Dolan, Frith Cer Cor 10:308-17 2000

Dissociable functions in the medial and lateral orbitofrontal cortex: evidence from human neuroimaging studies

Sentence completion, story comprehension, guessing, DMTS, hypothesis testing tasks. mOFC monitoring and holding in mind reward values. Lat OFC activeated when a response previously associated with reward has to be suppressed. [but physiology says that ACC does this, OFC only stimulus-reward pairings]


Elliott, Rees G, Dolan Neuropsychologia 37:403-11 1999

Ventromedial prefrontal cortex mediates guessing

Card-playing task, predicting vs reporting in a task with no difference in expected reward.


Rogers... Pickard, Sahakian, Robbins JNeurosci 20:9029-38 1999

Choosing between small, likely rewards and large, unlikely rewards activates inferior and orbital prefrontal cortex

Lateral OFC in generic risk-taking processes but ACC in processing relative risk in reward-related decisions?


Yokoyama, Jennings...Boller Neurology 37:4:624 1987

Lack of heart rate changes during an attention-demanding task after right hemisphere lesions

Normal HR deceleration during foreperiod of warned reaction task. Blunted in Rt hemisphere lesions.


Lebedev, Messinger, Kralik, Wise SP PLoS Biol 2:e365 2004

Representation of attended versus remembered locations in prefrontal cortex

Monkeys simultaneously remember one location, and attend to another location. VLPFC cells encode attentional locus, DLPFC encode both locations.


Roediger, HL Amer psychologist 45:1043 1990

Implicit memory: retention without remembering

Viewing vs generating words – better for recognition, but not fragment completion priming?















Naccache & Dehaene Cer Cor 11:10:966 2001

Bilingual subjects trained with 2AFC exact addition, approximate addition, exact base 6/base8 addition, approximate logarithms & cube roots. Significant ↓RT. Measured cost of changing language or of generalising to unseen test sums: no generalisation with exact tasks, good generalisation with approximate tasks.

fMRI (approx>exact) at bilateral LIP & IPL, (exact>approx) at LIFG.

EEG: respective areas different at 220ms exact, 270ms approx.


Marshall, Halligan, ... Frakowiak Cognition 64:1:B1 1997

The functional anatomy of a hysterical paralysis

Normal leg causes activity in contralateral dlPFC, premotor and motor cortex. Abnormal leg causes prefrontal, premotor activation, but not primary motoro cortex; plus right orbitofrontal and right anterior cingulate. No changes seen in the baseline condition.


Pardo, ... Raichle PNAS 87:1:256 1990

The anterior cingulate cortex mediates processing selection in the Stroop attentional conglic paradigm

PET during colour-word stroop, comparing incongruent (“red” in green ink) minus congruent trials. Increase in ACC, left premotor cortex, left postcentral cortex, left putamen, SMA, right STG, bilateral peristriate cortex.


Faw Consciousn & Cognit 12:1:83 2003

Prefrontal executive committee for perception, working memory, attention, long-term memory, motor control and thinking: A tutorial review

5-member committee metaphor – right VLPFC perceiver, left VLPFC verbaliser, VM/OFC motivator, dorsal-medial/cingulate attender, DLPFC coordinator/chair. Different coordinate systems. Subcommittee: posterior input areas, PL, VSS, basal ganglia loop (long timeframe), cerebellar loop (short timeframe), response-control loop from SMA via brainstem/cord.


Sternberg Acta Psychologica 30:276 1969

The discovery of processing stages: Extensions of Donders' method**

Additive factor method of modelling reaction times


Spivey M

The continuity of mind

Rather than a sequence of logical operations performed on discrete symbols, cognition is better described as continuously changing patterns of neuronal activity. Recognition is ‘close visitations of labelled attractors’.


Slovic, Fischoff, Lichtenstein Ann Rev Psychol 28:1-39 1977

Behavioral decision theory [Theory, Decision]

Review of 310 papers between 1971 and 1975 on psychological individual decision making. Specialised books: Kozielecki 1975 “Behavioral decision theory”; Lee 1971 “Decision theory and human behavior”.

Conservatism: posterior probabilities are nearer the prior probabilities than Bayes (bookbag & pokerchip task)

Kahneman & Tversky 1972: judgemental heuristics –


availability (frequency biased by recallability),

anchoring and adjustment;

law of small numbers: small sample is highly representative of population;

overconfidence in own ability for tasks involving skill (illusion of control).

Tversky 1972: Elimination by aspects (EBA): serial selection of criteria by importance, then elimination.

Corbin & Marley 1974: Random utility model (includes EBA as special case)

Svenson & Montgomery 1976: multiple strategies at different times in choice

Conjunctive / disjunctive (independent evaluation on each dimension then conjunction)

Lexicographic (ordering of dimensions)

Compensatory (weighting and summation of dimensions)


Counting of dimensions, ignoring small differences in certain dimensions

Slovic 1975: Preference for simpler rules (all else being equal)



Krahmer Games Econ Behavior 59:105 2007

Equilibrium learning in simple contexts [Theory, Game]

A model in which repeated contests of agents with complementarity leads to positive feedback and lack of learning, with one agent continually winning.

Mentions Markov strategies – a mapping from histories onto {0,1}


Beach & Connolly 2004

The Psychology of Decision Making [Theory, Decision]

Paradigm=ideological framework; Theory=provides predictions based on paradigm; Model=gives mechanism by which events occur.

First generation – behavioural decision theory:

Phelps & Chanteau: Hog judges’ criteria given categorically in a table, or pictorially, then ANOVA shows clustering of visual cues but not categorical cues. → multi-stage inference

Edwards 1955: students gambling on difficulty of maths problems deviates from normative.

Second generation – organisational theory:

Herbert Simon 1945: Bounded rationality – subjects simplify with limited information but standard decision rules.; Satisficing – subjects accept first option that meets all criteria, thus deviating from rational decision.


1960s: book-bag and poker-chips experiments – conservatism compared with Bayes

Beach & Barnes 1986: Use of aleatory or epistemic rules depends on:

Framing: is chance an important component? Events regarded as unique or repeated? Is statistical or causal logic the norm for the domain?

Demands: payoff for accuracy or speed? Can judgements be revised later? Reliability of information? Is the subject’s credibility at stake?

Yates 1990: Value function versus utility function: do feelings of risk (probability of an outcome) influence choice?

Gary Klein 1993: recognition-primed decision model – simple match, modelling of variations, major modifications and rational simulation

Brunswik 1947: lens model of evaluating options – list all input cue factors for each option, regress against options’ actual value to obtain beta weights; then for a given subject regress the factors against their decision to determine the policy beta weights. Correlation of the two sets of betas gives ‘achievement coefficient’, first R2 gives task difficulty, second R2 gives subject’s consistency.




Paré, Munoz J Neurophys 76:6:3666 1996

Saccadic reaction time in Monkey – advance preparation of oculomotor programs in express saccades [Saccade]

Express saccades are:

Spatially selective and training dependent

Increased error rate

Increased by gap effect, predictability of target location, longer preparatory period, eccentric fixation opposite to saccade direction

Increased if previous trial was an express saccade, with no-gap-period, to the same side.


Munoz & Wurtz J Neurophys 70:576 1993

Injecting rostral pole of monkey SC with muscimol (GABA inhibitor) or bicuculline (agonist) causes speeding or slowing of saccadic RT [Neuron, Saccade]


Dorris & Munoz J Neurophys 73:6:2558 1995

A neural correlate for the gap effect on Saccadic RT in Monkey [Neuron, Saccade]

Activity of fixation cells in SC decreased during fixation and increased when target appeared. ?errors also correlated - not sure.


Sgroi & Zizzo Physics A: Stat Mech 375:2:717 2007

Neural networks and bounded rationality [Theory, Game]

Comparison of errors made by feedforward back-propagation network trained on Nash equilibria games, with human errors. Other ‘strategies’ are more likely than Nash.

Minimax; rationalizability; 0-level strict dominance (0SD); 1-level strict dominance (1SD); pure sum of payoff dominance (L1); best response to pure sum of payoff dominance (L2); maximum payoff dominance (MPD); nearest neighbor (NNG)


Rumelhart, Hinton, Williams Nature 323:9:533 1986

Learning representations by backpropagating errors [Theory, Learning]


Costa-Gomez, Crawford, Broseta Econometrica 69:5:1193 2001

Cognition and behaviour in normal-form games: and experimental study [Theory, Game]

Humans played against Altruistic, Pessimistic (maximin), Naive (L1), Optimistic (maximax), and strategic L2, D1 deletes decisions dominated by pure decisions and responds best to a uniform prior over the opponent’s remaining decisions, D2, Equilibrium (Naive Nash), Sophisticated (Perfect foresight) estimates opponent’s response distribution.

Subjects allowed to ask computer for payoff


Stahl & Wilson Games & Economic Beh 10:218 1995

On players’ models of other players: Theory and experimental evidence [Theory, Game]

McKelvey & Palfrey 1992: Bayesian vs Altruist model with error component to model data

S&W 1994: 3x3 symmetric games: boundedly rational behaviour types based on Nagel (1993): L0 (random) 0%, L1 24%, L2 49%, Nash 27%

This expt: adds Worldly (rational expectations) model – that believes some opponents are Nash, others are L0, L1, L2, (and some Worldly).

Camerer & Johnson (1993, 2002): two-person, three-period alternating-offers bargaining game; allow subjects to look up hidden payoffs. Theoretically ‘backward induction’ is best strategy, but deviations in optimal behaviour correlated with systematic deviations from lookup strategy.

Type → search strategy; search results + type → decision

Heterogeneous decisions 70%~L2/D1, rest Naive. Search strategy: Naive 50%, L2 50%


Current list of possible areas:

Game rule learning: have simple games (e.g. PD) been investigated where the player does not know the rules in advance? **D. Fudenberg and D.K. Levine, The Theory of Learning in Games, MIT Press, Cambridge, MA (1998).

Rules are given: what are the chances of obeying them when the rewards are stochastic? (this has probably been done)

First order strategies: are they all learnable? What are the priors for each strategy?

are second order strategies a good model for humans playing a simple game (for which they do know the rules) – compare this with saccade-to-cue task and latency/error rate depending on previous trial(s)




Perhaps better to look at tasks where eye movements are used for searching for the best value.

Hypotheses to disprove: A spatial ‘value’ map used to guide saccades – is this independent from an attentional map. How to differentiate attention and value?

Divided attention vs splitting of value – incommensurable values of tasks

Attention usually described as Probability driven. But cf expected value: do we just direct attention to the higher valued location?




Anderson & Carpenter J Vision 6:8:822 2006

Changes in expectation consequent on experience modelled by a simple, forgetful neural circuit [Behaviour, Saccade]

Gradual change in saccadic latency when prior probability of target suddenly changes. Gap task; rarer leftward targets have longer latency. Time constant l=0.05



What about top-down influences on this kind of change-situation? If we’re not told there is a sudden change in probability we’ll get a gradual shift of expectation with an exponential decay timecourse; if we are expecting a sudden change, the shape of this shift could be modulated (in the direction of a step-function) representing a conscious decision.

LATER predicts that if you expect a stimulus at a location, your prior is higher but there is no increase in the rate of information accumulation. Could this be compatible with an account where d’ increases where expectation is higher? Increased d’ is essentially amplification of the rate of rise. Whether the prior level or the rate of rise is affected, must depend on whether discrimination or detection is the criterion for eye movement.

peripheral go/no-go signal. Location is probabilistic. Prior probability towards expected location, but in addition to a higher prior there should be a higher rate of rise compared to the unexpected side. We will see shorter latency on expected side, plus decreased variability, and reduced error rate.

better still would be peripheral precue valid 80% (exogenous attention), then target which can be go/no-go 50%. Are we better at avoiding erroneous no-go saccades on the cued side? – this might be different if the go/no-go is preattentional (eg. colour) or post-attentional e.g. a shape

evidence for and against the hypothesis should be distinct from non-contributory evidence, in the rate of learning a new expectation (which involves forgetting the old expectation).



What is the source of the variability in saccades? Normal distributions occur as a result of summing over a large number of identical independent random processes with any given distribution of responses. If the ‘rate of rise’ of information is Gaussian, then the rate is the sum of many independent signals, whose distributions are constant over the period of time.


Role of feedback in learning- positive versus negative evidence?

Adaptation to stimulus affects the rate of detection or the threshold?

Make intertrial interval same irrespective of previous RT – will this reduce saccadic latency variability?

What is the alternative to the theory that decision information grows from a baseline to a criterion?


Sinha, Brown & Carpenter J Neurophys 95:3146 2006

Task switching as a two stage decision process [Behaviour, Saccade]

Presents a blue and a red dot on left and right. Saccade to the ‘current task’ colour. Task switches when a central dot is presented of the new colour. Latencies reflect two-stage model; first (task detection) stage has more intersubject variability in values of mu & sigma, second decision stage identical to control condition (when ignoring the task-switching stimulus).


Weber, Biscaldi & Fischer Vis Res 35:18:2615 1995

Intertrial effects of randomisation on saccadic reaction times in human observers [Saccade, Behaviour]

Random direction, intertrial interval, fixation foreperiod, or gap duration between trials. Significant ↑RT if foreperiod randomised in overlap (no-gap) block. Varying gap duration had variable effects.

Express saccade latency not affected by any of these factors, but proportion increases with predictability.


Sohn & Lee Neural Networks 19:8:1181 2006

Effects of reward expectancy on sequential eye movements in monkeys [Saccade, Reward]

Reward for a trained sequence. Error less for later movements in sequence. First saccade of sequence often erroneous – towards reward. → implies reward representation independent of


Jentzsch & Leuthold QJEP 59:8:1329 2006

Control over speeded actions: a common processing locus for micro- and macro-trade offs? [Behaviour, RT]

Speed-accuracy trade-off controlled by instruction between blocks. Measured effect of event history on RT: post-error slowing.

Osman (2000) – partitioned RTs into fast & slow, and calculated accuracy. Clear effect of RT and of trade-off instruction. No interaction between trial-to-trial RT and instruction.

**Rinkenauer, Osman, Ulrich 2004

Brewer & Smith 1984: post-error slowing accounted for by a ‘monitor’ unit that exerts top down inhibition proportional to response-conflict in previous trial. (does this control the rate of rise or the criterion??)


Important factors to control for:

intertrial interval variation (might promote express saccades though)?

Add stimulus position variability to reduce express saccades


Regressors to explain trial-to-trial variability in RT or accuracy

Error on previous trial (Ridderinkhof, 2002)

Last 3 trials requiring cognitive inhibition (Schall & Carpenter, 2007)

Response same as previous trial?

Accuracy on this trial – i.e. micro-SAT (Jenzsch & Leuthold 2006)

Attention - ? pupil size, running average of RT


Advantages of saccades:

Ballistic all-or-none

Fast, large number of trials

Less cognitively penetrable

Electrophysiology known

Theory of control largely known




Thomas J Math Psychol 50:441 2006

Processing time predictions of current models of perception in the classic additive factors paradigm [Theory, RT]

Compared signal detection theory (SDT) with random walk model (EBRW) and stochastic general recognition theory. They all predict RT and accuracy tradeoff; they predict different changes in RT for 2-factor interactions depending on their effect on si, mi, tmDi, tsDi (infinitesimal mean and variance).





Valentin, Dickinson, O’Doherty J Neurosci 27:15:4019 2007

Determining the Neural Substrates of Goal-Directed Learning in the Human Brain [fMRI, Reward]

Human model of the rat findings that dissociate associating stimuli with incentive values from associating stimuli with responses.

Gave chocolate, orange, tomato as probabilistic cued rewards, good learning after 5 blocks. Devaluation by satiation causes rapid falloff in pleasantness.

fMR pre-satiation > post-satiation = orbitofrontal cortex



Dias R, Robbins TW, Roberts Nature 380:69-72 1996

Dissociation in prefrontal cortex of affective and attentional shifts [Lesion, Executive]

9 marmosets trained on discriminating displays for reward. Displays of compound stimuli (blue polygon plus black lines). 3 OFC lesions, 3 lateral PFC lesions, 3 sham.

  • Retention of previous discriminations: no difference.
  • Learning novel discriminations on same stimulus dimension: no difference
  • Learning novel discriminations along other stimulus dimension: only lateral PFC lesions showed impairment
  • Reversing reward association for same stimuli: OFC lesions impaired

Conclude lateral lesions impair extradimensional shifting, whereas orbital lesions impair reversal learning.


Jensen J,...Kapur S Hum brain mapping 28:294-302 2007

Separate brain regions code for salience vs valence during reward prediction in humans [fMRI, Reward]

fMRI with appetitive (5$) vs aversive (electrical) stimuli in Pavlovian 33% partial reinforcement schedule. 3 CS’s →$5 (33% reliable) or shock (33% reliable) or neutral.

Salience is 'error prediction' (ventral striatum), used for learning;

Results: CS(aversive) > CS(appetitive) gives activity in right anterior insula. CS(appetitive) > CS(aversive) gives medial OFC. Reward prediction error only found in left PFC. Ventral striatum activation does not encode valence (active for both appetitive and aversive CS) i.e. ‘general salience prediction error’.

Conclude: not fully explained by temporal difference model; ‘Coding of valence is done elsewhere’. Amydgala was silent, perhaps because no learning or extinction.



Experiment 1: Block 1: trials 1-30: Left V4, P0.5

Right V1, P0.5

Block 2: trials 1-30: Left V1, P0.5

Right V4, P0.5

Block 3: = block 1

Expermient 2: Block 1: trials 1-30: Left V1, P0.25

Right V1, P0.75

Block 2: trials 1-30: Left V1, P0.75

Right V1, P0.25



Vadillo & Matute QJEP 60:3:433 2007

Predictions and causal estimations are not supported by the same associative structure [Behaviour, Learning]

Contingency learning cognitive versus associative (Rescorla-Wagner) models, using post-hoc question framing – does c predict o? does c cause o? what is p(o|c)? Found answers have different relations to Dp versus p(outcome | cue), and different learning-order sensitivity.



Watanabe, Hikosaka EBR 152:361 2003

Effects of motivational conflicts on visually elicited saccades in monkeys [Behaviour, Saccade, Reward]

Asymmetrical reward schedule (1-direction-reward or all-direction-reward) in 4 directions. No choice. Speeding of saccades to rewarded location (265→230ms) and slowing for orthogonal (280ms) and opposite (290ms) directions. Effect on latency, peak velocity, amplitude.



Takikawa, Kawagoe, Hikosaka EBR 2002

Modulation of saccadic eye movements by predicted reward outcome [Behaviour, Saccade, Reward]

Four directions with asymmetrical reward schedule. Speeding of latency and velocity to reward: learnt within a block after ~5 trials. Speed and accuracy covary not trade-off. Errors (and slight slowing) after a rewarded trial. Errors towards reward; curved → conflict between motivational and cognitive control.


Harris & Wolpert Biol Cyber 95:21 2006

The main sequence of saccades optimizes speed-accuracy trade-off [Behaviour, Saccade]

Explains general covariance of saccade duration, amplitude, peak velocity. Motor command perturbed by proportional signal-dependent noise; optimal duration and trajectory for given amplitude. Movement cost = time from onset to offset; Position cost = error (a function of eccentricity).


Gersch, Schnitzer, Sanghvi... Visual Cognition 14:1:104 2005

Attentional enhancement along the path of a sequence of saccades [Behaviour, Attention, Saccade]

2AFC orientation discrimination of a 70ms Gabor at various locations during intersaccadic pause on sequences of visually- or memory-guided saccades. Compared with steady fixation, there was visual suppression at all locations, except along the saccadic path. Suppression greatest at start of sequence, and stops at completion. With memory-guided saccades, there was suppression for path-locations later than the next saccade in the sequence.


Gersch, Kowler, Dosher VisRes 44:1469 2004

Dynamic allocation of visual attention during the execution of sequences of saccades [Behaviour, Attention, Saccade]

Endogenous saccades to alternate boxes of 6 arranged in a hexagon. 2AFC orientation discrimination of flickering masked 90ms Gabor at random location; uncued/cued at start of trial. Finding: suppression of 1.5--3.0 times steady fixation. Much less suppression for next target (early) or current target. Slightly less suppression for previous target.




Logit & Logistic regression