Is a Linear Conception of Time Sufficient in Computational Psychology?

 

Is a Linear Conception of Time Sufficient in Computational Psychology?

A Critical Analytical Report

 

 

 

 

 

 

 

 

Analytical research report

20 March 2026

 

Abstract

This report examines whether a linear conception of time is sufficient in computational psychology and, if not, in what precise sense it becomes insufficient. The central finding is that the answer depends on which aspect of “linearity” is under discussion. A single linear temporal order - earlier versus later - is almost always indispensable, and a single external time axis is often a useful modeling convention. In several major model families, including diffusion decision models, Kalman-style state-space models, and standard reinforcement learning, linear time indexing is not itself the limiting assumption. The more substantive problems arise when one assumes that a single homogeneous clock is enough for all relevant processes, that mathematical time can be identified with experienced psychological time, or that linear temporal order implies linearly evolving process dynamics. These assumptions are often too strong.

The report shows that linear time is usually adequate as a practical idealization in tightly constrained perceptual decisions, short-horizon sensorimotor control, and many trial-based learning tasks. It becomes inadequate when the target phenomena depend on nested temporal scales, event boundaries, rhythmic organization, recurrent feedback, history-dependent context effects, or explicit representations of subjective duration. In such cases, the empirical problem is typically not that time ceases to be ordered, but that the modeled processes are multiscale, nonlinear, hierarchical, or event-structured relative to that order.

Methodologically, the report is based on a critical analytical review of selected peer-reviewed literature spanning decision models, reinforcement learning, timing, dynamical systems, event segmentation, active inference, and motor control. The conclusion is therefore qualified: linear time remains a necessary baseline and often a justified approximation, but richer temporal formalisms are required when the phenomenon itself contains multiple clocks, latent temporal hierarchies, or endogenous event structure.

Keywords: computational psychology; time representation; dynamical systems; reinforcement learning; Bayesian models; subjective time; multiscale dynamics

 

Table of Contents

Abstract.................................................................................................... 2

1 Introduction.......................................................................................... 4

2 Theoretical Background and Previous Research............................ 6

3 Method.................................................................................................. 9

4 Results................................................................................................. 10

5 Discussion........................................................................................... 14

6 Conclusion.......................................................................................... 16

7 Uncertainties and Open Questions................................................. 17

References.............................................................................................. 18

 

1 Introduction

This report addresses the question: Is a linear conception of time sufficient in computational psychology, and if not, in what precise sense is it insufficient? The question is more intricate than a simple yes-or-no judgment, because “linear time” can refer to several distinct assumptions. It may mean only that events can be ordered as earlier or later; it may mean that time is represented as a single continuous or discrete axis; it may mean that model dynamics unfold against an external clock; or it may mean that psychological change itself is assumed to vary linearly as a function of time. These assumptions are often conflated, even though they have different consequences for theory and model adequacy.

Computational psychology is likewise heterogeneous. It includes models of perceptual decision-making, perceptual inference, learning, memory, time estimation, motor control, and broader dynamical or Bayesian accounts of cognition. Across these domains, time enters models in different ways: as an index of state updates, as elapsed duration, as an endogenous control variable, as an event boundary structure, or as an explicitly represented latent variable. A single verdict about the sufficiency of linear time is therefore unlikely to generalize across all subfields.

The present report advances a differentiated claim. Linear temporal order is almost always necessary, and a single linear time axis often remains a serviceable idealization. However, this does not imply that all psychologically relevant temporal structure is exhausted by such an axis. In many models, the main limitation lies not in temporal ordering itself but in the assumption that one homogeneous clock can organize all relevant processes without loss, distortion, or theoretical ambiguity. The analysis therefore distinguishes four questions throughout: where linear time functions only as a modeling convention; where it is an empirically supported approximation; where it begins to distort the phenomenon; and where richer temporal structure is required.

The practical importance of this distinction is methodological. If one mislocates the source of model inadequacy, one may wrongly conclude that “time” itself must be reconceived, when the deeper problem is actually nonlinear state dynamics, hidden temporal context, event-based segmentation, or insufficient measurement resolution. Conversely, one may mistake a mathematically convenient time index for a psychologically realistic temporal representation. The report aims to separate these possibilities with as much precision as the current literature permits.

 

2 Theoretical Background and Previous Research

2.1 What “linear time” can mean

In the present context, a linear conception of time should be disaggregated into at least four senses. First, time may be treated as ordinal sequence: event A occurs before event B. This is the weakest and most general sense. Second, time may be represented as a single continuous or discrete axis on which system states are indexed. Third, model evolution may be assumed to unfold against a uniform external clock, whether physical clock time or trial number. Fourth, one may assume that the process of psychological change is itself linear in time, for example by expecting proportional accumulation, constant-rate decay, or stationary transition structure.

These four senses should not be collapsed. A model may preserve linear temporal order while allowing highly nonlinear dynamics. A Bayesian filter, for example, may update beliefs sequentially on a discrete time axis without implying that the underlying process changes linearly. Similarly, recurrent neural or dynamical-systems models often retain an ordinary time variable while generating attractors, oscillations, bifurcations, and multistable behavior that are explicitly nonlinear. Therefore, linear temporal indexing does not entail linear temporal dynamics.

A further distinction is required between physical time, model time, and experienced time. Physical time refers to external elapsed time. Model time refers to the parameter or index with respect to which equations are solved. Experienced time refers to the organism’s temporal representation or phenomenology. These are sometimes aligned, but they are not identical by default. In interval timing and subjective duration research, the divergence between external duration and perceived duration is a central explanandum rather than a nuisance variable (Merchant et al., 2013; Sadibolova & Terhune, 2022).

2.2 What counts as computational psychology here

For the purposes of this report, computational psychology includes at least six broad model families or application domains. First are decision models, especially diffusion and sequential sampling models, in which evidence is accumulated over time to a threshold (Ratcliff et al., 2015, 2016). Second are reinforcement-learning models, especially temporal-difference and state-space variants, which learn from sequential outcomes and must solve temporal credit assignment (Lee et al., 2012; Ludvig et al., 2014). Third are Bayesian and state-space models of perception, prediction, and motor control, in which latent states are updated over time under uncertainty (de Xivry et al., 2013). Fourth are dynamical-systems and neural population approaches, which emphasize coupled nonlinear trajectories, feedback loops, and multiple intrinsic timescales (Chaudhuri et al., 2014; Papo, 2013; Stine & Jazayeri, 2025). Fifth are models of event perception and memory, which explain how continuous activity is segmented into nested events (Zacks & Swallow, 2007; Geerligs et al., 2022). Sixth are models of temporal cognition itself, including interval timing, subjective duration, and temporal priors (Merchant et al., 2013; Sadibolova & Terhune, 2022).

2.3 Core assumptions that require scrutiny

Three assumptions recur across the literature and are especially important for the present question. The first is that physical time and psychological time can be straightforwardly identified. This is often false in domains where expectation, attention, uncertainty, and context reshape temporal estimation. The second is that one time axis is sufficient for all relevant cognitive processes. This can be too restrictive when processes unfold at nested or interacting scales, such as milliseconds for neural dynamics, seconds for action selection, and longer spans for narrative integration or learning history (Papo, 2013; Friston et al., 2017; Golesorkhi et al., 2021). The third is that a linearly ordered time axis commits one to linear process laws. This is a category error: nonlinear, feedback-governed, or event-triggered dynamics can be built on an ordinary time line.

2.4 Previous research relevant to the question

Several literatures point toward temporal heterogeneity rather than temporal abolishment. Work on diffusion models shows that simple choices can often be captured by monotonic evidence accumulation on a one-dimensional time axis, making linear clock time highly effective as a practical approximation in constrained tasks (Ratcliff et al., 2016). Reinforcement-learning theory, however, has long highlighted the problem of temporal credit assignment and the need for eligibility traces, discounting, and time-sensitive state representation, which already complicate a naïve single-clock view (Lee et al., 2012; Ludvig et al., 2014). Research on interval timing shows that organisms do not merely read off objective duration; temporal judgments are shaped by priors, context, and modality, indicating that experienced time is inferentially constructed (Merchant et al., 2013; Sadibolova & Terhune, 2022).

A different line of work argues that cognition is organized across multiple intrinsic timescales. Neural and cognitive systems exhibit hierarchies of temporal integration rather than a single uniform temporal window (Papo, 2013; Chaudhuri et al., 2014; Golesorkhi et al., 2021). In auditory and event perception, evidence suggests processing channels and segmentation mechanisms operating at distinct scales and boundaries rather than a seamless homogeneous stream (Teng et al., 2017; Zacks & Swallow, 2007; Geerligs et al., 2022). Active-inference work extends this point by modeling nested state transitions over deep temporal hierarchies, where inference at one level summarizes sequences at lower levels (Friston et al., 2017). Taken together, prior work suggests that the strongest pressure against simple linear time arises not from any denial of temporal order, but from the insufficiency of one undifferentiated temporal parameter to capture context-sensitive and hierarchically structured cognition.

 

3 Method

This report uses a critical analytical review method rather than a statistical meta-analysis. The material was selected to cover model families that bear directly on the question of temporal representation in computational psychology: diffusion decision models, reinforcement-learning models, Bayesian and Kalman-style state-space models, interval-timing models, dynamical-systems approaches, event-segmentation research, and active-inference accounts. Preference was given to peer-reviewed review articles, theoretically central papers, and empirical studies that clarify what kind of temporal structure a model assumes or requires.

The review was delimited in three ways. First, it focuses on computational and formal modeling traditions relevant to psychology and cognitive neuroscience, rather than on purely philosophical theories of time or on physics-based theories with no psychological modeling role. Second, it concentrates on cases where the temporal assumptions are theoretically load-bearing, not merely incidental features of data collection. Third, it does not attempt exhaustive coverage of all subfields. The goal is analytical precision, not encyclopedic completeness.

The analysis proceeded by distinguishing four candidate meanings of “linear time,” then evaluating for each model family whether the relevant temporal assumption was: (a) a modeling convention, (b) an empirically justified approximation, (c) a potential source of distortion, or (d) insufficient and in need of replacement or supplementation. Special attention was given to whether apparent failures of linear time were actually failures of state representation, process nonlinearity, event structure, or scale separation.

The evaluation of reliability and validity follows standard principles for conceptual synthesis. Reliability is strengthened by relying primarily on established peer-reviewed sources, especially reviews and well-cited theoretical contributions. Construct validity depends on keeping distinct concepts separate: temporal order, clock structure, dynamical law, and subjective time. Internal validity in this type of report concerns the coherence of the inferential chain from literature to conclusion. External validity is limited because conclusions about “computational psychology” as a whole cannot be stronger than the breadth and comparability of the source literatures. The report therefore avoids treating findings from one model family as automatically generalizable to all others.

 

4 Results

4.1 Where linear time is a useful and often sufficient idealization

In several domains, linear clock time is sufficient in the limited but important sense that models indexed by a single time variable capture the behavior of interest without major explanatory loss. The clearest case is simple perceptual decision-making under tightly controlled conditions. Diffusion decision models assume that noisy evidence evolves over time until a response boundary is reached. Here, a single temporal axis is not merely convenient; it is deeply informative because response times and accuracy jointly constrain the model (Ratcliff et al., 2015, 2016). The success of these models suggests that for brief, well-specified tasks with stable evidence streams and limited contextual structure, linear time is often an empirically justified approximation.

A similar point applies to many state-space and Kalman-style models in motor control and perceptual tracking. In such models, latent states are updated sequentially in discrete or continuous time, and the main burden falls on uncertainty estimation and prediction, not on revising the temporal order itself. For visually guided and predictive pursuit, Kalman filtering can unify sensory prediction and motion extrapolation under a single temporal axis with considerable empirical success (de Xivry et al., 2013). Again, linear time is not claimed to mirror phenomenology; it functions as a useful parameterization of ongoing control.

Standard reinforcement-learning models also often work with a linear sequence of time steps or trials. In many conditioning and action-value settings, this is adequate so long as the relevant temporal structure can be encoded into states, discounting, or eligibility traces (Lee et al., 2012). In this restricted sense, linear time frequently remains sufficient as a backbone. The model need not represent multiple clocks explicitly if the task structure is short-horizon, stationarized, and externally paced.

4.2 Where linear time is insufficient because one homogeneous clock is not enough

Linear time becomes insufficient when cognition depends on interacting temporal scales that cannot be collapsed into one homogeneous update schedule without explanatory loss. Research on neural and cognitive timescales suggests that systems differ in how long they integrate information and maintain context. Sensory areas often operate on shorter windows than higher-order association regions, supporting a temporal hierarchy rather than a uniform integration horizon (Chaudhuri et al., 2014; Golesorkhi et al., 2021). In such cases, a single external clock may still exist mathematically, but it is not the right descriptive level for the phenomenon. What matters is that different subsystems effectively operate with different temporal receptive windows.

The same problem appears in deep temporal and hierarchical generative models. Active-inference accounts explicitly model nested temporal structure: higher levels infer slower-changing patterns over sequences of lower-level transitions (Friston et al., 2017). This does not reject linear order, but it rejects the adequacy of a flat time axis as the sole temporal representational resource. A model that uses only one scale can miss how context at longer horizons constrains interpretation at shorter horizons, as in language, narrative understanding, or planning.

Event perception offers a further example. People do not treat continuous activity as a uniform stream only indexed by elapsed seconds. They segment experience into events and subevents, often hierarchically, and these boundaries have consequences for working memory, learning, and long-term memory (Zacks & Swallow, 2007; Geerligs et al., 2022). A purely linear-time description can list when boundaries occur, but boundary structure itself becomes a psychologically real organizing principle. In that respect, temporal representation is partly event-based, not only metrically elapsed.

4.3 Where the key problem is nonlinear dynamics rather than time structure itself

In many cases, the real inadequacy does not lie in linear time per se, but in the mistaken expectation that process change should be linear with respect to time. Cognitive and neural systems often exhibit recurrence, feedback, phase transitions, attractor-like stability, oscillations, and sudden reorganizations. These are nonlinear dynamics unfolding on an ordinary time axis. Dynamical-systems approaches make this explicit: the temporal parameter can remain conventional while the state trajectory becomes history-dependent and context-sensitive (Papo, 2013; Stine & Jazayeri, 2025).

This distinction matters because it blocks an overreaction. One might observe nonlinear learning curves, abrupt shifts in attention, or context-sensitive timing behavior and conclude that linear time has failed. More cautiously, one should first ask whether the failure lies in the process law, not in the temporal axis. For example, drift-diffusion models have been extended with collapsing bounds, urgency signals, or time-varying gain, showing that a linear temporal axis can coexist with nonstationary decision dynamics. Likewise, reinforcement-learning models can incorporate eligibility traces, semi-Markov timing, or richer state representations without abandoning temporal sequence.

Thus, some apparent evidence against linear time is actually evidence against linear dynamics, stationarity, or impoverished state description. This is a central interpretive result of the review.

4.4 Where subjective time specifically exceeds simple linear external time

The strongest case against a simple linear conception arises when the target phenomenon is subjective time itself. Interval timing, duration estimation, temporal reproduction, and related tasks show that perceived duration is systematically shaped by priors, context, modality, uncertainty, attention, and action demands (Merchant et al., 2013; Sadibolova & Terhune, 2022). Here, the difference between physical time and psychological time is not residual error but an empirical phenomenon requiring explanation.

Bayesian models of timing formalize this by treating elapsed duration as uncertain evidence combined with prior expectations. Such models can account for central tendency effects and context sensitivity, but they also reveal a deeper point: time, as represented by the organism, is inferential rather than passively read from an external clock (Sadibolova & Terhune, 2022). Therefore, a simple one-to-one mapping from external linear time to internal temporal experience is inadequate.

Motor and control contexts reinforce this point. Organisms often time actions relative to opportunities, predictions, and sensorimotor contingencies rather than by explicitly representing neutral elapsed duration. Timing is embedded in control, not merely measured by a clock (Merchant et al., 2013; Stine & Jazayeri, 2025). The richer the task structure, the less plausible it becomes that one homogeneous external clock fully captures psychologically relevant temporal organization.

4.5 A precise answer to the main question

The evidence reviewed supports a qualified answer. Linear time is sufficient when it means ordered sequence plus a single time index for model updates in tasks that are short-horizon, externally paced, and weakly hierarchical. It becomes insufficient in a more precise sense under four conditions.

First, it is insufficient when one axis is asked to capture multiple interacting timescales without additional temporal structure. Second, it is insufficient when event boundaries or rhythms organize cognition in ways not reducible to uniform elapsed duration. Third, it is insufficient when the explanandum is subjective time itself, because experienced duration is context-sensitive and inferential. Fourth, it is insufficient when modelers tacitly equate linear temporal order with linear process dynamics or stationarity.

In short, the main inadequacy is usually not that linear temporal order is false, but that a flat, homogeneous, externally clocked temporal framework is too poor to represent hierarchical, event-based, recurrent, and subjectively constructed temporal organization.

 

5 Discussion

A strong alternative view is that linear time remains entirely sufficient for most of computational psychology, and that apparent failures can always be absorbed by adding richer latent states, nonlinear transition functions, or task-dependent observation models. There is real force to this position. Many successful models already do exactly that. One can preserve a single time axis while introducing hidden states, context variables, semi-Markov dwell times, recurrent connections, or hierarchical levels. From this perspective, the bottleneck is not the temporal framework but the rest of the model architecture.

This alternative is partly persuasive. In many practical applications, retaining a single temporal backbone is beneficial because it preserves tractability, comparability, and identifiability. Indeed, one should not multiply temporal formalisms without need. If a state-space or recurrent model with one time parameter explains the data, replacing the time axis with a more exotic formalism may add complexity without explanatory gain.

However, this alternative view has limits. First, once multiple nested timescales, event boundaries, or temporal priors become indispensable to prediction and interpretation, saying that “linear time still suffices” risks becoming purely formal rather than substantively explanatory. A model may retain one mathematical time variable while effectively encoding richer temporal structure elsewhere. In that case, the important theoretical issue is not whether the symbol t still appears in the equations, but whether the temporal ontology of the model has become hierarchical, event-based, or context-sensitive. Second, if the phenomenon under study is subjective time, preserving only external clock time as the primary temporal variable may obscure the fact that the model’s key inferential work concerns internal temporal representation.

Another counterview is that time is being blamed for problems that are really due to measurement level or task design. This concern is well founded. Some failures attributed to temporal structure may reflect overly coarse sampling, trial averaging, or laboratory paradigms that suppress naturalistic temporal organization. For example, a model might look inadequate because it ignores latent context or environmental volatility, not because its conception of time is wrong. This possibility weakens any sweeping anti-linear-time conclusion.

The most defensible position is therefore intermediate. One linear time axis is often an adequate base structure, but only if one is explicit about what it does and does not represent. It is most defensible when used as a computational scaffold. It is less defensible when reified into a claim that cognition itself unfolds on one homogeneous clock or that subjective temporal organization is reducible to external elapsed duration.

Several richer temporal frameworks appear promising. Multiscale and hierarchical state-space models allow nested update rates. Semi-Markov and event-driven models represent variable dwell times and boundary-triggered transitions. Oscillatory and rhythmic frameworks capture periodic structure more naturally than homogeneous elapsed time alone (Teng et al., 2017). Dynamical-systems models capture recurrence and phase-sensitive change. Bayesian timing models represent subjective duration as latent inference rather than direct clock reading. Active-inference frameworks explicitly organize state transitions over deep temporal hierarchies (Friston et al., 2017). These approaches do not all reject linear time, but they do treat it as insufficiently rich on its own for many phenomena.

The broader theoretical implication is that temporal adequacy should be judged relative to the target phenomenon. For reaction-time tasks, linear time can be a strong approximation. For narrative cognition, long-horizon planning, temporal credit assignment across delayed outcomes, or subjective time estimation, adequacy increasingly depends on event structure, memory depth, and interacting timescales. Therefore, temporal representation is not one problem but a family of problems.

 

6 Conclusion

This report has argued that a linear conception of time is not simply either sufficient or insufficient in computational psychology. The answer depends on which meaning of “linear” is at issue and on which class of phenomena is being modeled.

What can be treated as established is the following. Earlier-later order remains fundamental, and a single continuous or discrete time axis is often a useful modeling convention. In perceptual decision models, many motor-control models, and some reinforcement-learning settings, this convention is often empirically adequate for the explanatory target (Ratcliff et al., 2016; de Xivry et al., 2013; Lee et al., 2012). It is also established that one should not confuse linear temporal indexing with linear dynamics.

What is plausible but still partly contested is that many important cognitive phenomena are better characterized by temporal hierarchies, multiple integration windows, event-based segmentation, and endogenous temporal context rather than by one homogeneous clock. The relevant literatures strongly suggest this, but the precise formal unification across model families remains incomplete (Papo, 2013; Friston et al., 2017; Golesorkhi et al., 2021; Stine & Jazayeri, 2025).

What remains open is how far one can go by enriching models built on a single time axis before a genuinely different temporal formalism becomes necessary. This is partly a substantive question about cognition and partly a modeling question about abstraction level. A single axis may remain mathematically present while the model’s operative temporal organization becomes multiscale, event-based, or inferential.

The most precise answer, then, is this: linear time is usually sufficient as a baseline ordering device and often sufficient as a pragmatic modeling scaffold, but it is insufficient when treated as a complete account of temporal organization in cognition. Its inadequacy lies not mainly in the falsity of sequence itself, but in the failure of a flat, homogeneous, externally clocked temporal framework to capture nonlinear, hierarchical, recurrent, event-structured, and subjectively constructed temporal processes.

 

7 Uncertainties and Open Questions

The present conclusions should be interpreted with caution for several reasons. First, the source literatures are methodologically heterogeneous. A successful decision model, a neural timescale study, and a theoretical review of active inference do not test the same proposition under commensurable conditions. Second, some arguments concern explanatory adequacy rather than direct falsification. Showing that a richer temporal framework is illuminating does not by itself prove that simpler linear-time models are unusable. Third, naturalistic cognition remains more difficult to model and validate than tightly controlled laboratory tasks, so some of the strongest claims about multiscale temporal structure rest on converging but still incomplete evidence.

Further research should address at least five questions. First, under what conditions can multiscale temporal structure be compressed into a single-axis latent-state model without significant loss of predictive or explanatory power? Second, how should subjective time be linked formally to task time, neural time, and model time in a way that permits empirical discrimination among theories? Third, when do event-based or semi-Markov formalisms outperform trial-based and uniformly clocked models in real psychological data? Fourth, can diffusion, reinforcement-learning, and Bayesian hierarchical models be compared under common benchmarks that manipulate temporal hierarchy and delay explicitly? Fifth, what measurement regimes are needed to distinguish true temporal multiscale structure from artifacts of sampling resolution, preprocessing, or model misspecification?

These questions remain open. Accordingly, the strongest defensible conclusion is not that linear time must be abandoned, but that its scope conditions must be stated explicitly and tested rather than assumed.

 

References

Chaudhuri, R., Bernacchia, A., & Wang, X.-J. (2014). A diversity of localized timescales in network activity. eLife, 3, e01239. https://doi.org/10.7554/eLife.01239

de Xivry, J.-J. O., Coppe, S., Blohm, G., & Lefevre, P. (2013). Kalman filtering naturally accounts for visually guided and predictive smooth pursuit dynamics. Journal of Neuroscience, 33(44), 17301-17313. https://doi.org/10.1523/JNEUROSCI.2321-13.2013

Friston, K. J., Rosch, R., Parr, T., Price, C., & Bowman, H. (2017). Deep temporal models and active inference. Neuroscience and Biobehavioral Reviews, 77, 388-402. https://doi.org/10.1016/j.neubiorev.2017.04.009

Geerligs, L., Gozukara, D., Oetringer, D., Campbell, K. L., van Gerven, M., & Guclu, U. (2022). A partially nested cortical hierarchy of neural states underlies event segmentation in the human brain. eLife, 11, e77430. https://doi.org/10.7554/eLife.77430

Golesorkhi, M., Gomez-Pilar, J., Zilio, F., Berberian, N., Wolff, A., Yagoub, M. C. E., & Northoff, G. (2021). The brain and its time: Intrinsic neural timescales are key for input processing. Communications Biology, 4, Article 970. https://doi.org/10.1038/s42003-021-02483-6

Lee, D., Seo, H., & Jung, M. W. (2012). Neural basis of reinforcement learning and decision making. Annual Review of Neuroscience, 35, 287-308. https://doi.org/10.1146/annurev-neuro-062111-150512

Ludvig, E. A., Balci, F., & Spetch, M. L. (2014). Time representation in reinforcement learning models of the basal ganglia. Frontiers in Integrative Neuroscience, 7, Article 114. https://doi.org/10.3389/fnint.2013.00114

Merchant, H., Harrington, D. L., & Meck, W. H. (2013). Neural basis of the perception and estimation of time. Annual Review of Neuroscience, 36, 313-336. https://doi.org/10.1146/annurev-neuro-062012-170349

Papo, D. (2013). Time scales in cognitive neuroscience. Frontiers in Physiology, 4, Article 86. https://doi.org/10.3389/fphys.2013.00086

Ratcliff, R., Smith, P. L., & McKoon, G. (2015). Modeling regularities in response time and accuracy data with the diffusion model. Current Directions in Psychological Science, 24(6), 458-470. https://doi.org/10.1177/0963721415596228

Ratcliff, R., Smith, P. L., Brown, S. D., & McKoon, G. (2016). Diffusion decision model: Current issues and history. Trends in Cognitive Sciences, 20(4), 260-281. https://doi.org/10.1016/j.tics.2016.01.007

Sadibolova, R., & Terhune, D. B. (2022). The temporal context in Bayesian models of interval timing: Recent advances and future directions. Behavioral Neuroscience, 136(5), 364-373. https://doi.org/10.1037/bne0000513

Stine, G. M., & Jazayeri, M. (2025). Control principles of neural dynamics revealed by the neurobiology of timing. Annual Review of Neuroscience, 48, 43-63. https://doi.org/10.1146/annurev-neuro-091724-015512

Teng, X., Tian, X., Rowland, J., & Poeppel, D. (2017). Concurrent temporal channels for auditory processing: Oscillatory neural entrainment reveals segregation of function at different scales. PLoS Biology, 15(11), e2000812. https://doi.org/10.1371/journal.pbio.2000812

Zacks, J. M., & Swallow, K. M. (2007). Event segmentation. Current Directions in Psychological Science, 16(2), 80-84. https://doi.org/10.1111/j.1467-8721.2007.00480.x