1  
 
 
 
Fronto-Parietal Cortex  
in Sequential Behaviour 
 
 
 
Ausaf Ahmed Farooqui 
Fitzwilliam College 
MRC – Cognition and Brain Sciences Unit 
 
 
 
This dissertation is submitted for the degree of  
Doctor of Philosophy at University of Cambridge 
 
 
August, 2011 
 
 
 
 
2 
Preface  
 
I declare that this dissertation is the result of my own work and includes nothing, 
which is the outcome of work done in collaboration. Where reference is made to the 
work of others, the extent to which that work has been used is indicated and duly 
acknowledged in the text and bibliography.     
 
This dissertation does not exceed 60,000 words in length. 
 
 
Ausaf Ahmed Farooqui 
 
 
 
 
 
 
 
 
                     
                
3 
Acknowledgment 
It is a pleasure to thank all those who made this thesis possible.  
My supervisor, John Duncan, for his valuable advice and guidance. Working under 
him has been a very instructive experience. I'm thankful for his encouragement and 
support for my varied ideas.  
My co-supervisor and friend, Russell Thompson, for patiently teaching me almost 
everything I know of experimental methods, and for being the most accessible and 
reliable help for almost everything. 
The wider community at the MRC – Cognition and Brain Sciences Unit. 
The Gates Cambridge Trust, for funding my stay. 
To my wife, shukriya. 
 
 
 
 
 
 
 
 
 
4 
Abstract  
 This dissertation investigates the fronto-parietal representation of the structure of 
organised mental episodes by studying its effect on the representation of cognitive events 
occurring at various positions within it. The experiments in chapter 2 look at the completion 
of hierarchically organized mental (task/subtask) episodes. Multiple identical target-
detection events were organized into a sequential task episode, and the individual events 
were connected in a means-to-end relationship. It is shown that events that are 
conceptualized as completing defined task episodes elicit greater activity compared to 
identical events lying within the episode; the magnitude of the end of episode activity 
depended on the hierarchical abstraction of the episode.  
In chapter 3, the effect of ordinal position of the cognitive events, making up the 
task episode, on their representation is investigated in the context of a biphasic task 
episode. The design further manipulated the cognitive load of the two phases 
independently. This allowed for a direct comparison of the effect of phase vis-à-vis the 
effect of cognitive load. The results showed that fronto-parietal regions that increased their 
activity in response to cognitive load, also increased their activity for the later phases of the 
task episode, even though the cognitive load associated with the later phase was, arguably, 
lower than the previous phase. 
Chapter 4 investigates if the characteristics of the higher-level representations, like 
organization of task descriptions, have a causal role in determining the structure of the 
ensuing mental episode. Results show this to be true. They also confirm the results of earlier 
chapters in a different framework. Chapter 5 shows that the effect of episode structure is 
not limited to the elicited activity, but also affects the information content of the 
representation of the events composing the episode. Specifically, the information content in 
many regions of later steps is higher than that of earlier steps.  
 Together, the results show widespread representation of the structure of organised 
mental episodes.          
 
5 
Table of Contents
Preface ........................................................................................................................................... 2 
Acknowledgment ............................................................................................................................ 3 
Abstract .......................................................................................................................................... 4 
Table of Contents ............................................................................................................................ 5 
CHAPTER 1 ...................................................................................................................................... 7 
I n t r o d u c t i o n  ................................................................................................................... 7 
1.1 Fronto-Parietal Regions .......................................................................................................... 7 
1.1.1 Prefrontal regions and Control .................................................................................. 7 
1.1.2 Parietal Regions and Control ................................................................................... 15 
1.2 Mental Organisation ............................................................................................................. 30 
1.2.1 Neurocognitive Dynamics at Mental Episode Boundaries ....................................... 35 
1.2.2 Task Phases .............................................................................................................. 38 
1.3 Précis ..................................................................................................................................... 42 
CHAPTER 2 .................................................................................................................................... 44 
C o m p l e t i o n  o f  H i e r a r c h i c a l  T a s k  E p i s o d e s  .......................................................... 44 
2.1 Introduction .......................................................................................................................... 44 
2.2 Materials and Methods ......................................................................................................... 46 
2.2.1 Tasks ........................................................................................................................ 46 
2.2.2 Participants .............................................................................................................. 50 
2.2.3 Acquisition ............................................................................................................... 51 
2.2.4 Analysis .................................................................................................................... 51 
2.3 Results ................................................................................................................................... 53 
2.3.1 Experiment 1 ............................................................................................................ 53 
2.3.2 Experiment 2 ............................................................................................................ 62 
2.4 Discussion ............................................................................................................................. 65 
CHAPTER 3 .................................................................................................................................... 73 
Phase of  Task Episode and Fronto-Parietal  Activity ...................................................... 73 
3.1 Introduction .......................................................................................................................... 73 
3.2 Methods ................................................................................................................................ 77 
3.2.1 Participants .............................................................................................................. 82 
3.2.2 Acquisition ............................................................................................................... 82 
3.2.3 Analysis .................................................................................................................... 82 
3.3 Results ................................................................................................................................... 84 
3.3.1 Cognitive Load ......................................................................................................... 84 
3.3.2 Phase ........................................................................................................................ 96 
3.4 Discussion ........................................................................................................................... 102 
3.4.1 Cognitive Load ....................................................................................................... 102 
3.4.2 Phase ...................................................................................................................... 104 
CHAPTER 4 .................................................................................................................................. 107 
Higher Level Representations & Organisation of Task Episodes ................................................... 107 
4.1 Introduction ........................................................................................................................ 107 
6 
4.2 Methods .............................................................................................................................. 118 
4.2.1 Stimulus ................................................................................................................. 118 
4.2.2 Pre-scan ................................................................................................................. 119 
4.2.3 Imaging session ..................................................................................................... 120 
4.2.4 Participants ............................................................................................................ 121 
4.2.5 Acquisition ............................................................................................................. 121 
4.2.6 Analysis .................................................................................................................. 121 
4.3 Results ................................................................................................................................. 123 
4.3.1 Trial Type 1............................................................................................................. 127 
4.3.2 Trial Type 2............................................................................................................. 133 
4.3.3 Comparison of the two trial types ......................................................................... 139 
4.3.4 Trial Type 3............................................................................................................. 141 
4.3.5 Comparison of trial types 1 and 3 .......................................................................... 144 
4.3.6 Cognitive Load and Phase Effect ............................................................................ 149 
4.3.7 Cues ........................................................................................................................ 152 
4.4 Discussion ........................................................................................................................... 154 
4.4.1 Effect of Organisation ............................................................................................ 155 
4.4.2 Higher Level Representations in Organisation ....................................................... 158 
CHAPTER 5 .................................................................................................................................. 163 
Representation of  Task Information across the Sequential  Phases of  the Task 
Episode .................................................................................................................................... 163 
5.1 Introduction ........................................................................................................................ 163 
5.2 Methods .............................................................................................................................. 169 
5.2.1 ROI analysis ............................................................................................................ 169 
5.2.2 Whole Brain Analysis ............................................................................................. 173 
5.3 Results ................................................................................................................................. 173 
5.3.1 Effect on Category 1 information .......................................................................... 173 
5.3.2 Effect on Category 2 information .......................................................................... 188 
5.4 Discussion ........................................................................................................................... 197 
CHAPTER 6 .................................................................................................................................. 202 
Discussion ................................................................................................................................... 202 
6.1 What are Task Episodes? .................................................................................................... 202 
6.2 End of Task Episode ............................................................................................................ 203 
6.3 Phase Effect ........................................................................................................................ 208 
6.4 Implications ......................................................................................................................... 214 
Appendix .................................................................................................................................... 218 
References .................................................................................................................................. 234 
 
 
 
 
 
 
7 
CHAPTER 1 
I n t r o d u c t i o n  
 
Understanding the fronto-parietal regions presents a challenge to 
neuroscience. Activity in these regions can be found in almost all neuroimaging 
studies across almost all kinds of task conditions. However, the deficits caused by 
their lesions are variable and difficult to characterise and range from subtle and 
barely characterisable deficits to gross disorganisations in various domains of 
behaviour. In the following review some of the salient findings concerning these 
regions are presented. It is argued that they make a strong case that these regions 
represent all kinds of currently attended task-related representations.  
In the subsequent section, the case of goal directed behaviour is then 
discussed; it is hypothesised that the representations organising episodes of such 
behaviour will also be coded in fronto-parietal regions and affect the representation 
of individual behavioural events occurring in the episode. Two aspects of organised 
behaviour are specifically reviewed – completion of discrete episodes and the effect 
of the ordinal position of task events within the sequence making up the episode.  
Finally, a summary of subsequent chapters, looking at the various issues 
raised, is presented. 
1.1 Fronto-Parietal Regions 
1.1.1 Prefrontal regions and Control 
 Lesion studies suggest that the prefrontal cortex (PFC) is necessary for 
performance in situations that require the operation of a number of cognitive 
processes. Bianchi (1922) characterised the behaviour of PFC lesioned monkeys as 
having lost the ability to coordinate the elements of a complex activity. Patients with 
damage to PFC are described as ‘having difficulty in grasping the whole of a 
8 
complicated state of affairs, well able to work along old routine lines. But they 
cannot master new types of tasks, in new situations’ (Rylanders, 1939). Luria (1966) 
documented that such patients had problems in programming non-routine aspects of 
behaviour including the preliminary analysis of situations, constructing a plan of 
action and monitoring their performance. These and other case studies (e.g. Milner, 
1964; Grafman et al., 1986) make a strong case for the conceptualisation of the PFC 
as an ‘executive’ or a ‘cognitive control centre’ controlling the representation and 
processing in other brain regions (Norman and Shallice, 1986; Desimone and Duncan, 
1995; Miller and Cohen, 2001).  
 More structured experimental studies of prefrontal lesion patients have 
substantiated the view that the deficit in such patients is that of an absence of 
control in their thought and behaviour. Such patients are unable to abstain from 
doing a habitual but irrelevant action. For example, they are impaired at making eye 
movements diametrically away from peripherally presented stimulus, and usually 
end up making a saccade to the visual stimulus. Plausibly, the antisaccade requires 
overriding the habitual tendency of making saccades (Milea et al., 2003). Similar 
tendencies are also manifest in the difficulty faced by such patients in the Stroop task 
(Stroop, 1935) that requires them to inhibit their tendency to read the word and 
instead name the colour of the ink in which the words are written (Perret, 1974; 
Vendrell et al., 1995), and in the Wisconsin Card Sorting Test (WCST; Grant and Berg, 
1948) that requires dismantling of the now irrelevant task set and the creation of a 
new one as per the now relevant rule (Milner, 1964). Severe prefrontal lesioned 
patients, at times, show frank utilisation behaviour wherein the mere presence of an 
object elicits its use, notwithstanding the social propriety of such an act; e.g. the 
presence of a toothbrush on the doctor’s table elicited the act of brushing one's 
teeth (Lhermitte, 1983). Although utilisation behaviour involves actions that have 
become habitual in the course of life, some prefrontal patients also show 
9 
reappearance of primitive reflexes (De Renzi and Barbieri, 1992) that are usually 
present early in life and disappear with the maturation of the brain. 
 Prefrontal patients have been found to be impaired on a host of other 
processes that would be important for purposive behaviour in novel and poorly 
characterised situations such as – sustained attention (Luria, 1966; Chao and Knight 
1995), attentional shifts (Windmann, 2006), controlling the interference from 
irrelevant distractors (Chao and Knight 1997), working memory (Milner, 1964; 
Lewinsohn et al., 1972), organising mnemonic information (Shimamura et al, 1991), 
temporal integration of cognitive activities (Fuster, 2008), monitoring (Swick and 
Turken, 2002) and inhibitory control (Aron et al., 2004). 
 Indeed prefrontal lesions may decrease the common cognitive capacity (‘g’; 
Spearman, 1904) required for the performance on a wide variety of tasks (Duncan et 
al., 1995). This could explain the deficit seen in prefrontal patients on a wide variety 
of tasks. In line with this, Roca et al. (2010) found that the deficit of frontal patients 
on a wide variety of executive processes could be explained by their concomitant 
deficits on tasks of fluid intelligence. In other words, the performance across the 
patients and controls was equal if they were matched on the measures of fluid 
intelligence. It is possible that prefrontal cortices are required for some domain-
general process that is required for control in all kinds of task requirements. 
10 
 
Figure 1.1. Description of the colour and letter tasks used by Koechlin et al (2003). See text 
for details. Reproduced from Koechlin and Summerfield (2007). 
 Neuroimaging studies routinely show greater activity in the PFC in conditions 
where lesions impair the performance (reviewed in Fuster, 2008). Specific to the 
discussion here, greater activity is seen when greater task contingencies are to be 
represented for correct task execution (Koechlin et al., 2003; Badre and D’Esposito, 
2007). Koechlin et al. designed a study in which they systematically varied the quality 
as well as the quantity of the information required for the execution of a trial block 
(Fig. 1.1). Subjects responded to coloured shapes and letters (by making simple 
11 
colour and letter judgements) or withheld their responses according to the rule cued 
at the beginning of the block. The rules could remain the same across the blocks or 
vary from block to block or even trial to trial. In the blocks depicted in Fig. 1.1a, the 
subjects respond to green colour by their left hand and to red by their right hand. 
They are not supposed to respond to the white colour at all. In low information load 
blocks of this kind (Fig. 1.1a.i), only one stimulus type appeared and hence the same 
response was to be made on all trials; on high information load blocks of this 
category both kinds of stimuli could appear (Fig. 1.1a.ii) and so potentially two kinds 
of responses could be made on any trial. Thus these blocks manipulated the number 
of valid stimulus-response associations.  
In the blocks depicted in Fig. 1.1b, higher-level information, i.e. the relevant 
rule, was manipulated. Green coloured letter stimuli were to be categorised as vowel 
or consonant, whereas the red coloured ones were to be categorised as small or 
capital letters. One (Fig. 1.1b.i) or both (Fig. 1.1b.ii) of these rules could be relevant 
on any of such blocks. Note that the rule has been referred to as higher-level 
information because it subsumes more than one of the previous category (stimulus-
response associations); hence, to make a decision on the valid stimulus-response 
association a decision on the relevant rule has to be made first. 
Finally in the blocks depicted in Fig. 1.1c, the colour-to-rule mapping changed 
across the blocks. Performance in such blocks required first resolving the valid 
colour-to-rule mapping, only then could the valid rule be decided about. Thus the 
information additionally manipulated in these blocks subsumed the rule information 
and hence can be regarded as belonging to an even higher level.  
The relation between these three categories of manipulated information is 
nested, such that increase in the information of the higher category simultaneously 
increased the information of the lower categories e.g. increasing the number of rules 
12 
linking the stimulus to the response would also increase the number of stimulus-
response combinations.  
 The authors found that blocks with higher content of relevant information led 
to higher prefrontal activity. They reported a hierarchical pattern of results wherein 
an increase in the magnitude of contingencies related to stimulus-response 
combinations (Fig. 1.1a) increased the activity only at the posterior loci of the PFC, 
while increasing the number of rules (Fig. 1.1b) increased the activity in middle and 
posterior prefrontal regions, and increasing the cue-to-rule contingencies (Fig. 1.1c) 
additionally increased the activity at anterior prefrontal regions.  These results, 
however, cannot be considered conclusive about the relation between the locus on 
the antero-posterior axis and the qualitative type of manipulated information load. 
For example, it is not clear if the anterior prefrontal cortex is so specific to the cue-
to-rule contingency that any amount of information load related to rules relevant on 
a trial would not modulate its activity. Nonetheless, these results do show that, as a 
general rule, increasing the information load of task-related contingencies led to 
greater and more widespread PFC activity. 
 
 
 
 
 
 
 
13 
 
 
 
 
Figure 1.2. Sample trial from a Goal Neglect task. Trials began with an instruction to attend 
to the left/right side of the fixation spot. The letters were to be read aloud, pairs of numbers 
were to be added, asterisks were to be ignored. ‘<’/’>’ signalled the side to be attended for 
the succeeding stimuli.  Reproduced from Duncan et al. (2008).  
 
Another direct demonstration of the role of PFC in the creation and utilisation 
of relevant task contingencies for the control of behaviour comes from the goal 
neglect studies of Duncan et al. (1996, 2008) (see also Dumontheil et al., 2010). They 
found that in prefrontal patients and healthy individuals with low fluid intelligence, 
instructions may be understood and remembered but yet don't exert control over 
behaviour. In a sample task (Fig. 1.2), subjects are shown a stream of pairs of stimuli 
(letters or numbers) presented simultaneously on the two sides of the centre of a 
14 
computer screen. The last three pairs of stimuli of the stream were preceded by a 
presentation of ‘>’ or  ‘<‘. The subjects are to follow a series of instructions in a 
particular sequence – attend to left (or right), read aloud the letters on the attended 
side, add the pairs of numbers, but ignore the asterisks, after the appearance of ‘>’ 
read aloud the letters on the right, if ‘<’ appears, read aloud the letters on the left 
side.  
They found that the last part of such instructions (in the above example, those 
pertaining to ‘>’ or ‘<’) was sometimes ignored by patients with lesions in their PFC, 
or by subjects with low fluid intelligence scores. They further found that the 
frequency of such neglect of a particular instruction was dependent on the 
complexity and load of task instructions given prior to the instruction in question, 
even when the rules relevant on the particular block were themselves not complex.  
Further, the ignored instructions, though neglected, were nonetheless remembered. 
These findings make a strong case that in prefrontal patients the capacity to 
construct task control representations is limited. Moreover, although particular 
information may be present in the mind, they are not incorporated in the relevant 
control structure, if the latter has reached a sufficient degree of complexity. 
In an fMRI study of a related design, the authors discovered that the 
presentation of each rule phasically activated multiple fronto-parietal regions and 
tonically increased the baseline activity in those regions (Dumontheil et al., 2010), as 
if the incorporation of a rule into the task control structure is related to an increase 
in the baseline activity in fronto-parietal regions. 
Goal neglect studies have also looked at the complexity of task control mental 
structures that can be created. Prefrontal lesion patients are also impaired at 
carrying out prospective memory tasks that required them to execute intended 
actions after a delay period filled with irrelevant tasks (Volle et al., 2011). While the 
experiments reviewed above (Duncan et al., 2008; Dumontheil et al., 2010) required 
15 
an immediate creation of a task control structure, prospective memory tasks require 
that some information be held till a future time when, plausibly, it will be used to 
create a control structure. Prefrontal representations thus seem important not only 
for immediate control of behaviour but also for control requirements that will come 
into play in the future (see also Fuster, 2008). 
 Finally, stimulation of prefrontal regions can enhance performance specifically 
in situations that require greater cognitive control. Stuphorn and Schall (2006) 
trained monkeys on a stop signal task that required them to make saccades to 
peripherally presented stimuli on most trials (called go trials), but on a fraction of 
trials (called stop trials) after the appearance of the go stimulus, they were signalled 
to cancel the prepared saccade. As can be expected, slowing down on the go action 
in general would improve performance since more time would be available for the 
stop action to come into play and cancel the go action that as yet has not been fully 
prepared. On the other hand, if the go action is executed quickly, the stop action will 
more often lose the race.  
The authors found that stimulating the supplementary eye fields delayed 
saccade initiation and hence improved the performance on the stop signal task. 
Importantly, initiation of simple visually guided saccades was not delayed when that 
was the only task to be done. This shows that the activation did not lead to a general 
slowing down of saccadic initiation, but enhanced cognitive control. 
1.1.2 Parietal Regions and Control 
While the above account focussed on the PFC, similar evidence exists for the 
role of the parietal regions in the control of behaviour. Parietal lesions have been 
extensively documented to cause attentional dysfunctions both spatial (Posner et al., 
1984; Duncan et al., 1997; Driver and Mattingley, 1998; Smania et al., 1998; 
Mesulam, 1999) and non-spatial (Husain and Rorden, 2003; Husain and Nachev, 
16 
2007) as well as a reduction in attentional capacity (Duncan et al., 1999) and a 
reduction in capacity for sustained attention (Robertson et al., 1997). Beyond 
attention, both spatial (Husain et al., 2001) and non-spatial (Koenigs et al., 2009) 
working memory deficits have been associated with these lesions. Classical tests of 
frontal dysfunction also show impairments in parietal lesioned patients, e.g. WCST 
(Anderson et al., 1991), Stroop (Pujol et al., 2001). Case reports of frank utilisation 
behaviour after parietal lesions also exist (Mizobuchi et al., 2011). In an interesting 
study Desmurget et al. (2009) stimulated inferior parietal regions in tumour patients 
undergoing surgery. They found that in the context of motor action, this elicited a 
feeling of intention for action but not the action itself. Intention being an abstract 
feeling has usually been associated with activity in the prefrontal regions (Libet et al., 
1983; Lau et al., 2004; Haynes et al., 2007). 
It is likely that frontal and parietal regions are part of the larger brain network 
that allows for behavioural control. Such a view gets strong support from 
neuroimaging studies, most of which find a co-occurrence of frontal and parietal 
activity (reviewed in Cabeza and Nyberg, 2000; Duncan and Owen, 2000; Corbetta 
and Shulman, 2002; Dosenbach et al., 2006; Duncan, 2006). While both frontal and 
parietal regions seem to be necessary for the controlled cognition, meta-analyses of 
the neuroimaging studies showing activation in frontal and parietal regions suggest 
that some regions within the fronto-parietal cortices are more important for control. 
Duncan (2006) reviewed 20 studies that manipulated what would be 
considered as different forms of control – response conflict, task novelty, number of 
elements in the working memory, working memory delay and perceptual difficulty. 
He found that reported peaks of a contrast between high versus low control 
conditions clustered in common regions of the fronto-parietal cortex (Fig 1.3a) – 
inferior frontal sulcus (IFS), anterior insula extending into frontal operculum (AI/FO), 
supplementary motor area/anterior cingulate (SMA/ACC) and intraparietal sulcus 
17 
(IPS). A number of other studies investigating as disparate forms or control as would 
manifest in visual attention (Hon et al., 2006; Jiang et al., 2000), auditory 
discrimination (Holcomb et al., 1998), self paced response production (Jahanshahi et 
al., 1997), response conflict (Bush et al., 1998), working memory (Rypma et al., 
1994), perceptual difficulty (Woolgar et al., 2011), task switching (Dove et al., 2000), 
episodic memory (Duncan and Owen, 2000), conscious perception (Dehaene and 
Changeaux, 2011), complex response mapping (Jiang and Kanwisher, 2003a), 
semantic (Thompson-Schill et al., 1997) and syntactic processing (Jiang and 
Kanwisher, 2003b), find activity clustered in these regions of the fronto-parietal 
cortices. On this basis these regions have been referred to variously as ‘multiple 
demands’ (Duncan, 2010), ‘task positive’ (Fox et al., 2005), ‘task-activation ensemble’ 
(Seelay et al., 2007) or a ‘task control’ network (Dosenbach et al., 2006) and (Cole 
and Schneider, 2007). 
18 
 
Figure 1.3 (a) Clustering of reported peaks of activation of studies investigating various 
forms of control; response conflict (green), task novelty (purple), number of elements in 
working memory (yellow), working memory delay (red) and perceptual difficulty (blue). IFS: 
inferior frontal sulcus. AI/FO: anterior insula/frontal operculum. SMA/ACC: supplementary 
motor area/anterior cingulate. IPS: intraparietal sulcus. Reproduced from Duncan, 2006. (b) 
Regions with activity common across different tasks in Dosenbach et al., 2006. Reproduced 
from Dosenbach et al., 2006. 
19 
Conceptually similar results were found by Dosenbach et al (2006), who 
conjointly analysed data from experiments using ten very different tasks letter 
identification, verb generation and reading, object naming, matching symbols and 
letters, living/nonliving judgements, reading aloud, motor timing, visual detection, 
abstract/concrete and physical/semantic judgement on nouns. They found that 
irrespective of task, discrete aspects of trial blocks were associated with common 
patterns of activity across the different tasks (Fig 1.3b). The starting cue in all tasks 
involved modulation of activity in ACC, AI/FO, intraparietal sulcus (IPS), precuneus, 
posterior cingulate cortex (PCC) and fusiform regions. The duration of the trial blocks 
across the tasks were associated with sustained change of activity in ACC, AI/FO, 
anterior prefrontal cortex (APFC), intraparietal lobule (IPL), middle and posterior 
temporal regions, PCC and the ventromedial prefrontal cortex. 
Evidence for a causal relation between these specific areas of the fronto-
parietal cortices and diverse forms of cognitive control was recently provided by 
Woolgar et al. (2010). They looked at the relation between these multiple demands 
regions and the general factor g that has been hypothesised to contribute to 
performance in diverse kinds of cognitive activity (Spearman, 1927), and found that 
damage to each of the above regions predicted loss in g, whereas damage to fronto-
parietal regions outside these multiple demands regions was not predictive. 
These findings make interesting suggestions. A set of fronto-parietal regions 
are always co-active in response to seemingly diverse conditions, and plausibly, these 
form a core component of diverse forms of cognitive control. Secondly, and as a 
corollary to the previous point, the seemingly diverse forms of control do not appear 
to be anatomically diverse. It is possible that the absence of expected diversity is due 
to the poor spatio-temporal resolution of fMRI. However, it is also possible that 
these regions are sensitive to some core component that is common to all of the 
above reviewed control processes, or perhaps that the apparently diverse control 
20 
processes are actually not diverse, but are the manifestations of a more abstract 
form of control. 
1.1.3 Representations in the Fronto-Parietal regions 
The above review demonstrates the role of fronto-parietal regions in 
behavioural control. However, the mechanism by which these regions lead to control 
is far from clear. A good starting point in this regard is to investigate what kind of 
mental/behavioural events are represented in these regions. 
Neurophysiological studies show that neurons in multiple fronto-parietal 
regions respond to all behavioural and mental events that are relevant and 
consequential in the context of the task at hand, irrespective of the actual identity of 
the task (Duncan, 2010). Kusunoki et al (2009) trained monkeys to covertly monitor a 
stream of different pictures presented on the right or left (varying across trials) of 
the fixation spot for a specific target, and to specify its occurrence by making a 
saccade to the relevant side at the target stimulus offset (Fig 1.4a). Across the task 
variants the identity of the target could remain constant across trials (called fixed 
target task) or could vary in case of the cued target trials. In the latter version of the 
task, monkeys learnt three cue-target associations. In such trials, the identity of the 
target stimulus varied across trials in such a way that the target on one trial could be 
a non-target on the other. Many of the non-target stimuli on any trial had served as 
targets on previous trials at some point during stimulus presentation. The firing 
profile of more than half of all neurons recorded from the lateral PFC discriminated 
the target from the non-target events (Fig 1.4b). Very few neurons discriminated 
individual non-targets, including those that had been targets in earlier trials.  
Similar coding of task-relevant events has been found in a number of studies, 
wherein the neurons have been shown to code for task rules (Asaad et al., 1998; 
Hoshi et al., 1998; White and Wise, 1999; Asaad et al., 2000; Wallis et al., 2001; 
21 
Wallis and Miller, 2003), relevant response (Quintana and Fuster, 1992; Asaad et al., 
1998), cues, rewards and cue-reward association (Watanabe, 1990, 1992, 1996). 
With regard to stimulus features, fronto-parietal neurons seem to code for any 
feature that is linked to a behaviourally relevant event – stimulus identity (di 
Pellegrino and Wise, 1991; Rao et al., 1997), location (Azuma and Suzuki, 1984; Rao 
et al., 1997), colour (Quintana and Fuster, 1992), orientation (Mikami et al., 1982), 
vibration frequency (Romo et al., 1999), category membership (Freedman et al., 
2001) and so forth.  
 
Figure 1.4. (a) Scheme of the task used by Kusunoki et al. (2009). Monkeys viewed a 
stream of stimuli, presented to one side of the fixation spot, and monitored for the presence 
22 
of the target picture, and at its occurrence waited for its offset and then made a saccade to 
the side it occurred. The targets were fixed across trials in the fixed target condition. In cued 
target condition, the target on any trial was determined by the nature of the cue at its 
beginning. Three cue-target combinations were used. (b) Percentages of cells in lateral 
prefrontal cortex that discriminated targets from non-targets (red – stronger response to 
targets; green – stronger response to non-targets). Heavy black line shows time of stimulus 
presentation; discrimination was tested in sliding windows beginning 200 ms before stimulus 
onset. Reproduced from Duncan, 2010. 
 
Indeed, an important property of these neurons is the flexibility of their 
coding. Often, neurons cease responding to an event that has lost its importance and 
assume responding to the now relevant event. This property was seen in the above 
mentioned study, wherein minimal response was seen for non-targets that had been 
targets in earlier trials. The rapidity of this adaptation was amply demonstrated in a 
study by Rao et al. (1997). In their task (Fig. 1.5a), monkeys were shown a sample 
target stimulus at fixation, the identity of which they had to remember during the 
following delay (‘what delay’). The screen succeeding the delay showed two objects 
at the periphery, one of which was the earlier shown target. Now the monkeys had 
to remember the location where the target appeared in the succeeding delay 
(‘where delay’). Finally they were shown a screen with four locations and had to 
make a saccade at the place where the target had appeared. Thus in this task, the 
monkeys had to first code for the identity of the target and maintain it during the 
‘what delay’, then represent the spatial location of the target during the ‘where 
delay’. Again roughly half of the neurons recorded coded first for the identity of the 
target and then its location (Fig. 1.5b).   
23 
 
Figure 1.5. (a) A sample trial from Rao et al., (1997); trials began with a sample stimulus 
presented at the fixation spot, the identity of which was to maintained during the ‘what’ delay, 
which was followed by the  peripheral presentation of two stimuli, one of which was the 
sample stimulus. This time the position of the sample stimulus was to be remembered 
across the ‘where’ delay, after which monkeys were probed for the position of the sample 
stimulus, to which they made a saccade. (b) Sample prefrontal neuron that codes for the 
identity of the object during ‘what’ delay and the location during ‘where’; the two grey bars 
represent the presentation of sample (left) and test objects (right), respectively . Reproduced 
from Rao et al., (1997). 
 
24 
A number of human fMRI studies also provide similar evidence (Hon et al., 
2006; Thompson and Duncan, 2009; Serences and Boynton, 2007, Haynes et al., 
2007). Hampshire et al (2006) designed a task that required subjects to monitor a 
sequence of visual stimuli and covertly detect the presence of a visual target that had 
been cued at the beginning of the trial. The stimulus sequence could terminate 
randomly with a probe asking if the previous stimulus in the stream was the target to 
which subjects made a yes/no response, otherwise, the detection of visual targets 
was covert. The actual identity of the target changed across trials. Thus, the target 
stimuli were made relevant in the context of that trial by the cue but they had the 
same decision-making and task load requirements as the non-target stimuli. Across 
many fronto-parietal regions a selective response was seen to the targets, proving 
that fronto-parietal regions have a predilection for currently relevant events. 
Interestingly, some prefrontal regions also showed a higher response to other stimuli 
belonging to the same category as the target. For example, if the target was a 
particular face, such regions additionally showed greater response to all faces. 
Other studies have used multivariate pattern analysis to show a broad 
representation of attended and task-relevant cognitive content across the fronto-
parietal regions (Dux et al., 2009; Eger et al., 2009; Esterman et al., 2009; Greenberg 
et al., 2010; Jenkins and Ranganath, 2010; ). The conceptual basis of this 
methodology is that if a category of cognitive events (e.g. relevant rule) is 
represented in a region then the different exemplars of that event (e.g. different 
rules) will be represented by different neuronal population codes. Since it is very 
likely that the distribution of such population codes would be different across the 
voxels composing the region, the pattern of fMRI activity elicited by the exemplars of 
the represented cognitive event can be expected to be different. So the 
discriminability of the patterns of activity elicited by the exemplars of the cognitive 
event can be considered as a measure of the representation of the cognitive event. 
25 
  
 
Figure 1.6. (a) Task used by Woolgar et al., (2011); subjects responded to one of the four 
positions of the stimulus by a key press. Two mapping rules were used, and were cued by 
the colour of the background. Two colours mapped to each rule. (b) t-score on a one-sample 
t-test of classification accuracy of patterns of elicited activity against chance. A score of 0 
indicates chance level classification and positive values indicate above chance classification. 
Dotted line: significance threshold p = 0.05; solid line: significance threshold corresponding 
to Bonferroni-corrected p-value of 0.0125. Reproduced from Woolgar et al., 2011. 
In one such study (Woolgar et al., 2011), participants made simple key press 
responses to stimuli presented at four different positions in a horizontal row on a 
computer screen. Two rules could be used to map the stimulus position to the 
response (Fig. 1.6a), which could be one of the four alternative key presses made 
with index and middle fingers of the two hands. The rule relevant on a trial was cued 
by the background colour. Two colours were mapped to each rule. This design 
allowed the authors to look into the discriminability of the pattern of activity elicited 
by the different colours, response fingers, stimulus positions and rules. The results 
showed that all features were discriminable in fronto-parietal regions (Fig. 1.6b). 
Evidence was strongest for the coding of rule followed by colour and stimulus 
position; responses were least discriminable and their discriminability was significant 
26 
only in left inferior frontal sulcus (IFS). Similar results were also obtained by Stiers et 
al. (2010). 
Other studies have demonstrated representation of more abstract task-
related information such as the outcome of a decision (Haynes et al., 2007; Kuhn et 
al., 2010; Bode et al., 2011). Haynes et al (2007) asked subjects to freely decide 
which of two operations (addition or subtraction) to perform on a forthcoming 
display of numbers. They were to hold this intention for a variable gap, after which 
they saw the numbers and performed the intended operation and thus indicated 
which of the two operations they had decided for. The authors showed that the two 
decisions were discriminable in a number of prefrontal regions before the actual 
execution. Hence prefrontal regions represent not only concrete task rules but also 
abstract task-related decisions.  
The representations in fronto-parietal regions are not only limited to those 
that are explicitly related to a task at hand. Rather, any conscious and attended 
mental representation seems to be represented in the fronto-parietal regions. Lumer 
and Rees (1999) found that switches in between bistable percepts (e.g. in binocular 
rivalry) in the absence of any task-related requirements of decision, control or 
response, caused activity in the fronto-parietal regions.  
A recent study made a strong case that fronto-parietal regions represent any 
conscious mental representation and activity in them is elicited whenever such  
representation is changed and a new one is created, irrespective of whether the 
change has been initiated by the self or exogenously. Knapen et al. (2011) related the 
magnitude of elicited fronto-parietal activity to the temporal duration of transitions 
in between the two percepts in bistable viewing conditions. They looked at two 
forms of bistable percepts. In bistable apparent motion, subjects were shown 
stimulus with equal motion energy in two opposite directions resulting in the 
wavering of perception in between the two stable motion percepts. The other 
27 
bistable percept they looked at was binocular rivalry, in which the two eyes were 
given dissimilar inputs that cannot be fused together, and so subjects’ percepts 
swayed alternatively between the two percepts projected to the eyes. In both, 
binocular rivalry and bistable apparent motion, subjects experience mixture states, 
when one percept is gradually shifting over onto the other and observers experience 
a mix of both perceptual interpretations. The duration of this transition period can 
be variable. They found fronto-parietal activity at the transitions between the 
percepts, the magnitude of which was dependent on the duration of the transition 
periods. 
 Further, they compared the activity elicited by such endogenous shifts 
between bistable percepts to that elicited when the subjects were exogenously 
replayed two monostable stimuli corresponding to the two perceptual 
interpretations they had of the bistable stimuli. In such conditions the cause of the 
transition of the percept was exogenous and its duration was manipulated by the 
experimenters. When the duration of transition was controlled for, no difference was 
found in the magnitude of elicited fronto-parietal activity in the two (exogenous and 
endogenous) conditions. This makes a strong case that the activity in the fronto-
parietal regions was not the cause of transition between the percepts in the bistable 
conditions, but was the effect of the transitions.  
A number of studies have shown greater fronto-parietal activity when a 
stimulus is perceived consciously compared to subliminal processing (reviewed in 
Dehaene et al., 2006). However, in most such studies consciousness of a stimulus is 
usually associated with better decisions made on the stimulus manifested in better 
performance on the associated task. Such studies do not make it clear if the 
associated increase in fronto-parietal activity is the result of better task performance 
on the stimulus or is specifically linked to the subjective conscious perception of it. 
To circumvent this confound, Lau and Passingham (2006) used metacontrast masking 
28 
to create experimental conditions in which the objective task performance on the 
stimulus was the same but the subjective awareness of that stimulus was not. They 
found that activity in prefrontal regions was associated specifically with conscious 
awareness even when the objective task performance on the stimulus had been 
controlled for.  
These regions not only represent the conscious perceptual representations 
but also the endogenously created contents of imagery. Stokes et al. (2009) asked 
subjects to visually imagine an ‘X’ or an ‘O’, and found that the resultant patterns of 
activity could be significantly discriminated in anterior insula. Indeed some 
vegetative state patients can flexibly and reliably elicit distinct patterns of fronto-
parietal activity when asked to create an image of common situations like playing 
tennis or moving around their house (Owen et al., 2006; Monti et al., 2010). 
While attention on and consciousness of the mental representation certainly 
is a feature associated with fronto-parietal representations, some studies suggest 
that behaviourally relevant information may be represented in these regions in the 
absence of conscious awareness. Lau and Passingham (2007) cued subjects to do one 
of two tasks on a displayed word – judge whether it is bisyllabic or whether it 
referred to something concrete. The cue (diamond or square shape) signalling which 
task to be done was displayed before the display of the word. On some trials an 
additional priming cue was displayed very briefly prior to the main cue display. 
Authors tried to achieve meta-contrast masking of the prime cue by the main cue, 
resulting in varying levels of perception of the former. On some trials subjects 
reported consciously seeing the first transient cue; on others it was entirely 
subliminal. The priming cue could be congruent or incongruent with the actual cue. 
Interestingly, greater prefrontal activity was elicited on the trials in which the main 
cue was preceded by incongruent but subliminal prime, suggesting that activity in 
this region could be triggered by an unconscious event.  
29 
However, this and other such studies (Boy et al., 2010; Gaal et al., 2010; Reuss 
et al., 2011) showing unconscious events triggering fronto-parietal activity do not 
conclusively prove if this activity is related to an unconscious event, since it is very 
possible that the greater activity reflects the conscious feeling of difficulty faced by 
the subjects due to the activation of task irrelevant representations. Soon et al. 
(2008) showed more conclusive evidence of the presence of information in the 
fronto-parietal regions to which subjects did not have a conscious access. They asked 
subjects to randomly choose to make one of two actions (left or right index finger 
button press), and then proceed on to making the decided response. To allow 
subjects to convey the time at which they became aware of their decision, a stream 
of letters, updated every 500 ms, was displayed on the screen. Subjects conveyed 
the time of their awareness of decision by taking note of the current letter on the 
screen. The authors discovered that the identity of the future decisions could be 
predicted from the pattern of activity in the anterior and medial prefrontal and 
medial parietal regions, seconds before the subjects reported becoming aware of 
them.  
The findings reviewed suggest that fronto-parietal activity is elicited by – and 
the pattern of elicited activity contains information about – any behaviourally 
relevant event. However, as discussed in the next section, behaviourally relevant 
events are situated in the larger context of a purposive mental (or task) episode, and 
play crucial roles in the organisation of such an episode. It is possible that the fronto-
parietal activity seen with behaviourally relevant events is the result of the 
concomitant changes in the structure of the mental episode.  
Consider, for example, the case of visual target detection. Subjects are 
required to search for a stimulus; the search ends when the target is detected. Can 
the neural activity found with the detection of the target be attributed solely to the 
event of target detection? Note that target detection also corresponds to the 
30 
boundary between two distinct mental episodes (that of visual search and rest). It is 
very much possible that the fronto-parietal activity seen, corresponds to such a 
change in the mental episode rather than attentional detection. Similarly, in many of 
the findings reviewed above, the behaviourally relevant event is associated with 
salient roles in the organisation of the task episode. Indeed a criterion for 
behavioural relevance would be that the event brings about a change in the current 
mental/behavioural state (or episode) or has clear implications for succeeding 
mental processes (or how the subsequent mental events are to be organised. 
1.2 Mental Organisation 
Much of our behaviour and mental life is organised. Consider a sequence of 
actions for preparing a hypothetical breakfast: 
pour water → add coffee → pour cream → toast bread → spread  butter    
Arguably, in real behaviour such a sequence of actions can only exist in the context of 
a hierarchical set of organising representations (Lashley 1951, Miller et al., 1960, 
Schank & Abelson, 1977, Reed et al. 1995), which ensure that the correct action is 
chosen at each step. Abstract representations defining temporally extended goals 
organise nested lower-level representations that have temporally more confined 
roles, define more concrete goals and help select more defined aspects of the 
cognitive schema; such a hierarchy eventually converges into the selection of the 
representation of definite actions. This results in a hierarchical parcellation of the 
mental episode. A consequence of this is the temporal organisation of mind and 
behaviour, since representations at different levels organise mental episodes of 
varying temporal abstraction and duration. In other words various representations 
carve out discrete mental episodes. 
An organised morning may be parsed into episodes of toiletries, getting 
dressed, preparing and eating breakfast etc., based on the different goal 
31 
representations organising these episodes. Each of such episodes may be further 
chunked into smaller episodes. Preparing breakfast, in the earlier example, may be 
parsed into subtasks of ‘preparing coffee’ and ‘preparing buttered toast’ (Fig. 1.7). 
Due to such an organisation, task-related representations occur in the context of 
higher over-arching representations. For example, in the case of hypothetical 
breakfast preparation, the representation of lower level actions (e.g. ‘add coffee’ or 
‘toast bread’) occurs in the context provided by the higher level representations 
(‘prepare coffee’ or ‘prepare toast’).  
 
Figure 1.7. Hypothetical organisation of an episode of preparing breakfast. 
 
The role of hierarchical representation was demonstrated in an experiment by 
Schneider and Logan (2007), wherein they asked subjects to memorise two task 
sequences (called alpha and beta). These sequences were built of various 
combinations of two task units – making a judgement about the size of the stimulus 
word (task A), or a judgement on its meaning semantic (task B), e.g. alpha = AABB; 
beta = ABBA. Subjects then performed trials on which they were randomly cued to 
perform a task at one of the serial positions in a sequence (e.g. ‘alpha 2’ meant the 
second task in the alpha list which is size). They found several effects that were 
consistent with a hierarchical representation of these two lists. One such effect was a 
large sequence repetition benefit, i.e. subjects were faster if they were cued on the 
same sequence consecutively than when they were cued on different sequences on 
consecutive trials. This was the case even when the actual task done in the former 
condition was different (e.g. beta 1 followed by beta 2), while that done in the latter 
32 
condition was the same (alpha 2 followed by beta 1). In fact task repetition benefits 
only occurred when the tasks were cued from the same list. 
In a different study (Schneider and Logan, 2006), the same authors had 
subjects memorise such task sequences and then execute them repeatedly. For 
example, if the subjects were executing a sequence ABBA, they performed the tasks 
cued by the elements of this sequence sequentially on trials 1 to 4, then repeated 
the whole sequence on trials 5 to 8 and so forth. They found that across all kinds of 
sequences executed, the RT on trials wherein they executed the first element (i.e. 
trials 1, 5, 9, 13, etc.) of the sequence was the highest (Fig. 1.8). Two features of this 
increased RT are noteworthy. In case of sequence ABBA, the first position involves a 
task repeat, i.e. A to A; whereas positions 2 and 4 involve a task switch, i.e. A to B. If 
the organisation was not hierarchical, RTs corresponding to the sequence positions 2 
and 4 would have been higher than 1, but in the actual RT profile, RT at position 1 
was always the highest.  
33 
 
Figure 1.8. Task elements at the beginning of task sequences have the longest reaction 
time, which depends upon the complexity of the sequence (see text for details). Reproduced 
from Schneider and Logan (2006). 
Secondly, this increase in RT at the first sequence position was not affected by 
the task at this position being a repetition or a switch with respect to the preceding 
task. For example, if the sequence was AABB, the first position in the sequence 
would involve a switch with respect to the preceding task at the fourth position in 
34 
the sequence i.e. from B to A : AABB-AABB; whereas if it was ABBA, the first position 
would involve a task repeat i.e. from A to A: ABBA-ABBA. Nonetheless, the first RT of 
sequence ABBA was higher than AABB (Fig. 1.8a), plausibly because the former is a 
more complicated sequence since it involves two task switches, while the latter 
involves just one. This notion was proven by the findings that the magnitude of the 
first RT was indeed dependent upon the number of switches within the sequence; 
hence first RT for a sequence like AABBBA was smaller than that for AABABB (Fig. 
1.8b). While these results certainly reinforce the inference of a hierarchical 
organisation of the resulting task episode, they also suggest that the beginning of the 
task episode involves instantiation of representations related to the entire length of 
the episode (and hence the dependence of the first RT on the complexity of the 
episode).  
As has been reviewed in the previous sections, the neural representation of 
task-related events has been worked upon. However most of these studies have 
largely ignored the fact that task-related events occur in the context of organised 
mental episodes. Take the above example of preparing breakfast. An action like ‘pour 
water’ can be looked upon as a task event on its own, or this can be looked upon as 
an event that occurs in the larger context of preparing breakfast. It is, therefore, not 
known how the larger context of an organised mental episode in which such task 
related events occur affects the representation of these events. How does the fact 
that there are underlying representations (‘prepare breakfast’, ‘prepare coffee’, 
‘prepare toast’) that have organised the context of lower level sequence of actions 
(pour water → … → spread butter), affect the representation of the various actions 
composing the sequence? 
An important consequence of the organisation of behaviour into discrete 
episodes is that specific mental/behavioural events mark the beginning and the end 
of episodes. The beginning would mark the creation of the episode specific cognitive 
35 
focus and the assembly of representations creating this focus, while at the end of 
episodes, these representations would be dismantled. In the earlier example, ‘pour 
cream’ marks the end of the episode of ‘preparing coffee’, while ‘spread butter’ 
completes the preparation of ‘buttered toast’. As is evident from Fig. 1.7, in such a 
scheme, ‘spread butter’ additionally completes the whole episode of ‘prepare 
breakfast’. This brings forth an important characteristic of organised behaviour. 
Because the parcellation of a behavioural episode is hierarchical and nested, some 
events would complete a smaller and hierarchically lower episode (e.g. subtask), 
while others would complete a hierarchically higher and temporally longer episode 
(e.g. task).  
1.2.1 Neurocognitive Dynamics at Mental Episode Boundaries 
It was seen earlier that fronto-parietal regions flexibly represent all kinds of 
task-relevant events. In a sequential task, the successive mental states can be 
expected to be represented in these regions. Old representations can be expected to 
dissolve away while new ones would be successively assembled. This issue was 
investigated by Sigala et al. (2008) using an experimental task described earlier (Fig. 
1.4a, Cued target task), wherein monkeys watched a series of pictures presented on 
one of the two sides (left/right) of the fixation spot, and waited covertly for the 
target specified by the cue at the beginning of the trial. At the appearance of the 
target, they waited for its offset and then made a saccade to the side on which target 
had appeared. Three possible cues were associated with three different targets. 
The authors compared the activity profiles of 324 neurons recorded from 
lateral PFC during eighteen possible events - three cues presented to right and left 
(i.e. six cue types); delay between two stimuli for the trials beginning with the six cue 
types; and the six targets associated with the six cue types. Mean firing rate of each 
neuron was obtained in response to these eighteen events; this was normalised for 
each neuron by dividing the rate of response to an event by mean activity across all 
36 
event types. Thus, a pattern of activity was obtained for each event across the neural 
population sample in the form of a vector. These eighteen vectors were then 
correlated.  
 
Figure 1.9. Correlation patterns in Sigala et al., (2008). Note that the correlation between the 
different types of events at the same phase (e.g. between the three cues presented in the 
same hemifield) are high, while correlation across phases (e.g. between the cues and delay) 
is minimal. Reproduced from Sigala et al., 2008. 
The results (Fig. 1.9) showed high correlations for activity patterns from the 
same task phase, particularly in the same hemifield, and negligible or even negative 
correlations for different task phases. This shows that the different steps of a 
sequential task are represented by different patterns of activity and hence form 
different neural representations. The structure of neural representation seems to 
care less about the content of representation at a particular step of the sequence – 
seen in the high correlations in the activity pattern across the different cue – target 
identities but at the same phase, e.g. the three targets in the same hemifield had 
highly correlated patterns. However, for the different phases of the task, the 
representation was essentially orthogonal, i.e. activity patterns for targets, cues and 
delays were minimally correlated with each other.  
37 
What would happen at the boundaries of hierarchically higher task episodes? 
Plausibly, such boundaries represent greater revisions in mental representations. In 
the case of Fig. 1.7, finishing the act of ‘pour water’ brings the subject to the next 
action ‘add coffee’, while the higher level representations of ‘prepare coffee’ and 
‘prepare breakfast’ under which this action took place are not affected. In contrast 
finishing the act of ‘pour cream’ not only changes the lower level action 
representation but also changes the higher level representation of ‘prepare coffee’. 
At the extreme, the event completing the whole task of ‘preparing breakfast’ 
changes the representations at all levels of hierarchy into which the task episode was 
organised.  
Arguably, brain regions representing such organising representations (like 
‘prepare coffee’ or ‘prepare breakfast’) can be expected to show a change in activity 
when these are revised at the end of mental episodes. Further, the activity can be 
expected to be greater when the revision is more intense, e.g. at the end of the 
overall episode (‘prepare breakfast’) versus completion of a sub-episode (‘prepare 
coffee’). Note that earlier studies focussing on task-related representations have 
usually focussed on task-related stimulus, action, rule or other related behavioural 
events like reward. To my knowledge none of these have explicitly focussed on the 
implicit representations organising the task episode. As has been mentioned earlier, 
such representations provide the organising framework that controls the lower level 
behavioural events like action and related decisions. 
38 
1.2.2 Task Phases 
 
Figure 1.10. (a) Task used by Saga et al., (2011); Monkeys were presented with four 
sequential cues interrupted with blank delays. After the fourth cue, monkeys released a lever 
to get a reward. (b) Examples of neurons specific for cues belonging to particular phases. 
Reproduced from Saga et al., 2011. 
39 
 Another consequence of organisation of mental/behavioural events into 
discrete episodes is that the representations at the various ordinal positions assume 
distinctness because of their rank position. Had the behaviour been a hierarchy-less 
flat chain of events, rank position would have no meaning and any step would be 
equivalent to any other. A number of experiments show that fronto-parietal neurons 
code for the rank position of the behavioural events (Hasegawa et al, 2004; 
Bardyyeva and Olson, 2010; Campos et al., 2010; Saga et al., 2011). In an experiment 
by Saga et al (2011), monkeys were to wait for four sequential cues (separated by 
intervening delays) before releasing a lever and gaining a reward (Fig. 1.10a). The 
four cues on any trial were identical and could be presented in one of three sensory 
modalities (visual, auditory and tactile). The sequential cues, therefore, can be 
regarded as sequential phases that differ only in terms of their position in the 
sequence. They found that more than 60% of the lateral prefrontal neurons, sensitive 
to task events, coded information about the phase of the trial. Such neurons, 
however, varied in the actual detail of this coding, for example, some were specific to 
the cue at a particular sequential position (Fig. 1.10b) while others coded for specific 
delay periods between the cues. Another group of neurons coded for a combination 
of sequential cues and delays.  
 While representation of the sequential steps of a multi-step task is 
qualitatively distinct (Sigala et al., 2008; Saga et al., 2011), does the representation of 
the various steps differ quantitatively as well? Phenomenologically, the various steps 
do seem to be different. The final steps appear to be more salient and valued than 
earlier ones. Subjective frustration experienced is greater when subjects commit a 
mistake closer to the completion of the task (Louro et al., 2007). Likewise, neural 
activity gets higher as subjects get closer to the reward in a multistep task (Rowe et 
al., 2008; Sohn and Lee, 2006; Mizuhiki et al., 2007; Shidara et al., 2005, Platt and 
Glimcher 1999). However, these studies do not make it clear if the effect is related to 
40 
the ordinal position of the step with respect to the completion or is due to the 
proximity and imminence of reward.  
 
Figure 1.11. (a) The two tasks used by Hasegawa et al. (2004). In the self ordered 
condition, monkeys performed three steps. In the first step, they were presented with the 
three stimuli, and were explicitly signalled to choose one of them by making a saccade. In 
the next two steps, the same three stimuli were presented but at different locations, and 
monkeys chose a stimulus that had as yet not been chosen by them. In the control task, they 
were presented with one stimulus at each step to which they made a saccade. In both tasks, 
41 
monkeys were rewarded after each step. (b) to (d) Samples of neuronal activity profiles 
showing effect of task phase at the different epochs of the step. Reproduced from Hasegawa 
et al., 2004 
A study by Hasegawa et al., (2004) showed a similar increase in activity across 
the sequential steps of a task even when the reward expectancy had been controlled 
for. Monkeys performed two distinct tasks (Fig. 1.11). In the self-ordered condition, 
the monkeys fixated at the centre of an array of three pictures, chosen from a set of 
6 pictures. On the first step they made a saccade to one of these three pictures. On 
the 2nd and 3rd steps (steps 2 and 3), the same 3 objects reappeared at different 
spatial positions, and the monkeys had to make saccades to objects that had not 
been chosen on previous steps. Monkeys received rewards after every step, which 
increased with the position of the step in the ratio of 1:2:4. To control for the effect 
of reward, the monkeys did a control task in which they made simple visually guided 
saccades in three similar steps without the requirements of identification or 
memory. The reward schedule was identical across the two kinds of tasks. 
These two tasks differ in two critical aspects. The three steps of the self-
ordered condition have a means-end relation and so form the three phases of the 
same task. On the contrary, in the control condition the individual steps were 
independent of each other, and so are less strongly cohered as belonging to the 
same task. Secondly, the working memory load increased across the three steps of 
the self ordered condition since the monkeys had to remember the pictures they had 
already chosen; there was no working memory load in the control condition. 
The results (Fig. 1.11, b to d) showed that activity in half of the recorded 
neurons was modulated with the step position in the self-ordered condition (dark 
lines in Fig 1.11, b to d). This effect was weak or absent in the control condition 
(dotted lines), which excluded the possibility that reward was the main cause of 
modulation. While the working memory load did increase across the steps of the 
self-ordered condition, many aspects of their results suggest that working memory 
42 
load was not the cause behind the modulation of activity. Many neurons showed 
step modulation of activity in the post saccadic epoch (Fig. 1.11b). Note that in step 
3, monkeys can relinquish their memory load after the saccade has been made and 
hence the post saccadic epoch of step 3 need not have any cognitive load at all. 
However, as is evident in Fig. 1.11b, neurons do show a modulation of activity across 
the post saccadic phase of the three steps. Secondly, since the working memory load 
is constant across the various epochs of the step, the neural effect of load should be 
constant across all epochs of the step. However, most neurons showed the effect of 
the step position exclusively at certain epochs of the step – the neurons exemplified 
in Fig 1.11b, show the effect exclusively at the time of the saccade, while those in Fig 
1.11d, showed a decrease in activity across successive steps and this decrease was 
limited to their activity before the onset of the cue.  
1.3 Précis 
 In this thesis several effects of task organisation on the representation of 
behavioural events are investigated. In chapter 2, two fMRI experiments are 
described that investigate the fronto-parietal activity elicited by the completion of 
task episodes at different levels of hierarchy (subtasks versus tasks). The experiments 
show that the completion of defined task episodes results in activity in a number of 
fronto-parietal regions, but especially in anterior prefrontal and anterior cingulate 
cortices. It is discussed that this activity cannot be attributed to the recall of new task 
rules, termination of cognitive activity, etc.; rather, they are solely related to change 
in the organising representations underlying the task episode. 
 The third chapter looks at the representation of the different phases of a task 
episode. The fronto-parietal activities elicited by early and late phases of the trial 
were investigated.  The design further manipulated the cognitive load of the two 
phases independently. This allowed for a direct comparison of the effect of phase vis-
à-vis the effect of cognitive load. The results showed that fronto-parietal regions that 
43 
increase their activity in response to cognitive load, also increase their activity for the 
later phases of the task episode. 
 The fourth chapter investigates if the characteristics of the higher-level 
representations have a causal role in determining the structure of the organisation of 
the task episode. Results show this to be true. They also confirm the results of earlier 
chapters in a different task framework.  
 The fifth chapter looks at the pattern of activity across the voxels of a region 
elicited by the different steps of a sequential task, to investigate if the 
discriminability of these patterns, and by extension, the content of the information in 
the region, depends upon the sequential position of the step. The results show that 
the representation of varied kinds of information about the step is affected by their 
ordinal position, all of them being higher for steps closer to the completion.  
 Finally, the sixth chapter discusses the various neural possibilities underlying 
these effects. 
 
  
  
 
 
 
 
 
44 
CHAPTER 2 
Compl et ion  of  H ie rarch i ca l  Task  Ep i sodes  
 
 
2.1 Introduction 
A central feature of purposive behaviour is parcellation of a main task episode 
(e.g. preparing breakfast) into smaller subtasks (preparing coffee and buttered 
toast), whose achievement in their proper sequence culminates in achievement of 
the main purpose. The whole sequence of behavioural events (e.g. pour water → add 
coffee → add cream → toast bread → spread butter) may be organised into a nested 
hierarchy of tasks and subtasks, corresponding to the abstractedness of purpose 
being maintained. The representations underpinning the subtask would be nested 
within those underpinning the task episode.  
Attempts at modelling cognitive operations underlying organised behaviour 
find such hierarchy of representations to be critical (Cooper and Shallice, 2000). 
Without higher level representations organising the events at the lower level, the 
creation of correct sequences tends to be slow, since lot of options need to 
considered before the correct one is selected, and inflexible, as minor changes can 
upset the process (Russell and Norvig, 1995). Further such systems find it difficult to 
follow long term goal or to follow multiple goals (Gat, 1998). 
In hierarchically organised episodes, discrete behavioural events would 
correspond to the creation, revision and dismantling of representations at different 
levels of hierarchy. In the current example, the event ‘add cream’ completes the 
subtask of preparing coffee and hence marks the revision of subtask level organising 
representations, while the event ‘spread butter’ completes the whole task of 
preparing breakfast and changes the representations up till the highest level. 
45 
Correspondingly, the different behavioural events making up the sequence would 
differ in terms of the change in mental representations that takes place at them – 
with greater changes occurring at the behavioural events marking the end of 
hierarchically abstract and temporally extensive episodes.  
A key role in purposive behaviour is played by prefrontal cortex (Luria, 1966; 
Duncan, 1986).  In the behaving monkey, cells of lateral and medial prefrontal cortex 
code many kinds of information relevant to a current cognitive operation, including 
stimuli, responses, rules, working memory contents etc. (Duncan, 2001; Miller and 
Cohen, 2001; Procyk et al., 2000).  In functional magnetic resonance imaging (fMRI) 
studies of the human brain, using either adaptation or multivoxel pattern analysis to 
examine the detailed content of task representations, extensive coding of task-
relevant information is seen across multiple regions of frontal and parietal cortex 
(e.g. Li et al., 2007; Thompson and Duncan, 2009; Woolgar et al., 2011).  It is 
frequently proposed that adaptive fronto-parietal representations are a source of 
cognitive control, biasing processing in other brain regions (Desimone and Duncan, 
1995; Miller and Cohen, 2001; Rigotti et al., 2010). In such a framework, completion 
of each task step signals that a previous control representation should be dissolved 
and the next put into place (Sigala et al., 2008). In a hierarchical task that requires 
control at numerous levels, task steps crossing across hierarchical episodes would 
result in revision of representations at multiple levels . 
We know little of how fronto-parietal control activity is influenced by 
hierarchical task-subtask structure. While several studies have linked hierarchical 
task control to the organisation of prefrontal cortex, especially along the antero-
posterior axis, these studies have not been explicitly concerned with task-subtask 
structure (Koechlin, 2003; Badre, 2007, Christoff, 2006). For example, Badre (2007) 
showed more posterior frontal activity for simple one-level decisions (e.g. red > left 
key) than for two-level decisions in which one stimulus feature indicated how the 
46 
other should be processed.  Similarly, Koechlin (2003) showed anterior frontal 
activity when the context of a whole task block determines how a stimulus should be 
interpreted. Though control in such tasks is certainly hierarchical, this sense of 
hierarchy is different from a comparison of task and subtask completion in complex, 
multi–step behaviour. 
To investigate this issue, we devised a sequential target detection task with a 
hierarchical structure, such that some target detections represent the achievement 
of subgoals, others the achievement of a higher level goal.  We reasoned that, when 
a subtask is completed, its specific content loses relevance, but higher level 
representations pertaining to the overarching task episode must remain in 
behavioural control. By comparison, completion of a more substantial behavioural 
episode requires a more substantial revision of task related representations, and 
perhaps producing a stronger or more extensive pattern of fronto-parietal activity. 
2.2 Materials and Methods 
2.2.1 Tasks 
Experiment 1 
Two experiments were conducted. In the main experiment (Experiment 1), 
participants monitored a series of letters presented at a rate of 1/1.3s on a computer 
screen (Fig. 2.1). The task was to detect four target letters in turn. A three - letter 
word (e.g. ‘CAT’) presented at the start of the trial indicated the first three targets 
(e.g. Fig. 2.1, T1 ‘C’, T2 ‘A’, T3 ‘T’).  The fourth target was always the letter X. At the 
end of the 52 s letter stream, the participants were probed to indicate whether X had 
appeared (true on 50% of trials called complete trials, remaining trials referred to as 
incomplete trials). 
 
47 
 
 
 
Figure 2.1. Structure of a typical trial (Experiment 1). Trials began with a 3-letter cue word. 
The three letters of this word were to be covertly detected, in the correct order, in the 
ensuing letter stream; after all three had been detected, search for the letter ‘X’ began. The 
complete sequence of four target letters appeared in only half of the trials. The letter stream 
ended with a probe asking whether the letter ‘X’ had appeared in the correct sequential 
position, i.e. following the three letters of the target word. Thus the first two targets (T1 and 
T2) changed the representations at the lowest level (component letters of the first target 
word; level 1); the third target, T3, completed the representations at the next highest level 
(complete target word; level 2), while the fourth target, X, completed the whole goal of the 
task, and changed the representation at the highest level (level 3). 
 
At the start of each trial, the instruction screen specifying the first three 
targets (Fig. 2.1, ‘CAT’) was presented for 3500 ms.  The letter sequence began after 
a jittered gap of 1000 – 5000 ms, each letter being presented for 800 ms with a gap 
of 500 ms in between consecutive letters. The letter stream consisted of a total of 40 
letter presentations and lasted 52 s. Participants began by monitoring for T1 (Fig. 2.1, 
‘C’), at the detection of which search for T2 started, and so on.  Targets were only 
48 
relevant once the appropriate preceding targets had already been detected; for 
example, while Xs might appear at any point in the sequence, only an X appearing 
after sequential detection of T1, T2 and T3 was relevant.  In this way participants 
were obliged to search for the specified four targets in turn, with a positive response 
at trial end only if all four targets had appeared in the correct order. Within each 
trial, the inter-target interval varied randomly between 1.3 and 32 s. Responses were 
made when the letter stream terminated with a probe (‘X?’ displayed for 2000 ms) 
asking whether the letter ‘X’ had appeared at the relevant position i.e. after the 
detection of T1-T3.  Responses were made on a button box positioned under the 
participant’s right hand (index finger for ‘yes’, middle finger for ‘no’). A variable inter-
trial interval of 2000 to 7000 ms preceded onset of the cue for the next trial. (See 
Appendix B.2 for exact instructions). 
All stimuli were centred on the screen, visible from the participant’s position 
in the scanner via a mirror mounted within the head coil. Letters subtended a visual 
angle of 2° vertically.  The experiment was controlled by a program written in Visual 
Basic. Subjects learnt the task in a ten minute pre-scan practice session and 
proceeded to a scanning session of an hour, which was divided into three separate 
scanning runs, each consisting of 20 trials.   
 Task-subtask structure was emphasised in the initial instructions that 
mentioned “two tasks are to be done on each trial; search the letters of the cued 
word, and then search for X”. This was reinforced by the distinction of the sources 
that guided these two tasks - a cue-word to be kept in working memory as it changed 
from trial to trial, and the letter ‘X’ that was constant across trials.  According to such 
a structure, T1 and T2 completed subtasks at the lowest level (component letters of 
the initial target word; level 1), T3 completed a task at the next higher level (target 
word completed; level 2), while ‘X’ completed the entire task episode (level 3). 
Participants were informed that  correct ‘yes’ responses on complete trials increased 
49 
the current score by +1, whereas correct ‘no’ response on an incomplete trials did 
nothing to the score; an incorrect response on trial decreased the current score by -
1.  
 
 
Figure 2.2. Structure of a typical trial in Experiment 2, similar to Experiment 1 but with only 
three targets T1-T3. 
 
Experiment 2 
In Experiment 1, when each of targets T1 to T3 was detected, the previous 
search was to be abandoned and a new one began. In contrast, detection of the final 
target (X) was associated with termination of the search and an increase in the 
current score, which could be interpreted as a form of reward. To examine the 
significance of these features, in a follow-up experiment (Experiment 2) the task was 
slightly modified - T3 detection completed the task, i.e. there was no requirement to 
50 
search for a final ‘X’ (Fig. 2.2). Visual search now terminated at T3 rather than X; 
further an explicit reward (of 10p) was given at the completion of the task. 
Comparison between experiments allowed a direct examination of the importance of 
these factors. At the end of each trial participants were probed to indicate whether 
all the three target letters had appeared in the correct order (Fig. 2.2, ‘DAT?’); as 
before this was the case in only half the trials, with 25%, 18% and 6% of the trials 
having two, one and no target letters respectively. In Experiment 2 each letter 
stream was composed of just 40 letters, lasting a total of 40 s.  The scoring system 
was same as before, but with an explicit addition of monetary reward. 
An inherent problem in experimental designs focussing on sequential events is 
that the latter events will necessarily occur after the earlier ones and hence there is a 
possibility of such event regressors being correlated. This, however, was obviated in 
the design of the current experiments by having a large temporal jitter between the 
different target events (1.3 to 32 s). In the second experiment an additional feature 
was added. Here the number of target events appearing in a trial varied from none to 
all, thus allowing for greater dissociation between them as the earlier target events 
were not always followed by the later target events. Finally, a distinct set of FIR 
regressors modelled the entire duration of the letter stream (see below) and ensured 
that effects related to the temporal positions within the letter stream were 
regressed out and did not affect the estimates of target related activity. Lastly, the 
effect of expectancy was obviated by ensuring that the probability of any instance of 
the letter stream being a target was constant across the entire sequence of the letter 
stream. 
2.2.2 Participants 
18 participants (10 female, mean age = 22.5 ± 3.6 years) in the first 
experiment and 21 (15 females, mean age 24.5 ± 4.1years) in the second experiment 
were recruited from the MRC-CBU volunteer panel. Participants were right handed 
51 
and had normal or corrected vision. Informed consent was taken and the participants 
were reimbursed for their time. The study had the approval of Hertfordshire Local 
Research Ethics Committee. 
2.2.3 Acquisition 
fMRI data were acquired using a Siemens 3T TimTrio scanner with a 12 
channel head coil. A sequential descending T2*-weighted echo planar imaging (EPI) 
acquisition sequence was used with the following parameters: Acquisition time 2000 
ms; echo time 30 ms; 32 oblique slices with slice thickness of 3 mm and a 0.75 mm 
interslice gap; inplane resolution 3.0x3.0 mm; matrix 64x64; field of view 192mm; flip 
angle 78 deg. T1-weighted MPRAGE structural images were also acquired for all 
participants (slice thickness 1.0 mm, resolution 1.0x1.0x1.5 mm, field of view 256 
mm, 160 slices).  
2.2.4 Analysis 
The fMRI data were analysed using SPM5 (Wellcome Department of Imaging 
Neuroscience, London, England; www.fil.ion.ucl.ac.uk). Prior to statistical analysis, all EPI 
volumes were slice-time corrected using the first slice as a reference, and then 
realigned into a standard orientation using the first volume as a reference. These 
realigned were then normalised into the Montreal Neurological Institute (MNI) 
space, and spatially smoothed using an 8 mm full-width half-maximum (FWHM) 
Gaussian kernel. During the normalisation stage, voxels were resampled to a size of 
3 × 3 × 3 mm. The time course of each voxel was high pass filtered with a cutoff 
period of 90 s. 
Statistical analysis was carried out using a general linear model. To capture 
activity related to target detection, a 16 sec epoch starting from each target onset 
was modelled using an FIR basis set of eight 2 s long boxcar regressors. In this way 
the response to target detections could be modelled without making any 
52 
assumptions about the shape of the BOLD response. To capture and regress out the 
possible effects of position of target events in the letter stream, the entire letter 
stream was modelled using another FIR set of 25 2 s long boxcar regressors (in 
experiment 2, the duration of letter stream required only 20 such regressors, since 
the letter stream were shorter in this experiment). Additionally, the cue and probe 
were modelled using epoch regressors, of width equal to the duration of respective 
events, convolved with a basis function representing the canonical hemodynamic 
response. Movement parameters and block means were included as covariates of no 
interest. Parameter estimates for each regressor were calculated from the least 
squares fit of the model to the data, and estimates for individual participants were 
entered into a random effects group analyses.  
Whole-brain comparisons were performed using paired t-tests on the relevant 
contrast values from each participant’s first-level analysis. Unless otherwise 
specified, all results are reported at a threshold of p < .01, corrected for multiple 
comparisons using the false discovery rate (FDR). Coordinates for peak activation are 
reported using an MNI template. 
To capture fronto-parietal regions widely engaged in cognitive control, ten 
regions of interest (ROIs) were created as spheres of 10 mm radius at coordinates 
(Table 2.1) that have been shown to be consistently active in varied tasks (Duncan, 
2006; Dosenbach et al., 2006). The ROIs (in MNI space) were bilateral inferior frontal 
sulcus (IFS; central coordinate ±41 23 29), bilateral intra-parietal sulcus (IPS; ±37 -56 
41), bilateral anterior insula extending into frontal operculum (AI/FO; ±35 18 3), 
anterior cingulate (ACC; 0 31 24), and pre-supplementary motor area (pre-SMA; 0 18 
50), all taken from Duncan (2006); along with bilateral anterior prefrontal cortex 
(APFC; 27 50 23 and -28 51 15) taken from Dosenbach (2006). ROIs were constructed 
using the MarsBar toolbox for SPM (http://marsbar.sourceforge.net; Brett, 
Johnsrude & Owen, 2002). Estimated data were averaged across voxels within each 
53 
ROI using the MarsBar toolbox and the mean values were exported for analysis using 
SPSS. 
ROI Coordinates 
IFS  ±41 23 29 
IPS ±37 -56 41 
AI/FO ±35 18 3 
APFC  R 27 50 23 
              L -28 51 15 
ACC 0 31 24 
pre-SMA 0 18 50 
 
Table 2.1 List of ROIs with their coordinates in MNI space. Note that for midline regions 
(ACC and pre-SMA), a single ROI centred on the midline was used. 
2.3 Results 
2.3.1 Experiment 1 
Average response time to the probe was 723 ± 15 ms, and accuracies for 
responses to the probe exceeded 97% for most subjects (mean 97.1 ± 1.4), 
suggesting that they were indeed covertly detecting target letters. 
Behaviour 
ROI analyses  
Imaging 
54 
 
Figure 2.3. (Exp 1) Time course of activity in the different ROIs, constructed from the 
estimates of eight FIR regressors linked to each target event (blue: T1; green: T2; red: T3; 
purple: X).  
 
The FIR model provided estimates of activity in eight successive 2-second long 
windows starting from each target onset, which were plotted to construct an 
estimate of the time course of activity following each target detection. Fig. 2.3 shows 
these plotted for the ROIs; the difference across the various target detections is 
evident. Note that the time courses for T1 and T2 are nearly identical in many ROIs. 
For further analysis, an index of phasic BOLD response was calculated by subtracting 
the estimate of first FIR bin from the average of the second and third.  Mean values 
of this index for each ROI appear in Fig. 2.4. It is evident that X, which completes the 
overall goal (level 3) shows the highest activity in all ROIs. In contrast, T1 and T2 
(level 1 subgoals) show the least; indeed in ACC, pre-SMA and APFC, as is evident 
from the error bars depicting one standard error, their activation index does not 
differ from zero. 
55 
 
Figure 2.4. (Exp 1) Comparison of phasic activity in response to various target events. An 
index of phasic activity was derived for each target event by subtracting the estimate of the 
first FIR regressor from the average of the second and third. The error bars represent one 
standard error of the mean. All ROIs showed a main effect of target position on the 
magnitude of this index (Table 2.2a). 
 
 
 
(a) 
 
 
 
 
 
 
 
 
 
ROI Target 
position 
F(3,51) 
Hemisphere 
 
F(1,17) 
Target x 
Hemi 
F(3,51) 
IFS  4.9* 11.4* 14.2* 
IPS 6.3* 9.9* 2.7 
AI/FO 7.1* 2.6 1.4 
APFC  9.1* 4.9* 3.4* 
ACC 9.0* - - 
Pre-SMA 11.2* - - 
  
56 
 
(b) 
Table 2.2. [a] Effect of the position of target detection, laterality of the region and their 
interaction in different ROIs. Values in bold were significant at p <0.05.  [b] t(17)
 
 values from 
the pairwise comparisons of different target detections at various ROIs. Values in bold were 
significant at a Holm-Bonferonni corrected threshold of 0.05 for multiple comparisons. 
 
To confirm that activity across the various target type, a repeated measures 
ANOVA was done with target type (T1, T2, T3, X) and ROIs as factors. As would be 
expected, there was a significant main effect of target type (F3, 51 = 13.5, p <0.001). 
The main effect of ROIs was significant as well (F9, 153 = 9.4, p <0.001), and so was an 
interaction between ROIs and target type (F27,459 = 2.0, p <0.01), showing that the 
differential response to the target types varied across the ROIs. 
ROI Laterality  T1 & T2 vs T3 T1 & T2 vs X T3 vs X 
IFS  L 1 1.2 0.3 
  R 0.67 4.9* 3.8* 
IPS L 1.9 2.6* 0.6 
  R 2.5* 4.4* 1.8 
AI/FO L 2 2.9* 1.6 
  R 1.2 3.7 * 3.2* 
APFC  L 2.8* 3.7* 0.8 
  R 3.3* 4.8* 1.7 
ACC   2.6* 4.4* 3.1* 
Pre-SMA   2.2 4.8* 3.0* 
  
57 
For more detailed investigation, activity indices for each ROI were examined 
by ANOVA, with factors target type and hemisphere (for lateral ROIs only).  All ROIs 
showed a main effect of target type (Table 2.2a). Amongst the bilateral ROIs, IFS, 
APFC and IPS showed a main effect of laterality reflecting greater activity on the 
right. In IFS and APFC significant interaction of target type with hemisphere showed 
that the differential activity was greater in the right hemisphere.   
To amplify the significant effect of target type, pairwise comparisons 
compared T3 (level 2 subgoal) with the average of T1 and T2 (level 1). APFC, ACC and 
right IPS showed significantly higher activity to T3 (Table 2.2b). Amongst other target 
events, X was significantly higher than T1 and T2 in all ROIs except left IFS. In all ROIs, 
the tendency was for greater activity to X than T3, however, this was significant only 
in right IFS, right AI/FO, ACC and pre-SMA. There was no significant difference 
between T1 and T2 in any ROI. 
  
 
 
 
 
 
 
 
58 
Whole Brain Analyses 
 
Figure 2.5. (Exp 1) Regions where the activation indices for the various targets were 
significantly greater than zero. As in the ROIs, X and T3 elicited activity in much wider 
network of regions than T1 and T2. 
  
In the first series of contrasts, the regions where the activation index for the 
different target detections differed significantly from zero were investigated. As in 
the ROIs, the involvement of other regions varied with the position of the targets. T1 
and T2 (Fig. 2.5a) did not elicit much activity beyond few foci in posterior frontal 
(dorsal premotor and intrafrontal junction) and IPS. T3 (Fig. 2.5b) elicited much more 
widespread and intense activity spreading onto anterior and medial prefrontal 
59 
regions, anterior insula, precuneus; activity was most widespread for X (Fig. 2.5c) and 
additionally involved interparietal lobule, temporo-parietal junction and cuneus.  
 
Figure 2.6. (Exp 1) [a] Whole brain render showing areas that had higher activation index for 
‘T3’ compared to ‘T1’ and ‘T2’ ; [b], areas with greater activity for ‘X’ than ‘T1’ and ‘T2’. 
 
To check if these differences were significant, two contrasts compared the 
activity during T3 and X with that during the earlier target detections. Consonant 
with the ROI results, the first contrast (Fig. 2.6a) showed that T3 elicited greater 
activity in APFC, ACC and parts of pre-SMA along with lateral parietal and precuneus 
(for list of coordinates see Appendix, table A2.1). Results of the second contrast 
showed greater activity for X in widespread frontal, parietal and occipital regions 
with a dominance on the right (table A2.2). In the right prefrontal cortex it involved 
inferior frontal sulcus, frontal operculum and the anterior insula, extending 
anteriorly through the middle frontal gyrus up to anterior prefrontal region. In the 
60 
left prefrontal cortex only the anterior prefrontal regions and rolandic operculum 
were involved. Medially, stronger activity for X compared to T1/T2 was seen in the 
anterior cingulate and the pre-SMA on both sides. Posteriorly, significant activity was 
found in the intra-parietal sulcus extending into the temporo-parietal junctions, 
precuneus, cuneus and parts of calcarine and lingual regions. 
To summarise activity was strongest and most widespread in response to the 
completion of the final goal (level 3). Selected regions especially APFC and ACC 
showed greater activity for change in representations at level 2 compared to that at 
level 1. 
Default Mode Network 
Apart from the MD regions, there are other fronto-parietal regions that show 
modulation of BOLD signal during task phase, however these regions that are part of 
what is commonly called the default mode network show a decrease in BOLD signal 
during task periods, when MD regions show increased BOLD signal. It is commonly 
assumed that this decrease in DMN regions is an undifferentiated response to 
cognitive load during the task periods compared to periods of relative cognitive 
inactivity (Fox et al., 2005; Buckner et al., 2008). However, some recent evidence 
suggests that these regions do have a task dependent modulation of activity. For 
example, Mayer et al., (2009) showed that qualitatively different regions of the DMN 
responded to visual attention load versus working memory load. 
A recent study gives evidence of possible role of DMN regions in the accessing 
hierarchically organised information (Rogers et al., 2010). Subjects were trained to 
do discriminate between image pairs using transitive inference. Some image pairs 
were derived from an implicit image sequence (e.g. A>B>C>D>E), whereas others 
were independent (e.g. F>G, H>J). The authors found greater functional connectivity 
between DMN regions (like posterior cingulate) and task related (multiple demand) 
61 
fronto-parietal regions when subjects were making judgements on pairs belonged to 
the implicit sequence than when the image pairs belonged to independent sets.  
 It is not known whether activity in the DMN fronto-parietal ROIs is modulated 
by the organisation of the task episode. Since the four target detections in the 
current task are matched in their attentional and working memory demands, a 
variance in their modulation of activity across the four targets would suggest that 
these regions are indeed sensitive to the organisation of the task episode. 
 
 
Figure 2.7. (Exp 1) Estimated time course of activity in the Default Mode ROIs (blue: T1; 
green: T2; red: T3; purple: X). Note that the change in activity is greater for T3 and X. 
62 
Three ROIs were constructed based on Dosenbach (2007) – ventromedial 
prefrontal cortex (1, 31, -2), and bilateral posterior cingulate (10, 56, 16; -11, 77, 13). 
The time course of activity in these ROIs for the different target events (Fig. 2.7) 
show that while first and second target detection barely cause any decrease in 
activity, the third and the final target detections caused mark decrease in BOLD 
signal. To compare the phasic change in activity in response to the various target 
detections, the first five FIR bins were compared for each ROI across the four target 
detection events to see if the effect of time interacted with that of the target type.  
All three DMN ROIs showed a significant interaction between time bin and 
target type, suggesting that modulation of activity in these regions was different for 
different target types (vmPFC: F(12, 204) = 2.07 p<0.05; right posterior cingulate: F(12, 
204)= 4.03, p<0.001, left posterior cingulate: F(12, 204)
 
= 2.74, p<0.01). The effect of task 
episode organisation is not just limited to fronto-parietal regions that show an 
increase in activity during task episodes. The result further suggests that the 
modulation of activity in the DMN regions during task episodes is not 
undifferentiated inhibition rather they also seem to carry information about the 
structure of organisation of the episode. 
2.3.2 Experiment 2 
In Experiment 2, only the 3 letters of the target word (T1 to T3) were to be 
found on each trial.  By associating T3 with task termination and reward – which in 
Experiment 1 were linked to the final target ‘X’ – we planned to investigate the 
importance of these factors. For brain regions where activity is sensitive only to 
task/subtask structure, the contrast of T3 vs. T1/T2 should show similar results in the 
two experiments. Additional influence of task termination and reward should be 
63 
shown by additional T3 activity in Experiment 2, corresponding to the strong activity 
seen, in Experiment 1, for the final ‘X’. 
Average response time to the probe was 770 ± 18 ms, and accuracies for 
responses to the probe exceeded 95% for most subjects (mean 96.2 ± 2.7). 
Behaviour 
 
Imaging 
 
Figure 2.8. (Exp 2) Comparison of phasic activity in response to the three target events in 
experiment 2. Only APFC and ACC showed effect of target position on the magnitude of 
activity.  
 
64 
Fig. 2.8 shows plots of index of activation for the three target events from the 
various MD ROIs. A repeated measures ANOVA with target type (T1, T2, T3) and 
hemisphere (for lateral ROIs only) as factors showed the significant effect of target 
type only in APFC and ACC, where responses to T1 and T2 were not significant (one 
sample t(20) 
To directly compare the two experiments, we carried out an ANOVA with 
factors target type (T1, T2, T3), hemisphere (for lateral ROIs only) and experiment. 
No ROI showed a significant interaction, either between experiment and target type 
(p >0.3 in all comparisons) or between experiment, target type and hemisphere (p > 
0.08 in all comparisons). The results suggest that associating T3 with task termination 
and reward has little effect on the fronto-parietal activity. Instead it is the 
hierarchical position of the event in the plan of the task that determines activity, in 
particular the strong response to T3 in the APFC and ACC.  
< 1.5, p>0.12) 
 
Figure 2.9. (Exp 2) The contrast of T3 against T1 & T2 showed significant activity in anterior 
prefrontal, pre-SMA, ACC along with regions near TPJ, precuneus and cuneus. 
 
Lastly, the activity during T3 was contrasted with that during the earlier target 
detections. The results in fronto-parietal regions were very similar to those in the 
contrast of T3 and earlier targets in the previous experiment, although the current 
contrast additionally involved precuneus, cuneus and TPJ. Recall that the latter set of 
65 
regions were seen in the contrast of earlier targets with X. This makes it likely that 
these regions were specifically active at the end of the overall task episode.  
2.4 Discussion 
These experiments show that in various regions of the fronto-parietal cortices, 
activities generated by task events vary with the hierarchical status of the episode 
they complete. Events completing the lowest level subtasks (level 1), led to the 
weakest fronto-parietal activity. At the next higher level (level 2), stronger activity 
was seen especially in anterior and medial prefrontal regions. Completion of the top-
level task or the whole trial (level 3) resulted in greatest and most widespread 
activity. These results show that the fronto-parietal activity elicited by a task event 
depends upon the role played by it in the overall organization of the behaviour.  
Why should completion of task episode elicit fronto-parietal activity?  A task 
episode is characterized by the organization of multiple mental processes to enable 
task-relevant processing in various cognitive domains. While the exact nature of 
mental structures that organize such an episode is a matter of ongoing investigation, 
task descriptions, representations of purpose/goal (Miller and Cohen, 2001; Duncan, 
2006), and task rules (Duncan et al., 2008; Sakai, 2008; Dumontheil et al., 2010) are 
considered important components that could act as the source of control and bias 
representations in diverse brain systems to achieve relevant neural processing. At 
the end of a task episode, these structures would be changed or dismantled 
depending upon the nature of the future behavioural episode. A plausible 
interpretation of our result is that the restructuring of these mental representations 
at the end of the corresponding episode is the elicitor of fronto-parietal activity.  
Purposive behaviour is further characterized by a hierarchical organization in 
which the overall task episode is further organized into smaller nested episodes, 
each one having its own cognitive focus and organized to achieve a smaller goal 
66 
(Lashley, 1951). This organisation is achieved through a similar nested hierarchy of 
mental representations (Miller et al., 1960; Schank and Abelson, 1977; Barsalou 
1988; Reed et al. 1995; Schneider and Logan, 2007), wherein the lower level 
representations organize smaller episodes of behaviour which in turn are organized 
by higher level representations into larger and more complex behavioural episodes. 
As lower level nested episodes are completed, the lower level representations lose 
their relevance, while the higher level representations organizing larger episodes 
remain in control; on the other hand completion of the overall goal changes the 
representations at all levels and brings about a greater change in the mental 
organisation. It can hence be expected that the activity elicited by the changes in the 
mental representations organizing the behaviour would be more intense/widespread 
at the completion of higher level episodes. Our results above show that this was 
indeed the case in fronto-parietal regions.  
Apart from a change in the underlying representations, completion of 
purposive episodes additionally corresponds to the termination of the ongoing task 
(and so a change in the task set) and to a feeling of reward corresponding to the 
achievement of the goal. These, however, appeared to be less significant. T3 of 
Experiment 2 was similar to 'X' of Experiment 1 in terms of these factors, since visual 
search terminated at both, and, plausibly, both resulted in a feeling of a successful 
completion of the trial. T3 of Experiment 2 additionally had a monetary reward 
attached to it. However, in terms of the level of episode completed, this T3 was 
identical to the T3 of experiment 1 (Fig. 2.1 & 2.2). As would be expected if the 
hierarchy of episode completed was the major determinant of fronto-parietal 
activity, T3 in the two experiments elicited very similar patterns of fronto-parietal 
activity (Fig. 2.4 & 2.8). Hierarchical changes in the representations organizing the 
behaviour, therefore, seem to be the most plausible explanation of our results. 
67 
Earlier investigations into changes in task related representations have 
approached it with the concept of task set (Sakai, 2008), understood as a 
neurocognitive structure linking the stimulus to the relevant response as per the task 
rules and have shown that restructuring the task set in response to changes in task 
rules corresponds to activity in the fronto-parietal regions. Our concept of the 
organising representations subsumes this concept of task set and also includes other 
representations like task and goal descriptions. End of task episodes will always 
revise the representations corresponding to the task and goal descriptions even if 
the relevant rules linking stimulus to response remains the same, as was the case in 
our experiments. As discussed above changes in the task set that occurred at the 
termination of the visual search were less important than changes in task 
description. 
Our results also showed a functional distinction amongst the fronto-parietal 
areas. Regions like IPS, IFS and pre-SMA were active during all target events, whereas 
APFC and ACC were active only at the completion of task episodes, suggesting that 
while any change in the underlying representations elicited activity in the former 
areas, only changes in higher level representations activated the latter.  APFC and 
ACC have been shown to be active at the beginning of the trial blocks (e.g. 
Dosenbach et al., 2006; Sakai and Passingham, 2006; Haynes et al.,, 2007) and to 
have sustained activity for the duration of task episodes (e.g. Koechlin et al., 2003; 
Braver et al., 2003; Dosenbach et al., 2006; Marklund et al., 2007). In the light of our 
findings, it is plausible that this pattern of activity reflects the role of these regions in 
the creation and maintenance, respectively, of the higher level task representations 
that organize the whole episode.  
Compared to our finding that anterior prefrontal activity can be elicited in 
very simple mental conditions without any serious requirement of control or working 
memory organisation and manipulation, a number of other studies have found such 
68 
activity in situations that require maintenance of information across periods of 
unrelated tasks (Koechlin et al., 1999; Pisapia et al., 2007; Reynolds et al., 2009; 
Gilbert, 2011). Typically, such studies compares trials in which more than one task 
are executed such that information about one of the task has to be maintained while 
a different task is being executed and which is followed by the task that required the 
maintained information, to trials in which the same tasks are executed sequentially 
but without there being a requirement to maintain any information across a different 
task period.  
In a task by Koechlin et al., (1999) subjects did four kinds of trials (Fig. 2.10). In 
the control condition, they judged whether two successively presented letters were 
also in immediate succession in the word ‘tablet’ (only upper-case letters were 
presented). In the delay conditions, subjects had to ignore lower-case letters which 
were used to occasionally delay the response required by the control condition. In 
the dual task condition, subjects had to respond as in the control condition for both 
upper and lower case letter series, but at every first letter showing a case change 
they judged if the letter was ‘T’ (or ‘t’). Finally in the branched condition, they had to 
respond to the upper case as in the delay condition and to the lower case as in the 
dual task condition. This meant that specifically in the branch conditions, subjects 
had to remember the identity of the last upper case letter presented before the case 
switch, while they were responding to the lower case letters. Anterior prefrontal 
activity was found associated with the branching task condition. 
69 
 
Figure 2.10. Design of task used by Koechlin et al., (1999). Four kinds of trials were used. 
Control condition: subjects had to decide whether two successively presented letters were 
also in immediate succession in the word `tablet' (only upper-case letters were presented). 
Delay condition: subjects had to ignore lower-case letters which were used to occasionally 
delay the response required by the control condition. Dual task condition: subjects had to 
respond as in the control condition for both upper- and lower-case letter series with one 
exception. Subjects had to decide whether every first letter indicating a case change was the 
letter T (or t). Branching condition: subjects had to respond to upper-case letters exactly as 
in the delay condition and to lower-case letters exactly as in the dual-task condition. 
Reproduced from Koechlin et al., (1999) 
 
Conceptually similar task design was used by Pisapia et al, (2007), Reynolds et 
al., (2009) and Gilbert (2011), in which the condition resulting in the anterior 
prefrontal activity involved a task episode during which some not immediately 
relevant information was to be maintained, which was used to execute a task after 
the current episode was over.  
It can be argued that the requirement to maintain information across an 
irrelevant episode binds what would have been two unconnected task episodes into 
a common task. Binding the separate task episodes into a common task would create 
a task plan that has higher levels of organisation than were present in the separate 
70 
task episodes. In the dual task condition of Koechlin et al., (1999), the lower case and 
higher case episodes were disconnected to each other. However, in the branching 
task condition, the requirement to maintain the upper case information across the 
lower case episode means that the lower case episode be represented as part of the 
same higher level mental episode as the upper case one, resulting in a more 
hierarchical mental plan.  
Another support for this thesis comes from studies on reasoning (Smith et al., 
2007). Compared to the trials that require separate inferences of the relation 
amongst two separate pairs of objects, trials that additionally require a comparison 
of these relations elicited greater activity in APFC. Again, making separate inferences 
on two separate pairs creates two separate mental episodes, while the requirement 
to compare the two inferences could result in a common organisation of these 
hitherto separate episodes.  
The current results show that a hierarchical mental plan without any 
requirement of maintaining and integrating information across branched and 
unrelated task episodes can result in anterior prefrontal activity.  
Other fronto-parietal regions – IFS, IPS and pre-SMA, in contrast, are active 
with any cognitive challenge (Duncan, 2010). Regions like IFS (DLPFC in many studies) 
show transient activity in situations that show sustained activity in APFC (Braver et 
al., 2003; Dosenbach et al., 2006; Reynolds et al., 2009). Interestingly, change in task 
representations within an episode of behaviour, for example, switching between the 
different rules active in the block, frequently result in fronto-parietal activity which in 
lateral PFC tends to be limited to the IFS (e.g. Dove et al., 2000; Braver et al., 2003; 
Brass et al., 2005; Savine and Braver, 2010; Kim et al., 2011).  
Additionally our results also suggest a functional distinction along the antero-
posterior axis of the lateral and medial prefrontal cortex. Consonant with our finding 
71 
that the behaviour of IFS and pre-SMA were similar, while that of APFC paralleled 
ACC, Taren et al. (2011) found that the posterior regions of medial PFC had maximal 
resting state functional connectivity with the posterior parts of lateral PFC, whereas 
anterior regions of the two were maximally connected with each other.  
Earlier studies have tried to link different aspects of hierarchical behaviour 
with different regions along the antero-posterior axis of the lateral PFC. Badre and 
D’Esposito (2007) noted that tasks with increased levels of decisions were 
accompanied with activity in more anterior regions of lateral PFC. Compared to 
single level decisions where one feature (e.g. colour) determined the response, two 
level decisions, where one feature (e.g. shape) determined how the other feature 
(e.g. colour) would determine the response, was associated with more anterior locus 
of prefrontal activity. APFC activity was found in tasks that required four level 
decisions (i.e. when the context of the trial block determined the relevant 
dimensions which in-turn determined the relevant features that determined how the 
colour would map to the response). Koechlin et al. (2003) found that increasing the 
temporal abstraction of the context that determined a response led to more anterior 
locus of prefrontal activity, for e.g. compared to the trials when the trial information 
was sufficient to allow response selection, trials that required the consideration of 
the context of the block for response selection to occur elicited activity at more 
anterior loci in the lateral PFC. 
Clearly in both of these studies, as noted earlier for studies showing activity in 
anterior prefrontal regions, the kinds of trials that result in anterior activity require a 
more complex behavioural plan for execution. For example, in Badre and D’Esposito 
(2007), compared to trials with single level decision wherein the colour directly 
allows for the selection of response, trials with two-level decisions have got an 
additional level of hierarchy - subjects first determine what colour to response 
mapping is valid in the light of the shape (subgoal), and then determine the relevant 
72 
response based on the colour (completion of goal). Similarly, in Koechlin et al. (2003), 
the trials wherein the context of the block determined how the stimuli will map onto 
the response, clearly required an additional level of representation (context of the 
trial block) compared to the trials where the response could be determined by the 
stimulus itself.  
A difference between earlier studies and our study is worth noting. Earlier 
studies had found APFC activity in trials that had very high cognitive demands 
compared to trials that had activity limited to posterior prefrontal regions. For e.g. 
four nested levels of decision processes (Badre and D’Esposito, 2007) or information 
integration across three/four levels (Koechlin and Summerfield, 2007). In our 
experiments simple events marking the end of trial, without concomitant complex 
decision process or complex information integration, produced activity in these 
regions (cf. Hon et al., 2006).  
Neurophysiological evidence is also supportive of our findings. Neurons from 
around the principal sulcus have been found to be active during all kinds of task 
events (reviewed in Duncan, 2010) and seem to code for all stages of the task 
episode (Sigala et al., 2008). In contrast, in the only neurophysiological study of the 
anterior prefrontal cortex, Tsujimoto et al., (2010) found its neurons to be active only 
at the end of the task episode. 
An important suggestion of the current results is that mental organisation is 
itself an elicitor of fronto-parietal activity. It is obvious in the light of above results 
that to answer a question like – what neural regions underpin visual target 
detection? – requires consideration of the role of the act of target detection in the 
mental organisation underpinning the experimental design. An experimental trial is 
itself an organised mental episode, this fact needs to be kept in mind while making 
conclusions about the neural correlates of cognitive events.  
73 
CHAPTER 3 
Phase of Task Episode and Fronto-Parietal Activity 
 
3.1 Introduction 
 
Figure 3.1. Scheme of a typical plan underlying a goal directed behavioural episode. The 
sequence of cognitive/task events are organised into discrete subtasks, which are further 
grouped into a higher level task.  
 
Consider a behavioural episode consisting of a sequence of cognitive events 
(‘A’ to ‘h’), which is organised as a single task episode comprising of two subtasks 
(Fig. 3.1). In this behavioural episode, apart from their intrinsic identities (perception, 
decision making, response selection, task switch and so forth), the cognitive events 
also have unique roles due to their position in the plan of the behaviour. These roles 
would be the hierarchy of goal they achieve (‘D’ completes a subtask, while ‘H’ 
completes a task and the whole plan) and the phase of the plan they represent 
(events in ‘subtask1’ represent the early phase of the plan whereas those of 
‘subtask2’ represent the final phase since they culminate in the completion of the 
plan). These roles of the cognitive events come from the peculiarities of the 
74 
organisation of the mental episode, which, amongst others, would depend upon the 
goals to be achieved, the trajectory planned to achieve the goal, the organisation of 
the semantic knowledge of the world, the conceptual structure of the actions to be 
done, habits and the environmental/social constraints of the subject.  
How does the context of the plan affect the neural activity elicited by a 
cognitive event within it? Does the role that the cognitive event has in the plan of the 
behaviour affect its representation in the fronto-parietal regions? In the previous 
chapter, it was shown that compared to the target detections within a subtask, those 
completing the respective subtask and task led to increasingly greater fronto-parietal 
activity, and the target detection that completed the whole plan resulted in maximal 
activity. This chapter will explore whether the phase of the plan that the cognitive 
events represent affects their elicited neural activity. This should be the case, if the 
fronto-parietal activity representation of an event depends upon the larger context. 
During the early phases of the episode, representations about the upcoming phases 
or steps have to be additionally maintained, while during the final phase only the 
representations relevant to that the current step need to be maintained. It is 
possible that this might affect the fronto-parietal activity elicited during the different 
phases of the task episode. 
Our experimental design required the subjects to do two blocks of tasks on 
every trial (fig. 3.2). One block consisted of arithmetic tasks and the other of word 
tasks, their order in the trials was balanced, such that the first and the second task 
blocks composing the trial consisted of both kinds of tasks and were identical when 
averaged across the experimental session. To ensure that the subjects conceive the 
two blocks as part of the same mental plan, information relevant to both blocks was 
cued simultaneously at the beginning of the trial and the subjects executed the first 
block using the information relevant to that block, while separately maintaining the 
information relevant to the second block, and subsequently executed the second 
75 
block. Due to this feature of the design, while completing the first block, the subjects 
still had to maintain the relevant information for the second block and only at the 
completion of the second block could they rest and wait for the next trial to begin. 
Thus, the plan of a trial can be considered to be biphasic –the completion of the first 
task block completes the initial phase, and the completion of the second completes 
the final phase. Since subjects always maintained information relevant to the second 
block while executing the first, when averaged across the length of the experimental 
session, the cognitive load of the first block was higher than that of the second. 
Hence from this aspect of the design it could be expected that brain areas sensitive 
to cognitive load would be more active during the first block, whereas brain areas 
that prefer the final phase of a mental plan would be more active during the second.  
The main aim was to see the effect of the phase of the mental plan 
represented by the task on its elicited fronto-parietal activity. In current 
understanding, the most reliable determinant of task-related fronto-parietal activity 
is its associated cognitive load (Miller and Cohen, 2001), so to get a relative picture of 
the effect of phase, the level of difficulty of the two task blocks was additionally 
manipulated independently between two levels – ‘easy’ and ‘hard’. The execution of 
the harder blocks was more complex and required maintenance of greater amount of 
information from the cue displayed at the start of the trial.  
The experimental design allowed for two distinct kinds of cognitive load 
manipulations during the first task block (Fig. 3.2). In the trials in which this was the 
hard block, greater information load maintained from the cue was relevant to it 
(hence, in the context of the first block, referred to as the current load); however, 
when the second block was the hard block, the extra information load maintained 
during the first block pertained not to the current block, but the future block (hence 
referred to as the prospective load).  
76 
 
Figure 3.2. Scheme of the different kinds of trials. Thin lines refer to periods with low 
working memory load, thick lines to high working memory load relevant to the imminent or 
the current task and dashed lines to high working memory load relevant to the future task. 
 
It is plausible that these two kinds of information are represented separately. 
Greater current load could be represented as part of the current task set (Sakai, 
2008), whereas prospective load might require a separate and distinct 
representation. This view gets support from some studies (Koechlin et al., 1999, 
2007; Pisapia et al., 2007; Reynolds et al., 2009) that show that blocks requiring 
maintenance of prospective information were associated with activity in the anterior 
prefrontal cortex, compared to blocks with current load only, which were associated 
with activity in the dorsolateral prefrontal cortex. Hence, it was predicted that the 
modulation of prospective load would be associated with a different pattern of 
activity than that elicited by the modulation of the current load. Further, since 
prospective load was limited to the first block while the second block had no such 
requirement, regions involved in maintaining such prospective information would 
77 
show greater activity during the first compared to the second block. On the contrary, 
regions that prefer the final phase of the task plan would elicit greater activity during 
the second block compared to the first.  
It was found that increased current cognitive load resulted in an increase in 
activity in the fronto-parietal cortices, specifically in the multiple demand regions 
(Duncan, 2010); however, increased prospective load did not result in significant 
activity in any brain region. Importantly, we found that the phase of the task indeed 
had an effect on the elicited fronto-parietal activity: the second task block, although 
identical to the first and with lesser load, was associated with much greater activity 
in widespread regions of the fronto-parietal cortex, including those that had shown 
an effect of current cognitive load.   
3.2 Methods 
 
Figure 3.3. Scheme of trials. Trials began with a cue display of numerical information (for 
the arithmetic block) and pseudoword(s) for the word block. In trials with an easy arithmetic 
block a single number was displayed, whereas two numbers and an arithmetic operator were 
78 
displayed on trials with a hard arithmetic block; one and two word(s) were displayed prior to 
easy and hard word blocks respectively. The information displayed on top signalled the 
nature of first task. The cue was followed by the first interval when subjects prepared for the 
first block. An outlining square the coloured margins of which reflected the upcoming block 
(blue: word; green: arithmetic) was displayed throughout this interval. The two task blocks 
consisted of a series of stimulus presentation to which subjects made the relevant response 
using the relevant cued information (see text). The two task blocks were always different 
within a trial but identical when averaged across the experimental session. 
 
As schematised in fig. 3.3, trials began with a cue display (on for 3500 ms) of 
numerical information for the arithmetic task and word(s) for the word task. The 
numerical information could be a single number in case of easy blocks e.g. “5”; and 
two numbers and an arithmetic operator, + or -, in the format “x±  y” (e.g. “2+  3”) 
for difficult blocks. One word was presented before easy blocks e.g. “kaf” and two 
before the difficult blocks e.g. “daj & gaz”. The information relevant to the first task 
block was displayed on top, e.g. numerical information above the word(s) meant that 
the arithmetic task would be the first block followed by the word block.  
The cue display was followed by a jittered gap, hence referred to as the first 
preparatory interval (1000 to 5000 ms) during which an outline square whose 
coloured margins signalled the nature of the forthcoming task block (green- 
arithmetic block, blue- word block) was displayed on the screen to reinforce the 
relevant task set. The jittered temporal duration kept the subjects in a state of 
readiness throughout the interval. At the end of this interval the stimuli relevant to 
the first task block appeared in this square. In the arithmetic blocks these were a 
series of one to three 4000ms displays of a number and an arithmetic operator, 
preceded and succeeded by blanks (e.g. “...8-...”); in word blocks a series of two- to 
nine-word stimuli (“baf”) were displayed (1000 ms per display). The inter stimulus 
interval (ISI) was 500 ms in all task blocks. 
79 
During easy arithmetic blocks, subjects mentally put the cued number (“5”) in 
the second blank (“...8-...”), solved the resulting expression (“8-5”) and then 
answered through the button box positioned under their right hand if the solution 
was less than 10 (index finger: 'yes', middle finger: 'no'). In difficult blocks, when the 
cued information maintained consisted of two numbers and an arithmetic operator 
(“2+  3”), the subjects placed the number and the operator that were displayed on 
the left side in the cue, in the first blank of the stimuli (“2+8-...”), and the number 
that was displayed on the right side in the cue in the second blank (“2+8-3”), and, as 
in easy blocks, mentally solved the resulting expression and responded if the solution 
was less than 10; for e.g. if the cue was “2+   3” and the displayed stimuli “...8-...”, 
the resulting expression becomes “2+8-3”.The arithmetic operations always involved 
numbers less than 20. The stimuli in these blocks were displayed for 4000 ms and a 
valid response could be made during this period. Number of sequential stimuli 
presented in an arithmetic block varied randomly from one to three with an ISI of 
500 ms, giving the duration of such blocks a jitter between 4 to 13 seconds. In the 
word task blocks, subjects responded if the displayed word was the same as the cued 
word in easy blocks or was one of the cued words in difficult blocks (index finger: 
yes; middle finger no).  Such blocks consisted of two to nine sequential stimuli, each 
displayed for 1000 ms with an ISI of 500 ms, giving their duration a jitter of 2.5 to 13 
seconds.  
The first task block was followed by the second preparatory interval (which 
was identical to the first interval) during which a blank margin was displayed, the 
colour of which signalled the nature of the second task block; within a trial the 
nature of the task in the two blocks was always different. This interval lasted for 
1000 to 5000 ms during which the subjects prepared for the second block. The 
structure of the ensuing second block was same as the first, except that it ended with 
a black margined square at which subjects could relax and wait for the next trial to 
begin  (inter trial interval: 1000 to 5000 ms).  
80 
In fifteen percent of the trials the first task block did not appear. In such cases 
the first preparatory interval ended with a change in colour of the margin which 
marked the beginning of the second preparatory interval. At this point the subjects 
switched to preparing for the second task block. In another fifteen percent of trials 
the second task block did not appear. In such cases the second preparatory interval 
ended with the margins of the blank square turning black signalling that the trial had 
ended. (See Appendix B.3 for exact instructions). 
The design of trials described above thus created two distinct phases in each 
trial wherein subjects prepared for and then executed two distinct tasks. Averaged 
across the experimental session, the first and the second task blocks were identical 
except for their position in the trial. The second task block can be considered to 
represent the final phase of the trial since it culminated in its completion. The overall 
experimental design of two kinds of task blocks (arithmetic and word) with two levels 
of difficulty (easy and hard) at two phases (initial and final) of a trial potentially 
created eight kinds of trials (depicted here by the identity of their first and second 
task blocks) –  
I. Easy Arithmetic → Easy Word 
II. Easy Word → Easy Arithmetic 
III. Hard Arithmetic → Easy Word 
IV. Hard Word → Easy Arithmetic 
V. Easy Arithmetic → Hard Word 
VI. Easy Word → Hard Arithmetic 
VII. Hard Arithmetic → Hard Word 
VIII. Hard Word→ Hard Arithmetic 
81 
The last two kinds of trials were not used since behavioural pilots showed too 
many errors in such trials due to excessive working memory load. Our experiment 
therefore had six kinds of trials (I to VI). 
The requirement that the information relevant to both task blocks be encoded 
at the cue display meant that the subjects had to maintain information relevant to 
the second task block while executing the first, resulting in greater cognitive load in 
the initial phase of the trial compared to the final phase. This cognitive load, hence 
called prospective load, depended upon the level of difficulty of the second block 
since the amount of information to be maintained differed for easy and difficult 
blocks. Accordingly the first blocks preceding easy second blocks had lower 
prospective load compared to the first blocks preceding difficult second blocks.  
The first task blocks were therefore be classified as easy (in trial types I and II), 
hard (in trial types III and IV) and easy with greater prospective load (in trial types V 
and VI). The 1st intervals preceding these task blocks were likewise classified as 
having low or high information load relevant to the first task block (i.e. low or high 
current load) and low or high information relevant to the second task block (low or 
high prospective load). The second task blocks and their preceding intervals could 
only be classified as easy (I to IV) and hard (V and VI), since they had no prospective 
load.  
All stimuli (visual angle ½ deg x 1 deg) were centred on the screen, visible from 
the participant’s position in the scanner via a mirror mounted within the head coil. 
The experiment was controlled by a program written in Visual Basic. Subjects did a 
20 min pre-scan practise session. The scan lasted for an hour and was divided into 
three separate 20-minutes sessions, each having 55 trials. 
82 
3.2.1 Participants 
15 participants (6 females, mean age = 23.5 ± 3.6 years) were recruited from 
the MRC-CBU volunteer panel. Participants were right handed and had normal or 
corrected vision. Informed consent was taken and the participants were reimbursed 
for their time. The study had the approval of the Hertfordshire Local Research Ethics 
Committee. 
3.2.2 Acquisition 
fMRI data were acquired using a Siemens 3T TimTrio scanner with a 12 
channel head coil. A sequential descending T2*-weighed echo planar imaging (EPI) 
acquisition sequence was used with the following parameters: Acquisition time 2000 
ms; echo time 30 ms; 32 oblique slices with slice thickness of 3 mm and a 0.75 mm 
interslice gap; in-plane resolution 3.0x3.0 mm; matrix 64x64; field of view 192mm 
and a flip angle of 78 deg. T1-weighted MPRAGE structural images were also 
acquired for all participants (slice thickness 1.0 mm, resolution 1.0x1.0x1.5 mm, field 
of view 256 mm, 160 slices).  
3.2.3 Analysis 
The fMRI data were analysed using SPM5 (Wellcome Department of Imaging 
Neuroscience, London, England; www.fil.ion.ucl.ac.uk). Prior to statistical analysis, all EPI 
volumes were slice-time corrected using the first slice as a reference, and then 
realigned into a standard orientation using the first volume as a reference. These 
realigned images were then normalised into the Montreal Neurological Institute 
(MNI) space, and spatially smoothed using an 8 mm full-width half-maximum 
(FWHM) Gaussian kernel. During the normalisation stage, voxels were resampled to 
a size of 3 × 3 × 3 mm. The time course of each voxel was high-pass filtered with a 
cutoff period of 90 s. 
83 
The various stages of the trial – cue, first preparatory interval, first task block, 
second preparatory interval and the second task block – were modelled using epoch 
regressors, of width equal to their duration. Further, the different categories of these 
stages based on their associated cognitive load (see above) were separately 
modelled. Event regressors of no duration modelled the completion of the two task 
blocks. All regressors were convolved with a basis function representing the 
canonical hemodynamic response. Movement parameters and block means were 
included as covariates of no interest. Parameter estimates for each regressor were 
calculated from the least squares fit of the model to the data, and estimates for 
individual participants were entered into a random effects group analyses. The 
results were rendered on to a whole brain template displayed at a false discovery 
rate corrected threshold of p<0.01 (unless specified otherwise). 
To capture fronto-parietal regions widely engaged in cognitive control, ten 
regions of interest (ROIs; Table 1, Figure 2) were created as 10 mm diameter spheres 
at coordinates that have been shown to be consistently active in varied tasks 
(Duncan, 2006; Dosenbach, 2006). The ROIs (in MNI space) were bilateral inferior 
frontal sulcus (IFS; central coordinate ±41 23 29), bilateral intra-parietal sulcus (IPS; 
±37 56 41), bilateral anterior insula extending into frontal operculum (AI/FO; ±35 18 
3), anterior cingulate (ACC; 0 31 24), and pre-supplementary motor area (pre-SMA; 0 
18 50), all taken from Duncan (2006); along with bilateral anterior prefrontal cortex 
(APFC; 27 50 23 and -28 51 15) taken from Dosenbach (2006). ROIs were constructed 
using the MarsBar toolbox for SPM (http://marsbar.sourceforge.net; Brett, 
Johnsrude & Owen, 2002). Estimated data were averaged across voxels within each 
ROI using the MarsBar toolbox and the mean values were exported for analysis using 
SPSS. 
84 
3.3 Results 
In the first part I discuss the results pertaining to the effects of cognitive load. 
In light of these results the second part examines the effect of the phase of the trial.  
3.3.1 Cognitive Load 
Overall performance was fairly accurate with error rates being less than 3% of 
the total responses for most subjects (mean 2.8 ± 0.8). Trials with any error were 
excluded from further analysis. For further analysis the behavioural parameters on 
the arithmetic and word blocks were separated. 
Behaviour 
Arithmetic blocks  
Arithmetic  1 2 3 
E1 
  
RT (ms) 1541 (69) 1149 (90) 1166 (93) 
Error rate (%) 2.2 (1) 2.6 (1.2) 1.1 (0.6) 
E2 
  
RT(ms) 1418 (78) 1127 (80) 1059 (61) 
Error rate (%) 2.4 (0.9) 2.2 (0.7) 1.3 (0.6) 
H1 
  
RT(ms) 2312 (101) 1873 (109) 1871 (161) 
Error rate (%) 7.6 (2.1) 4.8 (1.9) 4.8 (2.6) 
H2 
  
RT(ms) 2303 (87) 1912 (74) 1697 (153) 
Error rate (%) 8.7 (2.3) 5.1 (2.1) 4.7 (2.7) 
E1 + P RT(ms) 1613 (93) 1167 (69) 1104 (67) 
Error rate (%) 4.1 (1.6) 3.4 (1.5) 0 (0) 
Table 3.1. Reaction times and error rates at the various positions within the arithmetic 
blocks, values in parenthesis represent standard error. E1 - Easy block during phase 1; E2 - 
85 
Easy block during phase 2; H1 – Difficult block during phase 1; H2 – Difficult block during 
phase 2; E1 + P – Easy block with high prospective load  
 
Table 3.1 shows the mean RTs and error rates at different positions in the easy 
and hard arithmetic blocks from the two phases, and in the easy blocks with high 
prospective load in the first phase. Performance during the different arithmetic 
blocks was compared to examine the effect of task difficulty (performance on I & II 
vs. performance on III & VI) and position of the block in the trial (performance on I & 
III vs. performance in II & VI). Repeated measures ANOVA with level of current task 
difficulty (easy or hard), task phase (1st or 2nd), and position of the stimulus within 
the block (1-3) as factors showed a significant effect of the level of difficulty of the 
block on RT (F (1, 12) = 131.72, p<0.001), but did not show any effect of the phase of 
the block (F (1, 12) = 2.7, p=0.12). Position of the stimuli in the block also affected the 
RT (F (2, 24) =48.4, p<0.001). Pairwise comparisons showed that this was driven mainly 
by the first RT being higher than the rest (mean difference > 378 ms; Holm-
Bonferroni corrected p<0.001 in all cases) as would be expected from a blocked set 
of responses (Allport and Wylie, 2000). The second and third RTs did not differ (mean 
difference = 67.1 ms; 95% CI: -86.9 to 221.1). A similar ANOVA for errors, showed 
more errors in difficult blocks (F(1,13) = 10.62, p<0.01), but did not show any 
significant difference in error rates across the two phases (F(1,13) = 0.11, p=0.74); 
error rates varied marginally with the position in the block (F(2,26)
A separate ANOVA compared the two easy arithmetic blocks that differed in 
their prospective load trial type I vs. trial type V. Reaction times (1294 ± 75 vs. 1285 ± 
71; F
 = 3.05, p = 0.05), 
again reflecting higher error rates at the first position. 
(1, 11)
 
 = 0.65, p=0.8) and error rates (1.3 ± 0.3 vs. 1.2 ± 0.3) were not significantly 
different between the two. 
86 
Word blocks 
Word   1 2 3 4 5 6 7 8 
E1 
  
RT(ms) 683 
(13) 
538 
(16) 
500 
(12) 
499 
(14) 
491 
(16) 
510 
(20) 
521 
(22) 
508 
(22) 
Error rate 
(%) 
0.0  1.2 
(0.3) 
1.0 
(0.2) 
1.6 
(0.3) 
1.2 
(0.4) 
1.3 
(0.4) 
1.7 
(0.6) 
1.6 
(0.8) 
E2 
  
RT(ms) 666 
(12) 
544 
(15) 
506 
(16) 
496 
(20) 
481 
(16) 
502 
(20) 
519 
(17) 
509 
(28) 
Error rate 
(%) 
2.1 
(0.9) 
1.9 
(0.3) 
1.7 
(0.2) 
1.4 
(0.4) 
1.8 
(0.7) 
1.4 
(0.5) 
1.8 
(0.7) 
1.5 
(0.7) 
H1 
  
RT(ms) 735 
(14) 
600 
(13) 
568 
(12) 
571 
(14) 
582 
(13) 
586 
(18) 
564 
(18) 
574 
(14) 
Error rate 
(%) 
1.8 
(0.8) 
2.0 
(0.5) 
1.3 
(0.4) 
1.9 
(0.6) 
2.7 
(1.1) 
2.0 
(0.6) 
1.2 
(0.3) 
0.0    
(0) 
H2 
  
RT(ms) 723 
(13) 
604 
(18) 
580 
(14) 
574 
(14) 
586 
(16) 
569 
(18) 
546 
(15) 
560 
(29) 
Error rate 
(%) 
2.0 
(0.8) 
1.3 
(0.4) 
1.6 
(0.7) 
1.2 
(0.3) 
1.4 
(0.4) 
1.6 
(0.6) 
1.2 
(0.5) 
0.0    
(0) 
E1 + 
P 
RT(ms) 711 
(18) 
552 
(19) 
524 
(17) 
520 
(16) 
524 
(18) 
523 
(19) 
533 
(21) 
535 
(30) 
Error rate 
(%) 
1.6 
(0.7) 
1.6 
(0.4) 
1.7 
(0.6) 
1.2 
(0.7) 
1.8 
(0.5) 
1.6 
(0.8) 
1.8 
(0.7) 
1.9 
(0.6) 
 
Table 3.2. Reaction times and error rates at the various positions within the word blocks, 
values in parenthesis represent standard error. E1 - Easy block during phase 1; E2 - Easy 
block during phase 2; H1 – Difficult block during phase 1; H2 – Difficult block during phase 2; 
E1 + P – Easy block with high prospective load 
 
87 
Table 3.2 tabulates the reaction times and the error rates in easy and difficult 
word blocks, across the two phases. Performance in word blocks was compared for 
effect of difficulty (I & II vs. IV & V) and position in the trial (I & V vs. II & IV). 
Repeated measures ANOVA with the level of current task difficulty, phase and the 
position of the stimulus in the block as factors showed a significant effect of difficulty 
on RT (F (1, 12) = 65.8, p<0.001), but no effect of the phase of the trial (F (1, 12) = 0.6, 
p=0.7). RTs varied significantly with the position of the stimulus within the block    
(F(7, 84) =58.9, p<0.001). Similar to the case in arithmetic blocks, pairwise comparisons 
showed the first RT to be the highest (mean difference > 140 ms, Holm-Bonferroni 
corrected p<0.001 in all cases). None of the other comparisons reached significance 
(mean difference <50.1 ms, Holm-Bonferroni corrected p>0.05 in all cases). Error 
rates were not significantly affected by the difficulty (F (1, 10) = 0.01, p=0.9) or phase (F 
(1, 10) = 0.001, p=0.87) or by position within block (F (7,84)
Performance on the word block from trial type II was compared with the word 
block from trial type VI to look at the effect of greater prospective load. While mean 
RTs during the task blocks with greater prospective load was higher (558 ± 16 vs. 537 
± 14), the difference failed to reach significance (F 
=0.6, p =0.4). 
(1, 11) = 3.62, p=0.08). Error rates 
(2.1 ± 0.8 vs. 1.9 ± 0.9) did not differ significantly between the two blocks either, (F (1, 
12) 
These data show that behavioural indices were affected by the level of 
difficulty of the current task but were not different across the two phases of the trial. 
The manipulation of prospective load did not have a significant effect on the 
behaviour.  
= 0.46, p=0.45).  
Each trial consisted of discrete stages (fig. 3.2): cue → interval1 → taskblock1 
→ interval2 → taskblock2. The effect of cognitive load was studied at each stage. 
Imaging 
88 
Cue 
The cues showed information relevant to the arithmetic and the word blocks 
to be maintained in working memory until the relevant block was complete. The 
amount of information to be internalised at the cue display varied depending upon 
the difficulty of the task blocks of that trial. The cues in the trials with two ‘easy’ task 
blocks (trial types I and II), called ‘easy’ cues, showed a single number relevant to the 
easy arithmetic block and a single word relevant to the easy word block. In other 
trials, ‘hard’ cues consisted either of - two numbers with an arithmetical operation 
and one word, or of two words and one number.  
 
Figure 3.4. The brain render shows areas with greater activity during the hard compared to 
that during easy cues. Significant activity was present in IPS, left premotor and pre-SMA 
(marked by asterisk). The plots show estimates of activity in the MD regions during these 
cues. (E: easy; H: hard)  
The brain activity in response to easy versus hard cues was compared. The 
whole brain render in fig. 3.4, shows regions more active during hard cues. 
89 
Significantly greater activity was present in left premotor cortex, left IPS and left pre-
SMA.  
In the ROI analysis, MD regions were investigated for the effect of cue 
difficulty. A repeated measures ANOVA with difficulty (low and high) and laterality 
(for the bilateral ROIs only) as factors found significant effect of difficulty in IFS (F (1, 
12) = 8.5, p<0.01), IPS (F (1, 12) = 15.5, p<0.001) and pre-SMA (F (1, 12) = 7.5, p<0.01). The 
effect of laterality (right > left) reached significance only in IPS (F (1, 12) 
First interval 
= 5.4, p<0.01).  
During the gap between the cue and the first block of the task, two kinds of 
information were maintained – those pertaining to the first block i.e. the imminent 
task and those pertaining to the second block i.e. the prospective task. Working 
memory maintenance, in general, has been shown to be associated with fronto-
parietal activity (Courtney et al., 1998). It has also been shown that different kinds of 
working memory (spatial vs. verbal) relate to activity in different regions of the 
prefrontal cortex (Sakai and Passingham, 2003). In the current design the working 
memory contents differed by their relevance to the current and prospective tasks. It 
is plausible that the working memory contents relevant to the imminent task would 
be incorporated in the current task set (Sakai, 2008) especially since this information 
formed the template on which further task-related processing was to be done - the 
cued numbers were to be used for the processing of each of the presented numerical 
problems and the cued words were the template against which each presented word 
was to be judged. On the other hand the information pertaining to the prospective 
task had to be maintained separately against potential interference from the 
currently relevant information (Koechlin, 1999, 2007). During different kinds of trials 
the load of these two kinds of information was separately manipulated – in trials III 
and IV the current load was high, whereas in trials V and VI the prospective load was 
high. It was expected that the regions representing this information would be 
90 
associated with greater activity when the load of the respective information was 
high, and this could potentially reveal regions that represent current versus 
prospectively relevant task information.  
Estimates of activity in intervals with greater load were compared with those 
of intervals with lower load in two separate contrasts (III & IV vs. I & II; V & VI vs. I & 
II). No brain region showed differential activity in any of these comparisons. Fig. 3.5 
shows the estimates of activity during the three kinds of intervals in the MD regions. 
Only IPS showed differential activity, which curiously was greater for the interval 
with low load (i.e. from trials I and II), and was only marginally significant (F (2, 24) = 
3.5, p=0.048). Estimates of activity for the three kinds of interval were not 
significantly different in other ROIs (F (2, 24)
Although IPS has been mostly shown to have greater activity with increased 
memory load (Duncan and Owen, 2000), a decrease in activity in this region has been 
reported for situations that overwhelm working memory capacity (Linden et al., 
2003). Although the difficult trials in the current experiment did require maintenance 
of three to four items, an overload is unlikely given that the performance was 
uniformly good. 
 < 2.9, p>0.08, for other ROIs).  
 
 
91 
 
Figure 3.5. None of the brain regions showed differential activity for the different kinds of 
intervals. Amongst MD regions, IPS showed greater activity for the intervals with low working 
memory load. (E: Easy; H: High current load; P: High prospective load) 
 
First Task Block 
 The first task blocks were categorised as ‘easy’ (I and II), ‘hard’ (III and IV) and 
‘easy with high prospective load’ (V and VI). The effects of the two kinds of cognitive 
load (prospective and current) were assessed separately. 
Easy and hard first task blocks (I & II and III & IV) differed in the working 
memory content (one number vs. two numbers & an arithmetic operator, in 
arithmetic blocks; one word vs. two words, in word blocks) and complexity of 
computation (one step operation vs. two step operation). Both contributed to the 
enhanced cognitive load in the hard block. 
92 
 
Figure 3.6. Brain render showing areas with greater activity during the hard first task blocks 
compared to the corresponding easy blocks. These included lateral prefrontal cortex, insula, 
pre-SMA and ACC, lateral parietal cortex and the precuneus. Lateral prefrontal and parietal 
activities were centred on the MD regions (circles) 
 
The hard first task blocks were contrasted against the corresponding easy 
blocks. Greater activity during hard task blocks was found in bilateral fronto-parietal 
regions (fig. 3.6; peak coordinates in Appendix, table A3.1). Bilaterally in prefrontal 
cortex, these areas extended antero-posteriorly along the inferior frontal sulcus, 
from precentral regions to the anterior lateral prefrontal cortex, along with insula; 
medially, areas between anterior cingulate to the pre-supplementary region were 
involved. Bilaterally in the parietal cortex, areas with greater activity were centred 
laterally on the intraparietal sulcus (and extended into superior and inferior parietal 
lobules and into the middle occipital regions) and medially on the precuneus. Non 
cortical areas that showed greater activity for difficult blocks included lateral 
cerebellar cortex, anterior thalamus, and parts of basal ganglia.  
In the second contrast the two kinds of easy blocks differing in their 
prospective load were compared. No brain region showed greater activity for task 
blocks with greater prospective load even at a liberal threshold of uncorrected p 
<0.001.  
 
93 
 
Figure 3.7. Estimates of activity during the three kinds of first task blocks (E: Easy; H: Hard; 
P: Easy with high prospective load). The estimates of activity of hard task blocks were 
significantly greater than that of the easy blocks in all ROIs. This effect was bilaterally similar 
in all ROIs except IFS, where it was greater in the left; Estimates of activity of easy blocks 
with high prospective load were numerically greater than that of easy blocks, but reached 
statistical significance only in IFS-R and IPS-R, and close to significance in APFC-L (p=0.06) 
 
The activities in the MD regions during the three kinds of task blocks were 
then compared (fig. 3.7). A repeated measures ANOVA with load and ROIs as factors 
showed a main effect for both (F (9,108) = 8.8, p <0.001 and F (2,24) = 14.6, p <0.001, 
respectively), along with a significant interaction between them (F (18,216) = 7.1, p 
<0.001), suggesting that the effect of load varied across the different regions. A 
second model investigated the different ROIs separately to look at the effect of 
cognitive load (easy, hard, and easy with high prospective load) and laterality (for 
bilateral ROIs only). All MD regions showed a significant effect of cognitive load (F (2, 
24) > 5.9, p <0.03, for all ROIs), which was not affected by the laterality in any ROI 
94 
except IFS, where the effect was greater on the left (F (2, 24)
Second interval 
 = 6.3, p <0.01). Pairwise 
comparisons showed that the hard task blocks had significantly higher activity 
compared to the easy blocks in all ROIs (paired t(12) > 2.8, Holm-Bonferroni 
corrected p <0.05 in all ROIs). Activity for higher prospective load blocks was greater 
than in the blocks with low prospective load only in right IFS (paired t(12) = 2.6, 
Holm-Bonferroni corrected p <0.05) and right IPS (paired t(12)= 3.6, Holm-
Bonferroni corrected p <0.01) and nearly so in left APFC (paired t(12)= 2.3, Holm-
Bonferroni corrected p =0.07). 
A comparison of the estimates of activity preceding the easy and hard second 
task blocks was made. These intervals like the case in the first intervals (see above) 
differed in the amount of information maintained in working memory. Note that this 
interval was in the second phase of the trial and so did not have any prospective 
load.  
Similar to the case in the interval before the first task block (see above), no 
brain region showed differential activity across the intervals with different working 
memory load. The estimates of activity in the MD regions during these intervals were 
not different either (F (1, 12)
 
 < 1.4, p > 0.22, for all ROIs).  
 
 
 
 
 
95 
 
Second Task Block 
 
Figure 3.8. The brain render shows areas with greater activity during the hard second task 
blocks. These included regions along the IFS and IPS, pre-SMA, ACC and precuneus. The 
plots show the estimates of activity during the easy and hard second task blocks. Estimates 
for the latter were significantly higher in all ROIs. 
 
Similar to the first task block, the hard second task blocks were contrasted 
against the easy ones (fig. 3.8). The results were similar to the effect of current task 
difficulty in the first block and the same set of brain regions showed greater activity 
for increased cognitive load – bilateral prefrontal regions along the inferior frontal 
sulci, insula, pre-SMA and ACC; bilateral parietal regions around the IPS, including 
parts of IPL and angular gyrus, and precuneus; parts of lateral cerebellum, thalamus 
and basal ganglia. Further, as was the case in the first task block, all of the MD ROIs 
showed a significant effect of the current task difficulty (F (2, 24) > 9.5, p < 0.01, for all 
96 
ROIs). This effect of difficulty was significantly greater in the left hemisphere for the 
anterior insula (F (1, 12) =9.8, p<0.01) and anterior prefrontal ROIs (F (1, 12)
The results so far show that hard blocks with greater working memory 
requirements and more difficult task were associated with greater activity in 
widespread areas of the fronto-parietal cortices (fig. 3.6 & 3.8) including all MD 
regions (fig. 3.7). Task blocks that differed only in their prospective information load 
(i.e. the two kinds of easy first blocks that differed only in their associated 
prospective load) resulted in minimal differential activity that was limited to right IFS 
and right IPS. Interestingly, the pre-block intervals with similar working memory 
disparity as the task blocks did not show any difference in associated fronto-parietal 
activity, which suggests that the activity seen in right IFS and IPS could have been 
due to the requirement to maintain the information and carry out a different task 
concurrently. 
 =7.3, 
p=0.02).  
3.3.2 Phase  
Having established the effect of cognitive load of task blocks on their elicited 
activity, the effect of the position of the block in the plan of the trial was 
investigated. As described earlier, trials had two discrete phases – the first interval 
and the first task block constituted the initial phase and the second interval and the 
second task block were part of the final phase. Across the experiment these two 
phases were identical to each other, except that the working memory load of the 
first phase was higher, since subjects additionally maintained information pertaining 
to the second task block. 
It was seen earlier that the behaviour in both kinds of task blocks were not 
affected significantly by the phase of the trial in which those blocks occurred. 
Behaviour 
97 
 
 
Imaging 
 
Figure 3.9. Regions showing greater activity for the second task block. Plots show estimates 
of activity for the first and second task blocks; the estimates for the second task blocks were 
higher in all ROIs. 
 
The two task blocks were compared to reveal brain regions sensitive to the 
phase of the trial. No brain region showed greater activity for the first task block 
compared to the second. This is interesting since in all trials subjects maintained 
extra prospective information during the first phase that was to be used during the 
second phase. On the contrary, the reverse contrast showed widespread regions 
with greater activity for the second block (fig. 3.9; peak coordinates in Appendix, 
98 
table A3.2), which included the right prefrontal cortex (right inferior frontal gyrus 
especially the frontal operculum/anterior insula, lateral anterior prefrontal cortex, 
premotor regions, posterior regions of superior frontal gyrus extending medially into 
superior medial regions and parts of dorsal ACC), left prefrontal cortex (inferior 
frontal gyrus especially frontal operculum/anterior insula extending on to the lateral 
anterior prefrontal region, posterior portions of middle frontal gyrus and superior 
medial frontal gyrus), bilateral parietal cortex (inferior parietal lobule, supramarginal 
and angular gyri) and anterior parts of the superior temporal gyrus.  
 
 
 
 
Figure 3.10. The contrasts – Hard vs. Easy blocks (red-yellow), and Second vs. First blocks 
(blue-green) overlaid together for comparison; in lateral frontal and parietal regions the peak 
voxels of the former contrast were more dorsal compared to the later. The two overlapped in 
anterior prefrontal cortex, anterior insula and inferior parietal lobule.  
 
This pattern of activity was qualitatively different from the pattern of activity 
in response to increased task difficulty (fig. 3.10). In the lateral frontal and parietal 
cortices, the peak voxels of the former were ventral to the latter. In the prefrontal 
regions, for example, the peak voxels showing phase related activity were centred on 
the inferior frontal gyri and parts of frontal operculum extending on to anterior 
prefrontal regions, whereas the regions showing increased activity with task difficulty 
were centred on the MD regions - IFS and extending along its axis to the anterior 
prefrontal regions. Prefrontal activity for both contrasts converged on the anterior 
prefrontal cortex and anterior insula.  
99 
While the pattern of activity showing the effect of phase was different from 
the MD pattern, the individual MD ROIs were affected by the phase of the trial (plots 
in fig. 3.9), being significantly greater for the second task block. Repeated measures 
ANOVA with the phase of the block and laterality of the ROI (for bilateral ROIs) as 
factors, showed a main effect of the phase in all MD regions (F (1, 12) > 9.5, p<0.01 for 
all ROIs). While all ROIs showed the effect of phase, individual ROIs differed in the 
extent of this effect. A comparison of the effect of phase across different ROIs 
showed a significant interaction between the two (F (9,108) = 5.5, p<0.001). Further, 
the effect was bilaterally symmetrical except in the IFS, where the difference was 
greater in the right hemisphere (F (1, 12)
The activity during the first and the second intervals were compared. Recall 
that the temporal separation between them was similar to that between the two 
task blocks. Both the whole brain contrast and the ROI analysis of the MD regions (F 
 = 16.8, p<0.001). These results show that the 
effect of the phase of the trial on MD regions was similar to that of cognitive load 
and affected the activity in all MD regions. However, there was one interesting 
difference. The effect of cognitive load was greater in the left prefrontal ROIs, 
whereas the effect of the phase was greater in the right IFS.  
(1, 12)
 
 < 1.1, p>0.3) showed null results (fig. 3.11), which suggests that there was no 
gradual increase in neural activity with the temporal progression of the trial and the 
difference in activity found between the two task blocks was likely to be due to 
increase in task-related activity during the final phase of the trial.  
100 
 
Figure 3.11. Comparison of activity during the first and the second intervals. No brain region 
showed any difference in the level of activity during these intervals. 
 
End of task blocks 
Finally the activities related to the end of the two task blocks were compared. 
The end of the first task block was related to a task switch (Monsell, 2003), wherein a 
new task set relevant to the second task block was configured, whereas the end of 
the second task block initiated a period of rest. However, from the perspective of 
their role in the plan of the trial, the completion of the first block completed a 
subtask, whereas the completion of the second block completed the whole plan (ref 
chapter 2).  
101 
 
Figure 3.12. Brain regions with greater activity at the completion of the second task block. 
Plots show the estimates of activity for the ends of the first and second task blocks; the 
estimates for the second task block were greater in most MD ROIs, but the difference 
reached statistical significance only in APFC and ACC. 
 
In whole brain analysis, no region showed greater activity for the end of the 
first task block compared to the second, even at a liberal threshold of uncorrected p 
< 0.01. On the contrary, widespread regions showed greater activity for the 
completion of the second task block than the first (fig. 3.12).  These included bilateral 
precentral, inferior, and superior frontal gyrii, anterior and middle cingulate cortices, 
medial and ventromedial prefrontal cortex; bilateral precuneii, posterior cingulate 
and regions around the temporo-parietal junction in parietal cortex; parts of inferior 
temporal cortex and middle and superior temporal gyri; and most of the occipital 
cortex, except for posterior most parts of calcarine and lingual gyri. Apart from these 
cortical regions – bilateral hippocampus, parahippocampus, amygdala, basal ganglia 
102 
and cerebellar regions also showed greater activity bilaterally for the end of the 
second task block compared to the first.  
Some of the regions above have been shown to be more active during periods 
of rest (medial frontal and parietal cortex, lateral temporal regions, hippocampii and 
parahippocampii). However, this as an immediate explanation seems less plausible 
for other regions (fig. 3.12) like inferior frontal gyrii, anterior and medial cingulate, 
precentral gyrus, early visual and subcortical regions.  
Estimates of activity related to the end of the second task block were higher in 
most ROIs (fig. 3.12), but this difference reached significance only in APFC (F(1, 12) = 
16.0, p<0.001) and ACC (F(1, 12) = 17.2, p<0.001). In APFC this difference was 
significantly greater on the right (F(1, 12)
3.4 Discussion 
 = 5.4, p=0.03). Interestingly this pattern of 
activity in the MD ROIs for the completion of the trial matches the results of earlier 
experiments (chapter 2) that showed APFC (right > left) and ACC to be most 
consistent in showing greater activity for task or plan completion. Indeed compared 
to their level of activity at the end of the second task block, earlier task stages had 
resulted in far less activity in these regions (compare figs. 3.9, 3.11, 3.12). 
 The above results show that the task elements in the final phases of a trial are 
associated with greater activity in widespread fronto-parietal regions compared to 
those in the initial phases; this effect was also present in those specific regions of 
these cortices where activity has been shown to reflect task complexity and cognitive 
control requirements (Duncan, 2010). Indeed the effect of phase in the MD regions 
was as widespread as the effect of task difficulty/cognitive load.  
3.4.1 Cognitive Load 
Hard task blocks showed greater activity in the extended regions of fronto-
parietal cortices. In the lateral prefrontal cortex, these regions extended antero-
103 
posteriorly from the premotor to the anterior prefrontal cortex (fig. 3.6 & 3.8). It is 
interesting to compare this result with a popular view that progressively anterior 
regions along the antero-posterior axis of the lateral prefrontal cortex are sensitive 
to hierarchically higher aspects of cognitive control (Badre, 2009). Badre and 
D’Esposito (2007) showed more posterior frontal activity for simple one-level 
decisions (e.g. red > left key) than for two-level decisions in which one stimulus 
feature indicated how the other should be processed, and more anterior frontal 
activity for higher level decisions, such that anterior prefrontal activity occurred 
during four-level decisions. Similarly, Koechlin et al., (2003) showed an increasingly 
anterior locus of prefrontal activity when increasingly greater extents of temporal 
context determined how a stimulus should be interpreted (for e.g. context of a trial 
vs. context of the block of trials). In our experiments the hard blocks producing 
activity in the entire length of the same antero-posterior axis of lateral prefrontal 
cortex, differed from the easy blocks in working memory load (2 items vs. 4 items) 
and in the number of computational steps (one vs. two). Since the pre-block intervals 
having similar working memory disparity did not differ in their elicited activity, it is 
likely that the difference between hard and easy blocks was related to the 
complexity of computation performed. Arguably, these blocks did not differ in the 
levels of decision hierarchy needed to reach the solution (c.f. Badre and D’Esposito, 
2007; Christoff et al., 2007) or in the levels of temporal contexts across which the 
information was to be integrated (c.f. Koechlin et al.,, 2003). This raises the 
possibility that the antero-posterior shift of prefrontal activity seen in these studies 
were just related to the greater number of computational steps and were not 
specific to the hierarchy of the control structures per se. 
Intriguingly, we did not find any effect of working memory load alone. Fronto-
parietal activity during the intervals preceding the task blocks did not vary with the 
amount of information maintained. The current results confirm a claim made in the 
last chapter, that the anterior prefrontal activity is not related directly to the 
104 
requirements of maintaining information while carrying out an unrelated task as has 
been hypothesised by earlier studies (Koechlin et al., 2007; Pisapia et al., 2007; 
Reynolds et al., 2009). Although the first task blocks had prospective information 
load, while second blocks had none, no region showed greater activity for the first 
task block compared to the second. Amongst the first task blocks those with greater 
prospective load only showed a minimal increase in right IFS and IPS. Further, 
anterior prefrontal activity was found in hard blocks compared to easy blocks, 
although they did not differ in their prospective information load. 
3.4.2 Phase 
The most significant result of the current study was that the phase of the plan 
in which the task blocks occurred determined their elicited neural activity in the 
fronto-parietal cortex. Greater activity for the second task block was most prominent 
in frontal operculum/anterior insula, anterior prefrontal cortex, medial prefrontal 
(medial superior frontal and dorsal ACC), and IPS extending into IPL. Interestingly, 
some of these regions viz. APFC, AI/FO, ACC were found by Dosenbach et al. (2006) 
as core regions showing sustained activity throughout a task block across a number 
of different experiments. Evidence of distinction between the two hemispheres was 
also discernible. Left prefrontal ROIs showed greater effect of task difficulty (IFS, 
AI/FO, APFC), whereas the effect of phase was greater in right IFS.  
What is different between the first and the second task blocks? The first 
possibility is expectation of completion/reward. It is interesting in this regard to 
compare the above results with the phenomenon of reward expectancy. Tasks 
requiring subjects to perform a number of trials to achieve a discrete reward show 
that the reaction time and error rates decrease for trials closer to the reward, with 
the last trial yielding the fastest and most accurate response (Shidara and Richmond, 
1998; Sohn and Lee, 2007). Likewise, the associated neural activity in prefrontal, 
105 
parietal, temporal and striatal neurons are seen to increase as trials get closer to the 
reward (Bowman et al., 1996; Shidara and Richmond, 1998).  
Reward expectancy, however, seems unlikely to be the explanation, primarily 
because there was no reward or achievement to be expected. Further, beyond a 
general instruction for being accurate and fast, the task instructions did not mention 
any end to be hoped for; no explicit scoring was done or mentioned. The profile of 
behavioural performance does not fit this picture either. Reaction time in both 
phases did not decrease after the second stimulus, whereas studies on reward 
expectancy show a progressive decline in reaction time till the end. Neither were the 
RTs different across the two phases. Furthermore, there was no difference in activity 
in between the two intervals preceding the two blocks, even though the temporal 
difference between them was similar to the temporal difference in between the two 
blocks. This also argues against a general increase in activity through the length of 
the trial and is evidence that the effect of phase was specific to the task blocks. 
A second possibility is that the decrease in the cognitive load that happens in 
between the first and the second block is the actual cause of increase in activity. 
Counterintuitive as this is in the light of the prevailing view about the relation 
between fronto-parietal activity and the cognitive load, it certainly is a possibility. 
The experimental design provides with a means to test this. If some fronto-parietal 
regions increase their activity only consequent to a decrease in task load that 
happens between the first and the second task block, it would be expected that an 
increase in the same kind of task load during the first task block would decrease the 
resulting fronto-parietal activity. However, contrary to this, increasing the 
prospective load during the first task block lead to an increase in activity (fig 3.7), 
thus, showing that decrease in task load in and of itself cannot be an explanation for 
the increase seen in fronto-parietal activity across the two phases.  
106 
A third possibility is that a decrease in the representational load pertaining to 
the structure of the episode, leads to an increase in activity elicited by the events of 
the later phase. During the first phase, in addition to the representations related to 
the current phase, subjects have to additionally maintain representations pertaining 
to the later phases. While in the later phases, those pertaining to the earlier phases 
can be dismantled, freeing up the representational space, which can then be utilised 
by the ongoing task events. This possibility suggests a kind of neurocognitive storage 
resource, loading which does not lead to an increase in fronto-parietal activity.  
This chapter thus adds onto the findings of the previous chapter that the 
context of the plan of the behaviour affects the fronto-parietal activity related to any 
task event or task episode within it. While the previous chapter showed the effect of 
the hierarchical position of the task episode completed on the elicited fronto-parietal 
activity, the current results show the effect of the phase in which the task events lie. 
However, the current experiment leaves it unclear if the effect seen is specific to the 
final phase, or the increase in activity would be a feature of any subsequent phase 
when compared to the previous phase. This issue is investigated in the next chapter. 
 
 
 
 
 
 
 
107 
CHAPTER 4 
 Higher Level Representations & Organisation of Task 
Episodes 
 
4.1 Introduction 
Purposive behaviour is organised into a nested hierarchy of subtasks and tasks 
(Miller et al., 1960). How this organisation is achieved is an open question. It was 
speculated in earlier chapters that the behavioural hierarchy stems from the 
hierarchical organisation of relevant representations in the mind such that abstract 
representations such as task or goal descriptions subsume and organise more 
concrete representations such as action plans and rules that in turn organise even 
more concrete representations like those of discrete actions (for a similar account, 
see Cooper and Shallice, 2000). Other accounts (Botivinick and Plaut, 2004), on the 
contrary, have tried to show that such hierarchical organisation of relevant 
representations are epiphenomenal to the output of action control systems which 
can be instantiated as a simple recurrent network that does not require 
hierarchically organised set of representations (see also Botivinick, 2009).  
 One of the ways to test the two accounts is to see if task episodes with 
identical lower level elements are organised differently when the organisation of 
their higher level representations is changed. Consider a hypothetical scheme of 
making tea: 
Boil water → Add tea → Brew → Add sugar → Add milk 
 Will the nature of the higher level representation under which this episode is 
organised affect the nature of its organisation? 
 
108 
 
 
Figure 4.1. Proposed organisation of a hypothetical sequence of actions for preparing tea 
under two different higher representations, if the nature of latter was causal to the details of 
organisation (see text). 
Consider the representation ‘tea with milk’ that emphasises the distinction of 
the last step (‘add milk’), versus the representation ‘milk tea’, which de-emphasises 
this distinction. Will the lower level sequence of task elements be organised 
differently in the two cases?  Fig. 4.1 depicts the two potentially different ways of 
organisation if the higher level label affected the organisation. On the other hand, if 
the properties of higher level label had no causal role, the organisation of the 
episodes would be identical in the two cases. 
109 
Results described in the chapter 2, furnished evidence in support of the 
former view by showing that the way the task was described, did affect the 
organisation of the task episode. Although subjects carried out four identical target 
detections, the first three had been described to them as being part of ‘one task’, 
and the fourth as part of the ‘other task’. The fronto-parietal activity at the third 
target detection was higher than the first two, and was similar to the case when the 
third target detection did actually complete the task i.e. when there were only three 
targets to detect.  
That the nature of task instruction determines its execution is a truism. But 
the way the same task instructions are organised by the subjects affects their 
performance. This is especially evident in complex tasks. Duncan et al. (unpublished 
data) instructed subjects to do a task with multiple contingent rules (chapter 1, Fig. 
1.2).  Subjects are shown a stream of pairs of stimuli (letters or numbers) presented 
simultaneously on the two sides of the centre of a computer screen. The last three 
pairs of stimuli of the stream were preceded by arrow heads pointing to left or right. 
The subjects are to follow a series of instructions in a particular sequence – attend to 
left (or right), read aloud the letters on the attended side, add pairs of numbers but 
ignore the asterisks, after the appearance of ‘>’ read aloud the letters on the right, if 
‘<’ appears, read aloud the letters on the left side. It had been shown by the authors 
earlier (Duncan et al., 2008) that the last part of the task (taking note of the arrow 
head and attending to the requisite side) tends to be frequently neglected. In the 
current study, the authors found that the way subjects were asked to recall the 
instructions, affected their performance. Subjects that were free to recall in any way 
they chose showed poorer performance than subjects that were explicitly instructed 
to organise the recall into a series of bullet points. Note that the instructions were 
the same in both groups, the difference lay in the way they were made to organise 
the instructions. Plausibly, the effect of different kinds of recall was the different 
110 
pattern of organisation of task instructions (higher level representations), which 
affected the assembly of task control structures.  
In the current task, the subjects were biased to organise the higher level 
representations into specific ways, then the organisation of the resulting task 
episode was judged from their behavioural profile and from the elicited neural 
effects of task organisation described in earlier chapters. Each trial in the current 
experiment, consisted of a sequence of eight steps (called subtrials), the rule 
relevant to them was cued to the subjects as an eight letter string composed of 
letters ‘A’ or ‘B’ (e.g. AABBBBAA). On the subsequent eight subtrials, responses were 
made according to the rule represented by the identity of the letter at respective 
position in the memorised string. ‘A’: categorise the letter as vowel/consonant; ‘B’: 
categorise the number as even/odd. For example, if the string was AABBBBAA, rule A 
was relevant on the first two subtrials, followed by rule B on the next four, followed 
by rule A again in the last two subtrials. 
 To bias the organisation of the representation of this sequence, subjects 
underwent a 20 minute pre-scan practise session during which they performed trials 
with four subtrials guided by four letter strings e.g. AABB. It was hence expected that 
in the main scanning session the subjects would be biased towards representing the 
eight letter string as two chunks of four letters (e.g. AABBBBAA as AABB and BBAA). 
Two kinds of strings were used in the scanning sessions. In one (trial type 1), 
the first and the second group of four letters were different e.g. AABBBBAA; they 
were the same in trial type 2, e.g. AABBAABB. It was hoped that these two kinds of 
strings would be represented differently. The sequence in trial type 1 required that 
all eight elements be kept in working memory from which individual letters at the 
relevant position were to be accessed. However, in the case of trial type 2, there was 
no need to remember all eight elements, and it was plausibly more parsimonious to 
111 
remember just four elements and then repeat them. If the properties of higher level 
representations causally determine how the behavioural episode is organised, it can 
be expected that these two trial types will be differently organised. 
 In trial type 1, since the entire sequence had to be represented as one unit 
from which individual elements were to be accessed, it was expected that the entire 
trial would be organised strongly as one episode albeit with some evidence of further 
organisation in two subtask episodes due to the bias created by the pre-scan session 
(Fig. 4.2a). In trial type 2, on the other hand, there was no need for the entire 
sequence to be represented, rather just representing the repeating four elements 
was enough, so this trial type was expected to be strongly chunked as two sub-
episodes (Fig. 4.2b). These expectations made specific predictions about the profile 
of behaviour and that of elicited fronto-parietal activity in the two trial types. 
112 
 
Figure 4.2. Possible schema of trial type 1 (a) and trial type 2 (b). In trial type 1, the entire 
trial was expected to be organised as a unified task episode with a strong representation of 
the highest sequence level representation, and with relatively weak evidence of the 
distinction between the two chunks; trial type 2, on the other hand, was expected to be 
organised more as two task episodes and the sequence representation was expected to be 
relatively weaker, while the chunk representations were expected to be relatively stronger 
than in trial type 1. 
The behaviour at the fifth subtrial, which marks transition across the two 
chunks, has different predictions depending upon the kind of organisation of the two 
trials. Transition across hierarchically organised episodes of behaviour has been 
shown to require additional time, compared to transits within the episode (Lien and 
Ruthruff, 2004; Schneider and Logan, 2006 and 2007). Therefore, the greater the 
hierarchical difference between the two episodes, the greater would be the cost of 
transiting across them. So if the subtrials 1 to 4 and 5 to 8 in trial type 2 are 
113 
organised more as separate task episodes, the reaction time (RT) on the fifth subtrial 
would be greater than in the case of trial type 1 where the two chunks are 
represented more as part of the same task episode.  
On the other hand, if the organisation of the two trial types was the same, the 
behaviour at the fifth step would depend upon how the cued sequence has been 
represented. If sequence has, nonetheless, been chunked then the fifth RT in trial 
type 1 would be expected to be higher than the corresponding RT in trial type 2 since 
it involves recalling from a different chunk and creating a new behavioural episode, 
whereas the relevant chunk remains the same in trial type 2, and the previous 
episode needs to be repeated. This assumption is supported by the findings of 
Schneider and Logan (2006 and 2007). They had subjects memorise two distinct 
sequences (called α and β) of four task units each. On every trial they cued the 
sequence and the position (e.g. α3) from which subjects were to access and execute 
the relevant task. They found a distinct switch cost whenever a different sequence 
was cued compared to the previous trial. In a different study, they had subjects 
execute these sequences in the order α α β β. Thus, the sequences were alternately 
repeated and switched. They found a distinct switch cost whenever the sequences 
switched. Finally, if the sequence was not chunked in either, the fifth RT would not 
be different from other RTs in both trial types.  
In earlier chapters it was shown that completion of defined task episodes 
correspond to an increase in activity across many fronto-parietal regions, and task 
phases closer to such completions have greater fronto-parietal activity compared to 
earlier phases. It was hence predicted that if trials are organised as two sub-
episodes, fronto-parietal activity at the fourth subtrial marking the completion of the 
first episode would be higher compared to other subtrials. Further, if this 
organisation as two sub-episodes is greater for trial type 2, then the fourth subtrial 
activity in it would be higher than in case of trial type 1.  
114 
What would be the pattern of phase effect across the two trial types? From 
the results of the previous chapter, it was not clear if the increase in activity is 
limited to the final phase or includes any subsequent phase compared to the earlier 
one. If the increase is pervasive and stepwise across the phases of the plan, trial type 
1, being a more unified episode, can be expected to show a pervasive stepwise 
increase in activity along the eight subtrials, while trial type 2 may show separate 
increase for the two sub-episodes i.e. parallel increase along subtrials 1 to 4 and 5 to 
8 (Fig. 4.3). If the increase was limited to the final phase, trial type 1 would show 
greater activity only at the eighth subtrial, while both, the fourth and the eighth 
subtrials can be expected to have relatively higher activities. 
 
 
Figure 4.3. Predicted pattern of activity over the eight subtrials of the two trial types if the 
phase effect was pervasive and not just limited to the final phase. Trial type 1, being a 
unified task episode, was predicted to show a gradual increase in activity across the length 
of the trial in fronto-parietal regions sensitive to the phase. Trial type 2, on the other hand, 
was expected to show separate increase in activity for the first and the second group of the 
four subtrials. 
While the two trial types (organised trials) described above were organised 
through specific learned sequences, a third kind of trials (called unorganised trials) 
also consisting of eight subtrials did not have an organising sequence cued at the 
beginning of the trial. Instead the relevant rule was cued on each subtrial by the 
115 
colour of the margin outlining the stimuli. These trials were identical to the organised 
trials in all other aspects. Such trials provided an opportunity to test a very 
speculative thesis. Since subjects had been biased to organise the subtrials in groups 
of four through the organising sequences, they might organise a series of eight steps 
in groups of four even in absence of explicit organising sequence. In this case it 
would be a case of abstract conceptualisation of the episode affecting its 
organisation. If the activity in some fronto-parietal regions was sensitive to such 
abstract conceptualisation of the structure of task episode, it can be expected that 
those regions would show an increase in activity during the fourth subtrial of this 
trial type. Note that the lack of organisation in case of the unorganised trials is only 
relative to the other two more (explicitly) organised trial types. Otherwise even the 
unorganised trials are organised. Subjects know that these have an explicit start, 
consist of eight steps and have an explicit end etc. 
Experiments in the previous two chapters showed patterns of fronto-parietal 
activity contrary to what would be expected from the requirements of cognitive load. 
In chapter 2, the last target detection event did not require the recall of a new target 
letter and beginning of a new visual search, unlike the previous target detections. 
Nevertheless, fronto-parietal activity was maximum for the last target detection. In 
chapter 3, as compared to the initial phases, the final phase had lower working 
memory load and did not require the maintenance of branched information 
(Koechlin, 2007) nevertheless activity was higher for the final phase in all areas of the 
fronto-parietal cortices. The current experimental design allowed for greater 
exploration of these issues. 
Hierarchical task episodes require hierarchical cognitive control (Schneider 
and Logan, 2006; Lien and Ruthruff, 2004) – higher level control accesses and 
maintains higher level representations while the lower level control accesses and 
executes the subtasks. For example, in an experiment by Schneider and Logan (2006) 
116 
multi-element chunks composed of two tasks (e.g. AABB, AABBAA) were repeatedly 
executed (e.g. AABB-AABB-AABB). The RT on the first element of the chunk was 
always found to be the highest and the degree of increase in RT at the first element 
was determined by the complexity of the chunk being executed (e.g. it was greater 
for AABABB compared to AABB), suggesting that the beginning of the chunk required 
an additional chunk level control, which was greater for more complicated chunks. 
That this control was hierarchically higher than that required for the execution of the 
individual elements was evident from the absence of any lower level task switch cost 
at the first element. For example, when AABB was repeatedly executed (AABB-AABB-
AABB) the first element involved a task switch (B to A), but the RT on this was not 
greater than in the case of the first element when ABBA was repeatedly executed 
(ABBA-ABBA-ABBA) even though the latter did not involve a task switch. Indeed the 
RT in the latter case was higher than the former, possibly because the latter 
sequence was more complicated as it involved a greater number of task switches 
within the chunk (two in ABBA compared to one in AABB). The higher chunk level 
control seems to subsume lower level control requirements.  
From this perspective, the organised trials required multiple levels of control. 
Lowest level control to perform the subtrials, chunk level control between the fourth 
and fifth subtrials to instantiate control structure required for the new chunk, and at 
the highest level, a control structure associated with the higher level sequence 
instantiated at the first subtrial. These control levels have a nested hierarchy since 
the highest sequence level representations (AABBBBAA) are required to access the 
chunk level representations (AABB or BBAA) from which in turn are accessed the 
elemental representations (A or B). Regions instantiating these higher order control 
can be predicted to have phasic activity at the first and fourth (or possibly fifth) 
subtrials.  
117 
Some recent theories on the functional organisation of the prefrontal cortex 
(PFC) suggest that control at progressively higher levels result in more anterior 
activity within the PFC (Koechlin and Jubault, 2006; Badre and D’Esposito, 2009). 
These would predict that the first subtrial would be associated with the most 
anterior locus of prefrontal activity since the highest sequence level control is 
instantiated at this point, followed by the fourth (or fifth) subtrial when chunk level 
control is instantiated, followed by other subtrials.   
The scheme of organised trials also made explicit predictions from the 
perspective of working memory. Working memory load decreased along the length 
of the trial as the representation relevant to the particular subtrial could be 
discarded after its completion. Brain regions sensitive to the working memory load 
can hence be expected to show a stepwise decrease in activity along the length of 
trial type 1. In trial type 2, such decrease may happen only between subtrials 5 to 8, 
since only during these can the relevant representations be discarded. 
In summary, divergent predictions exist from the perspectives of organisation, 
cognitive control and working memory load.  
The results showed that the two organised trial types were indeed differently 
organised. Compared to other subtrials, the fifth subtrial had longer RT in trial type 2 
than in trial type 1 suggesting that chunking was more intense in the former. 
Neuroimaging results agreed with this. Relative to other subtrials, the fourth subtrial 
completing the first chunk elicited greater activity in trial type 2. Most intriguingly, in 
trial type 3 a number of fronto-parietal regions showed increased activity for the 
fourth subtrial. These results provide compelling evidence that higher level 
representations are indeed causal in determining the organisation of the task 
episode. The eighth subtrial in all three trial types, predictably, elicited the greatest 
activity across most fronto-parietal regions. 
118 
 The results also confirmed that the phase effect is not specific to the final 
phase, rather all subsequent phases elicit greater activity compared to the earlier 
phases. The exact pattern of this was shown to depend on the organisation of the 
episode. Trial type 1, organised as a single episode, with poor evidence of chunking, 
showed most widespread evidence of a monotonic increase in activity across the 
length of the trial. This was weaker in trial type 2 and weakest in trial type 3. Indeed, 
the pattern of increase in activity in trial types 2 and 3 were qualitatively different 
from trial type 1. Instead of a monotonic increase across the length of the trial, these 
trial types showed parallel increase along subtrials 1 to 4 and 5 to 8, suggesting that 
the dominant pattern of organisation of these trial types was bi-episodic. 
Lastly, no brain region showed greatest activity during the first followed by 
the fourth (or fifth) subtrials of organised trials, as would have been predicted for 
regions involved in instantiating hierarchical control. Neither did any region show a 
decrease in activity across the length of the trial as would have been predicted for 
regions sensitive to working memory load. 
4.2 Methods 
4.2.1 Stimulus 
The stimulus consisted of a letter and a number placed at the different corners of a 
square (Fig. 4.4). Responses were made according to one of the two tasks: task A - 
categorise the letter as a vowel (index finger) or a consonant (middle finger); task B - 
categorise the number as even (ring finger) or odd (little finger) using the next two 
fingers. The exact position of the letter and the number within the square changed 
randomly. The letter and the number were in black Arial font and subtended a visual 
angle of half a degree. The outlining square was black in colour unless specified 
otherwise (see below) and had a visual angle of ~ 1.5 deg. 
 
119 
 
Figure 4.4. Scheme of trials during the first part of pre-scan practise session (see 
text for details). 
4.2.2 Pre-scan 
Prior to imaging, subjects did a 20 minute practice session (Fig. 4.4). Trials began with 
the presentation of a four-letter string (for 2 s) composed of all combination of 
letters A and B (e.g. ABAB, AABA, BBAA, BABB, AAAA etc.). This letter sequence was 
to be kept in working memory for the duration of the ensuing four subtrials. On each 
of the subtrials the stimulus described above appeared for 1s. The identity of the 
letter in the remembered sequence corresponding to the number of current subtrial 
determined the task-relevant to that subtrial; e.g. if the sequence was AABB, task A 
was to be done on the first two subtrials followed by task B on the next two. Inter 
subtrial interval was 1000 ms giving subjects 2 s to respond. After the fourth subtrial, 
the next trial began (inter-trial interval (ITI), 1s). Subjects did a total of 80 such trials. 
This aim of this session, apart from familiarising the subjects with some aspects of 
the task, was to bias the subjects towards representing four letter strings as a chunk.  
  
120 
 
Figure 4.5. Organised (top) trials began with a cue display of the eight letter sequence 
organising the rest of the trial. On the ensuing eight subtrials, the corresponding letter in the 
sequence was to be recalled and the subtrial executed accordingly (A: letter task; B: number 
task). In the unorganised trial (below) no letter sequence was cued, and the relevant task on 
any subtrial was determined by the colour of the square (blue: letter task; green: number 
task). 
4.2.3 Imaging session 
 Trials began with the presentation of the relevant eight letter sequence 
in the organised trials (Fig. 4.5) or with “…” in unorganised trials (presented for 3500 
ms). The sequence was made up of random combination of letters ‘A’ and ‘B’ (e.g. 
‘AABABBAB’). This was followed by a series of eight subtrials separated by an average 
interval of 3.5 s (jitter 0.5 to 6 s). On every subtrial the stimulus described above 
appeared for 1s. In organised trials (trial types 1 & 2) the colour of the margins of the 
square was black throughout the eight subtrials, while during the unorganised trials 
(trial type 3) the colour of the margins signalled the task-relevant on the particular 
subtrial, blue for task A and green for task B. Since in organised trials subjects had an 
a priori knowledge of the rule relevant on any subtrial, they could start preparing for 
the upcoming subtrial immediately after responding to the previous one. In 
unorganised trials, to allow for the same preparation time, the colour of the outlining 
121 
margin changed immediately after the offset of the stimuli of the previous subtrial 
(and after the offset of the cue for the first subtrial). Valid responses could be made 
up till 2 s following the stimulus appearance. At the end of the eighth subtrial a blank 
square with black margins remained on the screen till the beginning of the next trial. 
Average ITI was 4 s (jittered 2 to 7 s). (See Appendix B.4 for exact instructions). 
Trials were of three kinds – organised trials with different four-letter 
sequences (e.g. AABB and BBAA, presented as ‘AABBBBAA’), hence called type 1 
trials; organised trials with the same four-letter sequence repeated (e.g.  
‘ABABABAB’), hence called type 2 trials; and unorganised trials or type 3 trials. These 
three trial types were delivered in equal proportion. The total duration of the 
imaging session was 68 minutes, which was divided into four separate scanning runs. 
Subjects did 30 trials in each of the four sessions.  
4.2.4 Participants 
15 participants (7 female, mean age = 22.5 ± 3.6 years) were recruited from the 
MRC-CBU volunteer panel. Participants were right handed and had normal or 
corrected vision. Informed consent was taken and the participants were reimbursed 
for their time. The study had the approval of the Hertfordshire Local Research Ethics 
Committee. 
4.2.5 Acquisition 
Same as in previous experiments. 
4.2.6 Analysis 
The fMRI data were analysed using SPM5. Prior to statistical analysis, all EPI 
volumes were slice-time corrected using the first slice as a reference, and then 
realigned into a standard orientation using the first volume as a reference. These 
122 
realigned images were then normalised into the Montreal Neurological Institute 
(MNI) space and spatially smoothed using an 8 mm full-width half-maximum (FWHM) 
Gaussian kernel. During the normalisation stage, voxels were resampled to a size of 3 
× 3 × 3 mm. The time course of each voxel was high-pass filtered with a cutoff period 
of 90 s. 
For each trial, the cues and the subtrials [eight x three trial types] were 
modelled using epoch regressors, of width equal to the duration of the respective 
events. Each subtrial was modelled from the offset of the previous stimulus (or cue in 
case of subtrial 1) to the offset of the current stimulus. The stimulus events onsets 
within each subtrial were modelled with one event regressor of no duration to 
regress out the common phasic activity related to the presentation of the stimulus. 
All of these regressors were convolved with a basis function representing the 
canonical hemodynamic response. Movement parameters and block means were 
included as covariates of no interest. Parameter estimates for each regressor were 
calculated from the least squares fit of the model to the data, and estimates for 
individual participants were entered into a random effects group analysis.  
Whole-brain comparisons were performed using paired t-tests on the relevant 
contrast values from each participant’s first-level analysis. Unless otherwise 
specified, all results are reported at a threshold of p < 0.01, corrected for multiple 
comparisons using the false discovery rate (FDR) method. Coordinates for peak 
activation are reported using an MNI template. 
Apart from the MD ROIs used in the previous experiments, three additional ROIs 
were created from the three major clusters of prefrontal voxels that showed a 
significant effect of phase in the previous experiment (chapter 3, Fig. 3.9). These 
clusters were from the ventrolateral (centered at MNI coordinates 47, 32, 0 and -47, 
123 
20, -4) and medial superior regions (0, 37, 42) of the PFC. These ROIs will 
subsequently be referred to as the phase-sensitive ROIs. 
4.3 Results 
The hierarchical organisation of a task episode can be inferred from the time 
taken to access the representation at different points, since it is expected that more 
time will be needed to access representations from a different level of hierarchy than 
from the same level (Schneider and Logan, 2006 and 2007; Lien and Ruthruff, 2004). 
In the organised trials, if the cued sequence was chunked into groups of four, the RT 
on the fifth subtrial can be predicted to be higher than others (except the first RT). 
Further, the magnitude of this increase would depend upon the magnitude of 
chunking. 
Behaviour 
  
Figure 4.6. Profile of RTs across the eight subtrials in the three trial types. The first RT was 
highest of the eight subtrials in trial types 1 and 3, but not in trial type 2. Only trial type 2 
showed a high RT at the fifth position. As predicted the first RT was greater than fifth only in 
trial type 1. The fifth RT of trial type 2 was significantly greater than the fifth RT of trial type 1.  
124 
Fig. 4.6 shows the profile of reaction times in the three trial types. In trial type 
2, the first RT was higher than the other within-chunk reaction times (t14 > 4.2, 
p<0.001), however it was not higher than the fifth RT (t14=0.8, p=0.4). The fifth RT 
was higher than the other within-chunk reaction times (t14 > 4.1, p<0.01), unlike the 
case in trial type 1. As discussed earlier, this pattern of RTs is strongly suggestive of 
chunking of first and second groups of four subtrials. The results were different for 
trial type 1; the first RT was the highest (paired t14
The profiles of the RTs in the two trial types were directly compared to see if 
the effect of position on the RT was different in the two trial types. This was indeed 
the case (F
 > 3.2, p< 0.01, for all 
comparisons). The fifth RT was not higher than other reaction times, suggesting that 
chunking was poor in this trial type. 
(7,91)=6.1, p <0.01). Lastly, a direct comparison of the fifth RT on the two 
subtrials showed that this was significantly greater in the trial type 2 (t14
In trial type 3, there was an effect of position on the RT (F
 =3.5, p 
<0.01). The behavioural profile of the two trial types shows that these were 
organised differently. Indeed the pattern suggests that two chunks of subtrials in trial 
type 1 were organised as parts of the same episode, while those in trial type 2 were 
organised relatively independently. As discussed earlier, this profile is consistent with 
stronger chunking in trial type 2 than in trial type 1.  
(7, 91) = 3.2, p <0.01), 
which was driven by the first RT being higher than the rest (t14
 
 > 4.0, p<0.001); other 
RTs did not differ. 
Depending upon how the representation of the trial episode is organised in 
the particular brain region, distinct patterns of activity are to be expected. Regions, 
where the trial representation is organised strongly as two episodes (i.e. has a strong 
chunk level organisation), can be expected to show greater activity at the completion 
Imaging 
125 
of the first chunk i.e. the fourth subtrial (chapter 2). Regions that represent the trial 
as a single episode would show increased activity only at the eighth subtrial.  
Regions that are sensitive to the phase of the task can be expected to show an 
increase in activity across the sequential phases. However, the exact pattern of 
activity across the eight subtrials in these regions might depend upon the 
organisation of the trial, since which subtrials constitute the sequential phases of the 
same episode would get determined by the way the trial is organised. In those 
organised as a single episode, the eight subtrials constitute their eight sequential 
phases; whereas in those that are organised as two episodes, subtrials 5 to 8 do not 
belong to the same episode as subtrials 1 to 4. Regions representing the trial as a 
single episode may show a monotonic increase in activity across the eight subtrials, 
whereas those representing it as two episodes will may show parallel increase across 
the first and the second chunk of subtrials, but without any increase across the two 
chunks. 
The activities between analogous phases of the two chunks were compared. 
Subtrials 2 and 3 (first chunk) and 6 and 7 were chosen. Subtrials 4 and 8 were 
excluded to avoid confounds from activity related to the completion of episodes. 
Regions representing the trial as single episodes may show an increase across them 
such that activity in subtrial 7 > 6 > 3> 2, i.e. both within-chunk (3>2, 7>6) and across-
chunk phase effects (6 & 7 > 2 & 3). Regions representing trials as dual episodes may 
show 3 > 2 and 7 > 6, but not 6 & 7 > 2 & 3, i.e. only within-chunk but no across-
chunk increase in activity. This was tested using an ANOVA which modelled these as 
sequential phases within the two chunks, and looked at within-chunk effect i.e. 2 and 
6 vs. 3 and 7, and across-chunk effect i.e. 2 and 3 vs. 6 and 7. 
 
126 
 4 > 
int. 
8 > 
int. 
8 > 
4 
(3 & 7) > (2 
& 6) 
(6 & 7) >(2 
& 3) 
Single 
Episode 
- + + ± + 
Two 
Episodes 
+ + - + - 
 
Table 4.1. Different set of predictions for the two kinds of trial organisation. If the trial is 
organised as a single episode, subtrial 8, finishing the entire episode, can be expected to 
elicit maximum activity. Subtrial 4, in such cases, should not differ from other intermediate 
subtrials. Finally, the activity in some regions can be expected to increase monotonically 
across the eight subtrials, which can be quantified as greater activity during subtrials 6 and 7 
compared to that at 2 and 3. On the contrary, if the trial episode is chunked, both subtrials 4 
and 8, completing the two chunks, can be expected to have higher activities than other 
intermediate subtrials. Further, subtrials within a chunks may show increase in activity (3 > 2 
and 7 > 6) but this increase may not be present across the subtrials of the two chunks. 
 The various predictions are summarised in table 4.1. It is possible that a trial 
type may be organised in different ways in various regions. One or the other pattern 
may dominate across these regions depending upon the predominant way the 
representation of that trial type is organised. Some regions may show a mixture of 
these two patterns of representation. 
 
 
 
 
 
 
127 
4.3.1 Trial Type 1 
 
 
Figure 4.7. Pattern of activity across the eight subtrials of trial type 1 in the various ROIs. 
The three ROIs in light brown are the phase-sensitive ROIs created from the results of the 
last experiment (chapter 3). 
Figure 4.7 illustrates the profile of activity across the subtrials of trial type 1 in 
the different ROIs. As is evident, the level of activity varied across the eight subtrials, 
however, the actual pattern of variation seems to vary across the different ROIs. This 
was confirmed by a repeated measures ANOVA with subtrial position and ROI as 
main factors, which showed a significant interaction between the two (F(84,1092) = 4.6, 
p <0.001). Two distinct trends can be discerned. Some ROIs (e.g. medial frontal, ACC, 
bilateral ventrolateral frontal regions) show a gradual increase in activity over the 
128 
eight subtrials, but no additional increase in activity in the fourth subtrial. Activity 
during the fourth subtrial in these ROIs is not greater than other within-chunk 
subtrials (e.g. fifth and sixth subtrials). In contrast, some regions (e.g. IPS-L, IFS-L, and 
APFC-L) do not show an increase in activity across the eight subtrials, but have higher 
activity for subtrials marking the completion of the two chunks, i.e. the fourth and 
the eighth.  
To find areas affected by the chunk level organisation the activity in the fourth 
subtrial was compared with that during the within-chunk subtrials. This was 
significant in left APFC and IFS, bilateral IPS and pre-SMA (t(13)>2.7, p < 0.02). As can 
be expected from Fig. 4.7, the activity during the eighth subtrial was greater than 
that during the intermediate subtrials in all ROIs (t(13)
A repeated measures ANOVA compared the activities during subtrials 2, 3, 6 
and 7 to look at phase effects within chunks (3 & 7 > 2 & 6) and across chunks (6 & 7 
> 2 & 3). Only in ACC was the within-chunk phase effect significant (F
>3.1, p < 0.01), except left IFS 
and left IPS. Most of such ROIs also showed greater activity during the eighth 
compared to the fourth subtrial (table 4.2). 
(1,13) = 12.7, p 
<0.01), others (F(1,13) < 2.9, p > 0.1). Across-chunk phase effect was significant in right 
IFS, AI/FO, ACC and the phase-sensitive ROIs (F(1,13) > 6.8, p < 0.02), and nearly 
significant in left AI/FO (F(1,13)
These results confirm the two distinct pattern of behaviour in the ROIs. Left 
APFC, IFS, IPS and pre-SMA showed greater activity at the completion of the first 
chunk but did not show any effect consistent with regions representing the trial as a 
single episode - activity during subtrial 8 was not greater than during subtrial 4, there 
was no increase in activity across the two chunks. Other ROIs did not show any 
evidence of the chunk level organisation, but showed evidence consistent with 
sequence level organisation – Activity at subtrial 8 was higher than other subtrials, 
 = 4.5, p = 0.05). 
129 
but subtrial 4 was the same as intermediate subtrials; there was increase in activity 
across the two chunks. The summary of the above results are in table 4.2. 
 
 4 > 
int. 
t
8 > int. 
(13) 
t
8
(13) 
 
t
> 4 (3 & 7) > 
(2 & 6) 
(13) 
F
(6 & 7) 
>(2 & 3) 
(1,13) F(1,13) 
APFC-R 0.5 5.5** 3.9** 1.8 2.2 
APFC-L 2.7* 3.2** 1.0 2.5 0.3 
IFS-R 0.6 6.2** 3.6** 0.01 12.6** 
IFS-L 2.5* 0.6 -1.2 0.04 1.2 
AI/FO-R 0.7 5.5** 4.4** 0.5 6.8* 
AI/FO-L 1.5 4.4** -2.0 2.1 4.5 
Pre-SMA 2.7* 2.2* 0.4 0.1 0.8 
ACC 0.7 5.5** 4.3** 12.8** 10.9** 
IPS-R 1.7 3.0* 1.5 0.1 0.2 
IPS-L 2.4* 1.7 -0.23 0.2 0.3 
vFr-R 0.1 7.2** 5.7** 1.8 11.2** 
vFr-L 0.07 7.8** 5.0** 2.1 26.0** 
MF 0.2 7.6** 6.5** 2.9 18.8** 
 
Table 4.2. (Trial type 1). Summary of ROI analyses. The significant t and F statistics are in 
bold. (* 0.05 > p >0.01; **: p < 0.01)  
130 
None of the ROIs showed significantly greater activity at the first subtrial, 
neither did any ROI show a decrease in activity across the subtrials as would have 
been expected if their activity was commensurate with the working memory load. 
 
 
Figure 4.8. (Trial type 1) Whole brain render of a contrast looking at regions where activity 
during the fourth subtrial was greater than during other intermediate subtrials. Significant 
activity can discerned in left IFJ, pre-SMA and left APFC. (uncorrected p<0.001) 
  To look at other brain regions showing the above effects, a number of 
contrasts were done. The first contrast looked at brain regions that were affected by 
the chunk level organisation. Since this would predict higher activity during the 
fourth subtrial, the fourth subtrial was contrasted against the intermediate ones (2, 
3, 6, 7). No brain region survived the whole brain FDR correction at p<0.05. However, 
at an uncorrected threshold of p<0.001, the results paralleled those from the ROI 
analysis (Fig. 4.8); areas of significance were present in the left inferior frontal 
junction (slightly posterior to the left IFS ROI), left pre-SMA and left IPL, along with 
clusters of activity in the left APFC and the midbrain.  
131 
 
Figure 4.9. (Trial type 1) Areas with greater activity during the eighth subtrial compared to 
earlier subtrials.  
 In a second contrast, the final (eighth) subtrial was contrasted against the 
earlier ones. Significantly greater activity for the eighth subtrial was found in many 
brain regions (Fig. 4.9, coordinates in table A4.1 of appendix). In the frontal cortex, 
this was visibly greater on the right, where it involved the entire extent of the lateral 
PFC except for a small region in the middle of the middle frontal gyrus; medially the 
entire medial prefrontal cortex extending posteriorly to the anterior parts of ACC and 
pre-SMA was involved. In the left PFC, the effect was limited to the anterior insula, 
frontal operculum, parts of ventrolateral PFC and the superior frontal gyrus. In 
parietal cortices, bilateral inferior parietal lobules, angular gyrii, temporo-parietal 
junctions, precunei and posterior cingulate were involved along with parts of the 
right somatosensory regions. Huge swathes of areas in occipital and temporal 
cortices also showed this effect, along with hippocampus and parahippocampus, 
lateral cerebellum, the caudate nuclei and the amygdala.  
A contrast of average activity during subtrials 6 and 7 against that during 2 
and 3 revealed very similar results (Fig. 4.10), showing that the increase was not 
limited to the eighth subtrial but subsequent phases other than the final phase also 
showed greater activity than the earlier phases. The results were identical when a 
linear contrast was done across subtrials 2 to 7 which looked at regions where the 
activity increased in a stepwise manner.  
132 
In summary, the results of this trial type thus show that in widespread regions, 
this trial type was organised as a single coherent task episode. The evidence of chunk 
level organisation, though present, was weak. 
 
 
Figure 4.10. (Trial type 1) Regions with greater activity during the subtrials 6 & 7 compared 
to subtrials 2 & 3. Note that the regions are identical to those showing greater activity during 
the eighth compared to the earlier subtrials. 
An important feature of the above results is its confirmation that the phase 
effect, described in the last chapter, is not limited to the final phase, but extends 
across the entire sequence of phases making up the task episode. As in the previous 
chapter, this effect was widespread. Indeed in the fronto-parietal cortices, areas 
considered to be core regions of task control (Right APFC, bilateral AI/FO, ACC; 
Dosenbach et al., 2006) show robust increase in activity along the length of the trial 
along with others (e.g. posterior cingulate, IPL, TPJ, superior temporal sulcus) that 
are considered part of the default mode network and tend to deactivate during the 
trial blocks (Raichle, 2003). These two groups have hitherto been considered to be 
anti-correlated with each other (Fox et al., 2005), but in the above result behave in a 
very similar manner.  
No region showed a decrease in activity along the length of the trial, even at a 
very liberal threshold of uncorrected p<0.05, suggesting that nowhere was the 
133 
activity directly correlated with the working memory load. Lastly, compared to other 
subtrials, nowhere was the activity greater activity during the first or the fifth 
subtrials, even though, as manifested in RT profile, these were the points at which 
greater control was required. 
4.3.2 Trial Type 2 
 
Figure 4.11. Pattern of activity across the eight subtrials of trial type 2 in the various ROIs.  
 
Fig. 4.11 shows the profile of activity in the various ROIs during type 2 trials. 
As in the earlier trial type, the level of activity differed across the eight subtrials 
across ROIs (F(7,91) > 2.3, p < 0.03), but the actual pattern of this was different for the 
different ROIs (F(84,1092) = 3.7, p <0.001). The comparison of activity during the fourth 
subtrial with the within-chunk subtrials showed greater activity during the fourth 
subtrial in bilateral IFS, IPS, AI/FO, right APFC, vFr and pre-SMA  (t(13)>2.2, p<0.05), 
134 
showing that the effect of chunk completion was more widespread than trial type 1. 
The activity during the final subtrial was then compared to that during the earlier 
subtrials. The final subtrial had significantly greater activity in bilateral AI/FO, right 
IFS, APFC, ACC and the three phase-sensitive ROIs (t(13)>2.4, p<0.03). However, the 
activity during the eighth subtrial succeeded the fourth only in ACC and the phase-
sensitive ROIs (t(13)
As for the previous trial type, activities across subtrials 2, 3, 6 and 7 were 
compared for within and across-chunk phase effects. Unlike trial type 1, many ROIs 
showed a significant within-chunk phase effect without a concomitant across-chunk 
phase effect. This was the case in bilateral APFC, left IFS and IPS, and pre-SMA (F
>2.5, p<0.02). 
(1,13) 
> 5.6, p < 0.03). Whereas in right IFS, AI/FO, vFr, ACC and MF both kinds of phase 
effects were significant (within-chunk: F(1,13) >12.0, p < 0.01; across-chunk : F(1,13)
 
 > 
5.1, p < 0.05).  The results are summarised in table 4.3. 
4 > 
int. 
t
8 > 
int. 
(13) t
8
(13) 
 
t
> 4 (3 & 7) > 
(2 & 6) 
(13) 
F
(6 & 7) >(2 & 
3) 
(1,13) F(1,13) 
APFC-R 2.9* 2.4* 0.9 9.1* 1.5 
APFC-L 1.9 2.2 0.5 13.0** 1.0 
IFS-R 2.7* 4.4** 2 17.5** 7.4* 
IFS-L 2.4* 0.5 -1.2 8.4* 0.9 
AI/FO-R 4.2** 4.6** 1.5 13.0** 11.1* 
AI/FO-L 4.3** 3.3** 0.8 2.1 4.5 
Pre-SMA 3.1* 1.7 -0.2 18.1** 3.6 
ACC 1.1 3.6** 2.5* 13.2** 6.8** 
IPS-R 3.0* 1.5 -0.9 4.4 1.1 
135 
IPS-L 3.3* 0.4 -1.9 5.6* 0.3 
vFr-R 2.2* 3.8** 2.9* 12.2** 6.0* 
vFr-L 1.3 4.5** 3.2* 4.5 3.7 
MF 1.7 5.9** 3.5** 12.0** 5.1* 
 
Table 4.3 (Trial type 2). Summary of ROI analysis. (* 0.05 > p >0.01; **: p < 0.01)  
The pattern of results (subtrial 4 > intermediate, within-chunk phase effect, 
absence of subtrial 8 > 4, and lack of across-chunk phase effects) in bilateral APFC 
and IPS, left AI/FO and IFS and pre-SMA suggest that these represented trial type 2 as 
two distinct chunks. ACC, left vFr and MF show patterns consistent with regions 
representing the trial as a single episode (subtrial 8 > 4; across-chunk phase effects). 
Right IFS, AI/FO and vFr show greater activity at subtrial 4 and within-chunk phase 
effect, but also show across-chunk phase effect.  
 
Figure 4.12. (Trial type 2): Areas with greater activity during the fourth subtrial compared to 
the with-chunk subtrials. Note that the results are widespread and more intensive than those 
obtained from trial type 1 (Fig. 4.8). 
Whole brain contrasts were then done to look into other regions showing 
these effects. As in the previous trial type, to look at the brain regions affected by the 
chunk level organisation, activity for the fourth subtrial was contrasted against those 
during the intermediate subtrials. Fig. 4.12 shows a whole brain render of this 
contrast (coordinates in table A4.2). Compared to the case in trial type 1, where the 
136 
results were weak and could not survive correction for multiple comparison and 
could only be documented at an uncorrected p <0.001, the results here were more 
extensive. As in the ROI analysis, many regions in the fronto-parietal cortex showed 
greater activity during the fourth subtrial. These included AI/FO, areas along the 
antero-posterior axis of IFS up till APFC, premotor regions – ventral and dorsal 
extending medially onto SMA, pre-SMA and ACC. Notably, the locus of activity in ACC 
was posterior and dorsal to the locus of ACC that is one of the ROIs, which would 
explain the absence of this result in the ACC ROI. Posteriorly, the IPS was involved, 
along with superior parietal lobule, IPL, supramarginal and angular gyri, and the TPJ 
along with the posterior regions of superior and middle temporal gyrii. Medially, 
precuneus, cuneus and parts of the primary visual cortex were also involved.  
The brain regions where the activity was greater during the eighth compared 
to the intermediate subtrials differed from those showing greater activity for the 
fourth subtrial in interesting ways.  
 
Figure 4.13. (Trial type 2) Areas with greater activity during the eighth subtrial compared to 
the within-chunk subtrials. 
As is evident in Fig. 4.13, activity during the eighth subtrial was greater in the 
right prefrontal (premotor, AI/FO, VLPFC extending onto the APFC, and right superior 
medial regions) and the right parietal cortex (IPL, TPJ, angular gyrus, parts of primary 
sensory and motor cortex, medially mid and parts of posterior cingulate, paracentral 
137 
lobule and precuneus). Various regions in temporal, occipital and cerebellar cortex 
also showed greater activity during the eighth subtrial (Coordinates in table A4.3). 
 
Figure 4.14. (Trial type 2) Contrasts comparing areas with greater activity for the eighth 
subtrial over the within-chunk subtrials (green) rendered on a common template with areas 
having greater activity for the fourth over the within-chunk subtrials.  
 
The relation of these areas with those that showed greater activity for the 
fourth subtrial is interesting. Fig. 4.14 shows the two contrast images rendered on a 
common template (the images are at a slightly liberal threshold of FDR corrected 
p<0.05). Maximum overlap can be seen in the medial occipital regions (areas 17, 18, 
cuneus), right IPL and areas around the right TPJ, right IFS and the right AI/FO (MD 
region). Beyond this overlap, a clear cut separation is evident especially in the 
prefrontal cortex. Areas in the middle frontal gyrus show greater activity for the 
fourth subtrial, whereas those in the superior and inferior frontal gyri, especially in 
the right PFC, have greater activity for the eighth subtrial. Medially there is an 
antero-posterior distinction, with anterior regions of medial PFC being more sensitive 
to the eighth subtrial. Similarly anterior regions of the medial parietal regions have 
greater activity during the eighth subtrial whereas its posterior regions are more 
active during the fourth. All three phase-sensitive prefrontal regions (venterolateral 
frontal and superior medial regions), expectedly, prefer the eighth subtrial. Many of 
138 
these distinctions turned out to be statistically significant in directed contrasts 
between the fourth and the eighth subtrial activities (Fig. 4.15). 
 
Figure 4.15. (Trial type 2). Purple: areas where the eighth subtrial had greater activity than 
the fourth, blue: areas where the fourth subtrial elicited greater activity than the eighth.  
Lastly, the brain regions showing an increase across the two chunks were 
examined (Fig. 4.16). A comparison of mean activity between subtrials 2 & 3 and 6 & 
7 showed higher activity for the subtrials 6 & 7 in the bilateral APFC, and the right 
AI/FO, extending into VLPFC, and medial PFC. Other regions showing this effect 
included parts of the superior and middle temporal gyri including STS, posterior 
insula, pre and paracentral gyri (especially on the right), head of caudate nucleus and 
secondary visual areas. As would be expected, a reverse contrast showed that no 
region had greater activity for the first chunk compared to the second. Note that the 
locus of APFC in this contrast is anterior to the ROI based on APFC that was defined a 
priori and used in the ROI analysis.  
 
139 
Figure 4.16. (Trial type 2) Regions with greater average activity during the subtrials 6 & 7 
than during the subtrials 2 & 3. Note the paucity of the results compared to that in the earlier 
trial type (Fig. 4.10) 
4.3.3 Comparison of the two trial types 
The results from the analysis of the two trial types show that in a hierarchical 
task episode with multiple levels, brain regions are affected by organisation at all 
levels. Some regions seem to be affected only by the lower level organisation to the 
exclusion of any effect from the higher level organisation, and vice versa. Left APFC, 
IFS, IPS and pre-SMA in both trial types were only affected by the lower chunk level 
organisation, whereas the left vFr, MF and the ACC were only affected by the higher 
sequence level organisation i.e. these regions showed the effect expected if the 
entire trial was a single episode in both trial types. The behaviour of other regions, 
however, was variable across the two trial types. For example, right prefrontal MD 
ROIs were affected only by the sequence level organisation in trial type 1, but in trial 
type 2 they were mostly affected by the chunk level organisation.  
It was found that trial type 2 showed greater evidence of organisation as two 
separate chunked episodes than trial type 1 but weaker evidence of organisation as a 
single sequence of subtrials compared to trial type 1, and vice-versa. Right APFC, 
bilateral AI/FO and right IFS showed evidence of chunk level organisation in trial type 
2, but of sequence level organisation in trial type 1. To make a direct comparison 
between the two trial types, a repeated measures ANOVA looked at the interaction 
of the trial type with the effect of the position of the subtrial in the chunk (chunk 
level effect) and with the effect of the position of the chunk in the sequence 
(sequence level effect). The effect of position within the chunk differed across the 
two trial types in right IFS, right AI/FO and pre-SMA (F(1,13) > 4.9, p<0.01); however in 
none of the ROIs did the difference of the effect of the position of the chunk in the 
sequence, across the two trial types, reach statistical significance (F(1,13) < 3.4, 
p>0.06).  
140 
The whole brain comparisons supported these conclusions. The fourth subtrial 
showed greater activity in more widespread areas in trial type 2, whereas increase in 
activity across the two chunks was more evident in trial type 1.  
The results so far, show that the two trial types are differently organised, and 
that subtle difference in organisation of higher level representations do indeed 
change the organisation of the identical lower level behavioural events. The 
unorganised trial (trial type 3) allowed for the test of this hypothesis at an even more 
subtle level. Recall that there was no sequence guiding the performance on and 
organising the eight subtrials of this trial type. Instead the relevant rule was signalled 
on each subtrial by the colour of the margins of the stimulus. Thus there was no 
explicit organising higher level representation. However, since this trial had the same 
structure as the organised trials, it was predicted that there would be transfer of the 
implicit higher level representations from the organised trials. For example, as the 
subjects had become biased to chunk the organised trials into groups of four, it was 
expected that they would see the unorganised trial as having two groups of four 
subtrials, even though this had no role in the execution of the subtrials of this trial 
type.  
141 
4.3.4 Trial Type 3 
 
Figure 4.17. Pattern of activity across the eight subtrials of trial type 3 in the various ROIs.  
 
As in the earlier trial types, the activity varied across the eight subtrials (Fig. 
4.17) in all ROIs (F(7,91) > 2.2, p < 0.04), and the ROIs differed in their differential 
response to the eight subtrials (F(84,1092)
The fourth subtrial elicited greater activity compared to the intermediate 
subtrials in bilateral AI/FO (t
 = 3.2, p <0.001). Since, unlike the previous 
trial types, this trial episode was not further organised by chunked letter strings, the 
subtrials 2 to 7 were equivalent. However, as is evident in Fig. 4.17, the activity is, 
nonetheless, variable across them. The fourth and eighth subtrials were specifically 
investigated. 
(13)>2.4, p<0.02), and close to significance in left IFS 
142 
(t(13)>2.0, p<0.06). The activity at the eighth subtrial was the highest in most ROIs, it, 
however, reached significance only in ACC and right IPS, right IFS, right VLPFC, and 
MF (t(13)>2.2, p<0.04). Other aspects of organisation into two chunks were also 
present. Many regions showed within-chunk phase effect – bilateral IFS, AI/FO, IPS 
and pre-SMA (F(1,13)
 
> 6.3, p< 0.03), however none showed a significant effect across 
the two chunks. These are summarised in table 4.4. 
 4 > int. 
t
8 > int. 
(13) t
8
(13) 
 
t
> 4 (3 & 7) > 
(2 & 6) 
(13) 
F
(6 & 7) 
>(2 & 3) 
(1,13) F(1,13) 
APFC-R 1.4 1.6 1.3 1.3 0.3 
APFC-L 0.9 1.7 1.5 0.8 1.2 
IFS-R 0.7 3.2* 2.5* 9.3* 0.8 
IFS-L 2 0.4 -0.1 16.1** 0.6 
AI/FO-R 2.4* 1.9 1.5 9.7* 4.6 
AI/FO-L 2.6* 1.8 1.2 6.3* 1.1 
Pre-SMA 1.8 0.3 0 18.1** 3.6 
ACC 1.9 2.2* 2 0.3 0.1 
IPS-R 0.2 2.9* 2.5* 7.0* 0.0 
IPS-L 0.5 1.3 1 12.1** 0.0 
vFr-R 0.9 1.5 1.3 2.0 0.1 
vFr-L 0.7 3.0* 2.8* 1.6 0.1 
MF -0.6 2.8* 2.7* 1.9 0.3 
Table 4.4. (trial type 3) Summary of ROI analysis. (* 0.05 > p >0.01; **: p < 0.01) 
143 
 
Unlike the case in the previous trial types, the activity in the first subtrial was 
greater than that in the intermediate subtrials in a number of ROIs – ACC, pre-SMA 
and bilateral AI/FO (t(13)>2.8, p<0.01).  
 
Figure 4.18. (Trial type 3). Areas with greater activity at the fourth subtrial compared to the 
within-chunk subtrials. 
Following the ROI result, a contrast was made between the fourth against and 
the other intermediate subtrials (Fig. 4.18). Greater activity for the fourth subtrial 
was found in a number of brain regions - bilateral insula, left IFJ, left IFS (near the MD 
IFS region), left dorsal premotor extending into motor cortex, SMA and pre-SMA, and 
bilateral inferior occipital regions (coordinates in table A4.4). Thus, the abstract view 
of this trial as having two groups of four subtrials did indeed affect the activity in 
certain fronto-parietal and occipital regions. A noteworthy point in this regard is that 
this abstract conception of task plan did not affect activity in anterior prefrontal 
regions but rather in the posterior prefrontal regions which are usually thought to be 
concerned with concrete task demands. 
144 
 
Figure 4.19. (Trial type 3) Areas with greater activity at the eighth subtrial compared to the 
within-chunk subtrials. 
As in previous trial types, widespread areas showed greater activity for the 
last subtrial (Fig. 4.19), the extent of which, however, was limited compared to the 
previous trial types. In the prefrontal cortex this involved similar right lateralised 
regions– right inferior and superior frontal gyri, dorsal premotor and medial frontal 
regions. Similarly, the posterior clusters like in earlier trial types included regions 
around the TPJ, posteror to mid regions of the superior and middle temporal gyri, 
and posterior cingulate (table A4.5).  
4.3.5 Comparison of trial types 1 and 3 
Both of these trials consisted of eight sequential subtrials or phases, however 
the subjects’ knowledge of the phase they were in was less explicit in trial type 3, as 
their performance was not guided by an explicit sequence. Accordingly it was 
expected that the activity related to the phase of the trial, seen as a sequential 
increase in activity across the task phases would be more prominent in trial type 1. 
Trial type 3 being structurally similar to trial type 1 also becomes a control condition 
to demonstrate that the stepwise increase across the phases of a trial is not an 
artefact of modelling sequential events. 
 
145 
ROIs Trial Type 1 Trial Type 3 
MF-R 1.02 0.22 
VLPFC-R 0.92 0.33 
VLPFC-L 0.86 0.19 
ACC 0.63 -0.05 
IFS-R 0.70 0.45 
AI/FO-R 0.53 0.26 
pre-SMA 0.37 -0.05 
IPS-R 0.37 0.33 
APFC-R 0.35 0.13 
AI/FO-L 0.45 0.10 
APFC-L 0.14 -0.05 
IFS-L 0.05 0.03 
IPS-L 0.05 -0.06 
   
Table 4.5. z-transform of the Pearson's correlation coefficient between the subtrial number 
and its elicited activity. Note that the correlation coefficient was higher for trial type 1 in all 
ROIs. 
In this regard, the linear trend in the profile of activity across the eight 
subtrials was quantified. Regions that are most sensitive to the phase of the task 
element can be expected to show a stepwise increase in activity across the eight 
subtrials and so have a stronger linear trend than those less sensitive. A Pearson’s 
correlation was done between the subtrial number and its elicited activity to test the 
strength of the linear relationship. The correlation coefficients obtained were 
converted into z values using Fisher’s z-transform to enable parametric statistical 
comparisons (Table 4.5). As apparent in Table 4.5, the z-transform of the Pearson’s 
coefficients were greater for trial type 1 than for trial type 3 in most ROIs, being 
significant in all (t(13) > 2.3, p <0.03) except bilateral IFS and IPS. As is apparent, the 
linear trend was strongest in the phase-sensitive ROIs. Amongst the MD ROIs, right 
hemisphere regions had stronger linear trend than those on the left (F(1,13) = 56.6, 
p<0.001), again confirming that effect of the sequence level organisation was 
dominant in the right hemisphere MD regions. 
146 
A contrast looking at brain regions where the linear increase in activity across 
the length of the trial was greater for trial type 1 than trial type 3 revealed significant 
results in widespread regions (Fig. 4.20), which were largely similar to regions that 
showed greater activity at the completion of the trial episode or that showed the 
effect of sequence level organisation in the trial types 1 & 2 (c.f. Figs 4.9, 4.10 and 
4.13). It involved the phase-sensitive regions of the prefrontal cortex – ventrolateral 
PFC and superior medial frontal gyrii, along with superior frontal gyrus, parts of 
middle frontal gyrus, APFC, dorsal premotor, pre-SMA. Posteriorly, regions around 
the TPJ, posterior cingulate and the precuneus showed significant results.  
 
Figure 4.20. Regions where the linear increase in activity across the length of the trial was 
greater in trial type 1 compared to trial type 3. 
 
While linear increase in activity was more prominent in trial type 1, the 
disparity between the activities during the final versus intermediate subtrials was 
largely similar across the two trial types (Figs. 4.7 & 4.17). This was confirmed by 
repeated measures ANOVA that compared the activity in the eighth subtrial with that 
collapsed across the intermediate subtrials, across the two trial types, in the various 
ROIs. In none of the ROIs was there a significant main effect of the trial type (F(1,13) < 
3.1, p > 0.1) or an interaction of the disparity in activities between the final and 
intermediate subtrials with the trial type (F(1,13) < 1.7, p > 0.2). Neither did any brain 
region show a significant effect in the whole brain contrast looking for regions where 
147 
the difference between the activities of the final and intermediate subtrials was 
different across the two trial types. Thus the organisation of the task episode seems 
to affect only the increase in activity across the various phases of the episode but not 
the activity at the end of the episode. 
Results so far show that the pattern of fronto-parietal activity elicited by the 
various elements of the sequential task episode depends upon the way the episode 
has been organised. Trial types 1 and 2 were organised by an eight element 
sequence. Subjects had been primed to chunk this sequence into two four-element 
chunks. Behavioural evidence suggested that this chunking was stronger in the trial 
type 2. Contrary to these trial types, trial type 3 did not have any organising 
sequence, but like other trials consisted of eight sequential subtrials.  
Results so far 
The results showed three kinds of brain regions. The first set of regions (hence 
called Group I regions), which included the three phase-sensitive brain regions 
(bilateral ventral frontal, superior frontal and superior medial frontal regions) and 
ACC had the highest activity for the eighth phase, whereas their activity during the 
fourth phase was not higher than during intermediate phases (2nd, 3rd, 6th and 7th). 
Further these regions showed a trend towards sequential increase in activity across 
the eight phases in the first two trial types. Since this was the only pattern of results 
shown by these regions, it can be said that this set of regions was sensitive only to 
the sequence level organisation, but not to the chunk level organisation. On the 
other extreme were such brain regions (Group II regions), which only showed pattern 
of results consistent with chunked organisation of the trial, i.e. greater activity at 
subtrial 4 and increase in activity within a chunk, but not across the two chunks. 
These were the left hemisphere lateral prefrontal regions (IFS, APFC), IPS and the 
pre-SMA. Additionally, in both organised trial types, eighth subtrial activity in these 
148 
regions was not greater than the fourth. Thus, these regions were sensitive only to 
the chunk level organisation but never to the sequence level organisation.  
The behaviour of these two set of brain regions was more or less consistent 
across the two organised trial types, however, the behaviour of right hemisphere MD 
regions varied (Group III regions) across the two. These regions showed both – a 
weak chunk and but robust sequence level effect in trial type 1, but showed the 
chunk level and some sequence level effect in trial type 2. The variability in these 
regions was probably related to the way the representations of these trials were 
organised in the subjects’ minds. Subjects’ behaviour in these two trials showed that 
the effect of chunking was stronger in trial type 2 than in trial type 1. It is plausible 
that these regions are most sensitive to the subtleties of the plan of organisation. 
The results in trial type 3, showed three distinct points. The activity was the 
highest during the final phase in the first set of regions mentioned above, and in 
some of the third set of regions (right IFS and IPS). The sequential increase in activity 
across the phases was much less compared to the organised trials, suggesting that 
the increase in activity across the phases of a task episode is dependent upon the 
organisation of the episode. Finally and most interestingly, the activity across the 
intermediate (2nd to 7th) phases varied, and in certain regions (left IFS, left premotor, 
bilateral AI/FO and pre-SMA) was higher for the fourth phase than the average of 
other intermediate phases, which probably meant that there was some transference 
of the neurocognitive structures of the organised trials to the unorganised trial. The 
last result also makes it unlikely that the increase in activity in IFS, IPS and pre-SMA 
at the completion of the first chunk during the organised trials is solely the result of 
the control requirements for accessing and instantiating the representations related 
to the second chunk, as was speculated earlier.  
149 
4.3.6 Cognitive Load and Phase Effect 
An interesting aspect of the above results from whole brain analysis is the co-
activation of the MD regions with another set of fronto-parietal regions that have 
recently been referred to as the default mode network (DMN, reviewed in Buckner, 
2008) or task negative regions (Fox et al., 2005). Like MD regions, these regions also 
show modulation of activity across a wide variety of task situations (Dosenbach et al., 
2007). However, in most studies their modulation tends to be anti-correlated with 
the MD regions, hence the name ‘task negative’ as opposed to the ‘task positive’ MD 
regions (Fox et al., 2005). Supporting this view, mind wandering (Mason et al., 2007) 
and attentional distractions (Weissman et al., 2006) have been shown to increase 
activity in default mode regions while causing a decrease in MD region activity, 
whereas the execution of task blocks is associated with an increase in MD activity but 
with a decrease in the DMN activity (Dosenbach et al., 2007). Further, this anti-
correlation seems to increase with increased cognitive demands (Singh and Fawcett, 
2008). Thus, an increase in the task demand further increases the MD activity while 
further decreasing the activity in the DMN regions. In the light of these, the current 
findings of correlated pattern of activity in the MD and the DMN regions needs 
further investigation.  
 This becomes even more important when it is noted that some aspects of the 
above results seem to suggest that the increase in MD activity seen across the steps 
of organised trials could be the result of the accompanying decrease in the working 
memory load. For example, both trial type 1 and 3 consist of eight steps, but the 
increase in activity across these eight steps was greater in trial type 1 compared to 
trial type 3. Since these two trial types differ crucially in the decrease in WM load 
that occurs across the steps of trial type 1 but not of trial type 3, it is plausible that 
this increase in activity was due to the decrease in the WM load.  
150 
To investigate these issues in greater detail, additional ROIs were created at 
three regions considered to be the core of DMN (Greicius, 2011) - posterior cingulate 
(placed midline to include both hemispheres) and bilateral TPJ. As apparent in Fig. 
4.21, the pattern of activity in these regions on the three trial types was qualitatively 
similar to the fronto-parietal regions investigated earlier. In trial type 1, all regions 
showed a gradual increase along the length of the trial. In none of the regions was 
the activity during the fourth subtrial higher than that during the within-chunk 
subtrials, whereas all regions had significantly greater activity during the eighth 
compared to the earlier subtrials (t(13) >4.1, p<0.01). The average activity across the 
two chunks was also different (F(1,13)
In trial type 2, all ROIs showed a trend towards increased activity during the 
fourth subtrial, however, the direct comparison showed significantly higher activity 
only in bilateral TPJ (t
 > 11.2, p<0.001), showing that the increase in 
activity across the trial was significant.  
(13) >2.1, p<0.05). In all ROIs, there was a significant increase in 
activity between the second and the third phases of the chunk (t(13) >4.2, p<0.01). At 
the sequence level, however, there was no difference in average activity between 
the two chunks (F(1,13)
Together the above results suggest that these ROIs were affected by the 
sequence level organisation in trial type 1, but by the chunk level organisation in trial 
type 2. The behaviour across the two trial types was directly compared in a repeated 
measures ANOVA looking at the main effects of the position of the subtrial in the 
chunk, and of the position of the chunk in the trial, across the two trial types. The 
effect of position in the chunk of the subtrial was significantly different across the 
two trial types in posterior cingulate (F
 > 3.3, p>0.06).  
(3,39) = 2.8, p < 0.05), while the disparity 
between the average activities of the two chunks was different across the two trial 
types in bilateral TPJ (F(1,13) > 8.3, p<0.01); showing that the effects of sequence level 
organisation were significantly different for these two trial types in these regions.  
151 
The behaviour of these regions, thus, was very similar to that of right 
prefrontal MD ROIs and left AI/FO (group III regions). The above results also show 
that in the regions known to be sensitive to change in working memory load, load in 
and of itself is insufficient as an explanation of their behaviour in the two trial types. 
For example, the working memory load remains constant in the first four subtrials of 
the trial type 2, however, the activity in both TPJ and posterior cingulate vary across 
these subtrials. 
 
 
 
Figure 4.21. Pattern of activity in the three trial types (1, red; 2, blue; 3, green) in bilateral 
TPJ and posterior cingulate. Note the distinct patterns across the three trial types, especially 
1 and 2.  
 To further test for the possibility of change in the working memory load as an 
explanation of the various kinds of phase effects, following comparisons were made: 
152 
 Average activity in trial type 1 versus average activity in trial type 3: The 
former had high working memory load compared to the latter, and the thesis 
about cognitive load will predict that the activity during trial type 1 in these 
regions would be lower than in trial type 3. 
 Average activity during the first chunk of trial type 1 with the first chunk of 
trial type 2: subjects maintained two chunks during the former, but one during 
the latter; so the activity during the former can be predicted to be lower than 
the latter 
 Both of these analyses yielded null results across all ROIs. Average activity in 
trial type 1 did not differ from that during trial type 2 (F(1,13) <0.8, p>0.4). Average 
activity during trial type 1 was lower than that during trial type 3 only in the right TPJ 
(F(1,13) =11, p=0.005), suggesting that the working memory load did affect the activity 
of this region. This was not the case in any other region, indeed contrary to the 
above prediction, group II ROIs (IPS, IFS, APFC and pre-SMA) showed greater average 
activity for trial type 1 than trial type 3 (F(1,13) >4.7, p<0.05). Activity did not differ in 
other ROIs (F(1,13)
These findings suggest that it is unlikely that the increase in activity across the 
phases is related to the stepwise decrease in cognitive load per se. It is however 
possible that the stepwise decrease in cognitive load emphasises the knowledge of 
transition of phases to the subjects and that may contribute to the increase in 
activity. While further studies would be required to resolve this issue, the current 
conclusion is supported by the visibly present trend of increase in activity across the 
last four phases in trial type 3 (Fig. 17), where there was no change in cognitive load. 
 <2.4, p>0.12, ). 
4.3.7 Cues 
 The cues preceding the organised trials provided subjects with working 
memory load, whereas the cue preceding the unorganised trial (type 3) did not have 
153 
any working memory load. This provides another way to check if the increase in 
activity across the phases of the organised trials were caused by the stepwise 
decrease in cognitive load, since areas which deactivate with cognitive load can be 
expected to have lower activity during the cues preceding the trial types 1 and 2 
(referred to as Cues 1 and 2) than the cue preceding trial type (Cue 3). Note that 
since the cues were presented for 3.5 s and cue related activity would include both 
the encoding and maintenance related activity. The current design did not allow for 
the separation of these two. 
As is apparent in Fig. 4.21 the activity during Cues 1 and 2 were higher than 
Cue 3 in all MD and phase-sensitive ROIs. However, in DMN ROIs, the pattern was 
reversed and the activity during Cue 3 was the highest. This result shows that activity 
in the fronto-parietal regions considered in the above analyses increased during the 
encoding and maintenance of working memory. Further, the results also show that 
MD and phase-sensitive regions show qualitatively different response compared to 
DMN regions during these conditions.  
154 
 
Figure 4.22. Activity in response to the cues preceding the three trial types. The working 
memory load in the order Cue 1 > Cue 2 > Cue 3. MD regions (brown), phase-sensitive 
regions (light brown) and default mode regions (purple) gave very different response 
patterns. 
 
4.4 Discussion 
The results of this chapter add to those of earlier chapters and show that 
neural activity elicited by a task element is determined by its position in the overall 
organisation of the task episode. End of episodes were associated with maximal 
activity across many regions, steps closer to the end were associated with greater 
activity in many regions compared to those earlier. These effects were not limited to 
the fronto-parietal regions but included almost all major parts of the brain. The 
155 
results also support the thesis introduced in the introduction that the nature of 
higher level representations determine the nature of the task episode. 
4.4.1 Effect of Organisation 
The findings of earlier experiments are confirmed in a new experimental 
framework. Current findings confirm that the effect of phase is not limited to the 
final, but all subsequent phases elicit greater activity compared to the preceding 
ones. The set of fronto-parietal regions that showed this effect in the current 
experiment is very similar to those in the previous chapter. Both were lateralised to 
the right prefrontal cortex and involved the entire extent of inferior (including insula) 
and superior frontal gyri, premotor regions along with most parts of middle frontal 
gyrus and medial prefrontal cortex. The ROIs created from the core regions showing 
the effect of phase in the last chapter, showed very consistent phase effects in the 
current results.  
The pattern of additional activity elicited by the final subtrial of the trials was 
different from the pattern found for task episode completion in chapter 2. But it is 
obvious that the two cases are different. In chapter 2 the focus of investigation was 
additional phasic activity elicited by the event marking the transition across task 
episodes, while the current experimental design only allowed for an assessment of 
total activity at the final phase of the episode. Plausibly, the activity related to the 
last phase does include the activity related to the transition across episodes, but in 
addition also includes other activities, most prominent of which is that related to 
phases closer to the completion of the task episode. That might explain the very 
widespread nature of the current results.  
The experimental design does not allow for a definite conclusion regarding the 
reason behind the stepwise increase in activity. Nonetheless, it is worthwhile to 
consider some possibilities. A critical aspect of the current design was the sequence 
156 
that was to be kept in memory (in organised trials) and then executed out. This 
meant that, in such trials, at every step some representation would be discarded 
from the working memory. It is possible that the increase in activity across the trial 
was related to the parallel decrease in cognitive load. Indeed the whole brain results 
did include the various default mode regions that are known to deactivate with 
cognitive load (Fox et al., 2005). 
Many aspects of the results are difficult to reconcile with this view. Firstly, 
during cue presentation, most fronto-parietal regions showed increased activity with 
increasing load unlike the default mode regions that showed opposite pattern. Thus, 
the former do not normally deactivate with working memory load, consonant with 
the vast literature that shows increased activity in these regions during periods of 
maintaining greater working memory load (reviewed in Cabeza and Nyberg, 2000; 
Duncan, 2006). Secondly, many of these regions (including some of the DMN regions) 
showed increase in activity across the first four subtrials of trial type 2. However, 
since subjects, arguably, maintained only four letters and repeated them, none of 
these could be discarded across the first four subtrials. Hence, based on the working 
memory predictions, this increase should not occur. Thirdly, the magnitude of 
average activity during the various trial types was not inversely correlated with the 
accompanying working memory load. Trial type 1 required maintenance of eight 
letters, while trial type 3 required none. But the average activity across these trial 
types did not differ in any of these regions (including some DMN regions). Finally, 
many of these regions also showed increased activity across the first four and second 
four subtrials of trial type 3 (table 4.4), which had no such load.  
An important finding is the widespread nature of the regions that show the 
effect of phase. These included not just the multiple demand fronto-parietal region, 
but also the default mode regions, sensory and motor cortices, cerebellum and 
subcortical regions. The correlated activity in the MD and DMN regions is 
157 
noteworthy, since these regions are supposed to be anti-correlated (Fox et al., 2005). 
Some studies have noted correlated activity across parts of these networks during 
internally focussed but goal-directed cognition (Spreng et al., 2010; Christoff et al., 
2009). However to my knowledge their correlated activity in case of an externally 
directed task and involving such vast extent of the two networks has not been 
demonstrated. The current results cannot say, if the underlying reason for the 
modulation of activity in the various regions showing this effect is the same. It is 
quite possible that, for example, the DMN and MD regions activate across the trials 
for very different reasons.  
Current findings show some interhemispheric functional differences across 
the prefrontal regions. As noted earlier in the chapter, left prefrontal MD regions 
consistently showed a pattern of activity consonant with chunk level organisation of 
the trial across all three trial types. However, this was not the case for the right 
prefrontal MD regions, which were plastic in their behaviour. In trial type 1 they did 
not show any effect of chunking, but in trial type 2 and 3 they did. But in all trial 
types, in comparison to left prefrontal ROIs, these showed more effects consistent 
with the organisation of the trial as a single episode.  
This could point towards a speculative possibility. As discussed earlier (Fig. 
4.2), representation of a trial as chunked or as a single episode could involve 
representing different aspects of hierarchy. Temporally abstract episodes by 
definition constitute a higher rung in the behavioural hierarchy, for e.g. episode of 
preparing breakfast is more abstract than preparing coffee. Right prefrontal regions 
might be biased towards representing temporally more abstract episodes, and 
hence, in the current experiment, preferentially represented the trial as a single 
episode. This may also explain why the increase in activity at the end of the task 
episodes has been greater in the right prefrontal regions in all experiments described 
in this thesis. 
158 
Similar hemispheric differences have been reported earlier but with regard to 
the temporal abstraction of control. Braver et al., (2003) imaged subjects while they 
performed blocks of single task and blocks in which they switched between two 
tasks. They found left fronto-parietal activity on task switch trials, which was 
modulated by changes in response speed, suggesting that this activity was sensitive 
to the transient demands on switch trials. However, when pure trial blocks were 
compared with task switch blocks, sustained activity was found in right anterior 
prefrontal region, which co-varied with the mixing cost, suggesting that this 
sustained activity was related to more sustained control requirements like 
maintaining two task sets. Thus, while left fronto-parietal regions instantiated 
transient control across trials, sustained control extending over the trial block was 
instantiated by right anterior prefrontal region.  
Velanova et al., (2003) found very similar results on an episodic memory 
retrieval task. Transient left prefrontal activity was present on trials when subjects 
had to recall less recallable events compared to trials on which recall was easy. Right 
anterior prefrontal regions, on the other hand, showed sustained activity when 
blocks of less recallable events were compared with blocks of easily recallable 
events. A number of other studies have reported similar distinctions (Koechlin et al., 
1999, 2003; Badre & D’Esposito, 2006; Pisapia et al., 2007; Reynolds et al., 2009). 
4.4.2 Higher Level Representations in Organisation 
The current results show that organisation of higher level representations 
does determine the organisation of the ensuing task episode. To an extent it could be 
regarded as a truism. After all, the task instructions and higher level representations 
ensuing therefrom, do determine how the task is to be executed and hence how the 
behaviour is to be controlled (Logan and Gordon, 2001). But how will the ensuing 
behavioural episode be parsed into episodes of specific cognitive focus, is usually 
159 
ignored. The various mental/behavioural events which will constitute a task is always 
assumed, while its determinants have not been looked into. This issue was 
investigated in chapter 2, wherein it was shown that task instructions could 
determine the parsing of a set of target detections into distinct task episodes. The 
current experiment showed that biasing the way higher-level representation like the 
task list is organised changed the way the ensuing behaviour was organised. This 
raises the possibility that the behaviour could be organised by sequencing and 
arranging abstract task related concepts, and it is the nature of these concepts that 
determines the details of the ensuing task episode. 
Behaviour 
Although both of the organised trial types had a list of eight letter strings to be 
executed, the strings used in trial type 2 involved a repetition of a four-letter 
sequence. In the resulting behaviour chunking was concluded on the assumption that 
it would take longer to execute a step if it involved moving across hierarchically 
organised episodes (Schneider and Logan, 2006, 2007; Lien and Ruthruff, 2004). Fifth 
RT was more prolonged in trial type 2, suggesting that the chunking was greater in 
this trial type. It is likely that this was due to the fact that in case of trial type 2, 
representing just four letters was more parsimonious, while trial type 1 required the 
representation of all eight letters. It is interesting to note that a strategy to reduce 
working memory load, on its flip side, resulted in greater behavioural effort 
manifesting in greater RT at the fifth subtrial. 
The current results bear an interesting comparison to those of Schneider and 
Logan (2006). They had subjects memorise and execute two similar four letter 
chunks (‘AABB’ and ‘ABBA’) in the order –  
chunk1 chunk1 chunk2 chunk2 chunk1 chunk1 chunk2 chunk2… 
160 
 The chunks were thus alternately repeated and switched. They found that RT 
was greater on the first element of the chunk when the current chunk was a switch 
from the earlier one than when it was a repeat. The authors interpreted this as a 
chunk switch cost, akin to task switch cost seen when a new task has be to executed 
(Rogers and Monsell, 1995). The current experiment tried to replicate this finding 
using just two chunks. However, just the opposite was found, chunk repeats resulted 
in greater cost of transition than chunk switches.  
It is possible that the findings of Schneider and Logan (2006) do not stem from 
chunk switching per se, but from the peculiarities of the organisation of the resulting 
sequence. Their design required subjects to execute the chunks in the above-
mentioned sequence ad infinitum. In such a case subjects could have hyper-chunked 
the chunk repeats, and the series –  
chunk1 chunk1 chunk 2 chunk2 chunk1 chunk1 chunk2 chunk2 
could have been organised as –  
  chunk1 chunk1 – chunk2 chunk2 – … 
 It is possible that the same contributes to the task switch cost seen in many 
studies that use a similar design of alternate task switches and repeats (Rogers and 
Monsell, 1995). Indeed, the finding that organising such tasks into sequences alters 
task switch costs (De Jong, 1995; Lien and Ruthruff, 2004) supports this speculation. 
Imaging 
The elicited pattern of activity in the two trial types were different in a way 
that would be expected from their different organisation, with trial type 2 being 
organised more as two episodes, and trial type 1 as a single episode. Greater and 
more widespread activity was seen at the fourth subtrial in trial type 2 than in trial 
161 
type 1. More widespread result was seen in trial type 1 for regions showing 
maximum activity at subtrial 8. Secondly, more areas in trial type 1 showed increase 
in activity across the eight phases, whereas trial type 2 had more frequent cases of 
areas showing increase in activity across the four phases of the chunk, but no 
increase in activity across the two chunks. 
Why does the organisational structure of the episode affect the fronto-
parietal activity? The immediate answer would be that the organisation determines 
the nature of control structure required to execute the episode. The behavioural 
profile suggests that the control requirements did indeed vary with the trial 
organisation. However, the pattern of activity across the trial does not seem to agree 
with what would be expected if fronto-parietal regions were sensitive to the control 
demands resulting from the organisation. For example, the peak of activity occurred 
at the eighth or the fourth subtrial and not at the first or the fifth. While it could be 
argued that the activity at the fourth subtrial represents the control demands related 
to switching across the chunked episodes (which also manifests in increased RT at 
subtrial 5), it does not explain why the activity in many of such regions is equally high 
(if not higher) at subtrial 8. Even more difficult to reconcile is the finding that many 
of these regions also show an increase in activity between the second and third 
phases of these chunks, and in some cases, across the two chunks. 
The findings most inexplicable from this perspective are those of trial type 3. 
Recall that trial type 3 was different from others since it had no higher-level 
representations or list of task elements to organise. Subjects just faced a series of 
eight subtrials and were told on each of them what to do. However, given that it had 
the same eight subtrials as components, the bias of representing them in groups of 
four acquired in other trial types, plausibly, led subjects to perceive these trials as 
having two episodes of four subtrials each. They showed two effects that were also 
seen in trial type 2 – increased activity at the fourth subtrial and an increase in 
162 
activity between phases 2 and 3 of the two chunks. Even more interesting is the fact 
that these effects were most manifest in IFS, AI/FO, IPS, pre-SMA - regions that are 
mostly associated with representing more concrete task related information 
(Woolgar et al, 2011; Sigala, 2008; Badre, 2009), unlike more anterior regions that 
have usually been implicated for abstract representations (Christoff et al., 2009). 
 
 
 
 
 
 
 
 
 
 
 
 
163 
CHAPTER 5 
Representation of Task Information across the 
Sequential Phases of the Task Episode 
 
5.1 Introduction 
Previous chapters showed that the neural activity elicited by a task element 
depends upon the position of that element in the overall task organization. In this 
chapter, the influence of task organization on the pattern of elicited activity across 
voxels of the region will be investigated, which, arguably, can be considered as a 
better reflection of actual representation of the task element. 
  Conventional fMRI methodologies look at the measures of the magnitude of 
activity at individual voxels; these are averaged across all the voxels of the region to 
conclude if it is sensitive to the particular mental/behavioural event. While this 
allows one to conclude if the region is involved in the processing of the mental event, 
it gives relatively little clue about the actual representational content of the region 
during the processing of that mental event. This limits the epistemic value of such 
methodologies since any mental event has many cognitive concomitants, and 
conventional analyses cannot tell which one of these was actually represented by the 
neurons of the region in question. For example, a particular task event may be lead 
to a switch in rule, a switch in response, completion of a task episode, completion of 
a goal and so forth. Univariate analysis can tell if the task event, as a whole, elicited 
activity in the region of interest, but not which of the associated cognitive event was 
actually represented in that region. 
 Methodologies based on the analysis of voxel patterns offer relative 
advantages in this regard. Neural representations in a variety of studied domains are 
164 
based on the activity of a population of neurons (Georgopoulos et al., 1986; Young et 
al., 1992; Averbeck, et al., 2006). Given the relative size of brain areas with respect to 
the voxel size used in fMRI studies, in most cases the relevant population of neurons 
would be distributed over a number of voxels. Hence qualitatively different neural 
population codes could elicit qualitatively different patterns of activity across the 
voxels of the region. In other words, qualitatively distinct representations in a region 
will elicit qualitatively distinct patterns across voxels.  
This allows for greater inferential judgements about the kind of information 
represented in the region of interest. Consider the previous example of the task 
event with many cognitive accompaniments; if the region represents relevant task 
rules, distinct rules will have different population codes and hence will elicit 
qualitatively distinct patterns of voxel activity. On the other hand, if the region does 
not have information about the response made, different responses will not elicit 
different patterns of activity. Thus, the degree of distinction between the patterns 
elicited by the exemplars of a cognitive category can be a measure of the 
representation of that category. This distinction can be measured in terms of the 
accuracy with which the patterns in question can be differentiated by learning 
algorithms that have been trained to distinguish such patterns on a separate but 
representative part of the data (Norman et al., 2006), or in the measures of 
dissimilarity between the different patterns like 1-correlation, Euclidean distance and 
Mahalanobis distance (Kriegeskorte et al., 2008).  
 Earlier studies using pattern analysis methods have shown that the fronto-
parietal regions represent working memory contents (Harrison et al., 2010), decision 
results (Haynes et al., 2006), planned action (Haynes et al., 2008), task rules (Woolgar 
et al., 2011), contents of imagery (Stokes et al., 2009) and so forth. Indeed these 
regions seem to represent any information that is relevant to the task at hand 
(Duncan, 2010). More importantly, the magnitude of such representations increases 
165 
with increase in cognitive control and attentional resources invested (Woolgar et al., 
in press). It was seen earlier (chapter 3) that the effect of phase on the overall 
activity of these regions was similar to the effect of cognitive control. This raises the 
question whether the effect of phase on the representation of task elements will also 
be similar.  
 In this chapter, different aspects of the effect of task organization on the 
representation of the individual task elements will be discussed. The first issue 
pertains to the effect of phase on the representation of the task elements. Recall 
that the task elements representing phases closer to completion elicited greater 
activity in many regions. However, the conclusion that the increase in activity was 
due to the phase per se was confounded. It is possible that the increase in activity 
was the result of some non-specific modulatory effect resulting from increasing 
expectation of the imminent goal completion, or the increase was unrelated to the 
task event and was the result of disinhibition due to the accompanying decrease in 
cognitive load, akin to default mode activity. However, in both of these possibilities, 
the actual representation of the task elements would not be affected, since these 
predict a mass effect in terms of an increase in activity of the whole population of 
neurons, which would not affect the informational content of the task elements’ 
representations. Thus these possibilities would get ruled out if regions showing 
increased activity were also shown to have greater information about the later 
compared to the earlier phases. 
 In organized tasks, the information associated with the constituent task 
elements can be considered to belong to two distinct categories. The first category 
pertains to the information that is associated with a task element by virtue of its 
position in the task plan, e.g. its being the penultimate step of the task plan or it 
completing the task. For example, in the series of actions that constitutes the steps 
of a hypothetical breakfast: boil water, add coffee, add cream, toast bread, and 
166 
spread butter; ‘spread butter’ represents the last step of the task of preparing 
breakfast and ‘toast bread’ represents the penultimate step. Information of this kind 
associated with the event constitutes the first category of information. Earlier 
chapters have shown that task events that differ in this category also differ in elicited 
neural activity. 
  Task elements also have another kind of information associated with them 
which pertains to the information about how they are to be executed, e.g. in case of 
‘spread butter’ how the butter is to be spread, the rules relevant to it and so forth. 
This constitutes the second category of information. It is an open question if the 
representation of both of the above kinds of information is affected by the position 
of the task element in the plan of the task episode.  
Another issue pertains to the change in the representation of a cognitive 
event by the very fact of its being a part of an overarching task compared to the case 
when the cognitive event is assessed as a task in itself. For example, consider the 
series of actions: boil water, add coffee, add cream, toast bread, and spread butter. 
Each of the individual steps can be conceived as a task in itself (e.g. ‘boil water’) or as 
one of the steps in the overall task of preparing breakfast. Will the representation of 
these events be different in these two cases? 
 Varying predictions can be made in this regard. The representation of a task 
event as part of an overall task means that its representation would be ensconced 
under higher level representations organizing more extended task episodes. If the 
same limited capacity neurocognitive reserves are required for representing both the 
higher and the lower level representations, it can be predicted that less resources 
would be available for representing lower level task elements, and hence, the 
representation of these would decrease when they are part of a strongly organized 
hierarchical task episode, compared to the situation when they are weakly 
organized.  
167 
 From a different perspective, the individual task elements of an organized task 
episode may require more distinct representation since the individual steps now 
have to be more thoroughly distinguished from other steps. Indeed 
electrophysiological studies have shown that successive phases of behaviour are 
represented by very distinct neural states (Abeles et al., 1995; Sigala et al., 2008). 
Recording from a population of prefrontal neurons, Sigala et al. found that the neural 
activity across the sequential phases of a trial (cue – delay – target) was orthogonal. 
From this perspective it can be predicted that the representation of the task 
elements representing successive phases of a higher order task would be more 
distinct than successive task elements which do not represent successive phases of a 
higher order task.  
  The experimental design discussed in the last chapter allowed for the 
investigation of these issues. Recall that each trial consisted of eight subtrials which 
were identical in all aspects except for their sequential position. Accordingly, the 
distinctness of the pattern of activity elicited by subtrials at different position can be 
the measure of the representation of the phase information associated with the 
subtrial (category 1 information). Further, the subtrials at each position could be of 
one of the two types (referred to as A and B) depending upon the relevant rule that 
decided the correct response. The distinctness of the patterns associated with 
subtrials A and B at any position was therefore the measure of the representation of 
category 2 information, which was not related to the phase per se but was related to 
the execution of the individual subtrial. These measures of information could be 
compared across the eight positions to get the effect of position on the 
representation of information related to the task elements. If the increase in activity 
across the phases of the task plan, seen in earlier chapters, were accompanied by an 
increase in the representation of the corresponding task elements, the informational 
content in the relevant brain region can be expected to increase across the eight 
positions.  
168 
The different kinds of trial types that differed in organization, allowed for an 
investigation into the effect of organization on the representation of the individual 
subtrials. The organized trials were characterized by organizing (higher level) 
representations that sequenced the eight subtrials, while in the unorganized trials 
such higher level representations were absent and these subtrials were not explicitly 
organized. Note that the latter trials were not unorganized in the true sense of the 
term. The subjects did have an explicit knowledge that these trials consist of the 
eight subtrials. However, since there was no organizing representation and subjects’ 
knowledge of their position in the trial was less explicit, these trials can be said to 
have been less organized than the organized trials. Thus, the term ‘unorganized’ is 
used in a relative sense. Given the difference between these two kinds of trials, 
comparison of the representation of different aspects of the lower level task 
elements across these trial types would show the effect of organization on them. 
Note that even though the rule relevant on any subtrial was cued differently across 
the organized and the unorganized trials (identity of the letter in the memorized 
string sequence in the organized trials, colour of the stimulus margin in the 
unorganized trials), this was irrelevant, since the critical point is the distinctness of a 
subtrial compared to other subtrials of the same trial type - whatever the mode of 
conveying the relevant rule, it was the same for all the subtrials of a given trial type. 
The results, in summary, showed that the representation of information 
related to the subtrials increased across the eight positions in many brain regions. 
This increase was seen in both categories of information. Organization of the trial 
seemed to decrease the representation of individual subtrials, since the distinctness 
of pattern elicited by them was greater in trial type 3 than trial types 2 and 1. 
169 
5.2 Methods 
The details of experimental design and fMRI acquisition have been described 
in the previous chapter. The multivariate analysis described here is based on 
representational similarity analysis (described in Kriegeskorte et al., 2008). Two 
distinct general linear models were used to get the univariate estimate of activity in 
response to each event at each voxel.  
The first general linear model (GLM) was the same as the one used in the 
analysis described in the previous chapter. The individual subtrials were separately 
modelled according to the trial type they belonged to and the position they 
occupied. They were modelled with epoch regressors having a width equal to the 
duration of the subtrial (beginning from the moment the stimulus of the previous 
subtrial disappeared and ending at the moment when the stimulus of the current 
subtrial disappeared) convolved with the canonical hemodynamic basis function. For 
all subtrials, the common activity related to the appearance of a stimulus was 
modelled out using a common event regressor of no duration.   
In the second GLM, each subtrial was additionally characterised by the rule 
relevant to it beyond being characterised by their trial type and the sequential 
position they occupied. Other aspects of this GLM were the same as the previous 
one. In both cases, the first level analysis was done using unnormalised and 
unsmoothed pre-processed images.  
5.2.1 ROI analysis 
 These were done at the same ROIs that were used in the previous 
analysis – multiple demand regions, phase sensitive regions and two nodes from the 
default mode network. For all ROIs, a corresponding ROI in subjects’ native space 
was created. The parameter estimates (β values) of activity for the different subtrials 
were extracted for each voxel of the ROIs and a vector array of length equal to the 
170 
number of voxels was created. The individual cells of the array had a value equal to 
the β value of the corresponding voxel. These arrays then quantified the pattern of 
activity elicited by the various subtrials in that ROI. To measure the distinctness of 
activity pattern elicited by the subtrials, correlation distance (1-linear correlation 
between the corresponding arrays) was calculated across every pair of subtrials. 
Correlation distance was used as the measure of dissimilarity since it normalized for 
the mean level of activity, which, as seen in the last chapter, was variable across the 
subtrials. The dissimilarity values for all pairs of subtrials were assembled in a matrix 
(called representational dissimilarity matrix, RDM), the rows and columns of which 
represent the subtrials, and the value in the cell represented the dissimilarity 
between the corresponding subtrials (Fig. 5.1). The matrix was symmetric about the 
diagonal that represented the dissimilarity of a subtrial with itself and so had a value 
equal to zero. For visualisation, the dissimilarity values within each matrix were rank 
transformed and scaled between values of 0 and 1, and represented by a colour 
code. This was constructed for each ROI and for each subject and then subjected to 
further tests.  
The example in Fig. 5.1a depicts an ROI RDM for the eight subtrials of trial 
type 1. The diagonal represents the dissimilarity between each subtrial and itself 
which obviously would be zero. The first column represents the dissimilarities 
between the first and the other subtrials, the second column likewise has the 
dissimilarities between the second and the other subtrials. Since the matrix is 
symmetric across the diagonal, the eight rows represent the same dissimilarities as 
the eight columns. Such ROI RDMs allow for the visualisation of the pattern of 
dissimilarities. For example, it is apparent in this sample that the dissimilarities 
between subtrial 8 and the rest are higher than dissimilarities between earlier 
subtrials. This example also shows that the dissimilarity between a subtrial and the 
following one (i.e. between subtrials 2 and 3, 3 and 4, 4 and 5 and so forth) increases 
with the position of the subtrial  
171 
 
 
Figure 5.1. (a) A sample representation dissimilarity matrix (RDM) of left IPS from a subject 
representing the pattern of dissimilarities between the eight subtrials of a trial type. The rows 
and columns represent the eight subtrials and each cell represents the dissimilarity (1-
correlation) between the corresponding two subtrials. For each matrix, the values have been 
rank transformed and scaled between 0 and 1. The diagonal represents a vector of zero 
(represented by deep blue) corresponding to the dissimilarity between each subtrial and 
itself. The ROI RDM represents the actual pattern of dissimilarities extracted from an ROI. 
(b) This model RDM represents the predicted pattern of dissimilarity based on the 
hypothesis that the later subtrials were more dissimilar compared to the earlier ones i.e. the 
dissimilarities increase with the position of subtrials in the trial e.g. dissimilarity between 1 &2 
< 1 & 3 < 2& 3. (c) Model RDM (2) with dissimilarities increasing between subtrials 1 to 4 
and 5 to 8 in parallel. Note that the dissimilarities between the subtrials of 1st and 2nd
 
 chunk 
have been masked and excluded from the analysis.  
 Two methodologies were used for statistical analysis of the pattern 
within the ROI RDMs. For analysis involving populations of values that were not 
completely independent of each other or for testing a pattern that involved a broad 
aspect of the ROI RDM, model RDMs were constructed. They had the same structure 
as the ROI RDM that they were being used to test i.e. they had the same rows and 
columns. The values in the cells of the model RDMs were based on the idealised case 
of the hypothesized pattern. For example, consider the model RDM1 in Fig. 5.1b, that 
models the idealised case when the dissimilarity between the subtrials increases with 
the position of the subtrial, so the values in the rows and columns of this RDM 
increase from row 1 to row 8 and from column 1 to column 8. Since the rows and 
columns stand for the eight subtrials, this pattern means that in the idealised case of 
this hypothesis, the dissimilarity between subtrial 1 and others will increase with the 
172 
position of the subtrials i.e. greater dissimilarity between subtrials 1 and 8 compared 
to that between 1 and 7, which is in turn greater than 1 and 6, and so on; similarly 
the set of dissimilarity between subtrials 2 and others will be higher than that 
between subtrials 1 and others. Like the ROI RDM, the model RDM is also rank 
transformed and scaled between 0 and 1.  
The model RDM need not make a hypothesis about the entire extent of the 
ROI RDM, instead it can test for a pattern over a part of the ROI RDM. In such cases 
the model RDM is congruent only with the part of the ROI RDM that it seeks to test, 
while other parts are masked. For example, the model RDM in Fig 5.1c seeks to test 
the pattern in the top left and bottom right quadrant of the ROI RDM in Fig 5.1a.  
To measure the likeness between the patterns in the ROI and the model 
RDMs, Spearman’s rank correlation coefficient was obtained between the two. Note 
that Spearman’s rank correlation was used because a linear relationship cannot be 
assumed between the model and the ROI RDMs. To assess the statistical strength of 
this coefficient, the rows and columns of the ROI RDM were randomly reordered and 
then correlated with the model RDM. This step was repeated 105 times, to get a 
distribution of correlation coefficients expected if the two RDMs were unrelated. If 
the correlation between the actual ROI and model RDMs fell within the top 5% of the 
simulated null distributions, the null hypothesis that the matrices were unrelated 
was rejected.  
The second methodology was used for dissimilarity values that were 
independent of each other. In such cases the relevant correlation coefficients were 
extracted from the ROI RDMs, z-transformed (using Fisher’s z transform) and then 
subjected to further statistical parametric tests. 
 
173 
5.2.2 Whole Brain Analysis 
Whole brain searchlight analyses were done to look for other brain areas 
showing predicted patterns of representation. A spherical searchlight (9 mm radius) 
was sequentially moved across the entire set of brain voxels (brain images used were 
unsmoothed and unnormalised). At each step, the pattern of the activity across the 
voxels within the searchlight was computed for each event. The relation between the 
patterns elicited by the various subtrials was computed as described above for ROI 
analysis, and a searchlight RDM assembled. This RDM was then correlated with the 
model RDM. The resulting correlation coefficient was z-transformed. This value was 
assigned to the voxel that served as the centre of the searchlight. A brain map was 
thus obtained for each subject with voxel values equal to the z- transformed 
correlation coefficient between the model RDM and the searchlight RDM based on 
that voxel. These were then normalised into MNI space, smoothed and fed into a 
random effects analysis to look for the behaviour of brain regions across the subject 
population. The results were thresholded to correspond to an FDR correction of p < 
0.05. 
5.3 Results 
5.3.1 Effect on Category 1 information 
ROI analyses 
The representations of the eight subtrials making up a trial were compared. As 
described earlier, the dissimilarities between each subtrial with the other seven 
subtrials were assembled in RDMs. Fig. 5.2 shows such RDMs of the various ROIs for 
the three trial types, averaged across the subjects. It is apparent that across all three 
trial types, many ROIs show greater dissimilarities for the later subtrials with the 
eighth subtrial being the most dissimilar. This result thus parallels the earlier results 
of an increase in activity for the later subtrials, with the eighth eliciting the greatest 
174 
activity, and suggests that the increase in activity seen earlier was accompanied by 
an increase in representation of the information associated with the subtrials. 
 
 
175 
 
Figure 5.2. RDMs for the three trial types across various ROIs. In most ROIs, across the 
three trial types, the dissimilarities between the subtrials increased with the position of the 
subtrial in the trial, showing that the distinctness of the representation of latter subtrials was 
greater than earlier ones across different ROIs and trial types. APFC – Anterior prefrontal 
cortex, AI/FO – Anterior insula extending into frontal operculum, IFS – Inferior frontal sulcus, 
IPS – Intra parietal sulcus, ACC- Anterior cingulate cortex, pre-SMA – Pre-Supplementary 
motor area, vFr- Ventrolateral frontal cortex, MF – Superior Medial prefrontal, pCng – 
Posterior cingulated, TPJ – Temporoparietal junction. 
 
  To confirm this trend, these RDMs were correlated with a model RDM (model 
1, Fig. 5.1b) that predicted a linear increase in dissimilarities between the subtrials 
across the length of the trial. In this model RDM as described earlier (see methods) 
the dissimilarity between any two subtrials increased linearly with their position in 
the trial. The pattern of values in this RDM was one that would be expected if the 
dissimilarity between subtrials was solely determined by their position in the trial.  
The model considered the entire trial to be cohered as one plan, and predicted that 
the chunk level organisation of the subtrials had no effect on their representation.  
176 
ROI 
  
Model 1 Model 2 
Trial  type 1 Trial type  2 Trial  type 3 Trial type1 Trial type 2 Trial type 3 
APFC-L 0.05 0.36 0.20 0.37* 0.20 0.02 
APFC-R 0.46* 0.4 0.37 0.5* 0.27 0.02 
AI/FO-L 0.48* 0.40* 0.42 0.47* 0.27 -0.03 
AI/FO-R 0.60** 0.52** 0.61*** 0.37 0.21 0.18 
IFS-L 0.74*** 0.55** 0.42 0.46 0.47* -0.14 
0.25 IFS-R 0.83*** 0.65*** 0.55** 0.4 0.27 
IPS-L 0.79*** 0.56** 0.62** 0.39 0.23 0.25 
IPS-R 0.68*** 0.55** 0.64** 0.4 0.01 0.31 
ACC 0.79*** 0.53* 0.42 0.44 0.23 0.17 
pre-SMA 0.81*** 0.61** 0.53** 0.46* 0.62* 0.24 
MF 0.84*** 0.71*** 0.55** 0.52* 0.43 0.26 
vFr-L 0.47* 0.43* 0.52** 0.43 0.28 0.10 
vFr-R  0.70*** 0.67*** 0.66*** 0.34 0.32 0.33 
pCng 0.63** 0.56** 0.70*** 0.45 0.17 0.43 
TPJ-L 0.59* 0.54* 0.60** 0.39 0.25 -0.05 
TPJ-R 0.73*** 0.62** 0.67*** 0.40 0.30 0.29 
 
Table 5.1. Spearman's correlation coefficients between the ROI and model RDMs 1(model 
1) and 2 (model 2). The coefficients for trial type 1 are the highest for both models. For all 
trial types model 1 is the better predictor. 
177 
Table 5.1 (model 1) shows the Spearman’s rank coefficients obtained at each 
ROI, across the three trial types. This model was well correlated with the pattern of 
dissimilarities from all trial types in all ROIs except APFC and left AI/FO. This shows 
that there indeed was a pattern of increasing dissimilarities across the length of the 
trial in all trial types.  In other words, there was a significant trend of increase in the 
informational content/representation of the subtrials with increase in their position.  
Notably, this trend was also present in trial type 3, even though it was not 
explicitly organised. Recall that this trial type had shown minimal evidence of 
increase in activity across its length, from which it was concluded that this trial type 
did not show the evidence of organisation that other trial types had shown. Current 
results show that the representation of subtrials in this trial type changed in manner 
similar to the trials that were more explicitly organised. Trial type 3 was organised in 
that it was known to the subjects that it consisted of eight subtrials, albeit this 
organisation was less explicit than other trial types. While such organisation did not 
affect the mean level of activity between subtrials 1 to 7, it did affect the information 
content in the pattern of voxels. (For direct comparison across trial types, see 
below). 
In earlier results the extent of increase in activity across the eight subtrials 
was affected by the organisation of the task episode; specifically, the evidence of 
such an increase was greater in trial type 1 than in 2. Evidence suggested the former 
to be organised more as a single task episode, while the latter seemed to be 
organised more as two task episodes. In the current results, although all trial types 
show evidence of increase in representation across the eight subtrials, this was 
strongest in trial type 1, the coefficients for which exceeded those of trial type 2 and 
3 in most ROIs. 
An additional result is the effect of laterality, with right hemisphere 
coefficients being generally higher in all three trial types.  
178 
While the above findings show parallels with increase in activity across the 
eight subtrials seen earlier, differences are to be noted. In the last chapter, ROIs like 
left IFS and IPS did not show any trend of increase in activity across the trial; however 
both regions were highly correlated with a trend towards increase in representation 
along the eight subtrials. In the same vein, as mentioned earlier, the third trial type 
had shown minimal trend of increase in activity across the subtrials, but show good 
evidence of increase in representation across the subtrials. These results suggest 
that, while increase in representation has a similar trend to increase in activity, the 
two are not identical. 
Next, the ROI RDMs were correlated with a second model RDM that predicted 
a different trend in the pattern of dissimilarities across the subtrials. This second 
model was based on another feature of the pattern of their elicited activities seen in 
the previous chapter. Recall that activity in some regions (especially. in trial type 2) 
increased in parallel across the subtrials one to four and five to eight. This was most 
prominent in trial type 2 in bilateral insula, IPS, left IFS and APFC, and pre-SMA.  
Since the pattern of monotonic increase of activity seen best across the eight 
subtrials of trial type 1 was strongly paralleled by a similar monotonic increase in 
representation, it is plausible to think that the biphasic increase in activity seen 
across ROIs in trial type 2 might also lead to a similar pattern of change in 
representation of the subtrials.  
To test this, a second model RDM (model RDM 2, Fig. 1c) was created. It 
predicted a stepwise increase in the dissimilarities of the subtrials one to four and 
five to eight, but the average representation of the subtrials composing the two 
chunks was identical, so the dissimilarities of subtrial 2 with subtrials 1, 3 and 4 was 
same as that of subtrial 6 with subtrials 5, 7 and 8, and similarly subtrial 3 vs 1, 2 and 
4  was the same as of 7 vs 5, 6 and 8; in other words, the pattern of dissimilarities 
across the two chunks were parallel. It was expected that this model RDM would be 
179 
more strongly correlated with trial type 2 than with trial type 1. Note that this model 
RDM did not make a prediction about the entire extent of the ROI RDM, since the 
hypothesis cannot predict what the dissimilarities between the subtrials of the two 
chunks would be, relative to the dissimilarities of the subtrials within a chunk. So the 
corresponding parts of ROI RDMs were masked prior to their correlation with this 
model RDM. 
The Spearman’s coefficients are listed in Table 1 (model 2). Note that these 
are lower compared to that for model 1 in most ROIs across the three trial types. In 
trial type 2, only pre-SMA and left IFS were significantly correlated with this model; 
interestingly, for this trial type, in none of the ROIs was this model a better predictor 
than model 1. However, in trial type 1 the bilateral APFC was better correlated with 
model 2 than model 1.  
Contrary to the expectation from the pattern of the level of activity across 
subtrials seen in the previous chapter, across most ROIs model 2 was a better 
predictor of pattern of dissimilarity in trial type 1 than in trial type 2 (pre-SMA being 
a notable exception). It is likely that this results from the fact that the two model 
RDMs were correlated with each other, and trial type 1 being highly correlated with 
model 1 was also better correlated with model 2.  This result raises a possibility that 
the pattern of change in representation of the subtrials was minimally affected by 
their organisation into distinct chunks.  
 
 
 
 
 
180 
Whole Brain Analyses 
  
Figure 5.3. Regions where the actual pattern of dissimilarities across the eight subtrials were 
highly correlated with those in model 1.  
 
To investigate other regions that showed similar effects of organisation on the 
representation of the subtrials, a whole brain searchlight looked at regions where the 
activity pattern across voxels was significantly correlated with the two models (see 
methods). Fig. 5.3 shows the whole brain render of the regions that were highly 
correlated with model 1, i.e. where the representation of subtrials increased across 
181 
the eight subtrials. Note that the results are most extensive for trial type 1 (Fig. 5.3a), 
and involve very widespread brain regions – bilateral lateral prefrontal, premotor, 
sensorimotor, and superior parietal along with medial frontal, and parietal regions, 
right middle and superior occipital, right middle and superior temporal gyri and right 
lateral cerebellar hemispheres. While these regions are largely coextensive with the 
regions that had shown increase in activity across the eight subtrials of trial type 1, 
the difference is nonetheless noteworthy (c.f. Fig 4.10). Left lateral prefrontal, parts 
of pre-SMA and cingulate and left sensorimotor regions had not shown an increase in 
activity across the subtrials of this trial type; however, all these regions show an 
increase in representation of the subtrials across the length of trial type 1. Anterior 
temporal regions and left TPJ on the other hand had shown increase in activity but 
do not show any evidence of increase in representation. 
The regions showing increase in the representation of the subtrials across trial 
type 2 (Fig. 5.3b) were far less extensive and were limited to the bilateral premotor 
cortex, right IFS, ventrolateral prefrontal cortex and APFC along with bilateral 
superior occipital cortex, middle cingulate, precuneus and cuneus.  
The results for trial type 3 were qualitatively most distinct (Fig. 5.3c). The 
region with the strongest correlation was the occipital lobe and adjoining parts of 
temporal and parietal cortices and curiously cerebellum. Parts of parietal (IPS, 
superior parietal lobule, precuneus), frontal (right premotor, IFJ, AI/FO, medial 
frontal) and right sensorimotor regions also showed significant results. 
The second model, on the other hand, did not yield results that were 
statistically significant at the whole brain level.  
 
 
182 
Comparison of Trial types 1 and 3 
Both trial types 1 and 3 consist of a single trial episode having a sequence of 
eight subtrials. However, trial type 1 had explicit higher level representations 
organising the lower level elements (subtrials). A comparison of the representations 
of the subtrials in these two trial types would therefore show the effect of the 
presence of higher level organising representations on the representation of lower 
level elements. In this regard, earlier results (Table 5.1) had shown that the evidence 
of increase in the representation of the subtrials with the progression of trial was 
stronger in trial type 1. However that analysis does not tell anything about the actual 
level of representation of the subtrials in these two trial types. In the analysis 
described in this section the two trial types were compared directly in a single model. 
183 
 
Figure 5.4. ROI RDMs with the dissimilarities between subtrials of trial type 1 and 3. The first 
8 rows and columns represent the eight subtrials of trial type 1, the next eight represent 
those of trial type 3. Note that the dissimilarities between the subtrials of trial type 1 (top left 
quadrant) are generally lower than the dissimilarities between the subtrials of trial type 3 
(bottom right quadrant) 
 
For each ROI the various dissimilarities were assembled into a 16 x 16 matrix, 
the first eight rows and columns of which represented the eight subtrials of trial type 
1, and the next eight rows and columns represented those of trial type 3 (Fig. 5.4). 
184 
The top left quadrant of this matrix thus contained the dissimilarities between the 
subtrials of trial type 1, while the bottom right quadrant contained the dissimilarities 
between the subtrials of trial type 3. It is visibly evident that the top left quadrant of 
many of these ROI RDMs is ‘bluer’ than the bottom right quadrant, suggesting that 
the dissimilarities in between subtrials of trial type 1 were in general lower than 
those between subtrials of trial type 3. 
 
Figure 5.5. Model RDMs for testing the various kinds of relation between the dissimilarities 
of trial type 1 and trial type 3. (a) to (c) Three models predicting dissimilarity in trial type 3 to 
be greater than those in trial type 1. (d) to (e) predict the reverse.   
 
Seven possibilities were tested using respective model RDMs  (Fig. 5.5). Note 
that all of these models made predictions only about the dissimilarities in between 
the subtrials of trial type 1 (top left quadrant of ROI RDMs) and those in between the 
subtrials of trial type 3 (bottom right quadrant of ROI RDMs). The other quadrants 
were masked from analysis.  
185 
Three of these models predicted the dissimilarities in between subtrials of 
trial type 3 to be greater than those between subtrials in trial type 1 (Fig. 5.5a to 
5.5c): 
Model 1: In this model RDM (Fig. 5.5a), the lowest dissimilarities i.e. between the 
earliest subtrials were same across the two trial types, but the dissimilarities 
involving the later subtrials were greater for trial type 3. This model therefore also 
predicted that the rate of increase in dissimilarities with their position is greater in 
trial type 3. 
Model 2: In this model RDM (Fig. 5.5b), the dissimilarities between the last subtrials 
were similar for the two trial types, but the dissimilarities between earlier subtrials 
were greater for trial type 3. Hence, as per this model, although the absolute level of 
dissimilarities was greater in trial type 3, the rate of increase was greater in trial type 
1. 
Model 3:  All dissimilarities in trial type 3 were evenly greater than those in trial type 
1 (Fig. 5.5c). 
 The next three models (4, 5 and 6) were analogous to the above but predicted 
the level of dissimilarities to be greater in case of trial type 1 (Fig. 5.5d to 5.5f). 
Model 7 (not shown) predicted the dissimilarities to be equal in the two trial types.  
 
186 
 
Figure 5.6. Cross correlation matrix depicting correlation coefficients (not dissimilarities) 
between the various ROI RDMs shown in fig. 5.4 and the model RDMs in fig. 5.5. Most ROI 
RDMs are intensely correlated with each other. The model RDMs predicting greater 
dissimilarities in trial type 3 (m1 to m3) are better correlated with ROI RDMs than those 
predicting the reverse to be the case (m4 to m6). For most ROIs, m2 and m3 are better 
correlated with ROI RDMs than m7 which predicted the dissimilarities to be identical in the 
two trial types. 
 
 Fig. 5.6 shows a cross correlation matrix showing correlation coefficients 
between the various ROI RDMs and the model RDMs (marked as m1 to m7). The first 
sixteen rows and columns represent the sixteen ROIs, the following seven represent 
the seven model RDMs. As is apparent the three model RDMs predicting greater 
dissimilarities between subtrials of trial type 3 (models 1 to 3) were far better 
187 
correlated with the ROI RDMs than those predicting the reverse (models 4 to 6). In 
most ROIs the models 1 to 3, especially models 2 and 3, were better correlated with 
ROI RDMs than model 7 that predicted the dissimilarities to be the same in the two 
trial types. These results suggest that the dissimilarities were higher in trial type 3 
than in trial type 1, and show that the presence of higher level representations may 
reduce the distinctness of the representation of the lower level elements. Also note 
that the ROI RDMs are strongly correlated amongst themselves suggesting that the 
pattern of dissimilarities was very similar across them.  
Results till now  
Patterns elicited by task events representing the later phases are more 
distinct than those in earlier phases, and task plans organised as a unified task 
episode by explicit higher level representations show greater and anatomically more 
widespread increase in the distinctness of the representation of task elements. While 
such task plans show greater increase, the actual magnitude of the distinctness of 
task elements in them is lower compared to those plans that do not have higher level 
organising representations, suggesting that the organising higher level 
representations decrease the distinctness of individual lower level task elements. 
Two possibilities exist – either the organising representation binds the lower level 
representations into a common representational code which results in them being 
less distinct, or the presence of higher level representations leaves little 
representational space in the limited capacity neurocognitive reserves for the lower 
level task elements. This latter possibility could potentially also explain the increase 
seen in the distinctness of the later task elements, since as greater portions of higher 
level representations are discarded, more space becomes available for the 
representation of lower task elements. However, this thesis does not explain the 
increase in the distinctness of lower level task elements seen across the phases of 
trial type 3. 
188 
 Analyses so far have looked into the changes in representation of task 
elements that represent the phase of the task (category 1 information). It is 
therefore not clear if the increase in representation also affects information 
contained in the subtrials that is unrelated to the phase of the task (category 2 
information). To look into this issue, the next set of analyses investigates change in 
the representation of the relevant rule across phases of the task. 
5.3.2 Effect on Category 2 information 
 At each subtrial one of the two rules was relevant (called ‘A’ and ‘B’, see 
methods); on this basis the subtrials at each of the eight positions could further be 
classified as subtrial A or B. The dissimilarity between the pattern of activity elicited 
across the voxels of an ROI by subtrials A and B at each of the eight positions was 
calculated for all three trial types, for all ROIs. These values were then z-transformed. 
Graphs in Fig. 5.7 shows the measures of dissimilarity between A and B at each of the 
eight positions of trial type 1. While there is a definite trend towards increase in 
dissimilarities between A and B across the eight position, the increase, in general, is 
not as smooth and monotonic as was the case with the level of activity elicited by the 
subtrial (Fig 4.7, chapter 4) and the dissimilarities in between the subtrials (Fig. 5.2).  
 
189 
 
Figure 5.7. Dissimilarities between subtrials A and B at the eight positions in trial type 1. In 
many ROIs, these were higher for the later positions. Asterisks marks the ROIs where the 
dissimilarities in the first four positions were significantly lower than those in last four (* : p < 
0.01; **: p < 0.001).  
 
To look for the effect of position on the dissimilarity between the two rules, 
the dissimilarities in the first four positions were compared with the second four in a 
repeated measures ANOVA. This was significant in medial ROIs (MF, pre-SMA, ACC), 
right vFr, and bilateral IFS (F(1,12)
 
>5.1, p<0.05). Recall that these ROIs were the ones 
that had shown the strongest evidence of increase in category 1 information (Table 
5.1), current result shows that these regions also show the greatest tendency for 
increase in category 2 information. 
190 
 
Figure 5.8. Dissimilarities between A and B across the eight positions in trial type 2.  
 
Similar analysis in trial type 2 (Fig 5.8) showed significant results in right IFS, 
AI/FO, vFr, left IPS and MF (F(1,12)>4.7, p<0.05). In contrast to these two trial types, in 
trial type 3 (Fig. 5.9) none of the ROIs showed significantly greater dissimilarities 
between A and B during the second half of the 
trial.
191 
  
Figure 5.9. Dissimilarities between A and B across the eight positions in trial type 3. 
 
 
 
 
 
 
 
 
 
192 
Whole brain analyses 
 
 
 
 
 
 
Figure 5.10. Model RDM used for whole brain searchlight for regions where the dissimilarity 
between subtrials A and B increased across the eight positions. The RDM assembled to 
correlate with such a model had the first eight rows and columns representing subtrials A 
across the eight positions, and the next eight rows and columns representing the subtrials B 
across the eight positions. 
 
 For each trial type, a whole brain searchlight was carried out to look for 
regions where the dissimilarities between rules A and B increased across the eight 
positions. At each step of the searchlight, an RDM (16 x 16) was created based on the 
pattern across voxels in the searchlight. The first eight rows and columns of this RDM 
belonged to the subtrials A across the eight positions, and the next eight rows and 
columns represented the subtrials B across the eight positions. The model RDM (Fig. 
5.10) used was congruent to this RDM. However, all positions in the model RDM 
were masked except the eighth diagonals representing the dissimilarities between 
subtrials A and B at the eight positions. The value in these cells increased linearly. 
This model RDM thus predicted that the dissimilarities between A and B increase 
across the eight positions. The model RDM was correlated with the searchlight RDM 
created at every step of the searchlight and the dissimilarity between the two was z-
transformed and assigned to the voxel forming the centre of the searchlight.  
 
193 
 
 
Figure 5.11. Regions showing significant increase in the dissimilarity between A and B 
across the eight positions in trial types 1 (a), 2 (b) and 3 (c). 
 
Figure 5.11 shows the regions where this model RDM was highly predictive of 
the pattern in the brain across the subject population. As was found in ROI analysis, 
the increase was most widespread in trial type 1 (Fig. 5.11a), where it involved the 
right AI/FO, vFr, pre-SMA, ACC and the precuneus, along with the right angular, 
posterior superior temporal gyrus, parts of right sensorimotor, superior occipital and 
194 
parieto-occipital cortices, and parts of the cuneus (coordinates in table A5.1). These 
results were right lateralised. 
Trial type 2 (Fig. 5.11b), on the other hand, showed significant results most 
strongly in medial parietal (precuneus and cuneus) and ACC. Loci of significant trend 
were also found bilaterally along the superior temporal gyrus and sulcus (coordinates 
in table A5.2). In the case of trial type 3 (Fig. 5.11c), the only region to show 
significant trend of increasing dissimilarity was the primary visual cortex along the 
calcarine sulcus.  
Comparison of trial types  
 Lastly, the level of dissimilarity between subtrials A and B was compared 
across the three trial types to see if the higher level organising representations 
affected the representation of the relevant rule. For the current analysis, the 
dissimilarities between subtrials A and B for the first four (1 to 4) and the second four 
(5 to 8) positions were averaged. These two values are referred to as the dissimilarity 
between the two relevant rules in the first and the second chunk respectively. Fig. 
5.12 shows these two values for the three trial types across the various ROIs, 
presented in lighter (first chunk) and darker shades (second chunk). 
195 
 
Figure 5.12. Comparison of dissimilarities between A and B averaged across positions 1 to 
4 (lighter shades) and 5 to 8 (darker shades) from the three trial types.  
 
 Note that in almost all ROIs the level of dissimilarities averaged across the two 
chunks is highest in trial type 3 followed by 2 and 1. The level of dissimilarity in the 
first and second chunk differs minimally in trial type 3, while in the other trial types 
the dissimilarity is higher in the second chunk in many ROIs. A repeated measures 
ANOVA looking for the effect of trial type was significant in all ROIs (F(2,24)>3.6, 
p<0.05) except for the left IFS. This finding parallels the earlier result (Fig. 5.7) of 
greater dissimilarities in trial type 3 than trial type 1, and adds that the organisation 
of the episode and the presence of higher level representation also affect the 
representation of rule information associated with the lower level task elements. 
196 
Further, as would be expected from earlier analysis, right AI/FO, IFS, vFr, left IPS, and 
MF showed a significant effect of chunk (F(1,12)>5.8, p<0.03). The interaction between 
trial type and chunk position reached significance only in MF and ACC (F(2,24)
As is apparent in Fig. 5.12, across all ROIs the average dissimilarities in trial 
type 2 were also greater than those in trial type 1. This was significant in right IFS and 
vFr, left IPS, pre-SMA and ACC (p < 0.02). Since both trial types 1 and 2 are organised 
(albeit differently), and differ in the cognitive load of the organising representations 
(8 letters vs. 4 letters), the above finding raises the possibility that the decrease in 
the distinctness of representations is a consequence of higher level representations 
occupying the limited capacity cognitive resources and leaving less neurocognitive 
representational space available for lower level task elements. This hypothesis can 
be tested. The first chunk had greater representational load in trial type 1, whereas 
the second chunk of the two trial types had similar loads. So if the decrease in 
distinctness is the consequence of load of higher level representations, the first 
chunks of the two trial types should differ, while the second chunks should not. 
>3.6, 
p<0.04).  
A repeated measures ANOVA between these trial types (1 and 2) with trial 
type and chunk as factors was carried out. The above hypothesis predicts a 
significant interaction between the effects of chunk and trial type. While the main 
effect of trial type was significant in right APFC, IFS, vFr, left IPS, ACC and pre-SMA 
(F(1,12) > 5.3, p < 0.02), and that of chunk in right IFS, vFr, left IPS, pre-SMA and MF 
(F(1,12) = 6.6, p < 0.02), none of the regions showed a significant interaction between 
the two (F(1,12) < 2.9, p > 0.1). A second ANOVA considered all ROIs in the same model 
along with trial type and chunk as factors. Again, the main effects of trial type (F(1,12) 
= 9.8, p < 0.01) and chunk (F(1,12) = 11.3, p < 0.01) were present. However, neither an 
interaction between trial type and chunk (F(1,12) = 0.2, p = 0.6), nor between trial 
type, ROIs and chunk (F(12,144) = 2.1, p = 0.3) reached significance.  
197 
5.4 Discussion 
The results described above show that the representation of task elements is 
affected by their organisation into coherent task episodes. Two aspects of this were 
seen. Firstly, the discriminability of information related to the component task 
elements increased as they got closer to the completion of their respective episodes. 
While this increase was best seen in trial type 1 which was organised strongly as a 
unified episode by higher level representations, other trial types that were organised 
differently – either as two distinct sub-episodes (trial type 2) or weakly organised 
(trial type 3) – also showed such increase. This finding thus links up with those of the 
previous chapters wherein many regions had shown an increase in activity for the 
later subtrials, and proves that this increase in fMRI activity is accompanied by an 
increase in the task element related information that can be derived from the region. 
It further makes it unlikely that the increase in activity seen earlier was the result of a 
non-specific disinhibition consequent to the accompanying decrease in cognitive load 
(Singh et al., 2008), or due to a non-specific increase in blood flow documented 
before expected events (Sirotin et al., 2009). 
The second aspect of the current results shows that the average level of 
distinctness of representation of the task elements is lower in organised trials with 
greater load of higher level representations compared to those less organised and 
having less organising representations. The distinctness of both categories of 
information was lowest in trial type 1 and highest in trial type 3.  
Many aspects of the current results and those of the previous chapter suggest 
that the effect of phase could be caused by the dynamics of the representational 
load of organised episodes. Since these representations form the control structures 
that control the execution of individual steps, as each step is executed, the 
representations and control machinery pertaining to that step can be discarded, 
freeing up the limited capacity neurocognitive resources that can then respond to 
198 
and represent information related to the subsequent new step. Likewise, such 
episodes with greater higher level representational load will have less 
representational space for lower level elements, resulting in lesser distinction 
between them.  
The amount of representations discarded is clearly maximal in trial type 1, 
hence the greater and more consistent increase seen in such trials. Similarly, the load 
of higher level representations was highest in this trial type, hence the lesser 
distinction between the task events in this trial. In trial type 3, the absence of 
organising representation meant that fewer representations were discarded at every 
step; hence less neural space got freed at every subsequent subtrial, therefore the 
smaller increase in activity across the trial.  
Not all aspects of results, however, are easily amenable to the above thesis. It 
is less clear, why there should be any increase in activity/distinctness across the 
subtrials of trial type 3. Another less clear finding is that in many ROIs the 
dissimilarities during the second chunk was greater in trial type 2 than in trial type 1 
(Fig. 5.12), in spite of the load of higher representations being identical in the two 
trial types at this stage. Although, in defence of this thesis, it can be argued that the 
representational load during the second chunk of trial type 2 is still less than in trial 
type 1. Even though there are equal number of letters maintained, these letters are a 
repeat in trial type 2, and hence their representation might be easier compared to 
the case in trial type 1 wherein the four letters are entirely new compared to the 
previous chunk. 
Widespread regions showing linear increase in the discriminability of subtrials 
(especially in trial type 1) is an intriguing result (Fig. 5.3a). If the above mentioned 
thesis has some truth, it will imply that very widespread regions are involved in 
representing task related representations. Previous analyses have attributed task 
related representation to relatively limited regions (Fox et al., 2005; Dosenbach et 
199 
al., 2006; Duncan, 2010). The current findings show that beyond such task positive 
regions, task related information is also present in the default mode network (medial 
prefrontal, posterior cingulate, precuneus, IPL, TPJ, STS), task-relevant sensory and 
motor regions (left somatosensory and motor, bilateral occipital) and task irrelevant 
sensory and motor regions (right somatosensory and motor, bilateral primary 
auditory) regions. Further, the pattern of change in the discriminability of such 
information is identical across these functionally varied regions.  
The evidence of linear increase was much limited in trial type 2 (Fig. 5.3b). 
Results of the previous chapter suggested that in many regions these trials were 
represented as two separate chunks. It is possible that the poorer linear increase in 
discriminability in such trials is related to their different pattern of organisation. In 
fact many of the regions that show increase in discriminability of the subtrials of trial 
type 2 in the current analysis had shown increase in activity across the two chunks in 
the previous chapter (Fig 4.16). 
Results of trial type 3 are least clear. The pattern of these results was very 
different from those of trial type 1. Intriguingly, these were centred on the occipital 
regions, although regions with significant effect could be found in frontal and 
temporal regions. Indeed peak of the effect was in the primary visual cortex, which, 
in case of trial type 1 had shown no effect in the whole brain analysis. This suggests a 
possibility that the brain regions involved in representing the organised and 
unorganised trial types were very different. 
Comparison of the two analyses  
The results described in the previous and the current chapter largely parallel 
each other, although interesting differences can be noted. The increase in activity 
across the length of the trial seen in the previous chapter is paralleled by the 
increase in representation seen in the current analyses. In both, the increase was 
200 
best seen in trial type 1 and relatively less in trial type 2. However, while none of the 
ROIs had shown an increase in activity across trial type 3, most of them showed an 
increase in the representation of category 1 information. Dissociations between the 
two can also be seen in trial type 1. A number of regions had not shown an increase 
in activity in the last chapter but do show an increase in information. For example, 
compare parts of left lateral PFC, superior parietal lobule, preSMA and ACC in Fig. 
5.3a and Fig. 4.10. It is possible that some of these dissociations could be related to 
the greater sensitivity of multivariate methods (c.f. Harrison and Tong, 2009).   
The results of these analyses have suggested new functional roles of many 
regions. Default mode network (medial prefrontal, posterior cingulated, precuneus, 
IPL and other areas around TPJ) is usually considered as a task negative region anti-
correlated with the task positive fronto-parietal regions (Fox et al, 2005, Uddin et al., 
2009). The current results differ from this view and show that in organised tasks the 
behaviour of the two can be similar. Indeed, as can be seen in Fig. 5.11, the similarity 
can go to the extent of both regions showing a parallel increase in the 
discriminability of the relevant rule. 
A related result comes from Rogers et al., (2010). They had subjects do a 
transitive inference task between two pairs of visual stimuli that were derived from a 
remembered implicit sequence or were independent. They found greater functional 
connectivity between regions of multiple demand and default mode networks when 
subjects had to access the underlying sequence to do the task, as if greater 
functional interaction is needed between these regions when accessing hierarchically 
organised representations! 
Some earlier studies have found correlated activity in the DMN and task 
positive regions. Christoff et al., (2009) found the two to be correlated during the 
episodes of mind wandering; Spreng et al., (2011) found this to be the case during 
autobiographical planning. In all such cases the increase in DMN activity was related 
201 
to the internal direction of mental focus, while the activity in the task positive 
regions was related to the control exerted on to this internally directed thought. The 
current results are remarkable in that the correlated activity between the two occur 
in the context of an external task. Additionally, the increase in the representation of 
task related information proves that this increase is not due to enhanced mind 
wandering or an internal shift in the focus of attention. 
The presence of task organisation related effect on neural activity and 
information in the primary sensory regions is another interesting finding. The old 
view that such regions process only the sensory information coming from external 
senses has in recent times seen revisions from findings that have shown the 
complexity of the functional properties of these regions.  A number of studies 
(reviewed in Driver and Noesselt, 2007) have shown that primary sensory regions can 
respond to a stimulus from a different modality. For example, the primary auditory 
cortex shows response to visual stimulation alone (Calvert and Campbell, 2003; 
Kayser el al., 2007), visual enhancement of auditory stimulation (Kayser et al., 2007; 
Martuzzi et al., 2007) and tactile enhancement of auditory stimulation (Kayser et al., 
2005). Primary sensory regions also show sensitivity to task related information, like 
the attentional context (Watanabe et al., 1998; Li et al., 2004), and the presence of 
reward (Shuler et al., 2006). In chapter 2, sensitivity of early visual cortex to the 
completion of task episode was shown. The current finding of the presence of 
information related to the task organisation in primary sensory regions, even those 
whose modality was completely irrelevant to the task (e.g. primary auditory cortex), 
is very unique. Further studies would perhaps clarify these issues. 
 
 
 
202 
CHAPTER 6 
Discussion 
 
The experiments described in this thesis suggest that the fronto-parietal 
representation of task events is dependent upon their position in the structure of the 
task episode. Experiments in chapter 2 showed that events completing defined task 
episodes elicited activity that depended upon the hierarchical level of the episode; 
for example, completion of tasks elicited greater activity than completion of 
subtasks. Later chapters showed that in addition to such end of episode activity, 
subsequent task phases elicit greater activity compared to the earlier phases. 
Further, such increase in activity was associated with increase in the content of 
discriminable information. Even more interesting has been the finding that such 
effects of episode structure were widespread across very diverse brain regions, 
which suggests that the information about the task structure is present in many 
regions. The current experiments, however, are insufficient to give a conclusive 
picture about the neural, cognitive and computational reasons behind these findings. 
This chapter will discuss the various issues and possibilities raised in the earlier 
chapters. 
6.1 What are Task Episodes? 
In much of the earlier discussions, the concept of what constitutes a task was 
not explicitly defined. It is important to mention that not all usages of ‘task episode’ 
would be synonymous with the sense in which this phrase has been used here. For 
example, the whole experimental session, the whole day (and by extension, all of 
life) could be regarded as one task episode. Clearly, the effects talked about in the 
current experiments do not apply to these; else the activity across sequential trials 
(and across the length of day or even across life) will continue to increase! 
203 
Task episode in the usage of this thesis refers to a continuous period of 
organised and sequential mental activity, elements of which have a means to end 
relationship and which proceeds towards a defined end. Another characteristic of 
such episodes is a continuous maintenance of higher level organising 
representations. The episodes end when such maintenance was no longer needed. 
Further, the episodes got parcelled into sub-episodes when only a part of these 
organising representations changed to another, while the rest remained constant 
across the sub-episodes.  
Recall that this was, indeed, the defining feature of all task episodes in the 
previous experiments. The trial in all of these was designed and instructed as a 
period of continuous and organised mental activity. Chunking of cued 
representations was used to parse this episode into sub-episodes in chapter 2 and 4, 
whereas in chapter 3, the trial was inherently designed as two blocks of tasks.  
This brings to light another way of conceptualising the current findings. While 
numerous studies have shown fronto-parietal representation of task-relevant stimuli 
and responses or information linking them like rules and decisions (reviewed in 
Duncan, 2010), the current experiments suggest that representations underlying the 
episode, in the context of which other task-relevant events take place, are also 
present in these regions. 
6.2 End of Task Episode  
 All experiments showed maximum and most widespread activity at the end of 
the task episode. This involved widespread cortical and subcortical regions and, 
interestingly, was right lateralised in the prefrontal cortex. In this regard, the results 
of the first set of experiments (described in chapter 2) were somewhat different 
from those of the last experiment (chapter 4 and 5). In the former, the phasic 
increase in activity at the completion of the task episode was largely centred on the 
204 
multiple demand regions, with additional involvement of other regions like 
precuneus, cuneus and early visual areas seen and when the entire visual search 
episode ended. The last experiment, looking at the final phase of the episode, found 
much more widespread activity which in addition to multiple demand regions also 
involved default mode and sensory and motor regions.  
 A possibility is that this difference stems from their different designs, as a 
result of which these experiments looked at slightly different qualitative issues. The 
first set of experiments looked at the transient increase in activity at the completion 
of the episode, while in the latter experiments the estimates of activity came from 
the entire duration of the last step. While, it seems plausible that the regions 
showing maximum activity at the last step will include those that show an increase at 
the completion of the episode, they will also include other regions such as those that 
show sequential increase across the various phases of the task episode. 
 Another possible source of difference could be that these experiments had 
different kinds of ends. Recall that the final target detections in the first set of 
experiments ended the visual search, but not the entire trial, which happened only 
after the probe was responded to. On the other hand, the final step of the trial in 
chapter 4 marked the completion of the entire trial, after which subjects could rest. 
It is possible that with respect to the structure of their respective trial episodes, the 
end investigated in chapter 2 completed a hierarchically lower episode than was the 
case with the end investigated in chapter 4.  
Findings do suggest that different kinds of ends could be represented 
differently. Consider, in the case of trial type 2 of chapter 4, the disparity between 
the activities at subtrial 4, the final step of the lower level episode, and subtrial 8, the 
final step of the entire episode (Figs. 4.12 & 4.13). Interestingly, regions showing 
greater activity for the fourth subtrial were largely centred on the multiple demand 
regions, and involved relatively fewer areas beyond them (Fig. 4.6). In fact the 
205 
pattern was close to that seen at the completion of task episode in chapter 2 (Fig. 
2.6). On the contrary regions where subtrial 8 elicited greater activity were very 
different (Fig. 4.13) and closer in pattern to those seen at the final phase of other 
task episodes like trial type 1 (Fig. 4.9), and the second task block in chapter 3 (Fig. 
3.9). 
 What causes the end of episode activity? Perhaps, the explanation might be 
different for the two kinds activities found at the end of task episodes. Some regions 
in certain task episodes showed increased activity only at the end; while in others 
this increase occurred in the context of a sequential increase in activity across the 
entire length of the episode. The former is dealt with here, while the latter would be 
discussed in detail in a subsequent section. While the current findings cannot specify 
amongst the various plausible reasons discussed below, it is also possible that 
different combinations of the following reasons are causative in different regions.  
Change in the Organising Representations 
 It was discussed in chapter 2, that the greater increase in activity at the third 
target detection of experiment 1 was best explained by the fact that the first three 
target detection events were implicitly thought of by the subjects as belonging to the 
same task episode, while the fourth was considered a separate task; this meant that 
the representations organising the first episode terminated at the third target 
detection, which elicited the increase in activity. This explanation could apply to the 
ending of all kinds of task episodes. Further, as the case of chapter 2 exemplifies, the 
representations in question need not even be explicit.  
 Similar conclusions can be made from the analysis of trial type 3 in chapter 4. 
Recall that in spite of this trial type not having any explicit letter string organising its 
execution, the fourth subtrial in such trials elicited greater activity in a number of 
regions. Here again the cause was likely to be the change in subtle representations 
206 
that implicitly organised the first and the second group of four subtrials as separate 
episodes.  
However, also note that the actual brain regions involved in these two cases 
were different. In the case of experiment 1 of chapter 2, the dominant region was 
right anterior prefrontal cortex, whereas in trial type 3 of chapter 4, left posterior 
prefrontal regions (around inferior frontal junction, Brass et al., 2005) showed the 
maximal effect. This suggests that the pattern of activity related to the change in 
organising representations is likely to depend upon the details of the task episode 
since the actual brain representation of these representations can be different across 
different kinds of episodes. 
Salience  
 Intuitively, the ends of episodes are more salient mental events compared to 
events within the episode. It is possible that some regions respond to this salience. 
Indeed many brain regions are known to respond to various kinds of salient mental 
events (Clark et al., 2000; Downar et al., 2000; Steven et al., 2000; Schall et al., 2003; 
Seelay et al., 2007). Across various studies this has included multiple demand 
regions, ventrolateral prefrontal cortex, medial parietal regions, TPJ, and non cortical 
regions like striatum and cerebellum. 
Other studies have looked at issues that can be conceptualised as salient 
boundaries of mental episodes. Zacks et al. (2001) investigated the phenomenon of 
event segmentation; they had subjects passively view movies of some episode of 
purposive behaviour like making a bed; after passive viewing, each subject 
segmented these videos into episodes of coherent events by tapping a button to 
mark what they considered were the points where coherent segments began and 
where they ended. The authors then applied these segment boundaries to the 
corresponding moments in the passive viewing fMRI data, to identify transient 
207 
activity time locked to these boundaries in the passive viewing condition. They found 
that these boundaries elicited activity in a distributed set of regions – posterior 
prefrontal, IPS, precuneus, STS and lateral occipital cortices; recall that this pattern of 
activity is nearly identical to that found in the first set of experiments during the 
early target detections. Clearly, such event segmentation would be hierarchical; 
however, the authors did not report the variation of boundary activity with the 
hierarchy of the episode. 
 In another study, Sridharan et al. (2008) had subjects listen to musical pieces 
that had transitions between movements of symphony works. They found activity in 
multiple demand regions along with deactivation in the DMN regions at such 
transitions; a pattern very reminiscent of the results obtained in chapter 2. 
Additionally, Granger causality analysis showed that some of regions from the 
multiple demand group (ACC and AI/FO) played a critical and causal role in switching 
between the multiple demand and the default mode pattern of activity.  
Default Mode Regions   
 A set of brain regions are known to decrease their activity during task blocks – 
medial prefrontal and medial parietal regions, TPJ, STS (Buckner et al., 2008); the end 
of such blocks would accordingly lead to an increase in activity in these regions. 
Some of these regions were indeed present in the current results. However, this 
explanation has to contend with the fact that this increased activity in default mode 
regions occurs along with increased activity in multiple demand regions. In the 
current understanding, what activates one deactivates the other. For example, if 
change in the organising representations and the accompanying salience of the 
mental event are the elicitor of multiple demand activity, these are also supposed to 
be the deactivators of default mode activity. Indeed, the default mode explanation 
for the increased activity at the end of the task episode presumes that these regions 
were deactivated earlier during the task blocks for precisely these reasons. A 
208 
potential, but currently unverifiable explanation could be that the default mode 
regions do transiently deactivate at the end of task episodes in response to events 
that cause activity in the multiple demand regions. However this transient 
deactivation in DMN is so transient that the fMRI estimate of activity is dominated by 
their succeeding sustained reactivation. 
6.3 Phase Effect 
 It is less clear why the activity in a number of regions should increase across 
the steps of the task. As discussed in the earlier chapters, none of the currently 
plausible explanations can account for the results completely. However, across the 
three sets of analyses on this issue (chapters 3, 4 and 5), the hypothesis that seems 
to account for most findings is that the decrease in some form of representational 
load with the progression of the task is responsible for the increase in activity. This 
hypothesis will now be explained in greater detail. 
 One of the foundational claims of the cognitive paradigm in psychology is that 
sequential tasks require representations at multiple levels (Lashley, 1951; Miller et 
al., 1960) – representation of individual behavioural/mental events making up the 
sequence at the lower level, which are organised by subsuming representations at 
higher levels. While the representations at the various levels can be conceptualised 
as separate, evidence suggests that the higher-level representations could be made 
up from the elements of lower level representations. For example, executing events 
marking the beginning of task sequence always takes much longer than executing 
identical events within the sequence (Allport and Wylie, 2000; Schneider and Logan, 
2006). Conceivably, this is due to the extra requirement of constructing the higher-
level representation at the beginning. However, across multiple domains it has been 
found that this increase at the beginning is dependent on the nature of the 
sequence, i.e. it takes longer to start executing more complicated sequences.  
209 
Schneider and Logan (2006) found that during the execution of a list of tasks, 
the reaction time at the first step is always the highest and is directly related to the 
complexity of the list. Across many studies looking at simple movements (like finger 
tapping), the reaction time is always the highest at the beginning of the chunk (Povel 
and Collard, 1982; Restle and Burnside, 1972; Rosenbaum et al., 1983). In the same 
vein, simple reaction time initiating a motor action tends to depend upon the 
complexity of the motor action (Henry and Rogers, 1960) and it takes longer to start 
speaking a word that has many syllables than a monosyllabic word, the effect being 
present even when pictures are being named (Klapp et al., 1973). Similar findings 
come from memory retrieval studies. Anderson and Matessa (1997; see also 
Anderson et al., 1998) found that when subjects memorise digit sequences that were 
grouped into chunks during presentation, the pattern of RT during recall showed 
prolongation at the start of the chunk, the magnitude of which depended upon the 
length of the sequence. 
Common across the findings reviewed above is the fact that complexity of the 
future sequence determines their initiation times, which suggests that planning the 
sequence may require access to, and representation of, some information related to 
each element of the sequence. Plausibly, these form the scaffold for the higher-level 
organising representation that will organise and control the execution of the 
episode. In other words the organising representations controlling the execution of 
the sequence could be built from the information related to individual steps of the 
sequence.  
The analogy of this view with the ‘memory drum’ theory of action of Henry 
and Rogers (1960), whose finding was mentioned above, is noteworthy for the 
current discussion. As per the their view, the information for performing the entire 
motor action is constructed and stored in the form of motor memory, and the time 
to initiate the action is the time taken to transmit this motor memory to the 
210 
effectors. More complicated actions require more complicated memory structure, 
which takes longer to be created and transmitted to the effectors.  
Given that across a variety of sequential behaviours – whether it is executing a 
sequence of motor acts or a sequence of abstract tasks or recalling a list of 
memorised items – the profile of higher effort and reaction time at the beginning of 
the sequence is common, it is suggestive that all sequential behaviours require 
construction of control structures that are based on representations pertaining to 
the entire sequence. The tasks used in the current experiments required sequencing, 
not of simple motor acts or stimuli, but of more abstract cognitive sets. It is likely 
that the beginning of such episodes were associated with the construction of a 
structure (or a plan) built on the representations from the entire sequence of 
cognitive sets. While the neural regions representing this structure are not known, it 
is likely that limited capacity neurocognitive resources that are thought to be 
engaged by other task control structures (Pashler, 1999), will be involved.  
What would be the dynamics of this structure across the task episode? It can 
again be speculated that as individual steps are executed, the representations 
pertaining to them can be removed from the main control structure. Consequently, 
the representational load of this structure would decrease as more and more steps 
of the sequence are executed, making this structure lighter with each step, and 
hence freeing up more and more of the limited capacity neurocognitive reserves. 
Hence, more of such resources would be available to respond to and represent the 
events of each step. 
Most of the current findings can be accounted for by this explanation. Not 
only did the sequential tasks show increase in activity across the sequential steps, 
the evidence for increase was greater in tasks where greater representational load 
would be discarded at each step. Hence in chapter 4, the increase was greater in 
trials with explicit organising representations, trial types 1 and 2, than in trial type 3 
211 
that had no such explicit representations. Task steps separated by a greater number 
of steps showed greater disparity in activity; in chapter 3, the two task blocks were 
consecutive, hence the disparity in activity between them (Fig. 3.9) is far less than 
between subtrials 2 and 8 of trial type 1 in chapter 4 (Fig. 4.7). Finally, the average 
estimates of representation of the individual task elements were lower on trials with 
greater load of organising representations. Thus in chapter 4, the average 
discriminability of information pertaining to a subtrial varied across the three trial 
types in inverse relation to the associated representational load: 
trial type 3 > trial type 2 > trial type 1 
Note that the term 'representational load' associated with the structure of a 
task episode in the current usage is not synonymous with the conception of working 
memory load (Baddley, 2007). As discussed in the earlier chapters, the dynamics of 
the phase effect cannot entirely be explained by the dynamics of the associated 
working memory load. A consideration of trial type 2 (chapter 4) makes this point 
clear. Recall that it had a single four letter string repeated twice, and the 
accompanying profile of behaviour showed increased RT at the fifth step consonant 
with chunking of the trial episode into groups of four subtrials. The activity in a 
number of multiple demand regions increased across the first four subtrials, but after 
the fourth subtrial showed a large decrease at the fifth subtrial and then increased 
again sequentially till the last subtrial, i.e. the pattern of activity in a number of 
regions was of a parallel increase between subtrials 1 to 4 and 5 to 8. Note that 
subjects had only four letters to remember and none of them could be forgotten 
across the first four subtrials since they were to be recalled again over the next four 
subtrials. Thus, clearly the working memory load stayed constant throughout the first 
four subtrials. Nonetheless, the activity in many multiple demand regions was seen 
to decrease across these four subtrials. The case of trial type 3 in chapter 4 further 
supports this notion. Recall that there was no working memory load across this trial 
212 
type, nonetheless some fronto-parietal regions showed parallel increase in activity 
across the subtrials 1 to 4 and 5 to 8, as was the case in trial type 2.  
While the dynamics of working memory load cannot explain these, the 
dynamics of the representational load can. In both of these trial types the subtrials 
were chunked in groups of four. Hence, the first four and the second four subtrials 
were likely to be represented as two distinct task episodes. It is plausible that the 
cognitive structures necessary for the execution of the first four subtrials were 
constructed at the beginning of the trial, in which case the representational load 
associated with it can be expected to be greatest at the first subtrial and would 
decrease stepwise after the execution of each subtrial as the structure relevant for 
each subtrial is disassembled. This would free up the fronto-parietal resources 
involved, might manifest as an increased disinhibition of these regions. After the 
fourth subtrial, a new cognitive structure for the execution of the second task 
episode would be constructed, the increased load of which manifested in the 
decrease in fronto-parietal activity seen after the fourth but before the fifth subtrial. 
Subsequently, each subtrial executed again freed up the corresponding fronto-
parietal resources, seen as increased disinhibition of these regions till the last 
subtrial.  
 Another aspect of the results that shows a difference between the 
representational load of episode structure and the working memory load is that 
increasing working memory load is known to increase the activity in multiple demand 
regions (Duncan, 2006), an effect that was replicated in our results as well. For 
example, in chapter 3, increasing the working memory load by making subjects 
maintain additional information pertaining to the second block, increased the activity 
during the first block. Similarly in chapter 4, cues requiring greater information to be 
maintained increased the activity in multiple demand regions. On the contrary, 
increased representational load decreases the activity in these regions. 
213 
Apart from the direction of neural modulation, the two kinds of loads also 
differ in the extent of brain regions affected. Working memory load in the current 
and earlier studies (Cabeza and Nyberg, 2000; Duncan, 2006) elicits activity that is 
mostly limited to the multiple demand regions of fronto-parietal cortices. However, 
the regions showing change in activity across the phases of sequential tasks and 
hence across the levels of representational load are much more extensive and 
included almost all major brain regions.  
 The above thesis suggests a distinction between two kinds of short term 
storage of information: those that are maintained actively in the working memory 
and increase the net activity in fronto-parietal regions, and others that are stored 
implicitly and decrease the activity in these regions. The information that creates the 
organisational structure of the task episode seems to belong to the latter category. 
Plausibly, task related information that cannot be maintained in the implicit 
representational structure is maintained in working memory by some other process 
like the phonological loop (Baddeley, 2007). 
 Some critical findings that were hitherto unexplainable seem to fit this 
dichotomy. Consider the case of the experiment described in chapter 4. At the cue 
stage, trial type 1 having the greatest working memory load elicited the greatest 
activity compared to other trial types. However, by the beginning of the subtrial 
sequence, the cue information has been incorporated into the structural framework 
of the trial episode, and the overall activity at the initial subtrials did not differ across 
the trial types differing in their working memory load. Similarly in chapter 3, the hard 
cues with greater working memory load elicited greater activity, but in the following 
interval, by the time of which the most of the cue information might have got 
incorporated into stable structure representations, the intervals preceding the hard 
task blocks did not have greater activity than those preceding easy blocks. On the 
214 
contrary, in some regions like IPS, the intervals preceding the easy blocks had greater 
activity (Fig 3.5).  
6.4 Implications 
Task Episodes 
The results of this thesis show that mental organisation is an important 
determinant of activity in widespread regions including many within the fronto-
parietal cortices, and that information about the structure of mental (or task) 
episode is represented widely across these regions.  
How control arises in cognition, is a central question in the neurosciences. 
Earlier formulations have suggested that representations of task-relevant 
information could act as the source of control by biasing the relevant processing 
across various brain systems (Miller and Cohen, 2001; Duncan, 2006). Earlier studies 
in this regard have shown fronto-parietal representation of information related to 
task stimuli (Hampshire et al., 2007; Woolgar et al., 2011), rules (Bunge and Wallis, 
2008), decisions (Haynes, 2007; Reverberi et al., 2011) and responses (Woolgar et al., 
2011; Soon et al., 2008). The representation of the structure of the task episode in 
the context of which such information was represented was not looked into by 
earlier studies. It is possible that this representation, which has been shown by the 
current results to be present in widespread brain regions, is the source of control 
required for organising extended sequential behaviour. Perhaps disruption of such 
representations after frontal lesions is the cause of behavioural disorganisation seen 
in them (Duncan, 1986; Shallice and Burgess, 1991). 
Interestingly, current results also show that the magnitude of representation 
of attended events (chapter 2) and cognitive set related to task events (chapter 5), 
are dependent on their position in the episode. The current results, thus, highlight an 
important issue, hitherto, ignored by many studies. Many previous studies have tried 
215 
to correlate a cognitive process with activity in the fronto-parietal regions e.g. 
conflict monitoring, (MacDonald et al., 2000), response inhibition (Aron, 2004), 
attentional shift (Tamber-Rosenau et al., 2011). However, as the results in chapter 2 
illustrate in the context of attentional detection, such an attempt requires a further 
qualification – where does that cognitive process occur in the extended mental 
episode, and what role does it fulfil in the organisation of that episode? 
It’s also worth pointing out that an experimental trial is also an organised task 
episode, and so the issues highlighted in this thesis pertain to it as well. Notably, in 
many studies the kind of trials that elicit greater prefrontal activity, require a more 
extended and complicated organisation of the mental episode. Consider, for 
example, the two trials from Koechlin et al. (2003), (described in detail in chapter 1). 
In the low prefrontal activity eliciting trials, subjects decided about the response 
directly based on the colour of the stimuli; whereas in the high activity eliciting trials, 
they had to decide first about the episode, then about the currently relevant rule 
and, finally, about the colour-to-response mapping based on that rule. Clearly the 
number of sequential steps required to reach the solution is greater in the latter 
trials. While the authors interpreted the results as showing that greater frontal 
activity is elicited when the context of the episode has to be used to select for the 
relevant context or rule, the results in chapter 2 suggest that the same can be 
elicited in any extended, multistep task.  
Conceptually similar issues can be speculated about the results of Dosenbach 
et al. (2006). They looked at the activity related to different aspects of task blocks in 
ten different experiments. However, the structure of the task blocks across these 
conditions was the same - extended blocks consisting of numerous individual trials 
with jittered inter-trial intervals. It is therefore possible that the qualitatively similar 
activity elicited by these blocks across the different experiments stemmed from their 
identical pattern of organisation. 
216 
Default Mode Regions  
Another noteworthy result has been the presence of episode structure 
information in regions that have been thought of as task negative or default mode 
regions (Fox et al., 2005). The current results show that beyond representing episode 
structure information, these regions even show an effect of this information on their 
representation of task rule, since their content of task rule information increased 
across the sequential phases of the trial episode (chapter 5). Thus, these regions not 
only have information about the gross structure of the episode, but they also seem 
to represent the specific content of the individual steps of the episode. Further, the 
pattern of activity in these regions can be correlated with that in multiple demand 
regions. Together, these results call into question the assumption that these regions 
do not actively participate in tasks requiring cognitive control and attention to the 
external world. 
Differentiation within MD Regions 
The results discussed in the thesis also point at some functional distinctions 
amongst the different MD regions. It was noted in chapter 2 that the behaviour of 
posterior prefrontal regions like IFS was in distinction to the anterior prefrontal 
regions. While the anterior regions were only active at the completion of higher level 
episodes, posterior regions were active at even the lowest level episode completions, 
suggesting that it is only the anterior prefrontal regions that specifically represent 
hierarchically higher level episode. The behaviour of left IFS was specifically notable. 
This region was active in response to the completion of lowest level subtasks, but did 
not show additional activity at the completion of higher level task episodes. Similarly, 
in chapter 4, it was seen that in all trial types left IFS represented lower chunk level 
organisation, but not the higher sequence level organisation. In contrast the right 
APFC only represented the higher sequence level organisation but not the lower 
chunk level organisation. Right APFC and left IFS thus seem to be at the opposite 
217 
ends of the spectrum. The former only represents higher level aspects of task 
organisation, while the latter exclusively represents the lower level aspects of task 
hierarchy. 
The results also pointed to a lateralisation in the representation of task 
episode structure. It was seen in chapter 2 that right prefrontal regions – right IFS 
and right APFC - showed greater activity in response to the completion of higher 
level task episodes than left prefrontal regions. Conceptually similar results were 
obtained in chapters 3 and 4, wherein the right prefrontal regions showed greater 
evidence of higher level task structure representation. Left prefrontal regions, on the 
other hand, showed greater modulation of activity with task difficulty (chapter 3). 
Taken together, these results suggest that the higher level structure of task episodes 
is preferentially represented in the right prefrontal regions, in contrast to lower level 
task structure which seems to have greater representation in the left prefrontal 
regions.  
Primary Sensory and Motor Regions  
 One of the most puzzling results of this thesis has been the evidence that 
episode structure information is present in the primary sensory and motor regions of 
the brain. Some earlier studies have shown that task related information is present in 
the primary sensory region of the relevant modality e.g. primary visual cortex in 
visual tasks (Mirabella et al., 2007; Woolgar et al., 2011). However, such effects can 
be accounted for by the fact that the pattern of top down control exerted on these 
regions will be different when different rules are relevant. However, our results show 
that not only is the relevant rule coded in these regions but that this coding is also 
affected by the structure of task episode in which the individual rules are to be 
applied. 
218 
Finally, the demonstration of episode structure information in sensory regions 
of a different modality or in the task irrelevant primary motor cortex is, to my 
knowledge, novel. This and other issues raised by the current thesis will require 
further investigation. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Appendix 
A. Tables of peak coordinates 
Table A2.1. Chapter 2, Exp 1:  T3 vs T2 & T1  
BA  Label Z XYZ(mm) 
46 R Mid Frontal 3.92 34,48,34 
46 R Mid Frontal 3.53 26,50,18 
46 L Mid Frontal 3.65 -40,28,38 
219 
10 L Mid Frontal 4.1 -26,54,10 
6 R Sup Frontal 4.15 26,4,58 
32 L Sup Med 4.84 -10,28,36 
32 L Ant Cingulate 4.63 -10,28,30 
32 R SMA 4.04 8,12,48 
32 L SMA 3.81 2,14,46 
32 R Mid Cingulate 3.98 6,14,46 
32 L Mid Cingulate 4.69 -10,26,34 
7 R Sup Parietal Lob 4.02 12,-68,54 
7 L Sup Parietal Lob 3.47 -14,-68,44 
40 R Inf Parietal Lob 3.85 34,-44,44 
40 L Inf Parietal Lob 3.46 -60,-52,38 
7 R Precuneus 4.1 10,-68,56 
7 L Precuneus 3.52 -14,-64,38 
7 R Sup Occipital 4.01 20,-60,40 
7 R Cuneus 3.88 18,-60,40 
18 L Calcarine 3.8 -4,-92,-12 
 R Caudate Nucleus 3.75 8,22,6 
 L Caudate Nucleus 4.02 -4,16,8 
 
Table A2.2 . Chapter 2, Exp 1:  X vs T2 & T1 
BA  Label Z XYZ(mm) 
8 R Sup Frontal 5.59 26,12,58 
6 L Sup Frontal 3.8 -20,4,68 
8 R Mid Frontal 5.41 28,10,54 
46 R Mid Frontal 4.93 30,44,30 
46 L Mid Frontal 4.77 -30,44,32 
45 R Inf Frontal (p. Tri) 5.15 54,22,18 
45 R Inf Frontal (p. Op) 5.13 56,22,14 
6 L Inf Frontal (p. Op) 4.11 -56,8,12 
6 R Precentral 4.89 44,6,52 
47 R Insula Lobe 4.06 28,22,-10 
47 L Insula Lobe 3.96 -28,26,-6 
24 R Ant Cingulate 4.95 4,26,30 
24 L Ant Cingulate 5.14 -2,26,32 
8 R SMA 5.09 8,26,54 
32 L SMA 4.32 2,16,46 
8 R Sup Med 4.7 8,28,54 
24 L Sup Med 5.02 -2,26,38 
24 R Mid Cingulate 5.41 2,26,32 
24 L Mid Cingulate 5.43 -6,18,36 
220 
40 R Inf Parietal Lob 4.51 40,-56,44 
40 L Inf Parietal Lob 4.29 -54,-44,38 
40 R Supramarginal 4.66 60,-42,32 
    
23 R Post Cingulate 4.62 4,-34,30 
23 L Post Cingulate 4.66 -4,-36,26 
7 R Precuneus 4.88 2,-56,52 
7 L Precuneus 5.25 -2,-58,54 
18 R Lingual 4.7 12,-74,2 
17 L Lingual 4.41 -8,-76,6 
6 R Rol Operculum 4.55 54,8,10 
6 L Rol Operculum 4.81 -46,-6,10 
18 R Cuneus 5.23 16,-66,36 
17 L Cuneus 4.42 -12,-70,32 
17 R Calcarine 4.69 10,-74,4 
17 L Calcarine 4.41 -8,-78,6 
 
 
 
 
 
 
Table A3.1: Chapter 3, Hard block – Easy Block 
BA  Label Z XYZ(mm) 
8 R Sup Frontal 5.12 24,8,52 
8 L Sup Frontal 4.99 -20,6,50 
6 R Mid Frontal 4.57 32,4,40 
8 R Mid Frontal 5.93 30,20,54 
10 R Mid Frontal 5.27 36,62,10 
46 R Mid Frontal 4.39 40,38,36 
221 
10 L Mid Frontal 5.56 -34,54,12 
6 L Mid Frontal 4.72 -40,4,54 
44 L Inf Frontal (p. Op) 4.82 -52,16,24 
45 L Inf Frontal (p. Tri) 5.26 -46,32,24 
10 R Mid Orbital 4.56 34,58,0 
32 L Sup Med 4.18 -8,26,40 
32 R SMA 4.4 8,22,46 
8 L SMA 4.65 -10,20,58 
7 R Sup Parietal Lob 5.23 12,-68,54 
7 L Sup Parietal Lob 4.87 -22,-68,42 
40 R Inf Parietal Lob 4.31 38,-50,48 
40 L Inf Parietal Lob 5.05 -38,-46,50 
7 L Inf Parietal Lob 4.32 -24,-66,44 
7 R Precuneus 5.66 4,-60,48 
7 L Precuneus 5.42 2,-62,46 
19 R Mid Occipital 4.65 30,-72,34 
 R Cerebelum (Crus 1) 5.75 32,-64,-28 
 L Cerebelum (Crus 1) 5.09 -32,-64,-30 
 Vermis (7) 5.02 0,-78,-26 
 
Table A3.2. Chapter 3, Final block – Initial Block 
BA  Label Z XYZ(mm) 
46 R Mid Orbital 6.49 42,54,-6 
47 L Insula Lobe 6.08 -30,24,-8 
8 R SMA 5.77 6,24,54 
45 R Inf Frontal (p. Op) 5.42 54,18,8 
8 R Sup Med 5.3 6,32,44 
8 R Mid Frontal 4.91 30,16,54 
40 L Inf Parietal Lob 4.9 -54,-48,38 
47 R Insula Lobe 4.88 46,22,-2 
46 R Mid Frontal 4.87 44,52,2 
47 L Mid Orbital 4.8 -36,44,-6 
6 L Sup Frontal 4.73 -24,-4,70 
40 R Supramarginal 4.72 52,-42,40 
8 R Sup Frontal 4.69 12,26,54 
44 R Precentral 4.49 48,12,40 
3 L Postcentral 4.44 -40,-32,48 
6 L Precentral 4.41 -32,-12,66 
24 L Mid Cingulate 4.3 0,30,36 
 L Cerebelum (VI) 4.26 -16,-72,-24 
45 L Inf Frontal (p. Tri) 4.25 -56,20,8 
222 
37 R Mid Temporal 4.22 46,-54,8 
3 R Postcentral 4.13 56,-22,44 
46 L Mid Frontal 4.13 -44,48,2 
2 L Sup Parietal Lob 4.09 -40,-42,58 
2 R Sup Parietal Lob 4.05 44,-40,58 
38 L Temporal Pole 3.96 -36,18,-28 
38 R Temporal Pole 3.94 48,20,-14 
6 R Rol Operculum 3.9 52,8,12 
44 L Mid Frontal 3.89 -40,22,40 
20 L Mid Temporal 3.64 -52,-34,-14 
 R Cerebelum (Crus 2) 3.63 14,-76,-34 
32 L Ant Cingulate 3.62 -2,36,32 
37 L Mid Occipital 3.57 -48,-68,6 
32 R Ant Cingulate 3.54 4,34,30 
 R Hippocampus 3.48 38,-12,-14 
22 L Sup Temporal 3.42 -62,-50,18 
42 R Sup Temporal 3.4 64,-36,16 
20 R Inf Temporal 3.31 58,-32,-18 
 R Caudate Nucleus 3.3 12,10,6 
7 R Cuneus 3.27 20,-66,34 
18 R Sup Occipital 3.15 22,-66,32 
 
 
 
 
 
Table A4.1. Chapter 4, Trial type 1: Subtrial 8 – intermediate subtrials 
BA  Label Z XYZ(mm) 
37 L Fusiform 6.05 -32,-44,-14 
4 R Postcentral 5.84 18,-36,70 
47 L Insula Lobe 5.6 -32,18,-12 
 L Cerebelum (IV-V) 5.53 -8,-44,-6 
18 L Cuneus 5.42 -10,-74,20 
21 L Mid Temporal 5.36 -56,4,-14 
223 
18 R Lingual 5.28 10,-78,-2 
23 R Mid Cingulate 5.25 6,-28,42 
47 R Inf Frontal (p. Orb) 5.25 46,24,-2 
47 R Inf Frontal (p. Tri) 5.24 50,24,0 
47 R Insula Lobe 5.17 32,18,-10 
23 L Mid Cingulate 5.16 -6,-36,46 
11 R Mid Orbital 5.15 12,58,-8 
20 R Mid Temporal 5.15 50,-16,-12 
 R ParaHippocampal 5.15 18,-36,-8 
3 L Postcentral 5.14 -42,-18,36 
37 R Fusiform 5.11 26,-40,-12 
45 R Inf Frontal (p. Op) 5.09 50,22,14 
39 L Inf Parietal Lob 5.06 -56,-54,40 
18 R Sup Occipital 5.02 12,-94,20 
17 R Calcarine 5.01 8,-80,2 
22 R Angular 4.99 62,-54,26 
38 L Inf Frontal (p. Orb) 4.99 -36,18,-14 
 L ParaHippocampal 4.99 -20,-40,-4 
 R Amygdala 4.99 32,4,-20 
5 R Precuneus 4.95 2,-40,48 
10 R Sup Frontal 4.95 28,64,10 
18 L Sup Occipital 4.95 -14,-90,18 
11 L Sup Frontal 4.91 -22,58,2 
8 R SMA 4.85 10,26,52 
 L Hippocampus 4.85 -36,-16,-16 
23 L Precuneus 4.83 -12,-38,46 
 R Caudate Nucleus 4.82 8,10,2 
9 R Mid Frontal 4.8 32,34,42 
23 R Post Cingulate 4.77 8,-44,32 
9 L Mid Frontal 4.75 -28,38,34 
40 R Inf Parietal Lob 4.67 46,-54,44 
20 L Inf Temporal 4.65 -40,-18,-18 
32 L Ant Cingulate 4.62 -2,36,32 
11 R Olfactory 4.57 20,8,-14 
39 R Mid Occipital 4.55 50,-74,26 
 R Cerebelum (IV-V) 4.55 22,-38,-18 
 R Putamen 4.54 28,16,-6 
3 R Precentral 4.53 36,-16,40 
32 R Ant Cingulate 4.53 4,38,30 
4 R Paracentral Lob 4.51 14,-40,50 
4 L Paracentral Lob 4.48 -10,-32,52 
6 L Precentral 4.48 -20,-24,70 
44 R Inf Frontal (p. Tri) 4.47 44,24,32 
 L Heschl 4.46 -38,-18,8 
224 
 R Hippocampus 4.45 20,-30,-6 
23 L Post Cingulate 4.39 -8,-34,34 
38 R Rol Operculum 4.38 58,16,0 
5 R Sup Parietal Lob 4.28 14,-42,66 
 
Table A4.2. Chapter 4, Trial type 2: Subtrial 4 – intermediate subtrials 
BA label AAL label Z XYZ(mm) 
6 L SMA 4.6 -12,10,50 
18 R Cuneus 4.31 8,-82,28 
 Vermis (3) 4.3 0,-40,0 
27 R Lingual 4.22 2,-38,0 
47 R Insula Lobe 4.2 32,18,4 
18 L Cuneus 4.16 -8,-84,20 
 L Cerebelum (IV-V) 4.16 -4,-40,0 
 R Putamen 4.12 30,18,4 
47 L Mid Frontal 4.1 -28,48,4 
11 L Mid Frontal 4.1 -26,48,2 
7 R Sup Parietal Lob 3.97 16,-66,58 
17 L Calcarine 3.94 -6,-90,0 
6 R Sup Frontal 3.9 18,4,54 
47 R Inf Frontal (p. Tri) 3.88 32,32,4 
19 L Sup Occipital 3.86 -14,-86,18 
7 L Sup Parietal Lob 3.85 -16,-70,52 
3 R Postcentral 3.83 56,-24,46 
47 L Insula Lobe 3.81 -36,26,4 
6 R SMA 3.79 2,-4,74 
32 L Mid Cingulate 3.78 -12,8,46 
2 R Supramarginal 3.75 56,-26,46 
18 L Mid Occipital 3.66 -18,-88,16 
45 R Inf Frontal (p. Op) 3.64 46,16,2 
22 R Mid Temporal 3.61 58,-46,12 
27 R Post Cingulate 3.53 4,-40,6 
21 L Mid Temporal 3.53 -58,-54,6 
45 L Inf Frontal (p. Op) 3.49 -46,12,2 
40 L Inf Parietal Lob 3.48 -28,-46,38 
19 R Mid Occipital 3.43 36,-72,30 
32 L Sup Med 3.42 -6,18,44 
27 R ParaHippocampal 3.42 16,-40,-4 
8 R Mid Frontal 3.41 26,12,50 
6 L Precentral 3.39 -34,2,56 
46 R Mid Frontal 3.33 24,42,24 
6 R Precentral 3.31 42,4,36 
41 R Angular 3.28 48,-46,26 
225 
32 R Mid Cingulate 3.21 10,18,38 
29 L Post Cingulate 3.21 -6,-42,8 
7 R Sup Occipital 3.21 32,-72,42 
 
Table A4.3.  Chapter 4, Trial type 2: Subtrial 8 – intermediate subtrials 
BA Label Z XYZ(mm) 
18 L Calcarine 5.63 -2,-92,14 
17 R Cuneus 5.5 12,-96,16 
18 R Calcarine 5.45 8,-90,10 
17 R Linual 5.07 6,-82,0 
18 L Sup Occipital 4.86 -12,-86,22 
19 L Linual 4.61 -26,-54,-4 
47 R Inf Frontal (p. Orb) 4.6 38,26,-8 
47 R Insula Lobe 4.57 30,18,-8 
7 L Precuneus 4.49 -10,-74,38 
45 R Inf Frontal (p. Op) 4.46 52,20,12 
40 R Angular 4.43 60,-48,32 
3 L Postcentral 4.43 -46,-18,34 
2 R Postcentral 4.41 28,-40,62 
21 R Mid Temporal 4.41 62,-26,-4 
19 L Fusiform 4.4 -28,-56,-6 
7 R Precuneus 4.38 6,-80,46 
27 R ParaHippocampal 4.38 18,-40,-6 
18 R Mid Occipital 4.38 26,-90,18 
18 L Cerebelum (VI) 4.3 -20,-70,-14 
18 Vermis (6) 4.26 2,-70,-4 
32 R Ant Cingulate 4.26 4,40,30 
41 R Sup Temporal 4.23 50,-42,24 
5 R Sup Parietal Lob 4.17 14,-46,64 
18 L Mid Occipital 4.14 -18,-88,16 
19 R Fusiform 4.14 28,-72,-10 
44 R Mid Frontal 4.14 40,20,40 
23 R Post Cingulate 4.13 4,-38,32 
6 R Precentral 4.13 34,-16,46 
3 R Postcentral 4.09 44,-18,34 
8 R SMA 4.07 6,26,56 
 L Cerebelum (Crus 2) 4.01 -22,-78,-34 
40 R Inf Parietal Lob 3.98 60,-48,38 
 R Putamen 3.93 28,16,-6 
 L Cerebelum (Crus 1) 3.91 -16,-84,-26 
27 L ParaHippocampal 3.86 -20,-40,-6 
32 L Ant Cingulate 3.86 -2,36,32 
18 R Cerebelum (VI) 3.85 18,-64,-12 
226 
 R Putamen 3.8 34,-8,4 
41 R Mid Temporal 3.77 46,-46,20 
23 L Post Cingulate 3.73 -4,-34,34 
 
Table A4.4. Chapter 4, Trial type 3: Subtrial 4 – intermediate subtrials 
BA  Label Z XYZ(mm) 
45 L Inf Frontal (p. Tri) 4.53 -36,22,10 
18 R Sup Occipital 4.34 20,-92,20 
24 L Ant Cingulate 4.29 -10,18,26 
19 R Mid Occipital 4.21 42,-86,2 
18 L Lingual 4.05 -10,-60,4 
17 L Calcarine 4.04 -10,-60,6 
19 R Inf Occipital 3.91 48,-74,-4 
6 L Sup Frontal 3.88 -30,-6,64 
6 R Mid Frontal 3.74 42,-2,60 
38 L Temporal Pole 3.72 -52,6,2 
47 R Inf Frontal (p. Tri) 3.68 38,26,8 
44 L Mid Frontal 3.67 -48,28,34 
47 R Insula Lobe 3.63 38,24,4 
32 L Sup Med 3.63 -12,32,32 
40 L Inf Parietal Lob 3.62 -38,-36,48 
37 L Mid Occipital 3.6 -42,-70,0 
24 L Mid Cingulate 3.53 -8,8,32 
19 L Sup Occipital 3.4 -24,-70,30 
27 R Lingual 3.36 8,-32,-6 
 L Cerebellum (IV-V) 3.26 -8,-56,-2 
24 R Mid Cingulate 3.25 6,14,34 
7 R Angular 3.24 28,-52,44 
40 R Inf Parietal Lob 3.23 30,-48,42 
 
 
 
Table A4.5. Chapter 4, Trial type 3: Subtrial 8 – intermediate subtrials 
BA Label Z XYZ(mm) 
23 R Precuneus 4.98 12,-50,36 
30 L Post Cingulate 4.97 -6,-48,22 
23 R Mid Cingulate 4.89 10,-48,34 
23 L Precuneus 4.83 -6,-54,20 
39 R Angular 4.71 40,-54,28 
22 R Sup Temporal 4.68 58,-54,24 
227 
 L Cerebelum (Crus 2) 4.61 -16,-80,-38 
39 R Mid Temporal 4.6 50,-60,20 
23 L Cuneus 4.55 -8,-62,28 
30 R Calcarine 4.45 14,-48,12 
39 R Angular 4.36 38,-56,28 
11 R Mid Orbital 4.32 10,58,-6 
10 L Mid Orbital 4.32 -8,62,-4 
37 L Fusiform 4.22 -26,-42,-12 
37 L Lingual 4.03 -24,-44,-8 
23 L Mid Cingulate 3.97 -8,-40,34 
 L Cerebelum (VII) 3.95 -20,-74,-40 
39 R Mid Occipital 3.88 44,-64,30 
 
Table A5.1. Chapter 5, Regions where discriminability of A vs B increased across trial type 1 
BA Label Z XYZ(mm) 
47 R Insula Lobe 4.16 27,26,-11 
11 R Sup Orbital 3.8 21,29,-14 
47 R Putamen 3.64 24,23,-5 
47 R Inf Frontal (p. Tri) 3.49 39,38,1 
11 R Mid Orbital 3.35 27,38,-11 
11 R Rectal 3.12 18,20,-14 
11 R Caudate Nucleus 3.11 18,26,-5 
37 R Fusiform 3.81 27,-52,-17 
37 R Linual 3.13 24,-49,-8 
5 R Precuneus 3.93 9,-40,58 
2 R Paracentral Lob 3.7 15,-40,55 
5 L Precuneus 3.59 -12,-43,49 
5 R Sup Parietal Lob 3.54 15,-43,61 
 R Insula Lobe 3.68 36,-13,22 
41 R Sup Temporal 3.63 42,-28,16 
3 R Postcentral 3.37 39,-19,37 
 R Putamen 3.29 30,-10,13 
18 R Cuneus 3.7 15,-79,31 
19 R Sup Occipital 3.69 27,-82,40 
23 R Precuneus 3.27 12,-67,28 
19 R Mid Occipital 3.22 36,-88,28 
8 R Sup Med 3.69 6,26,46 
32 R Ant Cingulate 3.66 12,35,19 
32 R Mid Cingulate 3.58 6,29,40 
8 R SMA 3.57 9,26,49 
32 L Sup Med 3.32 0,29,40 
9 R Sup Frontal 3.23 12,38,52 
 
228 
Table A5.2. Chapter 5, Regions where discriminability of A vs B increased across trial type 2 
BA  Label Z XYZ(mm) 
19 R Cuneus 4.62 12,-82,40 
17 R Calcarine 4.56 3,-64,16 
23 R Precuneus 4.07 6,-67,25 
19 R Sup Occipital 3.81 15,-85,37 
23 L Precuneus 3.81 0,-61,19 
18 L Cuneus 3.79 0,-70,22 
 R Cerebelum (IV-V) 3.21 15,-55,-11 
19 R Fusiform 3.12 27,-58,-2 
22 R Sup Temporal 4.18 66,-31,7 
22 R Mid Temporal 4.12 54,-49,16 
22 R Sup Temporal 3.15 66,-49,25 
32 L Sup Med 3.82 -9,47,31 
32 R Ant Cingulate 3.77 9,47,10 
32 L Ant Cingulate 3.58 0,47,13 
10 L Mid Orbital 3.3 -3,53,-5 
10 R Mid Orbital 3.22 0,53,-2 
 
 
 
 
 
229 
B. Task Instructions 
 B.2 Chapter 2 
At the beginning of each trial, you will see a cue three-letter word, followed by a series of 
letters presented in the middle of the computer screen. Two tasks are to be done on each 
trial: First Task is to detect the appearance of the letters of the word shown, in the correct 
order, amongst the other letters. The second task is to then search for the letter ‘X’ 
amongst the remaining letters. You only need to respond at the end of the trial when the 
cue “X?” appears: If you have seen the letters of the cue word and the letter ‘X’ in the 
correct order, then press the first button; if the two-word sequence was not completed in 
the correct order, then press the second button. You have two seconds to make this 
decision. 
 
Examples: 
  
 
CAT AWCFKIENJAJDXGIOPQMJSXNTBNCPSACTKXKQWLP X? 
 
 
 
CAT AWCFMIENJAJDXGIOPQMXSXNTBNCPSACTKDKQWLP X? 
(Note that ‘X’ did occur earlier in the sequence, but was irrelevant. Only the appearance of 
‘X’ after the appearance of the letters of the cued word is to be considered relevant) 
Correctly detecting ‘X’ will increment your score by one point. Any error will decrease your 
score by one point. Your total score will be presented briefly, followed by a short pause 
before the next trial starts automatically. 
Please try to remain still and look at the centre of the screen at all times. 
This sequence will be repeated throughout three blocks of about 20 minutes each, with 
short breaks in between. 
Search for ‘C’…‘A’…‘T’ Search for ‘X’ 
Press the index finger 
(‘YES’) button because ‘X’ 
did appear after the letters 
of ‘CAT’ had appeared  
Time 
Since ‘X’ did not appear 
after the letters of ‘CAT’, 
press the middle finger 
(‘NO’) button 
Search for ‘C’…‘A’…‘T’ 
 
Search for ‘X’ 
 
230 
B.3 Chapter 3 
At the beginning of each trial you will be presented with some arithmetic 
information (consisting of either a number e.g. ‘3’ or two numbers along with a ‘+’ or ‘-’, in 
the format ‘x+   y’ e.g. ‘6+   3’) and one or two words. These are the cues that you need to 
remember. In rest of the trial, you will do an arithmetic task block using the arithmetic 
information, and a word task block using the cued word(s). The two kinds of cued 
information will be displayed in different lines. The information related to the task block 
that will be done first will always be displayed on top.  
Consider the two cue displays below: 
 
In case of the left display, word task will be done first, followed by the arithmetic task. 
Reverse will be the case for cue display on the right. 
The cue screen will then disappear, and will be followed by an empty box whose 
margin colour will depend upon the nature of the 1st task block: Blue – word task; Green- 
arithmetic task.  
231 
 
Subsequently, the first task block will start. This will involve a series of stimulus 
presentations on each of which you will execute the relevant task (details below).  
The first task block will end with a blank interval, at which point the colour of the box 
margins will change, and will now signal the nature of the ensuing task block. Note that the 
two task blocks in a trial are always different i.e. the same task block is never repeated 
within a trial, thus if the first block was arithmetic, the second will be word, and vice-versa. 
You will then execute the second task block, at the end of which, the colour of the 
margins will become black. You can now relax till the next trial begins. 
Arithmetic Task: 
In arithmetic blocks, the presented stimulus consists of two blanks with a number followed 
by ‘+’ or ‘-’ in between (e.g. ‘...4+…’). If the arithmetic cue was a single number (e.g. ‘3’) then 
mentally put this number in the second blank and then solve the resulting expression (e.g. 
‘4+3’). If the solution is less than 10, press the index finger button, else press the middle 
finger button.  
In trials where arithmetic cue was in the pattern ‘x+  y’ e.g. ‘6+   3’, mentally put the first 
part of this (‘6+’) in the first blank, and the second part (‘3’) in the second blank, and then 
solve the resulting expression (‘6+4+3). Again, if the solution is less than 10, press the index 
finger button, else press the middle finger button. 
232 
Word Task: 
In word blocks, you will see a number of word presentations, at each respond with index 
finger button press if the presented word is the same as the word(s) cued at the beginning 
of the trial. 
 
 
B.4 Chapter 4 
Pre-Scan Session Instructions 
We will start with a practise session.   
Before each trial of this session, you will be presented with a four letter string consisting of As and 
Bs. This represents the sequence of four sequential tasks that you will execute on the ensuing four 
sequential stimulus presentations e.g. if the string is ‘ABAB’, you execute task A then task B then task 
A then task B. It is hence important that you keep this presented string in your mind for the rest of 
the trial. The trial will end after the fourth stimulus presentation. 
 
The stimuli consist of a letter and a number presented within a box. Task A requires you to 
categorise the letter as vowel or consonant. Press the index finger button if the letter is a vowel, else 
press the middle finger button. Task B is to categorise the number as even or odd; press the ring 
finger button if the number is even, else press the little finger button. 
Now, you will do a 20 minute practise session. 
Scan Session Instruction 
 In the main session, you will do a very similar experiment. You will now see an eight letter string at 
the beginning of each trial. The trial will now consist of eight sequential stimulus presentations 
instead of four, at each of which you will execute the task corresponding to the letters of the string.  
233 
 
 
In some trials, you will not be presented with the initial eight letter string; instead you will see a 
blank. In such trials, the colour of the margins of the enclosing box will tell you the task to be done at 
each step - blue: letter task; green: number task).  
 
After the response to the eighth stimulus presentation, the trial will end. You can now relax till the 
beginning of the next trial. 
 
234 
 
 
 
References  
 
Allport, A., & Wylie, G. (2000). Task- switching, stimulus-response bindings, and negative priming. 
 In S. Monsell & J. Driver (Eds.), Control of cognitive processes: attention and performance 
 XVIII. Cambridge, MA: MIT Press. 
Alvarez, J. A., & Emory, E. (2006). Executive Function and the Frontal Lobes: A Meta-Analytic 
 Review. Neuropsychology Review, 16(1), 17-42. doi:10.1007/s11065-006-9002-x 
Anderson, J. R., & Matessa, M. (1997). A Production System Theory of Serial Memory. 
 Psychological Review, 104(4), 728-748. 
Anderson, J. R., Anderson, J. F., Ferris, J. L., Fincham, J. M., & Jung, K.-J. (2009). Lateral inferior 
 prefrontal cortex and anterior cingulate cortex are engaged at different stages in the      
 solution of insight problems. Proceedings of the National Academy of Sciences, 106(26), 
 10799- 10804.  
Anderson, J. R., Bothell, D., Lebiere, C., & Matessa, M. (1998). An Integrated Theory of List 
 Memory, Journal of Memory and Language, 38(4), 341-380.  
Anderson, S. W., Damasio, H., Jones, R. D., & Tranel, D. (1991). Wisconsin Card Sorting Test 
 performance as a measure of frontal lobe damage. Journal of Clinical and Experimental 
 Neuropsychology, 13(6), 909-922.  
Aron, A. R., Robbins, T. W., & Poldrack, R. A. (2004). Inhibition and the right inferior frontal 
 cortex. Trends in Cognitive Sciences, 8(4), 170-177. 
Asaad, W. F., Rainer, Gregor, & Miller, Earl K. (1998). Neural Activity in the Primate Prefrontal 
 Cortex during Associative Learning. Neuron, 21(6), 1399-1407.  
Asaad, W. F., Rainer, Gregor, & Miller, Earl K. (2000). Task-Specific Neural Activity in the Primate 
 Prefrontal Cortex. Journal of Neurophysiology, 84(1), 451 -459. 
Azuma, M., & Suzuki, H. (1984). Properties and distribution of auditory neurons in the dorsolateral 
 prefrontal cortex of the alert monkey. Brain Research, 298(2), 343-346. 
Baddeley, A. (2007). Working memory. Oxford, England: Oxford University Press. 
 
235 
Badre, D., & D’Esposito, M. (2007). Functional Magnetic Resonance Imaging Evidence for a 
 Hierarchical Organization of the Prefrontal Cortex. Journal of Cognitive Neuroscience, 
 19(12), 2082-2099. 
Badre, D., & D’Esposito, M. (2009). Is the rostro-caudal axis of the frontal lobe hierarchical? Nat  Rev 
 Neurosci, 10(9), 659-669. 
Badre, D., Kayser, A. S., & D’Esposito, M. (2010). Frontal cortex and the discovery of abstract 
 action rules. Neuron, 66(2), 315-326. 
Bahlmann, J., Schubotz, R. I., Mueller, J. L., Koester, D., & Friederici, A. D. (2009). Neural circuits  of 
 hierarchical visuo-spatial sequence processing. Brain Research, 1298, 161-170.  
Berdyyeva, T. K., & Olson, C. R. (2009). Monkey Supplementary Eye Field Neurons Signal the 
 Ordinal Position of Both Actions and Objects. The Journal of Neuroscience, 29(3), 591  -
 599.  
Bianchi, L., & Macdonald, J. H. (1992). The Mechanism of the Brain: And the Function of the 
 Frontal Lobes. E. & S. Livingstone. 
Bode, S., & Haynes, J.-D. (2009). Decoding sequential stages of task preparation in the human 
 brain. NeuroImage, 45(2), 606-613. 
Bode, S., He, A. H., Soon, C.S., Trampel, R., Turner, R., & Haynes, J.-D. (2011). Tracking the 
 Unconscious Generation of Free Decisions Using Uitra-High Field fMRI. PLoS ONE, 6(6).  
Bode, Stefan, & Haynes, John-Dylan. (2009). Decoding sequential stages of task preparation in the 
 human brain. NeuroImage, 45(2), 606-613. 
Botvinick, M. M. (2008). Hierarchical models of behavior and prefrontal function. Trends in 
 Cognitive Sciences, 12(5), 201-208. 
Botvinick, M., & Plaut, D. C. (2004). Doing without schema hierarchies: a recurrent connectionist 
 approach to normal and impaired routine sequential action. Psychological Review, 111(2), 
 395-429.  
Bowman, E. M., Aigner, T. G., & Richmond, B. J. (1996). Neural signals in the monkey ventral 
 striatum related to motivation for juice and cocaine rewards. Journal of Neurophysiology, 
 75(3), 1061 -1073. 
Boy, F., Husain, M., & Sumner, P. (2010). Unconscious inhibition separates two forms of cognitive 
 control. Proceedings of the National Academy of Sciences, 107(24), 11134 -11139.   
Braver, T. S., Reynolds, J. R., & Donaldson, D. I. (2003). Neural Mechanisms of Transient and 
 Sustained Cognitive Control during Task Switching. Neuron, 39(4), 713-726.   
236 
Braver, T. S., Paxton, J. L., Locke, H. S., & Barch, D. M. (2009). Flexible neural mechanisms of 
 cognitive control within human prefrontal cortex. Proceedings of the National Academy of 
 Sciences, 106(18), 7351-7356. 
Brett, M., Johnsrude, I. S., & Owen, A. M. (2002). The problem of functional localization in the 
 human brain. Nat Rev Neurosci, 3(3), 243-249.  
Buckner, Randy L, Andrews-Hanna, J. R., & Schacter, D. L. (2008). The brain’s default network: 
 anatomy, function, and relevance to disease. Annals of the New York Academy of Sciences, 
 1124, 1-38.  
Bunge, S. A., Helskog, E. H., & Wendelken, C. (2009). Left, but not right, rostrolateral prefrontal 
 cortex meets a stringent test of the relational integration hypothesis. NeuroImage, 46(1), 
 338-342.  
Burgess, P. W., Dumontheil, I., & Gilbert, Sam J. (2007). The gateway hypothesis of rostral 
 prefrontal cortex (area 10) function. Trends in Cognitive Sciences, 11(7), 290-298.  
Buschman, T. J., & Miller, E. K. (2007). Top-down versus bottom-up control of attention in the 
 prefrontal and posterior parietal cortices. Science, 315(5820), 1860. 
Cabeza, R., & Nyberg, L. (2000). Imaging cognition II: An empirical review of 275 PET and fMRI 
 studies. Journal of cognitive neuroscience, 12(1), 1–47. 
Campos, M., Breznen, B., & Andersen, R. A. (2010). A Neural Representation of Sequential States 
 Within an Instructed Task. Journal of Neurophysiology, 104(5), 2831 -2849.  
Chikazoe, J., Jimura, K., Asari, T., Yamashita, K.-ichiro, Morimoto, H., Hirose, S., Miyashita, Y., et  al. 
 (2009). Functional Dissociation in Right Inferior Frontal Cortex during Performance of 
 Go/No-Go Task. Cereb. Cortex, 19(1), 146-152.  
Christoff, K., Keramatian, K., Gordon, A. M., Smith, R., & Mädler, B. (2009). Prefrontal 
 organization of cognitive control according to levels of abstraction. Brain Research, 1286, 
 94-105. 
Clark, V. P., Fannon, S., Lai, S., Benson, R., & Bauer, L. (2000). Responses to Rare Visual Target  and 
 Distractor Stimuli Using Event-Related fMRI. Journal of Neurophysiology, 83(5), 3133  -
 3139. 
Clegg, B. A., Digirolamo, G. J., & Keele, S. W. (1998). Sequence learning. Trends in Cognitive 
 Sciences, 2(8), 275-281. 
Cole, M. W., & Schneider, W. (2007). The cognitive control network: Integrated cortical regions with 
dissociable functions. NeuroImage, 37(1), 343–360.  
Corbetta, M., & Shulman, G. L. (2002). Control of goal-directed and stimulus-driven attention in  the 
 brain. Nature Reviews. Neuroscience, 3(3), 201-215.  
237 
Corbetta, M., Patel, G., & Shulman, G. L. (2008). The Reorienting System of the Human Brain: 
 From Environment to Theory of Mind. Neuron, 58(3), 306-324.   
Cusack, R., & Papadakis, N. (2002). New Robust 3-D Phase Unwrapping Algorithms: Application  to 
 Magnetic Field Mapping and Undistorting Echoplanar Images. NeuroImage, 16(3, Part 1), 
 754-764. 
De Renzi, E., & Barbieri, C. (1992). The incidence of the grasp reflex following hemispheric lesion  and 
 its relation to frontal damage. Brain: A Journal of Neurology, 115 Pt 1, 293-313. 
Dehaene, S., Changeux, J.-P., Naccache, L., Sackur, J., & Sergent, C. (2006). Conscious, 
 preconscious, and subliminal processing: a testable taxonomy. Trends in Cognitive Sciences, 
 10(5), 204-211.  
Dehaene, S., & Changeux, J.-P., Experimental and theoretical approaches to conscious 
processing. Neuron,  70  (2011), 200–227 
Desimone, Robert, & Duncan, John. (1995). Neural Mechanisms of Selective Visual Attention. 
 Annual Review of Neuroscience, 18(1), 193-222.  
Desmurget, M., Reilly, K. T., Richard, N., Szathmari, A., Mottolese, C., & Sirigu, Angela. (2009). 
 Movement Intention After Parietal Cortex Stimulation in Humans. Science, 324(5928), 811 
 -813.  
Dosenbach, N. U. F., Fair, D. A., Miezin, F. M., Cohen, A. L., Wenger, K. K., Dosenbach, R. A. T.,  Fox, 
 M. D., et al. (2007). Distinct brain networks for adaptive and stable task control in 
 humans. Proceedings of the National Academy of Sciences, 104(26), 11073-11078.  
Dosenbach, Nico U.F., Visscher, K. M., Palmer, E. D., Miezin, Francis M., Wenger, Kristin K., 
 Kang, H. C., Burgund, E. D., et al. (2006). A Core System for the Implementation of Task 
 Sets. Neuron, 50(5), 799-812.  
Dove, A., Pollmann, S., Schubert, T., Wiggins, C. J., & von Cramon, D. Y. (2000). Prefrontal cortex 
 activation in task switching: an event-related fMRI study. Cognitive Brain Research, 9(1), 
 103-109.  
Downar, J., Crawley, A. P., Mikulis, D. J., & Davis, K. D. (2001). The Effect of Task Relevance on  the 
 Cortical Response to Changes in Visual and Auditory Stimuli: An Event-Related fMRI 
 Study. NeuroImage, 14(6), 1256-1267.  
Downar, J., Crawley, A. P., Mikulis, D. J., & Davis, K. D. (2002). A Cortical Network Sensitive to 
 Stimulus Salience in a Neutral Behavioral Context Across Multiple Sensory Modalities. 
 Journal of Neurophysiology, 87(1), 615 -620. 
Driver, J., & Mattingley, J. B. (1998). Parietal neglect and visual awareness. Nat Neurosci, 1(1), 17-
 22.  
238 
Dumontheil, I., Thompson, Russell, & Duncan, John. (2010). Assembly and Use of New Task Rules 
 in Fronto-parietal Cortex. Journal of Cognitive Neuroscience.  
Duncan, John. (1986). Disorganisation of behaviour after frontal lobe damage. Cognitive 
 Neuropsychology, 3(3), 271-290.  
Duncan, John. (2001). An adaptive coding model of neural function in prefrontal cortex. Nat Rev 
 Neurosci, 2(11), 820-829.  
Duncan, John. (2006). EPS - Mid-Career Award 2004: Brain mechanisms of attention. The 
 Quarterly Journal of Experimental Psychology, 59(1), 2.  
Duncan, John. (2010). The multiple-demand (MD) system of the primate brain: mental programs 
 for intelligent behaviour. Trends in Cognitive Sciences, 14(4), 172-179.   
Duncan, John, Bundesen, C., Olson, A., Humphreys, G., Chavda, S., & Shibuya, H. (1999).  Systematic 
 analysis of deficits in visual attention. Journal of Experimental Psychology:  General, 
 128(4), 450-478.  
Duncan, John, Burgess, P., & Emslie, H. (1995). Fluid intelligence after frontal lobe lesions. 
 Neuropsychologia, 33(3), 261-268.  
Duncan, John, Humphreys, G., & Ward, R. (1997). Competitive brain activity in visual attention. 
 Current Opinion in Neurobiology, 7(2), 255-261.  
Duncan, John, Parr, A., Woolgar, A., Thompson, Russell, Bright, P., Cox, S., Bishop, S., et al. 
 (2008). Goal neglect and Spearman’s g: competing parts of a complex task. Journal of 
 Experimental Psychology. General, 137(1), 131-148.                                                         : 
Dux, P. E., Tombu, M. N., Harrison, S., Rogers, B. P., Tong, F., & Marois, R. (2009). Training 
 Improves Multitasking Performance by Increasing the Speed of Information Processing in 
 Human Prefrontal Cortex. Neuron, 63(1), 127-138.  
Eger, E., Michel, V., Thirion, B., Amadon, A., Dehaene, S., & Kleinschmidt, A. (2009). Deciphering 
 Cortical Number Coding from Human Brain Activity Patterns. Current Biology, 19(19), 
 1608-1615.  
Esterman, M., Chiu, Y.-C., Tamber-Rosenau, B. J., & Yantis, S. (2009). Decoding cognitive control  in 
 human parietal cortex. Proceedings of the National Academy of Sciences, 106(42), 17974 
 -17979. 
Farag, C., Troiani, V., Bonner, M., Powers, C., Avants, B., Gee, J., & Grossman, M. (2010).  Hierarchical 
 Organization of Scripts: Converging Evidence from fMRI and Frontotemporal 
 Degeneration. Cerebral Cortex, 20(10), 2453 -2463.  
Forstmann, B. U., Brass, Marcel, Koch, I., & Cramon, D. Y. von. (2005). Internally generated and 
 directly cued task sets: an investigation with fMRI. Neuropsychologia, 43(6), 943-952.   
239 
Fox, Michael D., Snyder, Abraham Z., Vincent, Justin L., Corbetta, M., Van Essen, D. C., &  Raichle, 
 Marcus E. (2005). The human brain is intrinsically organized into dynamic, 
 anticorrelated functional networks. Proceedings of the National Academy of Sciences of the 
 United States of America, 102(27), 9673 -9678.  
Freedman, D. J., Riesenhuber, M., Poggio, T., & Miller, E K. (2001). Categorical representation of 
 visual stimuli in the primate prefrontal cortex. Science (New York, N.Y.), 291(5502), 312-
 316. 
Fuster, Joaquín M. (2008). The prefrontal cortex. Elsevier. 
van Gaal, S., Ridderinkhof, K. R., Fahrenfort, J. J., Scholte, H. S., & Lamme, V. A. F. (2008). 
 Frontal Cortex Mediates Unconsciously Triggered Inhibitory Control. The Journal of 
 Neuroscience, 28(32), 8053 -8062. 
Gat, E. (1998). On Three-Layer Architectures. Artificial Intelligence and Mobile Robots. MIT Press. 
Georgopoulos, A., Schwartz, A., & Kettner, R. (1986). Neuronal population coding of movement 
 direction. Science, 233(4771), 1416 -1419. 
Gilbert, S.J. (2011). Decoding the content of delayed intentions. Journal of Neuroscience, 31(8), 
 2888-2894. 
Grafman, J, Salazar, A., Weingartner, H., Vance, S., & Amin, D. (1986). The relationship of brain-
 tissue loss volume and lesion location to cognitive deficit. The Journal of Neuroscience, 
 6(2), 301 -307. 
Grant, D. A., & Berg, E. (1948). A behavioral analysis of degree of reinforcement and ease of 
 shifting to new responses in a Weigl-type card-sorting problem. Journal of Experimental 
 Psychology, 38(4), 404-411.  
Greenberg, A. S., Esterman, M., Wilson, D., Serences, J. T., & Yantis, S. (2010). Control of Spatial  and 
 Feature-Based Attention in Frontoparietal Cortex. The Journal of Neuroscience, 30(43), 
 14330 -14339.  
Gregoriou, G. G., Gotts, S. J., Zhou, H., & Desimone, R. (2009). High-frequency, long-range 
 coupling between prefrontal and visual cortex during attention. Science, 324(5931), 1207. 
Greicius, M. D., & Menon, V. (2011). Default-Mode Activity during a Passive Sensory Task: 
 Uncoupled from Deactivation but Impacting Activation. Journal of Cognitive Neuroscience, 
 16(9), 1484-1492.  
Gruber, O., Karch, S., Schlueter, E. K., Falkai, P., & Goschke, T. (2006). Neural mechanisms of 
 advance preparation in task switching. NeuroImage, 31(2), 887-895.   
Hampshire, A., Duncan, John, & Owen, A. M. (2007). Selective Tuning of the Blood Oxygenation 
 Level-Dependent Response during Simple Target Detection Dissociates Human 
 Frontoparietal Subregions. J. Neurosci., 27(23), 6219-6223.   
240 
Harrison, S. A., & Tong, F. (2009). Decoding reveals the contents of visual working memory in 
 early visual areas. Nature, 458(7238), 632-635.  
Hasegawa, R. P., Blitz, A. M., & Goldberg, M. E. (2004). Neurons in Monkey Prefrontal Cortex 
 Whose Activity Tracks the Progress of a Three-Step Self-Ordered Task. Journal of 
 Neurophysiology, 92(3), 1524 -1535.  
Haynes, John-Dylan, Sakai, K., Rees, Geraint, Gilbert, S., Frith, C., & Passingham, R. E. (2007). 
 Reading Hidden Intentions in the Human Brain. Current Biology, 17(4), 323-328.   
Henry, H. M., & Rogers, D. E., Increased response latency for complicated movements and a 
 “memory  drum” theory of neuromotor reaction, Research Quarterly 31 (1960), pp. 448–
 458. 
Hon, N., Epstein, R. A., Owen, A. M., & Duncan, John. (2006). Frontoparietal Activity with 
 Minimal Decision and Control. J. Neurosci., 26(38), 9805-9809.   
Hoshi, E., Shima, K., & Tanji, J. (1998). Task-Dependent Selectivity of Movement-Related  Neuronal 
 Activity in the Primate Prefrontal Cortex. Journal of Neurophysiology, 80(6), 3392 -3397. 
Husain, M., & Nachev, P. (2007). Space and the parietal cortex. Trends in Cognitive Sciences, 11(1), 
 30-36. 
Husain, M., & Rorden, C. (2003). Non-spatially lateralized mechanisms in hemispatial neglect. 
 Nature Reviews. Neuroscience, 4(1), 26-36.  
Husain, M., Mannan, S., Hodgson, T., Wojciulik, E., Driver, J., & Kennard, C. (2001). Impaired 
 spatial working memory across saccades contributes to abnormal search in parietal neglect. 
 Brain, 124(5), 941 -952.  
Jenkins, L. J., & Ranganath, C. (2010). Prefrontal and Medial Temporal Lobe Activity at Encoding 
 Predicts Temporal Context Memory. The Journal of Neuroscience, 30(46), 15558 -15565.  
Jiang, Y, & Kanwisher, N. (2003a). Common neural mechanisms for response selection and 
perceptual processing. Journal of Cognitive Neuroscience , 15: 1095–1110 
Jiang, Y, & Kanwisher, N. (2003b). Common neural substrates for response selection across 
modalities and mapping paradigms. Journal of Cognitive Neuroscience , 15: 1080–1094.  
De Jong, R. (1995). The role of preparation in overlapping-task performance. The Quarterly 
 Journal of Experimental Psychology, 48(1), 2-25. 
Kesner, R. P., Hopkins, R. O., & Fineman, B. (1994). Item and order dissociation in humans with 
 prefrontal cortex damage. Neuropsychologia, 32(8), 881-891.  
Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jost, K., Philipp, A. M., & Koch, I. (2010). 
 Control and interference in task switching--a review. Psychological Bulletin, 136(5), 849-
 874. 
241 
Kim, C., Johnson, N. F., Cilles, S. E., & Gold, B. T. (2011). Common and Distinct Mechanisms of 
 Cognitive Flexibility in Prefrontal Cortex. The Journal of Neuroscience, 31(13), 4771 -4779.  
Klapp, S. T., Anderson, W. G., & Berrian, R. W. (1973). Implicit speech in reading: Reconsidered. 
 Journal of Experimental Psychology, 100(2), 368-374.  
Knapen, T., Brascamp, J., Pearson, J., van Ee, R., & Blake, R. (2011). The Role of Frontal and 
 Parietal Brain Areas in Bistable Perception. The Journal of Neuroscience: The Official 
 Journal of the Society for Neuroscience, 31(28), 10293-10301.  
Knight, R. T. (2007). Neural Networks Debunk Phrenology. Science, 316(5831), 1578 -1579. 
Koechlin, Etienne, & Jubault, T. (2006). Broca’s Area and the Hierarchical Organization of Human 
 Behavior. Neuron, 50(6), 963-974.  
Koechlin, Etienne, & Summerfield, Christopher. (2007). An information theoretical approach to 
 prefrontal executive function. Trends in Cognitive Sciences, 11(6), 229-235.   
Koechlin, Etienne, Basso, G., Pietrini, P., Panzer, S., & Grafman, Jordan. (1999). The role of the 
 anterior prefrontal cortex in human cognition. Nature, 399(6732), 148-151.   
Koechlin, Etienne, Corrado, G., Pietrini, P., & Grafman, Jordan. (2000). Dissociating the role of the 
 medial and lateral anterior prefrontal cortex in human planning. Proceedings of the National 
 Academy of Sciences of the United States of America, 97(13), 7651-7656.  
Koechlin, Etienne, Ody, C., & Kouneiher, F. (2003). The Architecture of Cognitive Control in the 
 Human Prefrontal Cortex. Science, 302(5648), 1181-1185.  
Koenigs, M., Barbey, A. K., Postle, B. R., & Grafman, Jordan. (2009). Superior Parietal Cortex Is 
 Critical for the Manipulation of Information in Working Memory. The Journal of 
 Neuroscience, 29(47), 14980 -14986.  
Kühn, S., Bodammer, N. C., & Brass, M. (2010). Dissociating mental states related to doing nothing 
 by means of fMRI pattern classification. NeuroImage, 53(4), 1294-1300. 
Lashley, K. (1951). The problem of serial order in behaviour. Cerebral Mechanisms in Behaviour (pp. 
 112-136). 
Levy, R., & Goldman-Rakic, P. S. (2000). Segregation of working memory functions within the 
 dorsolateral prefrontal cortex. Experimental Brain Research. Experimentelle Hirnforschung. 
 Expérimentation Cérébrale, 133(1), 23-32. 
Lhermitte, f. (1983). “Utilization behaviour” and its relation to lesions of the frontal lobes. Brain, 
 106(2), 237 -255.  
Li, S., Ostwald, D., Giese, M., & Kourtzi, Z. (2007). Flexible Coding for Categorical Decisions in the 
 Human Brain. J. Neurosci., 27(45), 12321-12330.  
242 
Libet, B., Gleason, C. A., Wright, E. W., & Pearl, D. K. (1983). Time of conscious intention to act in 
 relation to onset of cerebral activity (readiness-potential). The unconscious initiation of a 
 freely voluntary act. Brain: A Journal of Neurology, 106 (Pt 3), 623-642. 
Lien, M.-C., & Ruthruff, E. (2004). Task Switching in a Hierarchical Task Structure: Evidence for the 
 Fragility of the Task Repetition Benefit. Journal of Experimental Psychology: Learning, 
 Memory, and Cognition, 30(3), 697-713. 
Logan, G. D., & Gordon, R. D. (2001). Executive control of visual attention in dual-task situations. 
 Psychological Review, 108(2), 393-434. 
Louro, M.J.S., Pieters, R., & Zeelenberg, M. (2007). Dynamics of multiple goal pursuit. Journal of 
Personality and Social Psychology, 93(2), 174-193. 
Lumer, E. D., & Rees, G. (1999). Covariation of activity in visual and prefrontal cortex associated 
 with subjective visual perception. Proceedings of the National Academy of Sciences of the 
 United States of America, 96(4), 1669-1673. 
Luria, A. R. (1966). Higher cortical functions in man. Oxford, England: Basic Books. 
MacDonald, A. W., Cohen, J D, Stenger, V. A., & Carter, C. S. (2000). Dissociating the role of the 
 dorsolateral prefrontal and anterior cingulate cortex in cognitive control. Science (New York, 
 N.Y.), 288(5472), 1835-1838. 
Marklund, P., Fransson, P., Cabeza, Roberto, Petersson, K. M., Ingvar, M., & Nyberg, Lars. (2007). 
 Sustained and Transient Neural Modulations in Prefrontal Cortex Related to Declarative 
 Long-Term Memory, Working Memory, and Attention. Cortex, 43(1), 22-37.  
Mason, M. F., Norton, M. I., Van Horn, J. D., Wegner, D. M., Grafton, S. T., & Macrae, C. N. (2007). 
Wandering Minds: The Default Network and Stimulus-Independent Thought. Science, 
315(5810), 393–395.  
 
Mayer, J. S., Roebroeck, A., Maurer, K., & Linden, D. E. J. (2009). Specialization in the default 
 mode: Task-induced brain deactivations dissociate between visual working memory and 
 attention. Human Brain Mapping, NA-NA.  
Mesulam, M. M. (1999). Spatial attention and neglect: parietal, frontal and cingulate contributions 
 to the mental representation and attentional targeting of salient extrapersonal events. 
 Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 
 354(1387), 1325-1346.  
Mikami, A., Ito, S., & Kubota, K. (1982). Visual response properties of dorsolateral prefrontal 
 neurons during visual fixation task. Journal of Neurophysiology, 47(4), 593-605. 
Milea, D., Lehéricy, S., Rivaud-Péchoux, S., Duffau, H., Lobel, E., Capelle, L., Marsault, C., et al. 
 (2003). Antisaccade deficit after anterior cingulate cortex resection. Neuroreport, 14(2), 283-
 287.  
243 
Miller, E K, & Cohen, J D. (2001). An integrative theory of prefrontal cortex function. Annual 
 Review of Neuroscience, 24, 167-202.  
Miller, G. (1960). Plans and the structure of behavior. New York: Holt. 
B Milner (1964).  Some effects of frontal lobectomy in man. - The frontal granular cortex and 
behavior, 1964 
Mirabella, G., Bertini, G., Samengo, I., Kilavik, B. E., Frilli, D., Della Libera, C., & Chelazzi, L. 
 (2007). Neurons in area V4 of the macaque translate attended visual features into 
 behaviorally relevant categories. Neuron, 54(2), 303-318.  
Mizobuchi, K., Takahasi, N., & Ajima, A. (2011). Utilization and imitation behavior following 
 right parietotemporal lesions. Rinshō Shinkeigaku = Clinical Neurology, 51(5), 350-353. 
Monti, M. M., Vanhaudenhuyse, A., Coleman, M. R., Boly, M., Pickard, J. D., Tshibanda, L., Owen, 
 A. M., et al. (2010). Willful Modulation of Brain Activity in Disorders of Consciousness. 
 New England Journal of Medicine, 362(7), 579-589.  
Nachev, P., Kennard, C., & Husain, M. (2008). Functional role of the supplementary and pre-
 supplementary motor areas. Nature Reviews Neuroscience, 9(11), 856-869.   
Norman, D., & Shallice, T. (1986). Attention to Action: Willed and Automatic Control of Behavior. 
 In R. J. Davidson (Ed.), Consciousness and Self-regulation: Advances in Research and 
 Theory (Vol. 4, pp. 1-18). Plenum. 
O’Reilly, R. C. (2010). The What and How of prefrontal cortical organization. Trends in 
 Neurosciences, 33(8), 355-361.  
Owen, A. M., Coleman, M. R., Boly, M., Davis, M. H., Laureys, S., & Pickard, J. D. (2006). Detecting
 Awareness in the Vegetative State. Science, 313(5792), 1402.   
Pashler, H. E. (1999). The Psychology of Attention. MIT Press. 
di Pellegrino, G., & Wise, S P. (1991). A neurophysiological comparison of three distinct regions of 
 the primate frontal lobe. Brain: A Journal of Neurology, 114 ( Pt 2), 951-978. 
Perret, E. (1974). The left frontal lobe of man and the suppression of habitual responses in verbal 
 categorical behaviour. Neuropsychologia, 12(3), 323-330. 
De Pisapia, N., Slomski, J. A., & Braver, T. S. (2007). Functional Specializations in Lateral  Prefrontal 
 Cortex Associated with the Integration and Segregation of Information in Working 
 Memory. Cereb. Cortex, 17(5), 993-1006.  
Platt, M. L., & Glimcher, P. W. (1999). Neural correlates of decision variables in parietal cortex. 
 Nature, 400(6741), 233-238.  
244 
Posner, M., Walker, J., Friedrich, F., & Rafal, R. (1984). Effects of parietal injury on covert orienting of 
 attention. The Journal of Neuroscience, 4(7), 1863 -1874. 
Povel, D.-J., & Collard, R. (1982). Structural factors in patterned finger tapping. Acta Psychologica, 
 52(1-2), 107-123.  
Procyk, E., & Joseph, J. P. (2001). Characterization of serial order encoding in the monkey anterior 
 cingulate sulcus. European Journal of Neuroscience, 14(6), 1041-1046.  
Pujol, J., Vendrell, P., Deus, J., Junqué, C., Bello, J., MartI[combining acute accent]-Vilalta, J. L., & 
 Capdevila, A. (2001). The Effect of Medial Frontal and Posterior Parietal Demyelinating 
 Lesions on Stroop Interference. NeuroImage, 13(1), 68-75.  
Quintana, J., & Fuster, J M. (1992). Mnemonic and predictive functions of cortical neurons in a 
 memory task. Neuroreport, 3(8), 721-724. 
Race, E. A., Shanker, S., & Wagner, A. D. (2009). Neural Priming in Human Frontal Cortex: 
 Multiple Forms of Learning Reduce Demands on the Prefrontal Executive System. Journal 
 of Cognitive Neuroscience, 21(9), 1766-1781.  
Raichle, Marcus E. (2003). Functional brain imaging and human brain function. The Journal of 
 Neuroscience: The Official Journal of the Society for Neuroscience, 23(10), 3959-3962. 
Rao, S. C., Rainer, G, & Miller, E K. (1997). Integration of what and where in the primate prefrontal 
 cortex. Science (New York, N.Y.), 276(5313), 821-824. 
Reed, E. S., Montgomery, M., Palmer, C., & Pittenger, J. (1995). Method for Studying the Invariant 
 Knowledge Structure of Action: Conceptual Organization of an Everyday Action. The 
 American Journal of Psychology, 108(1), 37-65.  
Restle, F., & Burnside, B. L. (1972). Tracking of serial patterns. Journal of Experimental Psychology, 
 95(2), 299-307. doi:37/h0033619 
Reuss, H., Kiesel, A., Kunde, W., & Hommel, B. (n.d.). Unconscious activation of task sets. 
 Consciousness and Cognition, In Press, Corrected Proof.  
Reverberi, C., Görgen, K., & Haynes, John-Dylan. (2011). Compositionality of Rule 
 Representations in Human Prefrontal Cortex. Cerebral Cortex.  
Reynolds, J. R., West, R., & Braver, T. (2009). Distinct Neural Circuits Support Transient and 
 Sustained Processes in Prospective Memory and Working Memory. Cerebral Cortex, 19(5), 
 1208 -1221. 
Rigotti, M., Ben Dayan Rubin, D., Morrison, S. E., Salzman, C. D., & Fusi, S. (2010). Attractor 
 concretion as a mechanism for the formation of context representations. NeuroImage, 
52(3),  833-847.  
245 
Robertson, I. H., Manly, T., Beschin, N., Daini, R., Haeske-Dewick, H., Hömberg, V., Jehkonen, M., 
 et al. (1997). Auditory sustained attention is a marker of unilateral spatial neglect. 
 Neuropsychologia, 35(12), 1527-1532.  
Roca, M., Parr, A., Thompson, Russell, Woolgar, A., Torralva, T., Antoun, N., Manes, F., et al. 
 (2010). Executive function and fluid intelligence after frontal lobe lesions. Brain, 133(1),  234 
 -247.  
Rogers, B. P., Avery, S. N., & Heckers, S. (2010). Internal representation of hierarchical sequences 
 involves the default network. BMC Neuroscience, 11, 54.  
Romo, R., Brody, C. D., Hernández, A., & Lemus, L. (1999). Neuronal correlates of parametric 
 working memory in the prefrontal cortex. Nature, 399(6735), 470-473. 
Rosenbaum, D. A., Kenny, S. B., & Derr, M. A. (1983). Hierarchical control of rapid movement 
 sequences. Journal of Experimental Psychology: Human Perception and Performance, 9(1), 
 86-102.  
Rowe, J. B., Eckstein, D., Braver, T., & Owen, A. M. (2008). How Does Reward Expectation Influence
 Cognition in the Human Brain? Journal of Cognitive Neuroscience, 20(11), 1980- 1992.  
Russell, S., & Norvig, P. (2002). Artificial Intelligence: A Modern Approach (2nd Edition). Prentice 
 Hall.  
Rylander, G., (1939). Personality changes after operations on the frontal lobes: a clinical study of 32 
cases.  - E. Munksgaard 
Saalmann, Y. B., Pigarev, I. N., & Vidyasagar, T. R. (2007). Neural mechanisms of visual attention: 
 how top-down feedback highlights relevant locations. Science, 316(5831), 1612. 
Saga, Y., Iba, M., Tanji, J., & Hoshi, E. (2011). Development of Multidimensional Representations of 
 Task Phases in the Lateral Prefrontal Cortex. The Journal of Neuroscience, 31(29), 10648  -
 10665.  
Sakai, K. (2008). Task Set and Prefrontal Cortex. Annual Review of Neuroscience, 31(1), 219-245.  
Sakai, K., & Passingham, R. E. (2006). Prefrontal Set Activity Predicts Rule-Specific Neural Processing 
 during Subsequent Cognitive Performance. The Journal of Neuroscience, 26(4), 1211 -1218. 
Savine, A. C., & Braver, T. S. (2010). Motivated cognitive control: reward incentives modulate 
 preparatory neural activity during task-switching. The Journal of Neuroscience: The Official 
 Journal of the Society for Neuroscience, 30(31), 10294-10305.   
Schall, U., Johnston, P., Todd, J., Ward, P. B., & Michie, P. T. (2003). Functional neuroanatomy of 
 auditory mismatch processing: an event-related fMRI study of duration-deviant oddballs. 
 NeuroImage, 20(2), 729-736. 
246 
Schank, R., & Abelson, R. (1977). Script, Plans, Goals and Understanding: An Inquiry into Human 
 Knowledge Structures. Lawrence Erlbaum Associates. 
Schneider, D. W., & Logan, G. D. (2007). Retrieving information from a hierarchical plan. Journal of 
 Experimental Psychology. Learning, Memory, and Cognition, 33(6), 1076-1091.   
Schneider, D. W., & Logan, G. D. (2006). Hierarchical Control of Cognitive Processes: Switching 
 Tasks in Sequences. Journal of Experimental Psychology: General, 135(4), 623-640.   
Seeley, W. W., Menon, V., Schatzberg, A. F., Keller, J., Glover, G. H., Kenna, H., Reiss, A. L., et al. 
(2007). Dissociable Intrinsic Connectivity Networks for Salience Processing and Executive 
Control. J. Neurosci., 27(9), 2349–2356.  
 
Serences, J. T., & Boynton, G. M. (2007). Feature-Based Attentional Modulations in the Absence of 
 Direct Visual Stimulation. Neuron, 55(2), 301-312.  
Shaw, G. L., & McGaugh, J. L. (1990). Neurobiology of learning and memory. World Scientific. 
Shidara, M., & Richmond, Barry J. (2002). Anterior cingulate: single neuronal signals related to 
 degree of reward expectancy. Science (New York, N.Y.), 296(5573), 1709-1711.  
Shidara, M., Mizuhiki, T., & Richmond, Barry J. (2005). Neuronal firing in anterior cingulate 
 neurons changes modes across trials in single states of multitrial reward schedules. 
 Experimental Brain Research, 163(2), 242-245.  
Shimamura, Arthur P., Janowsky, Jeri S., Squire, Larry R. (1991). What is the role of frontal lobe 
 damage in memory disorders? Frontal lobe function and dysfunction. Levin, Harvey S. (Ed); 
 Eisenberg, Howard M. (Ed); Benton, Arthur L. (Ed), (1991). Frontal lobe function and 
 dysfunction, (pp. 173-195). New York, NY, US: Oxford University Press, xv, 427 pp. 
Sigala, N., Kusunoki, M., Nimmo-Smith, I., Gaffan, D., & Duncan, John. (2008). Hierarchical 
 coding for sequential task events in the monkey prefrontal cortex. Proceedings of the 
 National Academy of Sciences, 105(33), 11969 -11974. 
Singh, K. D., & Fawcett, I. P. (2008). Transient and linearly graded deactivation of the human 
 default-mode network by a visual detection task. NeuroImage, 41(1), 100-112.   
Sirigu, A, Cohen, L., Zalla, T., Pradat-Diehl, P., Van Eeckhout, P., Grafman, J, & Agid, Y. (1998). 
 Distinct frontal regions for processing sentence syntax and story grammar. Cortex; a 
 Journal Devoted to the Study of the Nervous System and Behavior, 34(5), 771-778. 
Sirotin, Y. B., & Das, A. (2009). Anticipatory haemodynamic signals in sensory cortex not predicted 
 by local neuronal activity. Nature, 457(7228), 475-479.  
Smania, N., Martini, M. C., Gambina, G., Tomelleri, G., Palamara, A., Natale, E., & Marzi, C. A. 
 (1998). The spatial distribution of visual attention in hemineglect and extinction patients. 
 Brain, 121(9), 1759 -1770. 
247 
Smith, R., Keramatian, K., & Christoff, K. (2007). Localizing the rostrolateral prefrontal cortex at the 
 individual level. NeuroImage, 36(4), 1387-1396.  
Sohn, J.-W., & Lee, D. (2007). Order-Dependent Modulation of Directional Signals in the 
 Supplementary and Presupplementary Motor Areas. The Journal of Neuroscience, 27(50), 
 13655 -13666.  
Soon, Chun Siong, Brass, Marcel, Heinze, H.-J., & Haynes, John-Dylan. (2008). Unconscious 
 determinants of free decisions in the human brain. Nat Neurosci, 11(5), 543-545.   
Spearman, C. E. (1927). The abilities of man: their nature and measurement. The Macmillan 
Company. 
Sridharan, D., Levitin, D. J., & Menon, V. (2008). A critical role for the right fronto-insular cortex in 
 switching between central-executive and default-mode networks. Proceedings of the 
 National Academy of Sciences, 105(34), 12569 -12574. 
Sridharan, D., Levitin, D. J., Chafe, C. H., Berger, J., & Menon, V. (2007). Neural Dynamics of 
 Event Segmentation in Music: Converging Evidence for Dissociable Ventral and Dorsal 
 Networks. Neuron, 55(3), 521-532.  
Stevens, A. A., Skudlarski, P., Gatenby, J. C., & Gore, J. C. (2000). Event-related fMRI of auditory and 
 visual oddball tasks. Magnetic Resonance Imaging, 18(5), 495-502.  
Stiers, P., Mennes, M., & Sunaert, S. (2010). Distributed task coding throughout the multiple 
 demand network of the human frontal-insular cortex. NeuroImage, 52(1), 252-262. 
Stokes, M., Thompson, R., Cusack, R., & Duncan, J. (2009). Top-down activation of shape-specific 
 population codes in visual cortex during mental imagery. Journal of Neuroscience, 29(5), 
 1565-1572. 
Stroop, J. R. (1935). The Basis of Ligon’s Theory. The American Journal of Psychology, 47(3), 499-
 504.  
Stuphorn, V., & Schall, J. D. (2006). Executive control of countermanding saccades by the 
 supplementary eye field. Nat Neurosci, 9(7), 925-931. 
Summerfield, C., Egner, T., Greene, M., Koechlin, E., Mangels, J., & Hirsch, J. (2006). Predictive 
 codes for forthcoming perception in the frontal cortex. Science, 314(5803), 1311. 
Swick, D., & Turken, A. U. (2002). Dissociation between conflict detection and error monitoring in 
 the human anterior cingulate cortex. Proceedings of the National Academy of Sciences, 
 99(25), 16354 -16359.  
Tamber-Rosenau, B. J., Esterman, M., Chiu, Y.-C., & Yantis, S. (2011). Cortical mechanisms of 
 cognitive control for shifting attention in vision and working memory. Journal of Cognitive 
 Neuroscience, 23(10), 2905-2919. 
248 
Taren, A. A., Venkatraman, V., & Huettel, S. A. (2011). A Parallel Functional Topography between 
 Medial and Lateral Prefrontal Cortex: Evidence and Implications for Cognitive Control. The 
 Journal of Neuroscience, 31(13), 5026 -5031.  
Thompson-Schill, S.L., D'Esposito, M.D., Aguirre, G.K. & Farah, M.J. (1997). Role of left inferior 
prefrontal cortex in retrieval of semantic knowledge: A reevaluation. Proceedings of the 
National Academy of Sciences USA , 94: 14792–14797 
Thompson, Russell, & Duncan, John. (2009). Attentional modulation of stimulus representation in 
 human fronto-parietal cortex. NeuroImage, 48(2), 436-448.   
Todd, J. J., Fougnie, D., & Marois, R. (2005). Visual Short-Term Memory Load Suppresses Temporo-
 Parietal Junction Activity and Induces Inattentional Blindness. Psychological Science, 
 16(12), 965 -972.  
Tsujimoto, S., Genovesio, A., & Wise, Steven P. (2010). Evaluating self-generated decisions in 
 frontal pole cortex of monkeys. Nat Neurosci, 13(1), 120-126.  
Uddin, L. Q., Clare Kelly, A. M., Biswal, B. B., Xavier Castellanos, F., & Milham, M. P. (2009). 
 Functional connectivity of default mode network components: Correlation, anticorrelation, 
 and causality. Human Brain Mapping, 30(2), 625-637.  
Uretzky, S., & Gilboa, A. (2010). Knowing your lines but missing your cue: rostral prefrontal 
 lesions impair prospective memory cue detection, but not action-intention superiority. 
 Journal of Cognitive Neuroscience, 22(12), 2745-2757.  
Velanova, K., Jacoby, L. L., Wheeler, M. E., McAvoy, M. P., Petersen, S. E, & Buckner, R. L. (2003). 
 Functional–anatomic correlates of sustained and transient processing components 
 engaged during controlled retrieval. The Journal of neuroscience, 23(24), 8460. 
Vendrell, P., Junqué, C., Pujol, J., Jurado, M. A., Molet, J., & Grafman, Jordan. (1995). The role of 
 prefrontal regions in the Stroop task. Neuropsychologia, 33(3), 341-352. 
Volle, E., Gonen-Yaacovi, G., de Lacy Costello, A., Gilbert, Sam J, & Burgess, P. W. (2011). The role of 
 rostral prefrontal cortex in prospective memory: A voxel-based lesion study. 
 Neuropsychologia, 49(8), 2185-2198.  
Wallis, J D, Anderson, K. C., & Miller, E K. (2001). Single neurons in prefrontal cortex encode 
 abstract rules. Nature, 411(6840), 953-956.  
Wallis, Jonathan D, & Miller, Earl K. (2003). From rule to response: neuronal processes in the 
 premotor and prefrontal cortex. Journal of Neurophysiology, 90(3), 1790-1806.   
Watanabe, M. (1990). Prefrontal unit activity during associative learning in the monkey. 
 Experimental Brain Research. Experimentelle Hirnforschung. Expérimentation Cérébrale, 
 80(2), 296-309. 
249 
Watanabe, M. (1992). Frontal units of the monkey coding the associative significance of visual and 
 auditory stimuli. Experimental Brain Research. Experimentelle Hirnforschung. 
 Expérimentation Cérébrale, 89(2), 233-247. 
Watanabe, M. (1996). Reward expectancy in primate prefrontal neurons. Nature, 382(6592), 629-
 632.  
Weissman, D. H., Roberts, K. C., Visscher, K. M., & Woldorff, M. G. (2006). The neural bases of 
momentary lapses in attention. Nat Neurosci, 9(7), 971–978.  
 
White, I. M., & Wise, S P. (1999). Rule-dependent neuronal activity in the prefrontal cortex. 
 Experimental Brain Research. Experimentelle Hirnforschung. Expérimentation Cérébrale, 
 126(3), 315-335. 
Womelsdorf, T., Schoffelen, J. M., Oostenveld, R., Singer, W., Desimone, R., Engel, A. K., & Fries, P. 
 (2007). Modulation of neuronal interactions through neuronal synchronization. Science, 
 316(5831), 1609. 
Woolgar, A., Parr, A., Cusack, R., Thompson, R., Nimmo-Smith, I., Torralva, T., Roca, M., et al. (2010). 
Fluid intelligence loss linked to restricted regions of damage within frontal and parietal 
cortex. Proceedings of the National Academy of Sciences of the United States of America, 
107(33), 14899–14902.  
 
Woolgar, A., Thompson, Russell, Bor, D., & Duncan, John. (2011). Multi-voxel coding of stimuli, 
 rules, and responses in human frontoparietal cortex. NeuroImage, 56(2), 744-752. 
Yeung, N., Nystrom, L. E., Aronson, J. A., & Cohen, Jonathan D. (2006). Between-task competition 
 and cognitive control in task switching. The Journal of Neuroscience: The Official Journal  of 
 the Society for Neuroscience, 26(5), 1429-1438.  
Yoshida, W., Funakoshi, H., & Ishii, S. (2010). Hierarchical rule switching in prefrontal cortex. 
 NeuroImage, 50(1), 314-322.  
Young, M., & Yamane, S. (1992). Sparse population coding of faces in the inferotemporal cortex. 
 Science, 256(5061), 1327 -1331.  
Zacks, J. M., Braver, T. S., Sheridan, M. A., Donaldson, D. I., Snyder, Abraham Z., Ollinger, J. M., 
 Buckner, Randy L., et al. (2001). Human brain activity time-locked to perceptual event 
 boundaries. Nat Neurosci, 4(6), 651-655.