Consciousness and Cognition 21 (2012) 695–703

Contents lists available at ScienceDirect

Consciousness and Cognition

journal homepage: www.elsevier .com/locate /concog

Neural correlates of temporality: Default mode variabilityand temporal awareness q

Dan LloydDepartment of Philosophy/Program in Neuroscience Trinity College, Connecticut 300 Summit Street, Hartford, CT 06106 USA

a r t i c l e i n f o

Article history:Received 10 October 2010Available online 21 March 2011

Keywords:ConsciousnessBrainTimeTemporalityNeural networksSupport vector machinesfMRI

1053-8100/$ - see front matter � 2011 Elsevier Incdoi:10.1016/j.concog.2011.02.016

q This article is part of a special issue of this jourE-mail address: dan.lloyd@trincoll.edu

a b s t r a c t

The continual background awareness of duration is an essential structure of consciousness,conferring temporal extension to the many objects of awareness within the evanescentsensory present. Seeking the possible neural correlates of ubiquitous temporal awareness,this article reexamines fMRI data from off-task ‘‘default mode’’ (DM) periods in 25 healthysubjects studied by Grady et al. (‘‘Age-related Changes in Brain Activity across the AdultLifespan,’’ Journal of Cognitive Neuroscience 18(2), 2005). ‘‘Brain reading’’ using support vec-tor machines detected information specifying elapsed time, and further analysis specifieddistributed networks encoding implicit time. These networks fluctuate; none are continu-ously active during DM. However, the aggregate regions of greatest variability closelyresemble the default mode network. It appears that the default mode network has animportant role as a state-dependent monitor of temporality.

� 2011 Elsevier Inc. All rights reserved.

1. Introduction: timing and temporality

Time is important in human behavior, cognition, and consciousness. Yet the role of time is differently conceived by dif-ferent disciplines. In cognitive science, time is a perceptual dimension exploited as needed in various and distinct time-dependent tasks, such as temporal order and simultaneity judgments, or duration estimation and reproduction tasks. Timein these contexts is attended time, time in the spotlighted foreground of perception and behavior. As such time is one prom-inent stimulus feature among many. In contrast, in philosophical phenomenology time is the foundation of every aspect ofconsciousness. It is a structural feature of consciousness, among ‘‘those properties that are common to most or all consciousexperiences’’ (Seth, 2009). This is time as temporality, ‘‘the infrastructure of reality’’ (Zahavi, 1999). Although capacities forjudging time and coordinating behavior in time are essential to human cognition, timing is nonetheless only a subset of phe-nomenal temporality. The latter concept, as developed in phenomenology, encompasses time as a fundamental dimension ofall percepts and all behavior, a dimension which may or may not be a center of explicit attention, but which is nonetheless anexperienced feature of all mental acts.

William James explicated phenomenal time as temporality the context of his most famous metaphor:

The knowledge of some other part of the stream [of consciousness], past or future, near or remote, is always mixed in with ourknowledge of the present thing. . . . [A]ll our concrete states of mind are representations of objects with some amount ofcomplexity. Part of the complexity is the echo of the objects just past, and, in a less degree, perhaps, the foretaste of thosejust to arrive. Objects fade out of consciousness slowly. If the present thought is of A B C D E F G, the next one will be of B CD E F G H, and the one after that of C D E F G H I – the lingerings of the past dropping successively away, and the incomings

. All rights reserved.

nal on Standing on the Verge: Lessons and Limits from the Empirical study of Consciousness.

696 D. Lloyd / Consciousness and Cognition 21 (2012) 695–703

of the future making up the loss. These lingerings of old objects, these incomings of new, are the germs of memory andexpectation, the retrospective and the prospective sense of time. They give that continuity to consciousness withoutwhich it could not be called a stream. James (1890, chap. 15) italics in original.

In this passage and throughout The Principles of Psychology, James presents two main features of experienced temporality:Our experience of the present is compounded with an awareness of experiences just past and intimations of experiencesabout to occur, and this temporal spread is constitutive of every state of consciousness. Temporality is also central in theworks of Edmund Husserl (esp. Husserl, 1966 (1928)), a close reader of James. Husserl elaborated the structures of tempo-rality, providing them with details not worked out in James, and arguing for their necessity. Consciousness would be incon-ceivable without temporality, according to Husserl. His account of temporal awareness has remained a centerpiece ofContinental phenomenology ever since, and will be the philosophical frame of reference in this paper.1

Overall, Husserl’s phenomenology is expansive, offering an intricate picture of the contents and dynamics of conscious-ness. Perception, for Husserl as for James, is not limited to sensory properties of the occurrent stimulus. Sensuous ‘‘givens’’are always supplemented by non-sensory ‘‘apprehensions,’’ and these together comprise appearances. The non-sensorycontents of consciousness are explicit, and may or may not be attended. One of Husserl’s examples is a solid cube, whichpresents some of its faces but hides others; but the hidden side is nonetheless in our awareness (even if it is indefinite) withthe same presence as the visible faces. Moreover, for Husserl the hidden faces are not just in awareness, but perceived. Otherexamples include the familiar cases of seeing-as, where identity of sensory features between two objects is compatible withdifferent interpretations. The objects appear differently, despite their visible indiscernibility. In short, perceptual awarenessincludes non-sensory features.

The ubiquitous exemplar of non-sensory perception is indeed temporality, which Husserl elaborates in two principalways, speaking of the experience of flow (or flux) and of the experience of temporal structure (the temporal context ofevents). All conscious experiences involve both flow and structure. Experiences, whether perceptual or purely endogenous,do not simply occur and then disappear. Rather, they occur and leave their traces in ‘‘retention,’’ like a comet’s tail. As PaulMcCartney lands on ‘‘Jude,’’ the ‘‘Hey’’ is retained though no longer sensed. Likewise, as ‘‘Jude’’ sounds, we anticipatesomething to follow (‘‘Don’t make it bad,’’ if one knows the song, or something less definite) – this is protention.Combined with protention and retention, the immediate sensation is called the ‘‘primal impression.’’ This is the tripartitestructure of consciousness at every moment, and at every moment this entire structure is sinking (flowing) into the past.2

To perceive a song requires that its notes be held in awareness in a temporal structure; no occurrent sensation of a particular

1 James and Husserl may differ over James’ concept of the ‘‘specious present,’’ a brief temporal window within which moments of experience are co-present,although identifiable as about-to-be, right-now, and just-past:

[W]e are constantly conscious of a certain duration – the specious present – varying in length from a few seconds to probably not more than a minute,and that this duration (with its content perceived as having one part earlier and the other part later) is the original intuition of time. (1890, chap. 15)

Although Husserl did not refer to the specious present, his account and that of James seem roughly congruent. However, the details of the speciouspresent are elusive. For example, James writes:

When many impressions follow in excessively rapid succession in time, although we may be distinctly aware that they occupy some duration, and arenot simultaneous, we may be quite at a loss to tell which comes first and which last; or we may even invert their real order in our judgment. (1890,chap. 15)

The interval within which this order confusion can occur is his operational definition of the specious present. Yet in other passages, the specious presentembraces its elements in an unambiguous temporal order. For example:

[T]he practically cognized [specious] present is no knife-edge, but a saddle-back, with a certain breadth of its own on which we sit perched, and from whichwe look in two directions into time. The unit of composition of our perception of time is a duration, with a bow and a stern, as it were – a rearward – and aforward-looking end. It is only as parts of this duration-block that the relation of succession of one end to the other is perceived. (1890, chap. 15)

Husserl would probably agree with this formulation, but even more so with the following, in which James anticipates a geometrical representation of experienced time that Husserl then makes concrete in his ‘‘diagram of time’’ in section 10 of his Lectures on the Phenomenology of Inner-time Consciousness(Husserl1966 (1928)). James writes:

I have shown, at the outset of the article, that what is past, to be known as past, must be known with what is present, and during the ’present’ spot oftime. . . If we represent the actual time-stream of our thinking by an horizontal line, the thought of the stream or of any segment of its length, past,present, or to come, might be figured in a perpendicular raised upon the horizontal at a certain point. The length of this perpendicular stands for a cer-tain object or content, which in this case is the time thought of, and all of which is thought of together at the actual moment of the stream upon whichthe perpendicular is raised. . . (James, 1890, Chap. 15)

However, for present purposes, these interpretive issues can be left open.2 Recently, Husserl’s ‘‘retentionalist’’ position has been contrasted with ‘‘extensional’’ models of the ubiquitous awareness of time (Dainton, 2000, 2008). The

extensional present features a temporally smeared experience of perception in which immediate past and future are included within the subjective ‘‘now.’’ Thetwo views differ in the richness and temporal extent of the contents of present, occurrent consciousness. Retentional theories envision present consciousness asthe exhaustive theater of temporal awareness; every moment of conscious awareness coordinates with past and future by virtue of its place in the retentionalstructure. Extensional theories require only enough retention and anticipation for each present moment to overlap with its temporal neighbors and therebyestablish continuity in the flow of experience. This paper assumes a retentionalist, Husserlian view of consciousness.

D. Lloyd / Consciousness and Cognition 21 (2012) 695–703 697

tone is the percept of the song. But this is no less true for a static percept. The immobility of a boulder is also perceivedtemporally. It’s there (as an object of awareness) now, and still there a second later, and still there many seconds after that.Thus, Husserl’s expansive concept of perception extends into the past and future, as these are realized in retention andprotention.

In short, for Husserl duration is experienced as a changing (non-sensory) element of every percept and indeed of any stateof consciousness whatsoever. Duration is an ever-lengthening dimension of present intentional objects, but it continues toaccrue after the occurrent percept has disappeared; then, it is the awareness of how much time has passed since the occur-rent phase of perception ended, or how far it has slipped into retention. Looking forward, duration appears as the anticipatedlength of time until a probable event occurs, a magnitude that is always shrinking.3 This awareness is ever-present, and thusoverlooked. But we encounter temporality when background expectations of duration are violated. For example, delays in con-versational turns are highly significant; apparent haste or tardiness in a reply is quickly read as an indicator of one’s attitudetoward the speaker or the exchange. This is common at various time scales, from short (‘‘Do you love me?’’ coupled with‘‘Of course,’’ as compared to ‘‘[pause] Of course.’’) to long, as in, for example, perceived slights in tardy email responses. One’sattention need not be cued to notice these timing anomalies. Thus, it seems that duration is continuously monitored at manyscales in a host of human activities.

For Husserl, temporality is a necessary structure of consciousness. It is presupposed in the very concept of humanexperience. We can only see objects as objects by simultaneously supposing they have hidden sides; we can onlyexperience situations as either static or changing by also maintaining a simultaneous awareness of their immediatelypreceding conditions. (Temporality is equally implicated in other psychological states – recollection, fantasy, etc.) ForHusserl, then, all intentional objects are temporal objects, experienced in the framework of protention/primal impres-sion/retention, a frame that is constantly sinking toward the past. Thus, time perception is not merely a species ofperception to be found among others, like color perception, tone perception, etc. Rather, it is an essential aspect ofall consciousness, giving it its characteristic thickness. We can readily update Husserl’s argument to bear on cognitivecapacities in general, with the observation that the most useful model of the perceptual world is not so much aconstruct of objects in scenes as trajectories through space/time (Grush, 2004). (Mastery of trajectory estimation confersobvious adaptive advantages.)

The explicit cognition of time is a lively area of research in cognitive neuroscience and functional neuroimaging. Tempo-rality, in contrast, is infrequently studied (Grush, 2004; Lloyd, 2002; Yoshimi, 2007). There are several reasons contributingto this comparative neglect. First, since temporal awareness is continuous but non-sensory, it is the ambient background toexplicit, articulated sensory experiences. Informal phenomenologies, of the sort that occupy brief chapters in books by phi-losophers in the analytic tradition, overlook temporality (along with other structural features of consciousness). This biastoward the sensory and atemporal present is also reflected in many aspects of 20th Century cognitive science. David Marr(1982), for example, classically regarded the problem of perception as the extraction of three-dimensional structure from theoccurrent retinal image, and a huge research literature in psychophysics and perception operates ‘‘in the present tense,’’addressing perception of present objects and events. These interests have been readily translated into experiments in func-tional neuroimaging.

This comparative neglect is not merely a contingent preference, however. Temporality eludes experiment by its very nat-ure. If the phenomenologists are right that temporality is a continuous aspect of awareness, it can’t be turned on and off byexperimental manipulation. Moreover, in any experiment in which time is a variable, subjects are likely to notice. Time, then,will move from background to foreground, and the general structural scaffold of temporality will yield to explicit timing.These obstacles are compounded by the constraints of neuroimaging, especially fMRI. ‘‘Raw’’ fMRI images reveal continuousmetabolic activity nearly everywhere in the brain, overlaid with a great deal of noise. Unlike other medical and scientificimagery, single functional images reveal little or nothing. Interpretation thus often rests on contrasts (to detect smallchanges between all-over activation patterns) and averages (to smooth away some of the noise). Both are inimical to detect-ing temporality. If temporality is continuous, it is present on both sides of any contrast, and won’t appear in a contrastiveimage. If, on the other hand, temporality involves change that is independent of the time course of conditions in an exper-iment, averaging will eliminate those changes along with the noise.

These considerations impose severe constraints on any scientific study of temporality, especially research based inneuroimaging. Nonetheless, this paper will re-examine data from one fMRI study as a case study in temporality (Grady,Springer, Hongwanishkul, McIntosh, & Winocur, 2006). We will consider several questions: (1) Can we demonstratethat specific images encode temporal information?4 (2) Can we identify regions of the brain that are particularlyinvolved in encoding temporal information? (3) Can we begin to characterize mechanisms supporting the awareness oftemporality?

3 The perception of duration could be a basis for the consciousness of succession and order for experienced events, which can be compared in their durationsas readily as they can be compared along many other dimensions. Duration also affords a perception of the passage of time. When retentional consciousness isladen with event-memories separated by short durations, time seems to flow rapidly. But when ‘‘not much is happening,’’ time drags, as both Husserl and Jamesdiscuss.

4 In this paper, ’’encoding’’ means ‘‘carrying information about features of a source’’ and ’’information’’ in turn is meant in its standard information-theoreticsense. (When the presence or absence of a signal affects the estimated probability of presence/absence of a feature of a source, the signal carries informationabout the source.) An alternative would be ’’representation,’’ but this term has multiple senses and remains contentious, especially with respect toconsciousness. ‘‘Encoding’’ will enable us to avoid a semantic side trip. ‘‘Temporal information’’ will be elaborated in the next section.

698 D. Lloyd / Consciousness and Cognition 21 (2012) 695–703

2. Encoding temporality

If temporality is a continuous structural component of consciousness then, in some sense, the contents of consciousnessat every moment include some form of temporal awareness. This entails (as Husserl realized) that all perception is in con-tinuous flux. Even if a scene is static, the experience of it is always changing as its intentional objects progressively extendtheir perceived duration. This in turn implies that the information encoding objects of perception is time-inflected. Even if noother stimulus property is changing, the brain activity that supports awareness must also support its temporal progression.

Lloyd (2002) reexamined five neuroimaging experiments involving several aspects of perception and cognition to confirmthat (in 89% of subjects) continuous change is a feature of fMRI image series. That is, as the lag between images increased,the measured differences between images also increased, whether this is measured as correlation between images or asEuclidean distance. Although this is consistent with a Husserlian conception of temporality, other processes might give riseto continuous ‘‘drift’’ in image series. What is needed, therefore, is a way to test for the presence of explicitly temporalinformation.

Because temporality is ubiquitous and continuous, it is not subject to experimental manipulation in normal, awake hu-man subjects. Nonetheless, if the Husserlian view is correct, temporal information must be present in the brain and neuro-images may, possibly, detect it. Helpfully, if temporality is a continuous structural feature of consciousness, it ought to bepresent in every conscious subject, regardless of the ‘‘official’’ task of interest. One might especially look for signs of tempo-rality during moments when subjects are performing no task, and their environment is not changing, or at least is withoutmarkers of passing time. Time may be unavoidably noticed by subjects in the noisy and confined space of a scanner, butwhen there is no need to judge or reproduce durations, temporality should appear in awareness similarly to its appearancesin other perceptual environments.

Functional MRI experiments typically contain many blocks off-task and without markers of time, often intervals when thebaseline images are collected. Cognition during these ‘‘off-task’’ intervals has been termed ‘‘default mode’’ (DM) activity,following the discovery that, regardless of the target tasks, the off-task brain exhibits spontaneous activity in a networkof regions that is quite stable from experiment to experiment (Buckner, Andrews-Hanna, & Schacter, 2008; Mason et al.,2007; McKiernan, Kaufman, Kucera-Thompson, & Binder, 2003; Raichle et al., 2001). The default network is generallydepicted as a network engaged in continuous activity throughout off-task intervals, but this may be an artifact of averagingthe images collected throughout the interval. Temporality, in contrast, implies an awareness of change, even in the absenceof changing tasks or contexts. At a minimum, one is aware that time is passing, despite the recurring sounds and unchangingvisual scene. Among the possible expressions of temporality, here we explore the encoding of elapsed time. That is, weask whether brain images include information that specifies how much time has elapsed during the default interval. Inan experiment where there are multiple off-task blocks, this can be generalized: Does the brain encode elapsed durationinformation in the same way at comparable time points in each DM interval?

Specifically, for a subject in this experiment, among the various temporal properties that might be in and out of con-sciousness is the awareness of duration. This can be variously characterized as the awareness that a task block has endedor as the awareness that another is approaching. But the landmark events need not be the block starting and ending points.In the Husserlian structure, subjects’ awareness of their environment includes a protention of events to come and a retentionof events just past, both constantly sliding toward to past. The whole structure moves at once, so the entire contents of con-sciousness are subject to the same continuous change as they slide into the past. At each moment, the timeline of salientevents, whatever it may be for each subject, is the landscape against which the now appears. The Husserlian hypothesis thusimplies that each momentary experience will have a different content than moments before or after, a difference in the slid-ing contents of retention and protention, even if every other aspect of the current consciousness is somehow fixed. In effect,these retentions and protentions are the landmarks of a temporal landscape, and therefore each momentary state of con-sciousness contains content that specifies its position on a timeline of subjective temporal experience. So, the brain mustencode that content as well. How, then, is this content to be detected?

The evidence bearing on this question will be indirect, using a ‘‘brain reading’’ method. Employing pattern recognitionmethods described below, we construct multivariate mathematical functions that can recognize how much time has passedfor each image from the DM off-task blocks. The success of this machine learning strategy (compared to appropriate controlconditions) provides evidence that the images indeed encode temporal information. Following this analysis a separate anal-ysis can approach the question of how this information is encoded.

The present analysis is based on data from Grady et al., ‘‘Age-related Changes in Brain Activity across the Adult Lifespan(Grady et al., 2006), archived at the fMRI Data Center (www.fmridc.org, fMRIDC Accession Number 2-2005-119CE). As thetitle implies, Grady was specifically interested in memory and cognition in different age groups. For present purposes, thefour tasks of the experiment are not of interest, but rather the default mode intervals that interweave among them. We ana-lyzed two runs of 384 s each in 25 subjects (20–87 years old). Each run comprised eight alternating blocks of tasks and base-line (DM) conditions, each 24 s long. During baseline conditions, subjects pushed a button each time a fixation crossappeared on screen. Each image series was preprocessed using Independent Component Analysis (ICA), a statistical methodfor reducing the dimensionality of data while preserving most of the variance (Calhoun, Adali, Pearlson, van Zijl, & Pekar,2002). ICA separated 20 components for each run; each component can be understood as a temporally coherent distributednetwork of activity, an ensemble of voxels that are active or inactive in unison. The figures below use individual and com-posite component images, presented against background anatomical images and outlines. Importantly, ICA is a ‘‘data driven’’

Fig. 1. Pattern recognition after SVM training, at each time point during off-task blocks. The X-axis shows elapsed time in seconds from the onset of eachoff-task interval. The Y-axis shows the percentage of sessions in which elapsed duration information could be recovered by SVM pattern detection aftertraining, as compared to control trials using surrogate data (p < 0.05, corrected for multiple comparisons). Four 24 s blocks were in the test set for eachsession.

D. Lloyd / Consciousness and Cognition 21 (2012) 695–703 699

method, which identifies activity without reference to the time-course of experimental conditions. While some componentsapproximate the experimental model, most do not. Whole-brain images were collected every 2.5 s; for the purpose of thisanalysis, component images were interpolated to produce a pseudo-image series separated by 1 s intervals. For further detailconcerning methods, see the original study (Grady et al., 2006).

The image series for each subject was used to train a series of pattern recognizers, using half of each image series fortraining and the remaining images for testing (that is, four default mode intervals for training and a different set of fourfor testing). Each subject data set was analyzed separately. The task for each detector was to identify the elapsed durationfrom the DM block onset to the current time point, using just the current image as input. (Elapsed duration was encodedusing a binary scheme, where images before the current time point were binned in one category, and images after the cur-rent time point were binned in another. For each time point, the pattern recognizer was trained to separate images beforethat point from images after. This was repeated at each time point.) Success in this training implies that the pattern recog-nizers detect information in the images that encodes the relative elapsed time within each DM block. We tested this sepa-rately for each of the 24 possible serial positions, using Support Vector Machines (SVMs) (Cox & Savoy, 2003; Haynes & Rees,2006; LaConte, Strother, Cherkassky, Anderson, & Hu, 2005; Ma, Zhao, & Ahalt, 2002; Norman, Polyn, Detre, & Haxby, 2006).Additionally, we encoded the basic distinction between task blocks (regardless of task) and DM blocks. To validate the train-ing results, we contrasted them with the same training regime using surrogate data generated by randomly permuting theoriginal image series and executing the same training and testing procedure (Schreiber & Schmitz, 1996, 2000). The point ofthis comparison was to control for ‘‘spurious success’’ due to the likelihood that some test outputs will match their targets bychance alone (Lloyd, 2002). (For example, if the relative proportion of one target value exceeds another, a pattern recognizercould simply produce that target value at all time points.) At each time point, we calculated a p value for the contrast be-tween the actual SVM output and a set of one hundred surrogate outputs. The figures below are based on these comparisons.Accordingly, they are statistical parametric maps in the temporal domain, or T-SPMs.

Fig. 1 displays the percentage of runs in our analysis in which serial position (elapsed time) could be recovered (p < 0.05,determined by a one-tail t-test, corrected for multiple comparisons), at each time point in the default mode blocks reservedfor testing (that is, the latter four of the eight DM blocks). In general, temporal information was recovered for most runs, withbetter detection after about five elapsed seconds. The figure thus confirms that temporal information is encoded in theensemble of independent components. Time in this situation was not implicated in any particular behavior during theseblocks, nor was time signaled by any environmental feature. (Subjects could count scanner pulses or their own rhythmic but-ton presses, but there would be little motivation to do so.) Rather, passing time is a spontaneous aspect of evolving brainactivity in these subjects, and in general is quite accurate as a relative measure of elapsed duration within each block.

3. Neural correlates of temporality

This result above naturally raises the question of which brain regions might be particularly involved in encoding temporalinformation, in this case information about elapsed time in each off-task time interval. Something in the distributed patternof component activity encodes elapsed time. The encoding could be embedded in a single component, activated at a differentintensity at each time point, or it could be encoded in distinct patterns involving all components, or it could be somewhere inthe middle of a spectrum from localized to distributed. To explore these possibilities, the contribution of each component to

700 D. Lloyd / Consciousness and Cognition 21 (2012) 695–703

temporal representation was calculated using a ‘‘leave-one-out’’ strategy (Chapelle & Vapnik, 2000; Opper & Winther, 2000),as follows: The SVM pattern learning analysis was repeated on each subject image series, however with each repetition onecomponent was omitted from the training and test set. That is, the analysis repeatedly used 19 of 20 components, recordingthe impact of omitting each component on the overall capacity of the image to represent elapsed time. (This impact wasmeasured as a change in p value from the 20-component baseline to each of the 19-component alternatives. The greatestincrease in p signaled the component whose omission had the greatest negative effect on learning performance. This com-ponent would accordingly be the largest contributor to the distributed pattern encoding temporal information.) The compo-nent of greatest impact was then removed, and the leave-one-out process repeated on the remaining 19 components,reducing them to 18, and so on until the six most important contributors to the encoding of temporal information had beenidentified.

The leave-one-out strategy offers a systematic way to explore the joint contribution of multiple components to the rep-resentation of information. The analysis revealed that, on average, three components generally accounted for the encoding oftemporal information. However, the intersubjective variation was large. After one component was removed, 92% of compo-nent series still achieved significant pattern detection (p < 0.05, corrected for multiple comparisons); minus 2 components:84%; minus 3 components: 68%; minus 4 components: 50%; minus 5: 36%; minus 6: 30%. Thus, no one component explainedthe recoverable temporal information, except in a few cases (4 runs out of 50). But neither is it the case that the informationwas spread evenly across all components: 35 out of 50 runs embedded the significant temporal information in six or fewercomponents. Fig. 2 displays aggregate results for all subjects, partitioning the default mode blocks into 83-s intervals. Foreach session, the ICA magnitudes of significant components at each time point were used as coefficients, and the significantcomponents combined as a single image representing the distributed brain regions active during that 3 s window. Note thatthe brain images combined in this figure were those found significant in the leave-one-out analysis; the figure does not showevolution of the default mode overall, just the evolution of areas involved in encoding temporal information. Significant vox-els were probed by comparing image series for all sessions. Dark areas in Fig. 2 represent brain areas with increased activity,relative to the mean, in all subjects (p < 0.0001). Light areas represent brain regions where activity was significantly dimin-ished, relative to the mean.

This ‘‘glass brain’’ presentation affords an overview of the dynamic evolution of brain activity relevant to temporalitythroughout the default mode blocks in the Grady study. In general, each aggregate image is more like its near neighbors thanthose at greater lags. But overall, the series displays broad changes over time, involving broadly distributed brain areas. Noregion is either continuously active or suppressed during the DM encoding of temporality.

Anatomical region identification is difficult in Fig. 2. Accordingly, Fig. 3A displays the standard deviation of the DM tem-porality image series, over time. That is, the figure shows areas of the brain where change in temporal information is great-est. Once again, individual variation is large, but some commonalities emerge, including changing activity in superior andmedial frontal regions, anterior and posterior cingulate, hippocampal regions, cerebellum, and others. While Fig. 3A may looklike a statistical parametric map of activated regions following an experimental time course, it is not. These are regionswhose changes encode temporal information during periods when no other external conditions are changing. With the pass-ing seconds during each DM block, these areas are waxing and waning in complex configurations, as implied by Fig. 2, above.Reflecting the broad distribution of the involved areas, this will be referred to as a Dynamic Temporality Network.

4. Temporal information and the default mode network

The Default Mode network generally encompasses the medial prefrontal cortex (MPFC), posterior cingulate cortex (PCC),inferior parietal cortex (IPC), lateral temporal cortex (LTC), and hippocampal regions, among others (Buckner et al., 2008;Mason et al., 2007; McKiernan et al., 2003; Raichle et al., 2001; Sheline et al., 2009). What we have called the DynamicTemporality Network (DTN) overlaps with the ‘‘traditional’’ Default network in several areas, most prominently the anteriorDM regions: MPFC, LTC, and hippocampal regions. Fig. 3B shows a more detailed image of DM components in the Grady et al.study. For each run, we determined the component whose time course most correlated with the sequence of off-task defaultmode blocks. The figure is a composite of these, showing the top 5% of voxels common to the subjects and runs studied.Fig. 3B generally fits within the outlines of a ‘‘classical’’ default mode network. More important, the DM network in this studycan be compared to the sites of dynamism associated with temporally specific information. This match is fairly close, save forthe weaker involvement of posterior areas (PCC and IPC; however, the DT network trends toward involvement of those areas,

Fig. 2. Dynamic evolution of temporal information during off-task time periods. Aggregate ‘‘glass brain’’ images showing independent components that aremost significantly involved in encoding temporal information (p < 0.0001), combined in 3-s blocks spanning the off-task intervals. Dark areas indicatepositive component activation; light areas indicate decreased activation.

A

B

Fig. 3. Changing regions underlying temporal information encoding, compared with aggregate default mode network in 25 subjects, 2 runs each. A.Standard deviation of dynamic change in temporal components in fifty runs. For each run (two per subject), voxels above the 95th percentile in standarddeviation were noted. The aggregate image here displays voxels above the 95th percentile from the first pass. Thus, voxels in the image represent the top0.25% of standard deviation values. Voxels in yellow fell in the top 0.25% standard deviation in one to fifteen runs. Voxels in red were in the top 0.25% inmore than 15 runs. B. Aggregate best-match default mode components in fifty runs, showing voxels above the 95th percentile, activated (red) anddeactivated (blue). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

D. Lloyd / Consciousness and Cognition 21 (2012) 695–703 701

just slightly below the threshold for Fig. 3A). The comparison suggests a functional role for the DM network that incorporatesthe function of the Dynamic Temporal Network, namely, the encoding of elapsed time.

Before considering the positive implications of this hypothesis, some potential objections should be entertained. SVMpattern learning worked in this analysis, and further analysis identified specific regions of the brain, a Dynamic TemporalNetwork, where variations were exploited by the pattern recognizers in order to recover temporal information from novelimages (i.e., images that were not part of the SVM training set). The conjecture of this paper is that the DTN is a ‘‘neuralcorrelate’’ of an aspect of experienced temporality, the awareness of duration. Specifically, dynamic changes in distributedpatterns of activity within the DTN encode durations (elapsed time) in each off-task period in this experiment. Objections tothis conclusion take two general forms: First, could the pattern recognizer be attuned to some other varying feature of thedata that is unrelated to cognition in general? Second, even if the DTN encodes temporal information, does this warrant thefurther conclusion that DTN activity is conscious?

First, by basing the analysis on off-task blocks, we already restrict consideration to regions that are active off-task, whichis the DM network itself. Accordingly (by this objection), temporality will be found in this network ‘‘by default.’’ In reply tothis objection, we note that the SVM analysis was not limited to DMN regions, and that the supposition that only DMN com-ponents are active in off-task blocks is false. Independent Component Analysis reveals complex oscillations in every compo-nent. Even the component most significantly correlated with the default mode is only weakly associated with it. In short,every brain region was surveyed to identify those regions from which temporal information could be extracted.

A further objection could challenge the identification of the Dynamic Temporal Network with an aspect of cognition. Themost general form of this objection would hold that any evolving process ‘‘encodes temporality’’ just because it is changingas time passes, and not due to any explicit cognition of time. By this objection, an SVM analysis should work, in principle, onany system that changes over time. SVM (holds the objection) casts its net too widely to identify exclusively cognitive pro-cesses. More specifically, it might be that the time-determining information driving SVM pattern detection is merely theintrinsic oscillation of the DMN (Greicius & Menon, 2004), and presumably not a correlate to any feature of consciousness.Greicius and Menon reported intrinsic DMN oscillations at approximately 0.005, 0.015, and 0.035 Hz, which oscillations wereunrelated to the task wave-form. Could the pattern recognizer succeed merely on the basis of these intrinsic oscillations?

This objection, both in general and in its more specific form, is undermined by the timing of the experiment itself. TheSVM pattern analysis is temporally ‘‘local’’ – what is recovered is elapsed time within each DM period. Specifically, fourDM blocks of 24 s each were used to train a network, which was tested against four additional DM blocks. For example,the network learned to recognize the general pattern of brain activity when 10 s had passed in each training set. It then usedthat learning to identify the same elapsed interval in the test DM blocks. Thus, the overall sequence of the image series can-not specify the patterns learned by the SVM. Rather, it is repetition of patterns (or at least pattern similarity) at different timepoints that drives the analysis. That leads to the second objection: Perhaps intrinsic oscillations, like those observed byGreicius and Menon, are sufficient for the SVM analysis to work. This is possible, but unlikely, as the intrinsic oscillationwould have to closely match the wave form of the experiment itself, the overall oscillation of task and rest, and remainentrained to this oscillation for more than 6 min. The more plausible explanation is that the experimental variable in factdetermines the brain activity that encodes elapsed duration.

But what of the more specific identification of the DTN with the conscious awareness of time? The alternative interpre-tation is that the DTN does encode elapsed time, but that information is not conscious. This is a general problem with anypurported neural correlate of consciousness. However, the Default Mode Network (largely overlapping with the DTN) hasbeen repeatedly identified with a wide variety of conscious experiences. It has been linked to an increase in task-indepen-dent thoughts (daydreaming) and inversely related to task difficulty (Mason et al., 2007; McKiernan et al., 2003; Sheline

702 D. Lloyd / Consciousness and Cognition 21 (2012) 695–703

et al., 2009). These ‘‘task independent thoughts’’ have been characterized as internal mental simulations, including planning,auto-biographical memory, ‘‘theory of mind’’ reasoning, and other forms of self-reflection (Addis, Wong, & Schacter, 2007;Amodio & Frith, 2006; Andreasen et al., 1995; Buckner et al., 2008; Greicius & Menon, 2004; Saxe & Kanwisher, 2003;Svoboda, McKinnon, & Levine, 2006).5 Alternatively, the DM network may serve to monitor the environment while focalattention is relaxed (Gilbert & Wilson, 2007; Gusnard & Raichle, 2001; Hahn, Ross, & Stein, 2007; Shulman et al., 1997).Temporal information does not exclude other functions expressed within a single complex state, even if only part of the brainis considered. The temporal analysis here does not contradict the other research on the cognitive functions of the default mode.Rather, temporality reconciles the tension between inner mentation and environmental monitoring. In this experiment, elapsedtime is not signaled by the environment, entailing that it must arise endogenously. But it is a property of environmental objects– an endogenous construct applied to the outside world.

Could some other DM function, like daydreaming, fully characterize the brain activity in the DTN? Here the most tellingconsiderations are already in Fig. 1. The SVM analysis is not indexed to the total temporal sweep of the experiment, but in-stead assumes that each off-task block will be accompanied by a dynamically similar awareness of time spent during that24-s block. For example, the analysis assumes that subjects will be in similar states of temporal awareness at 5 s, 10 s,and so forth during each off-task block. This cross-block similarity was confirmed. It is unlikely, then, that episodes unrelatedto time (like daydreaming) could be mapped so specifically to the same time point in each of eight blocks. The involved brainregions are certainly associated with conscious cognition in their other functions; it would be arbitrary to deny that this tem-poral mapping is not also an aspect of consciousness.

5. Conclusion

The successive analyses in this paper lend support to a new interpretation of the function of the Default Mode Network,relating it to the continuous monitoring of duration, a core structural feature temporal consciousness. The emerging pictureof default mode function, and the specific function of the overlapping Dynamic Temporal Network, conforms nicely to thephenomenology of time. In James’ and Husserl’s classic descriptions (and in the phenomenological tradition since), objectsare perceived as an amalgam of their static sensory properties and their dynamic non-sensory temporal properties. Onecannot perceive an apple without also perceiving its continually growing duration; even if the apple is immobile, onecan tell that it has been immobile for this long, now this, now this. The 3-dimensional object is inflected by the evermoving fourth dimension, even if the ‘‘object’’ is itself a psychological state, insofar as these too have readily perceptibledurations. The mental functions of the default mode seem also to be inflected by time. If the hypothesis proposed here isconfirmed in other experiments and other analyses, time may take its place as an omnipresent contextual dimension ofstates of conscious awareness, adding to an emerging science of consciousness an account of the thickness and effervescenceof the extended Now.

Acknowledgments

Thanks for Brian Castelluccio for data analysis and to the Trinity College Interdisciplinary Science Program for researchassistance support. An earlier draft of this paper was presented at a workshop on temporality sponsored by the EuropeanPlatform for Life Sciences, Mind Sciences, and the Humanities at the University of Turku, Finland; I thank the workshop hostsand participants for their comments. Thanks also to Goeff Lee, Michal Klincewicz, and Joe Neisser for their extensive andthoughtful comments at the Consciousness Online conference, and to two anonymous referees for this journal. IndependentComponent Analysis was implemented using the ICA Toolbox for Matlab (http://icatb.sourceforge.net/), developed by VinceCalhoun. SVM was implemented using the OSU-SVM toolbox (Ma et al., 2002).

References

Addis, D. R., Wong, A. T., & Schacter, D. L. (2007). Remembering the past and imagining the future: Common and distinct neural substrates during eventconstruction and elaboration. Neuropsychologia, 45(7), 1363–1377.

Amodio, D. M., & Frith, C. D. (2006). Meeting of minds: The medial frontal cortex and social cognition. Nature Reviews Neuroscience, 7(4), 268–277.Andreasen, N. C., O’Leary, D. S., Cizadlo, T., Arndt, S., Rezai, K., Watkins, G. L., et al (1995). Remembering the past: Two facets of episodic memory explored

with positron emission tomography. American Journal of Psychiatry, 152(11), 1576–1585.Buckner, R. 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.

5 Several recent articles speculate that DM activity embodies James’ ‘‘stream of consciousness’’ (Buckner et al., 2008; Fransson, 2006; Greicius & Menon,2004; McKiernan et al., 2003). This is a mistaken reading of James’ metaphor (James, 1890), which was intended to capture a feature of all conscious thought:

Consciousness, then, does not appear to itself chopped up in bits. Such words as ‘chain’ or ‘train’ do not describe it fitly as it presents itself in the first instance. Itis nothing jointed; it flows.

(James, 1890, Chap. 9) Most of James’ examples are perceptual and active. The DM authors are more likely thinking of the post-Jamesian literary meaning ofstream of consciousness, as in the novels of Dorothy Richardson, Virginia Woolf, Katherine Anne Porter, William Faulkner, James Joyce, and many others.These streams do have the disengagement and inner focus of default cognition.

D. Lloyd / Consciousness and Cognition 21 (2012) 695–703 703

Calhoun, V. D., Adali, T., Pearlson, G. D., van Zijl, P. C., & Pekar, J. J. (2002). Independent component analysis of fMRI data in the complex domain. MagneticResonance Medicine, 48(1), 180–192.

Chapelle, O., & Vapnik, V. (2000). Model selection for support vector machines. In S. A. Solla, T. K. Leen, & K.-R. Muller (Eds.). Advances in neural informationprocessing systems (Vol. 12). Cambridge, MA: MIT Press.

Cox, D. D., & Savoy, R. L. (2003). Functional magnetic resonance imaging (fMRI) brain reading: Detecting and classifying distributed patterns of fMRI activityin human visual cortex. Neuroimage, 19(2 Pt 1), 261–270.

Dainton, B. (2000). Stream of consciousness (2nd ed.). London: Routledge (2006).Dainton, B. (2008). Sensing change. Philosophical Issues, 18, 362–384.Fransson, P. (2006). How default is the default mode of brain function? Further evidence from intrinsic BOLD signal fluctuations. Neuropsychologia, 44(14),

2836–2845.Gilbert, D. T., & Wilson, T. D. (2007). Prospection: Experiencing the future. Science, 317(5843), 1351–1354.Grady, C. L., Springer, M. V., Hongwanishkul, D., McIntosh, A. R., & Winocur, G. (2006). Age-related changes in brain activity across the adult lifespan. Journal

of Cognitive Neuroscience, 18(2), 227–241.Greicius, M. D., & Menon, V. (2004). Default-mode activity during a passive sensory task: Uncoupled from deactivation but impacting activation. Journal of

Cognitive Neuroscience, 16(9), 1484–1492.Grush, R. (2004). The emulation theory of representation: Motor control, imagery, and perception. Behavioral and Brain Sciences, 27(3), 377–396 (discussion

396–442).Gusnard, D. A., & Raichle, M. E. (2001). Searching for a baseline: Functional imaging and the resting human brain. Nature Reviews Neuroscience, 2(10),

685–694.Hahn, B., Ross, T. J., & Stein, E. A. (2007). Cingulate activation increases dynamically with response speed under stimulus unpredictability. Cerebral Cortex,

17(7), 1664–1671.Haynes, J. D., & Rees, G. (2006). Decoding mental states from brain activity in humans. Nature Reviews Neuroscience, 7(7), 523–534.Husserl, E. (1966 (1928)). Zur Phänomenologie des inneren Zeitbewusstseins (Phenomenology of Inner Time Consciousness) (Vol. 10). The Hague: Martinus

Nijhoff.James, W. (1890). Principles of psychology. New York, NY: Henry Holt & Co.LaConte, S., Strother, S., Cherkassky, V., Anderson, J., & Hu, X. (2005). Support vector machines for temporal classification of block design fMRI data.

Neuroimage, 26(2), 317–329.Lloyd, D. (2002). Functional MRI and the study of human consciousness. Journal of Cognitive Neuroscience, 14(6), 818–831.Ma, J., Zhao, Y., & Ahalt, S. (2002). OSU-SVM Toolbox for Matlab.Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. San Francisco: W.H. Freeman.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.McKiernan, K. A., Kaufman, J. N., Kucera-Thompson, J., & Binder, J. R. (2003). A parametric manipulation of factors affecting task-induced deactivation in

functional neuroimaging. Journal of Cognitive Neuroscience, 15(3), 394–408.Norman, K. A., Polyn, S. M., Detre, G. J., & Haxby, J. V. (2006). Beyond mind-reading: Multi-voxel pattern analysis of fMRI data. Trends in Cognitive Sciences,

10(9), 424–430.Opper, M., & Winther, O. (2000). Gaussian processes and SVM: Mean field results and leave-one-out. In Advances in large margin classifiers (pp. 43–65).

Cambridge, MA: MIT Press.Raichle, M. E., MacLeod, A. M., Snyder, A. Z., Powers, W. J., Gusnard, D. A., & Shulman, G. L. (2001). A default mode of brain function. Proceedings of the

National Academy of Sciences of the United States of America, 98(2), 676–682.Saxe, R., & Kanwisher, N. (2003). People thinking about thinking people. The role of the temporo-parietal junction in ‘‘theory of mind. Neuroimage, 19(4),

1835–1842.Schreiber, T., & Schmitz, A. (1996). Improved surrogate data for nonlinearity tests. Physical Review Letters, 77(4), 635–638.Schreiber, T., & Schmitz, A. (2000). Surrogate time series. Physica D: Nonlinear Phenomena, 142(3-4), 346–382.Seth, A. (2009). Explanatory correlates of consciousness: Theoretical and computational challenges. Cognitive Computer, 1, 50–63.Sheline, Y. I., Barch, D. M., Price, J. L., Rundle, M. M., Vaishnavi, S. N., Snyder, A. Z., et al (2009). The default mode network and self-referential processes in

depression. Proceedings of the National Academy of Sciences of the United States of America, 106(6), 1942–1947.Shulman, G. L., Corbetta, M., Buckner, R. L., Raichle, M. E., Fiez, J. A., Miezin, F. M., et al (1997). Top-down modulation of early sensory cortex. Cerebral Cortex,

7(3), 193–206.Svoboda, E., McKinnon, M. C., & Levine, B. (2006). The functional neuroanatomy of autobiographical memory: A meta-analysis. Neuropsychologia, 44(12),

2189–2208.Yoshimi, J. (2007). Mathematizing phenomenology. Phenomenology and the Cognitive Sciences, 6(3), 271–291.Zahavi, D. (1999). Self-awareness and alterity. Evanston, IL: Northwestern University Press.