Adaptive neural models of queuing and timing in fluent action

doi:10.1016/j.tics.2004.07.003

Trends in Cognitive Sciences

Volume 8, Issue 9, September 2004, Pages 426-433

https://doi.org/10.1016/j.tics.2004.07.003 Get rights and content

In biological cognition, specialized representations and associated control processes solve the temporal problems inherent in skilled action. Recent data and neural circuit models highlight three distinct levels of temporal structure: sequence preparation, velocity scaling, and state-sensitive timing. Short sequences of actions are prepared collectively in prefrontal cortex, then queued for performance by a cyclic competitive process that operates on a parallel analog representation. Successful acts like ball-catching depend on coordinated scaling of effector velocities, and velocity scaling, mediated by the basal ganglia, may be coupled to perceived time-to-contact. Making acts accurate at high speeds requires state-sensitive and precisely timed activations of muscle forces in patterns that accelerate and decelerate the effectors. The cerebellum may provide a maximally efficient representational basis for learning to generate such timed activation patterns.

Section snippets

Three levels of temporal structure in skilled performance

A surprisingly demanding problem is the genesis of skilled behavior using complex effectors like a human's arm or speech articulators. Skilled behavior emerges in temporally structured episodes, and brain areas that use distinct representations contribute to this temporal structuring. This review examines computational models of neural circuits contributing to three levels of temporal structure in behavior. Level one is the fluent succession of acts prepared collectively as a sequence. This

Fluent succession of acts via competitive queuing

Fifty years ago, Lashley [1] used data on sequencing errors – in which early and later elements of a sequence mistakenly exchange positions – to infer that neural representations for all elements of a planned sequence are simultaneously active before sequence production. The proposal that sequences are represented by simultaneous parallel activation of representations of their elements differs from many classical and contemporary proposals. In most recurrent-state network models 2, 3, 4,

Neurophysiological evidence for the CQ model

Until 2002 there was no compelling electrophysiological evidence that the brain used the parallel sequence code and iterative choice cycle postulated by CQ theorists. New cell recordings by Averbeck et al. [8] plugged that evidential gap. They trained monkeys to draw a copy of a static geometric form using a routine, prescribed stroke sequence. Thus a form cued recall of sequence-representing information from long term memory. Recordings from area 46 of prefrontal cortex showed that before the

Progress of CQ models in explaining sequencing and timing

To motivate normalized CQ models, Grossberg [5] stressed that neurons exhibit finite activation ranges and noise. Both constrain the ability of neurons to use relative activations to reliably code the relative priority of a large number of sequence elements. Brains using this analog code should exhibit a small upper bound on the number of elements that can be reliably recalled in correct sequential order without secondary strategies, such as reloading chunks from long-term memory 5, 6, 12.

Coordination of rates and completion times in voluntary action

Many movement models, such as Equilibrium Point (EP) models ([31], Box 1), treat the temporal structure of actions from a biomechanical perspective. By contrast, some central pattern generation models, such as Vector Integration To Endpoint (VITE) models ([32], Box 1), treat timing from a cognitive dynamics perspective, with a focus on voluntary gating of plan execution and voluntary control of movement rates. VITE models have successfully simulated both the discharge patterns of diverse motor

Timed anticipatory responses

In a successful ball catch, the arm flicks out and ‘stops on a dime’ at whatever degree of arm extension enables the hand to catch the ball. Newtonian mechanics implies that an arm set in motion by extensor muscles would (disastrously) continue ‘past the mark’ unless braked by precisely timed, anticipatory action of opposing muscles. When driving a car, stomping the accelerator and hitting the brake are separate voluntary actions. When ‘driving’ our bodies, the braking contractions are

Conclusions

This review has focused on neural circuit models of fluent performance of discrete actions, and the implicit claim was that at least three kinds of temporal structuring must be acknowledged to exist as distinct factors in most episodes of skilled action. Competitive queuing theorists who endorsed Lashley's inference that some sequence planning involves parallel activation of all sequence elements can now point to compelling electrophysiological support, but this does not imply that other

Acknowledgements

Preparation of this article was partially supported by NIH R01 DC02852.

References (75)

C.C. Conway et al.
Sequential learning in non-human primates
Trends Cogn. Sci.
(2001)
S. Sternberg
The latency and duration of rapid movement sequences: Comparisons of speech and typewriting
T.A. Hartley et al.
A linguistically constrained model of short-term memory for nonwords
J. Mem. Lang.
(1996)
J.L. Contreras-Vidal
Neural dynamics of short and medium-term motor control effects of levodopa therapy in parkinson's disease
Artif. Intell. Med.
(1998)
C. Mateer
Asymmetric effects of thalamic stimulation on rate of speech
Neuropsychologia
(1978)
J. Brown
How laminar frontal cortex and basal ganglia circuits interact to control planned and reactive saccades
Neural Netw.
(2004)
J. Dessing
Prospective control of manual interceptive actions: Comparative simulations of extant and new model constructs
Neural Netw.
(2002)
D. Bullock
A vector-integration-to-endpoint model for performance of viapoint movements
Neural Netw.
(1999)
A. Ulloa et al.
A neural network simulating human reach-grasp coordination by continuous updating of vector positioning commands
Neural Netw.
(2003)
D. Bullock
A neural model of timed response learning in the cerebellum
Neural Netw.
(1994)

J.S. Albus

A theory of cerebellar function

Math. Biosci.

(1971)

N. Schweighofer

Unsupervised learning of granule cell sparse codes enhances cerebellar adaptive control

Neuroscience

(2001)

R.E. Clark

Classical conditioning, awareness, and brain systems

Trends Cogn. Sci.

(2002)

K.S. Lashley

The problem of serial order in behavior

J. Elman

Language processing

P.F. Dominey

Influences of temporal organization on sequence learning and transfer: Comments on Stadler (1995) and Curran and Keele (1993)

J. Exp. Psychol. Learn. Mem. Cogn.

(1998)

B. Rhodes et al.

A scalable model of cerebellar adaptive timing and sequencing: The recurrent slide and latch (RSL) model

Appl. Intell.

(2002)

Grossberg, S. (1978) A theory of human memory: Self-organization and performance of sensory-motor codes, maps, and...

Houghton, G. (1990) The problem of serial order: A neural network model of sequence learning and recall. In Current...

D. Bullock et al.

Competitive queuing for serial planning and performance

B.B. Averbeck

Parallel processing of serial movements in prefrontal cortex

Proc. Natl. Acad. Sci. U. S. A.

(2002)

Boardman, I. and Bullock, D. (1991). A neural network model of serial order recall from short-term memory. In...

G. Bradski

STORE working memory networks for storage and recall of arbitrary temporal sequences

Biol. Cybern.

(1994)

Rhodes, B. and Bullock, D. (2002). Neural dynamics of learning and performance of fixed sequences: Latency pattern...

M.D. Byrne et al.

Serial modules in parallel: The psychological refractory period and perfect time sharing

Psychol. Rev.

(2001)

M.A. Basso et al.

Modulation of neuronal activity in superior colliculus by changes in target probability

J. Neurosci.

(1998)

P. Cisek et al.

Simultaneous encoding of multiple potential reach directions in dorsal premotor cortex

J. Neurophysiol.

(2002)

G. Pellizzer et al.

Motor planning: Effect of directional uncertainty with discrete spatial cues

Exp. Brain Res.

(2003)

N. Cowan

The magical number 4 in short-term memory: A reconsideration of mental storage capacity

Behav. Brain Sci.

(2001)

M.P.A. Page et al.

The primacy model: A new model of immediate serial recall

Psychol. Rev.

(1998)

S.T. Klapp

Reaction time analysis of central motor control

W.B. Verwey

Buffer loading and chunking in sequential keypressing

J. Exp. Psychol. Hum. Percept. Perform.

(1996)

X. Lu

Role of monkey cerebellar nuclei in skill for sequential movement

J. Neurophysiol.

(1998)

R.P. Dum et al.

An unfolded map of the cerebellar dentate nucleus and its projections to the cerebral cortex

J. Neurophysiol.

(2003)

D. Bullock

A neural network model for cursive script production

Biol. Cybern.

(1993)

Page, M.P.A. (1999) Modeling the perception of musical sequences with self-organizing neural networks. In Musical...

N. Ward

A Connectionist Language Generator

(1994)

Cited by (75)

Internal speech is faster than external speech: Evidence for feedback-based temporal control
2024, Cognition
A recent model of temporal control in speech holds that speakers use sensory feedback to control speech rate and articulatory timing. An experiment was conducted to assess whether there is evidence in support of this hypothesis by comparing durations of phrases in external speech (with sensory feedback) and internal speech (without sensory feedback). Phrase lengths were varied by including one to three disyllabic nouns in a target phrase that was always surrounded by overt speech. An inferred duration method was used to estimate the durations of target phrases produced internally. The results showed that internal speech is faster than external speech, supporting the hypothesis. In addition, the results indicate that there is a slow-down associated with transitioning between internal and external modes of production.
How Beat Perception Co-opts Motor Neurophysiology
2021, Trends in Cognitive Sciences
Beat perception offers cognitive scientists an exciting opportunity to explore how cognition and action are intertwined in the brain even in the absence of movement. Many believe the motor system predicts the timing of beats, yet current models of beat perception do not specify how this is neurally implemented. Drawing on recent insights into the neurocomputational properties of the motor system, we propose that beat anticipation relies on action-like processes consisting of precisely patterned neural time-keeping activity in the supplementary motor area (SMA), orchestrated and sequenced by activity in the dorsal striatum. In addition to synthesizing recent advances in cognitive science and motor neuroscience, our framework provides testable predictions to guide future work.
Episodic memory: A hierarchy of spatiotemporal concepts
2019, Neural Networks
We propose a new model of episodic memory. It consists of a hierarchy of partial sequences of events, blended for consistency across space and time by feedforward/feedback links to concepts expressing their shared information. This blended concept hierarchy is acquired and represented incrementally through synaptic or connection-weighted adaptation in a neural network with episodic memory capability. We derive this model through argumentation, explain its derivation within a category-theoretic model of representation, and discuss the capabilities we propose for this model.
Moving in time: Simulating how neural circuits enable rhythmic enactment of planned sequences
2019, Neural Networks
Citation Excerpt :
The parallel planning layer in a CQ model corresponds to a working memory, in which multiple memory representations are kept active over an extended interval, to govern task performance for which there is insufficient on-line stimulus support (Chafee & Goldman-Rakic, 2000; Goldman-Rakic, 1990). Direct neurophysiological evidence for CQ-consistent working memory dynamics – graded parallel activations and one-by-one deactivations during sequence production – was discovered in monkeys during a figure-drawing experiment by Averbeck, Chafee, Crowe, & Georgopoulos (2002, 2003; reviewed in Bullock, 2004, and in Rhodes, Bullock, Verwey, Averbeck, & Page, 2004), and in later studies of human sequential actions (Kornysheva, Bush, Meyer, Sadnicka, Barnes, & Burgess, 2019). Although frontal, parietal, and superior temporal zones of the primate cerebral cortex may be strongly engaged during tasks requiring working memory, the most reliably engaged during serial action planning are the lateral prefrontal cortex (lPFC) and the strongly linked pre-supplementary motor area (preSMA).
Many complex actions are mentally pre-composed as plans that specify orderings of simpler actions. To be executed accurately, planned orderings must become active in working memory, and then enacted one-by-one until the sequence is complete. Examples include writing, typing, and speaking. In cases where the planned complex action is musical in nature (e.g. a choreographed dance or a piano melody), it appears to be possible to deploy two learned sequences at the same time, one composed from actions and a second composed from the time intervals between actions. Despite this added complexity, humans readily learn and perform rhythm-based action sequences. Notably, people can learn action sequences and rhythmic sequences separately, and then combine them with little trouble (Ullén & Bengtsson 2003). Related functional MRI data suggest that there are distinct neural regions responsible for the two different sequence types (Bengtsson et al. 2004). Although research on musical rhythm is extensive, few computational models exist to extend and inform our understanding of its neural bases. To that end, this article introduces the TAMSIN (Timing And Motor System Integration Network) model, a systems-level neural network model capable of performing arbitrary item sequences in accord with any rhythmic pattern that can be represented as a sequence of integer multiples of a base interval. In TAMSIN, two Competitive Queuing (CQ) modules operate in parallel. One represents and controls item order (the ORD module) and the second represents and controls the sequence of inter-onset-intervals (IOIs) that define a rhythmic pattern (RHY module). Further circuitry helps these modules coordinate their signal processing to enable performative output consistent with a desired beat and tempo.
Neural Competitive Queuing of Ordinal Structure Underlies Skilled Sequential Action
2019, Neuron
Citation Excerpt :
However, since Lashley’s seminal proposal (Lashley, 1951), there has been an alternative account, suggesting that all elements of a planned sequence are active simultaneously before execution, leading to the characteristic finding of transposition errors among nearby elements (Rhodes et al., 2004); e.g., as observed in speech or typing. So-called “competitive queuing” (CQ) models can formally explain this behavior by introducing a parallel preparation layer that determines serial order by competitive interactions between sequence elements driven by differing levels of excitation according to the sequence (Figures 1A–1C; see Bullock, 2004 for a review). The most active node wins the competition, generates the corresponding action, and is then self-inhibited through the planning layer, allowing the next most strongly activated node to generate the next action.
Fluent retrieval and execution of movement sequences is essential for daily activities, but the neural mechanisms underlying sequence planning remain elusive. Here participants learned finger press sequences with different orders and timings and reproduced them in a magneto-encephalography (MEG) scanner. We classified the MEG patterns for each press in the sequence and examined pattern dynamics during preparation and production. Our results demonstrate the “competitive queuing” (CQ) of upcoming action representations, extending previous computational and non-human primate recording studies to non-invasive measures in humans. In addition, we show that CQ reflects an ordinal template that generalizes across specific motor actions at each position. Finally, we demonstrate that CQ predicts participants’ production accuracy and originates from parahippocampal and cerebellar sources. These results suggest that the brain learns and controls multiple sequences by flexibly combining representations of specific actions and interval timing with high-level, parallel representations of sequence position.
Does articulatory rehearsal help immediate serial recall?
2018, Cognitive Psychology
Articulatory rehearsal is assumed to benefit verbal working memory. Yet, there is no experimental evidence supporting a causal link between rehearsal and serial-order memory, which is one of the hallmarks of working memory functioning. Across four experiments, we tested the hypothesis that rehearsal improves working memory by asking participants to rehearse overtly and by instructing different rehearsal schedules. In Experiments 1a, 1b, and 2, we compared an instructed cumulative-rehearsal condition against a free-rehearsal condition. The instruction increased the prevalence of cumulative rehearsal, but recall performance remained unchanged or decreased compared to the free-rehearsal baseline. Experiment 2 also tested the impact of a fixed rehearsal instruction; this condition yielded substantial performance costs compared to the baseline. Experiment 3 tested whether rehearsals (according to an experimenter-controlled protocol) are beneficial compared to a matched articulatory suppression condition that blocked rehearsals of the memoranda. Again, rehearsing the memoranda yielded no benefit compared to articulatory suppression. In sum, our results are incompatible with the notion that rehearsal is beneficial to working memory.

View all citing articles on Scopus

View full text

Trends in Cognitive Sciences

Adaptive neural models of queuing and timing in fluent action

Section snippets

Three levels of temporal structure in skilled performance

Fluent succession of acts via competitive queuing

Neurophysiological evidence for the CQ model

Progress of CQ models in explaining sequencing and timing

Coordination of rates and completion times in voluntary action

Timed anticipatory responses

Conclusions

Acknowledgements

Trends Cogn. Sci.

J. Mem. Lang.

Artif. Intell. Med.

Neuropsychologia

Neural Netw.

Neural Netw.

Neural Netw.

Neural Netw.

Neural Netw.

Math. Biosci.

Neuroscience

Trends Cogn. Sci.

The problem of serial order in behavior

Language processing

Influences of temporal organization on sequence learning and transfer: Comments on Stadler (1995) and Curran and Keele (1993)

J. Exp. Psychol. Learn. Mem. Cogn.

A scalable model of cerebellar adaptive timing and sequencing: The recurrent slide and latch (RSL) model

Appl. Intell.

Competitive queuing for serial planning and performance

Parallel processing of serial movements in prefrontal cortex

Proc. Natl. Acad. Sci. U. S. A.

STORE working memory networks for storage and recall of arbitrary temporal sequences

Biol. Cybern.

Serial modules in parallel: The psychological refractory period and perfect time sharing

Psychol. Rev.

Modulation of neuronal activity in superior colliculus by changes in target probability

J. Neurosci.

Simultaneous encoding of multiple potential reach directions in dorsal premotor cortex

J. Neurophysiol.

Motor planning: Effect of directional uncertainty with discrete spatial cues

Exp. Brain Res.

The magical number 4 in short-term memory: A reconsideration of mental storage capacity

Behav. Brain Sci.

The primacy model: A new model of immediate serial recall

Psychol. Rev.

Reaction time analysis of central motor control

Buffer loading and chunking in sequential keypressing

J. Exp. Psychol. Hum. Percept. Perform.

Role of monkey cerebellar nuclei in skill for sequential movement

J. Neurophysiol.

An unfolded map of the cerebellar dentate nucleus and its projections to the cerebral cortex

J. Neurophysiol.

A neural network model for cursive script production

Biol. Cybern.

A Connectionist Language Generator