hazeltine et al. - timing mechanisms neural mechanisms of ... · hazeltine et al.- timing...
TRANSCRIPT
Hazeltine et al. - Timing mechanisms
Neural mechanisms of timing
Eliot Hazeltine, Laura L. Helmuth and Richard B. Ivry
pmagmss has been made b&
ii&n@ &borate experimental designs t
sensory and mnemonic processes. Mark
temp@@ characteristics of the clock in perception an
@m&&g system. Similar conclusions can be reached from clinical stud
s,@th neocerebellar damage are impaired at discriminating and
B/s. However, other patient populations, especially those wit#) d
#a, also exhibit deficits in ti
v may be distributgd throulgti:~:Q~! x;i ‘y _,( ;,:j: :, _I Ljq “:L L ,I )* _ i >,-l_ i i j i /< .) ./ $.- ,i p ‘i” q.,, )i $ @ i ), :( i I ‘j$‘$:i:: ::_
:j : $.!& (, ‘.‘. (:::; :,:fil:)::‘i ;irlrlii)‘-:‘Ir~:-‘~.,i’ +;: :fiq
P recise timing is essential in many of our behaviors.
Reaching for an object requires a specific temporal pattern
of activity across the muscles of the shoulder, arm and wrist.
Similarly, predicting events, such as when a traffic light will
change to red, demands an accurate estimate of relevant
durations, such as how long the light has been yellow.
These examples emphasize that temporal information
underlies both sensory and motoric processes; in everyday
activities, temporal intervals must be perceived and pro-
duced correctly.
Component analysis of temporal variability
For both perception and production there is, of course, vari-
ability in temporal accuracy. The components of a move-
ment differ from trial to trial, and on perception tasks, we
make slight misestimations of elapsed time intervals.
However, errors can also emerge from a variety of sources
that do not necessarily involve timing per se but, instead, re-
sult from variability in other task-related processes. Indeed,
movement commands may be initiated at the correct time,
but delays in implementation may occur downstream, lead-
ing to poor performance. An analogous situation exists for
perception. Returning to the example of the traffic light, an
error in predicting the change to red can be caused by mis-
judging the relevant duration or by failing to detect the
light’s transition from green to yellow rapidly enough. Box 1
describes two approaches used to partition variability in
temporal tasks into component sources.
A common mechanism
Are perception and production subserved by the same
clock? Evidence for a common timing mechanism can be
observed in the tendency for movements to become en-
trained automatically with external stimuli. For instance,
when we enter a nightclub, we immediately find ourselves
moving to the beat.
Along this line, Treisman and colleagues examined how
trains of evenly spaced clicks affected time perception’ and
production task.?. In both cases, the influence of the clicks
was shown to be dependent on their frequency; particular
frequencies systematically increased or decreased perceived
or produced intervals. Moreover, frequencies that shortened
estimates of perceived time, presumably by slowing down
the clock, tended to lengthen movement times, and vice
versa. This significant negative correlation between the biases
observed on the perception and production tasks is consist-
ent with the hypothesis of a common timing mechanism.
Ivry and Hazeltine also found similarities between the
performance of time perception and time production using
the slope method (see Box 1). Subjects produced temporal
intervals by tapping and made temporal judgments across a
range ofdurations (325-550 ms). The increase in the variance
as a function of the duration was comparable for the two
tasks. Assuming that this duration-dependent component of
the variance provides an estimate of variability in an internal
timing mechanism, these results implicate a common timing
process in the two tasks. Studies measuring variation among
individuals’ perceptuomotor skills have also supported this
hypothesis. Keele eta/.* observed significant correlations be-
tween performance in a repetitive tapping task and a duration
discrimination task. Furthermore, acuity on the perception
task was not correlated with performance on a non-temporal
motor task (see also Ref. 5), suggesting that the correlation
reflects an overlap in the neural systems subserving these tasks.
Neural substrates
Given the correlational nature of these studies, it is important
to look for converging evidence supporting the common
Copyright 0 1997, Elsevier Science Ltd. All rights reserved. 1364.6613/97/$17.00 PII: 513646613(97)01058-9
Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997
Hazeltine et al. - Timing mechanisms
Box 1. Isolating timing variability
Wing and Krisrofferson’ developed a simple model to analyze
variability on a motor timing task. Initially, subjects matched
their movements to a pacing signal (e.g. 2.5 Hz), and then con-
tinued to make periodic movements after the signals termi-
nated. Variability during the unpaced phase was hypothesized
to arise from two sources: an inrernal clock providing an appro-
p&rely rimed trigger for each movement and a motor imple-
mentation system that rranslared this signal into a movcmenr.
Two assumprions arc important for this model. Firstly, the vari-
ability of the clock and motor implemenration systems must be
independent. Secondly, rhe operation of these processes must
OCCUI in an ‘open-loop’ fashion so chat no attempt is made to
correct errors on previous taps. These assumptions mean that
motor implementation variability causes a negative correlation
between the timing of successive inrervals whereas the effects of
clock variability are restricted co a single interval (see Fig. a).
Thus, implemenrarion variability can be identified by the co-
variance between successive intervals. By subrracting this esti-
mate from rhe total variability, an estimate can be made of clock
variability.
The slope method is an alcernarive approach thar is based on a
well-established phenomenon in the time perception literature:
rhar variance increases linearly as a function of the square of rhe
duration being estimated. In other words, time perception ad-
heres to Weber’s la+. Ivry and Hazeltine’applied this logic in
a time production task in which subjects capped over a range of
time intervals (Fig. b). The function relating the variance to the
target durarion was very nearly linear and had a positive y-inrer-
ccpt. Thus, regression analyses were used to separate duration-
dependent and durarion-independent processes, defined by the
slope and inrercepr rcrms, respectively. It was assumed rhar the
duration-dependent variance reflected the operation ofan internal
clock, as other processes, such as signal detection or motor im-
plementation, should remain constam across different durations.
Close agreement is found for estimates of implementation vari-
ability calculated from the slope method or via the Wing and
Kristofferson model (see also Ref. g).
a Wing, A. and Kristoffenon. A. (1973) Response delays and the
timing of discrete motor responses Percept Psychophys. 14, 512
b Allan. L. (1979) The perception of time Percept. Psychophys. 26.
340-354
c Bizo, L.A. and White, K.G. (1997) Timing with controlled reinforce
density: implications for models of timing J. Exp. Psycho/. Anim.
Behav. Proc. 23,44-55
d Getty, D. (1975) Discrimination of short temporal intervals: a
comparison of two models Percept Psychophys. 18, l-8
e Gibbon, 1. (1991) Origins of scalar timing Learn. MO&. 22,3-38
Time c
Central process C C C C C C+D C C
I I I I I I I I I \ \ \ \ \ \ \ \
Peripheral process
Interval
Overt response
I I+D I-D I I I+D I
Tap Tap Tap Tap Tap Tap Tap Tap
Fig. a The Wing and Kristofferson model. Tapping variability is attributed to two processes. The central process issues tap com-
mands about every C ms. The peripheral process requires on average M ms to implement each command. The two processes proceed
independently; delays in one have no effect on the progression of the other. If implementation is delayed by D ms in the motor process
(see left half of figure), the interval whose termination is defined by the tap will be lengthened and the next interval, whose onset is
defined by that tap, will be shortened. In contrast, the influence of clock variability is limited to a single interval (see right half of fig-
ure). Note that in a typical tapping task, the interval C is much longer than the motor delay M.
clock hypothesis. Neuropsychological research provides just
such an alternative method for investigating the neural
bases of internal timing. The cerebellum has been shown to
be critical for a wide variety of timing tasks. Some of the
cardinal symptoms of cerebellar dysfunction (dysmetria and
intentional tremor) have been attributed to a loss in coordi-
nation of the temporal pattern between antagonist mus-
cle&‘. Cerebellar dysarthria is most evident on sounds that
require precise timing between different sets of ardculator?.
Patients with cerebellar lesions also show increased variabil-
ity on a repetitive tapping rask9, and when these data are
analyzed with the Wing and Kristofferson model (see Box I),
a dissociation is observed as a function of the locus of the
pathology . ” Patients with medial cerebellar lesions suffered
from marked peripheral (motor) variability, while those
with lateral cerebellar lesions showed impaired central
(clock) variability.
Importantly, temporal deficits following cerebellar
damage have been reported in the perceptual domain as
well. Ivry and Keele” observed that such lesions not only in-
creased clock variability during tapping, but also impaired
performance on a duration discrimination cask. The pa-
tients’ performance was comparable to control subjects on a
loudness discrimination task, demonstrating that their
deficits were not related to the perception of the stimuli or
to general factors such as motivation. Cerebellar patients
Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997
f Ivry, R. and Hazeltine, RX. (1995) Perception and production of
temporal intervals across a range of durations: evidence for a
common timing mechanism J. Exp. Psycho/. Hum. Percept.
Perform. 21,3-18
p Wing, A. (1980) The long and short of timing in response
sequences, in Tutorials in Motor Behavior (Stelmach, G. and
Requin, 1.. eds). pp. 463-486, North-Holland
A 800
s 600 5 8 400 s ‘C 9 200
0
B 800
$- 600 .c.
I I 1 I
325 400 475 550 Duration (m$)
325 400 475 Duration (ms2)
550
Fig. b Evidence for a common mechanism for time percep- tion and production. The variability functions for both the time
production (0) and perception tasks (0) were best described when
the variance was plotted as a function of the duration squared.
Regression lines show the best linear fit. (A) The target interval
was presented once in both the production and perception tasks.
(6) The target interval was presented four times in both tasks. In
both experiments, the slope was comparable for the production
and perception tasks. Moreover, the magnitude of the slope was
affected by the number of intervals produced or presented. This
suggests that duration-dependent variability is reduced when
there are multiple presentations of the target interval.
have also been found to perform poorly on tests of motion
and velocity perception”,‘*, tasks that may rely on a repre-
sentation of precise temporal information. Furthermore, bi-
lateral increases in regional cerebral blood flow (rCBF) were
observed with positron emission tomography (PET) when
subjects judged the duration of tone inrervals’3. While these
imaging results are in accord with the cerebellar timing
hypothesis, the activation can be attributed to a variety of
sources. Indeed, the timing condition was compared with a
control task in which subjects were not required to make
any sensory judgments. Thus attention, motor selection
and other cognitive demands may have contributed to the
activations observed.
Structures within the basal ganglia have also been im-
plicated in timing functions. Jeuptner etaLL3 reported rCBF
increases in the basal ganglia (as well as cerebellar temporal,
prefrontal and cingulate cortex) during their time percep-
tion task. While Ivry and Keele9 failed to find performance
differences between a group of Parkinson’s patients and
controls on the repetitive tapping task, more recent studies
have found that patients with either Huntington’s or
Parkinson’s disease have more variable inter-tap intervals
than control subjects r4,15. Moreover, the performance of the
patients with Parkinson’s disease was improved by L-dopa
medication, and those with asymmetric symptoms produce
more variable time intervals with the more affected limb.
Decomposition of the time interval variability using the
Wing and Kristofferson”j model (see Box 1) attributed the
increased temporal variability to both the central (clock)
and peripheral (motor) sourcesIs.
Cortical structures seem to contribute different, per-
haps more integrative computations in timing tasks than
those attributed to the cerebellum or the basal ganglia. For
example, premotor or supplementary motor cortical lesions
have been associated with deficits in rhythm production”.
Furthermore, laterality effects have been reported on tests of
temporal reproduction: for intervals ranging from 1 to 5 s,
patients with precentral left hemisphere lesions tend to pro-
duce shorter time intervals than control subjects, whereas
patients with precentral right hemisphere lesions tend to
produce longer time intervals’s.
Component analysis of the neural systems for timing: animal models Reports of timing deficits in many different patient groups
have led some researchers to conclude that temporal pro-
cessing is distributed across a variety of cortical and subcor-
tical systems . I5 However, the plurality of participating brain
regions does not necessarily imply functional homogeneity.
Indeed, theoretical models of neural clocks typically have
discrete components each ofwhich is envisioned to perform
a specific operation2,‘9.
Pharmacological and lesion studies with rats have
helped to differentiate the functions of the cortical and sub-
cortical areas implicated in temporal processing. One widely-
used task measures the frequency of an animal’s response
(bar presses) over time. During training trials, reward is pro-
vided for the first response that is made after a set interval
has expired (e.g. 30s). In this paradigm the rats learn to
withhold responding until after the set interval has expired,
despite the fact that the end of an interval is not cued ex-
plicitly. The response rate typically increases when the rats
judge that the interval has expired. Performance is then
evaluated on trials in which no reward is given. Drugs that
target dopaminergic and cholinergic systems produce a shift
in the time at which the response rate peaks. Importantly,
these drugs appear to change temporal processing via differ-
ent mechanisms. The dopamine-related effects are initially
dramatic but diminish with prolonged drug exposure. This
pattern is consistent with the idea that the drug changes
the speed of an internal clock and over time the animal is
able to modify its memory of the reinforcement time. In
contrast, the acetylcholine-related effects develop slowly but
Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997
Fig. 1 A three-component clock-counter system. The substantia nigra sends regularly paced signals through the
striatum to the pallidurn. where the internal segment (GPi) acts as an accumulator. Efference from this accumulator
is sent via the thalamus to the frontal cortex, where it is compared with stored representations of task-relevant du-
rations. This model has been proposed to account for data from animal studies using durations of 20 s or longer”.
It is not clear how such a system could time the activations of different muscle groups in multijoint movements,
are more long lasting, suggesting that they affect the ani-
mal’s stored representation of the target interval (reviewed
in Ref. 20).
Meckz” proposed a multi-component model to account
for these findings (Fig. 1). A dopamine-dependent, basal
ganglia system forms the pacemaker-accumulator mecha-
nism, and an acetylcholine-dependent, frontal cortex sys-
tem is associated with temporal memory and attention.
Moreover, within the basal ganglia system, there is evidence
for further specialization. Lesions to the substantia nigra
(SN) prevent rats from learning temporal discriminations
when the time interval is more than 20 s but less than 60 s,
whereas damage to the dorsal striatum disrupts performance
predominantly on the 60 s intervals. To explain this disso-
ciation, Meek hypothesized that the SN provides a time-
keeping pulse which the dorsal striatum integrates for
longer interval discriminations. Output from this integrator
is compared with stored representations of the target inter-
val via the frontal lobes. This model is also consistent with
the timing behavior observed in patients with Parkinson’s
disease, who tend to underestimate time intervalsa’. To
date, the cerebellum has not been considered within the
framework of this model.
One might hypothesize that the cer-
ebellum is critical for short intervals and
other structures, such as the basal ganglia
and cortex, take over for longer intervals.
However, it remains possible that as the
target interval lengthens, additional pro-
cesses involved in memory and atten-
tional functions come into play. It is im-
portant to note that the majority of the
animal timing studies have not included
non-temporal control tasks. For example,
do dopaminergic agents distort perfor-
mance when the task requires discrimi-
nations along a dimension such as stimu-
lus intensity! Studies along this line
would help develop a component analysis of temporal pro-
cessing tasks, and in particular, help identify if a target
neural system is linked specifically to temporal processing.
This strategy has been explored in a recent study with hu-
mansZ6. Cerebellar and frontal patients were tested on du-
ration discrimination tasks and pitch discrimination tasks
in which the interval between the standard and comparison
interval varied between short and long. Cerebellar patients
were impaired on both duration discrimination tasks (inter-
vals centered on 400 ms and 4 s), while frontal patients were
impaired on both tasks, but only when the stimuli were sep-
arated by 4 s. This result suggests that the frontal contribu-
tion on these tasks may not be specific to timing, but may
manifest whenever the working memory or attentional re-
quirements are increased.
Do distinct mechanisms operate at different temporal
intervals?
With some exceptions’5, research examining the role of the basal ganglia in timing has involved time intervals spanning
several seconds or more2K22, while studies focusing on the
cerebellum have used events of less than 1 s (Refs 7,9). Does
this mean that these two systems mediate timing at two dif-
ferent areas along a temporal spectrum? Some animal research
supports this hypothesis. For example, cerebellar lesions in
the rat led to a selective deficit on a duration discrimination
task when the stimulus range was centered at 500 ms, but
did not affect performance when the range was centered at
30 s (Ref. 23). Human studies also sug-
gest a qualitative change in temporal pro-
cessing for intervals longer than 1.5-2 s
(Ref. 24). When attempting to tap in
synchrony with a series of tones, responses
are anticipatory when the inter-tone inter-
val is less than 1800 ms; for longer inter-
vals, the responses tend to be reactive to
the tones25.
A further problem is that investigations into the role of
the cerebellum and basal ganglia in timing have emphasized
different dependent variables. Timing abnormalities follow-
ing cerebellar lesions are manifest as increased variability,
whereas basal ganglia research has focused on changes in
bias. The former effects have been viewed as evidence of a
noisy timing system, the latter as a change in clock speed.
A multi-component timing mechanism can produce
such deficits in a variety of ways. According to the model
presented by Meckzo, the dorsal striatum acts as an accumu-
lator that stores output from other sources. A related idea
builds upon the hypothesis that the basal ganglia are critical
for shifting cognitive set2 ’ 30. Patients with Parkinson’s dis-
ease take longer to switch from one task to another, particu-
larly when both tasks use the same stimuli. In such tasks, the
internal ‘set’ as well as the external stimuli determine the
correct response. If timing long durations, in the absence of
external cues, requires the accumulation of short intervals
and the updating of internal states, then dopamine depletion
may slow this process, leading to timing problems in
parkinsonian patients. The pacemaker providing the short
Trends in Cognitive Sciences - Vol. 1. No. 5, August 1997
Box 2. How a cerebellar timing system might operate
Cerebellar Purkinje cells are the site of a tremendous conver- gence of information; a single Purkinje cell may receive inputs from as many as 200 000 parallel fibers. This architecture is ideal for pattern recognitior@. These cells might learn to recog- nize the duration-dependent pattern of activity along rhe paral- lel fibers to signal when expected events should occur. Input to Purkinje cells could code duration in several ways. For example, granule, stellate and basket may interact to operate much like the networks transducing temporal information to spatial infor- mation described by Buonomano and colleagues’,d. According to this explanation, relatively slow pre- and post- synaptic mechanisms allow neurons to create local represen- tations of temporal information. For example, Buonomano and Merzenichd have shown how temporal estimates can be derived from a neural network with random variation in the time course of paired pulse facilitation and slow inhibitory posrsynaptic po- tentials. Because of these properties, the network is in a con- tinuously changing state after firing in response to an initial stimulus. Therefore, responses to subsequent stimulation are contingent on the length of the intervening interval. A second hypothesis proposes that the cerebellum may house populations of interval-based timers, each coding a different duration for a particular effector’. Just as in the visual cortex, in which specific neurons code for orientarion in each region of the visual field, units in the cerebella cortex could code for a
specific time inrerval. These units could then be associated with specific muscle groups, or be linked with perceptual infor-
Hazeltine et al. - Timing mechanisms
mation. When more than one effector is used, such as when an in- dividual taps with both the left and right index fingers, output from all the units is averaged, resulting in a decrease in variabil- i$. These two accounrs are not mutually exclusive. Such timing mechanisms are unlikely to be adequate for inter- vals much greater than 1 s. Longer durations may be timed by the striatal system described above, or by the cerebellum in con- cert with orher neural suuctures. For example, the cerebellum might time srate changes in the frontal cortex with the basal ganglia required to effect these state changes. In this model, the cerebellum could provide the timed initiation signal while the frontal strucmres act as an accumulator.
References
a Albur. J.S. (1971) A theory of cerebellar function Math. Biosci. 10,
25-61
b Marr, D.A. (1969) A theory of cerebellar cortex 1. Physiol. 202,
437470
c Buonomano, D.V. and Mauk, M.D. (1994) Neural network model
of the cerebellum Neural Computat 6,38-55
d 8uonomano, D.V. and Merzenich, M.M. (1995) Temporal
information transformed into a spatial code by a neural network
with realistic properties Science 267, 1028-1030
8 Ivry, R. (1996) The representation of temporal information in
perception and motor control Curr. Opin. Neurobiol. 6, 851-857
f Helmuth, L.L. and Ivry. R.B. (1996) When two hands are better than
one: reduced timing variability during bimanual movements
1. Exp. Psycho/. Hum. Percept Perform. 22, 278-293
intervals could be instantiated by other structures such as
the cerebellum or substantia nigra, but the critical represen-
tation of time would be related to the updating or changing
of behavioral stateP.
How the cerebellum might perform timing
Beginning with Braitenbere33, many theorists have sug-
gested that the cerebellum’s architecture is well-suited for
calculating the precise temporal relationship between differ-
ent inputs and between input and output patterns (see
Box 2). In a recent review, Raymond etaL3* argued that the
cerebellum plays a critical role in timing based on the com-
putational overlap between two behaviors for which the
structure has been shown to be critical, eyeblink condition-
ing (reviewed in Ref. 35) and the vestibular ocular reflex
(VOR). In both cases, motor outputs (e.g. eyeblinks or eye
movements) must be timed precisely to be effective.
Why should the cerebellum, traditionally labeled as a
motor structure, be involved in the perception of time? In
many cases, the distinction between time production and
time perception is unclear. Consider, for example, when
one attempts to grab a moving object, a task widely consid-
.ered to be the purview of the cerebellum. Here, motor com-
mands must be integrated with dynamic sensory infor-
mation to perform the action properly. Timing is relevant
for both predicting the location of the target object and
scheduling the activation of component muscle groups.
A system that learns to solve such complex problems
might analyze correspondences between the two sets of
events rather than independently compute the requisite
temporal information for the sensory and motoric
processes. Strong evidence for the cerebellum’s ability to as-
sociate inputs from different sensory channels comes from
experiments establishing its critical role in eyeblink condi-
tioninti5, a learning task in which there is a need to encode
the precise temporal relationships between the uncondi-
tioned stimulus and the conditioned stimulus36,37. Perhaps
this system has evolved so that it can continue to operate
when there is only one source of input. For example, when
the duration of a short tone must be judged, the cerebellum
may compare its representation of the tone with some
internal standard instead of .a second stimulus or motor
signal.
In motor control theory, it has been proposed that the
cerebellum coordinates and fine-tunes cortical outputs by
performing complex pattern recognition across a wide range
of inputs 38,39. This sort of computation can be used to inter-
pret time-varying representations of sensorimotor patterns
so as to estimate elapsed durations4”,41. In other words, the
cerebellum may train other brain regions to recognize and
anticipate representational states by identifying activations
across sets of neurons that are associated with particular in-
tervals. This hypothesis has some similarity with that pro-
posed by Courschesne and colleagues42,43, who suggest that
the cerebellum coordinates internal operations with antici-
pated sensory information. Discrete motor actions are gen-
erally less than 1 s in duration, and the hypothesized mecha-
nisms for sustaining local neural activity are also thought to
be limited to relatively short durations*‘. Thus, for longer
durations, such a system is unlikely to be sufficient.
Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997
Outstanding questions
l Evidence suggests that perception and production share a common timing mechanism. How is such a system implemented?
l How do ensembles of neurons provide accurate temporal information? l What are the temporal ranges of these ensembles? l How do the basal ganglia, cerebellum and frontal lobes interact to
perform temporal computations?
A pattern recognition approach to timing could help
account for some notable phenomena in the timing litera-
ture. Indeed, despite the hypothesis that the cerebellum is
critical for timing, there is a little evidence that it behaves as
a pacemaker or oscillator44. Moreover, the character of cer-
ebellar timing deficits (an increase in variability without a
bias to shorten or lengthen intervals) is in accord with a
non-oscillatory form of computation. In addition, Helmuth
and Ivry4’ compared unimanual and bimanual finger tap-
ping, and observed that the temporal variability of each fin-
ger was reduced during the bimanual condition. When
Wing and Kristofferson analysis was applied to these data,
the advantage was restricted to the clock component of the
variability. The researchers interpreted this advantage as
evidence that independent clocks subserved the two fingers,
and that the clock signals were averaged to determine move-
ment initiation time. In the present framework, the im-
provement would be attributed to a gain in the information
available. The signal from the two fingers would have richer
dynamic properties than the signal from just one finger.
Conclusion
Temporal error takes many forms, which suggests that mul-
tiple operations are required to compute time. The results
of human and animal studies suggest that different neural
systems are essential for performance in timing tasks. The
question that remains is how do we characterize the func-
tions of these different systems? While it is possible that the
representation of temporal information is distributed across
multiple systems, a more parsimonious view is that each sys-
tem makes a distinct contribution. Therefore, the use of a
range of tasks and temporal intervals is essential for a com-
plete understanding of the relations between these neural
systems in both temporal and non-temporal processes.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References
1 Treisman, M. et a/. (1990) The internal clock: evidence for a temporal
oscillator underlying time perception with some estimates of its
characteristic frequency Perception 19, 705743
2 Treisman, M.. Faulkner, A. and Naish, P.L.N. (1992) On the relation
between time perception and the timing of motor action: evidence for
a temporal oscillator controlling the timing of movement Quart 1.
Exp. Psycho/. 45,235-263
3 Ivry, R. and Hazeltine, R.E. (1995) Perception and production of
temporal intervals across a range of durations: evidence for a common
timing mechanism J. Exp. Rychol. Hum. Percept. Perform. 21,3-18
4 Keele. 5. et al. (1985) Do perception and motor production share a
common timing mechanism? Acta Psycho/. 60, 173-193
5 Keele. S., Ivry, R. and Pokorny, R. (1987) Force control and its relation
to timing J. Motor Behav. 19,9&l 14
6 Hallett, M., Shahani, B. and Young, R. (1975) EMG analysis of
patients with cerebellar lesions J. Neural. Neurosurg. Psychiatry 38,
1163-1169
7 How. 1.. Wild, B. and Diener, H.C. (1991) Cerebellar dysmetria at the
elbow, wrist, and fingers J. Neurophysiol. 65, 563-571
8 Ivry. R. and Gopal. H. (1992) Speech perception and production in
patients with cerebellar lesions, in Attention and Performance Volume
XIV: Synergies in Experimental Psychology, Artificial Intelligence, and
Cognitive Neuroscience (Meyer, D.E. and Kornblum, S., eds), pp.
771-802, MIT Press
9 Ivry, R. and Keele. S. (1989) Timing functions of the cerebellum 1. Cogn.
Neurosci. 1, 136-l 52
10 Ivry. R.. Keele. S. and Diener. H.C. (1988) Dissociation of the lateral and
medial cerebellum in movement timing and movement execution Exp.
Brain Res. 73, 167-180
11 Ivry, R. and Diener, H.C. (1991) Impaired velocity perception in patients
with lesions of the cerebellum 1. Cogn. Neurosci. 3, 355-366
12 Nawrot, M. and Rizzo, M. (1995) Motion perception deficits from
midline cerebellar lesions in human Vis. ttes. 35, 723-731
13 Jeuptner, M. et a/. (1995) Localization of cerebellar timing processes
using PET Neurology 45, 1540-l 545
14 Freeman, J.S. et a/. Abnormalities of motor timing in Huntington’s
disease Parkinsonism R&t. Dis. (in press)
15 D’Boyle, D.J.. Freeman, J.S. and Cody, F.W.J. (1996) The accuracy and
precision of timing of self-paced, repetitive movements in subjects
with Parkinson’s disease Brain 119, 51-70
16 Wing, A. and Kristofferson. A. (1973) Response delays and the timing
of discrete motor responses Percept. Rychophys. 14. 5-12
17 H&band, U. et a/. (1993) The role of premotor and the supplementary
motor area in the temporal control of movement in man Brain 116,
243-266
16 van Steinbuchel. N., Wittman, M. and Poeppel, E. (1996) Timing in
perceptual and motor tasks after disturbances of the brain, in Time,
Internal Clocks, and Movement (Pastor, M.A. and Artieda, J., eds), pp.
281-304, Elsevier
19 Gibbon, J. and Church, R.M. (1990) Representation of time Cognition
37,23-54
20 Meek. W.H. (1996) Neuropharmacology of timing and time perception
Cognit. Brain Res. 3. 217-242
21 Pastor, M.A. et a/. (1992) Time estimation and reproduction is
abnormal in Parkinson’s disease Brain 115. 21 l-225
22 Maricq, A.V. and Church, R.M. (1983) The differential effects of
haloperidol and methamphetamine on time estimation in the rat
Psychopharmacology 79, 10-l 5
23 Clarke, 5. et a/. (1996) Exploring the domain of the cerebellar timing
system, in Time, internal Clocks, and Movement (Pastor, M.A. and
Artieda, J., edr), pp. 143-164, Elsevier
24 Fraisse, P. (1963) The Psychology of Time, Harper
25 Mates. J. et a/. (1994) Temporal integration in sensorimotor
synchronization 1. Cogn. Neurosci. 6. 332-340
26 Ivry, R. and Mangles, 1. (1997) The many manifestations of a cerebellar
timing mechanism. Presented at the Fourth Annual Meeting of the
Cognitive Neuroscience Society, March 23.
27 Brown, R.G. and Marsden. CD. (1988) Internal versusexternal cues and
the control of attention in Parkinson’s disease Brain 111, 323-345
28 Downea, J.J. et a/. (1993) Alternating fluency in Parkinson’s disease
Brain 116,887-902
29 Hayes, A. et a/. Toward a functional analysis of the basal ganglia
1. Cogn. Neurosci. (in press)
30 Fimm. 6. eta/. (1994) Different mechanisms underlying set on external
and internal cues in Parkinson’s disease Brain Cognit. 25. 287-304
31 Killeen, P.R. and Fetterman, J.G. (1988) A behavioral theory of timing
psycho/. Rev. 95,274-295
32 Braitenberg, V. and Atwood, R.P. (1958) Morphological observations
on the cerebellar cortex J. Comp. Neural. 109. l-34
33 Braitenberg. V. (1983) The cerebellum revisited J. Theor. Neurobiol. 2,
237-241
34 Raymond, J.L.. Lireberger, S.G. and Mauk, M.D. (1996) The cerebellum:
a neuronal learning machine? Science 272. 1126-1131
35 Kim, J.J. and Thompson, R.F. (1997) Cerebellar circuits and synaptic
mechanisms involved in classical eyeblink conditioning Trends
Neurosc;. 20, 177-181
36 IVV. R. (1993) Cerebellar involvement in the explicit representation of
temporal information Ann. New York Acad. Sci. 682, 214-230
37 Perret. 5.. Ruiz, B. and Mauk, M. (1993) Cerebellar cortex lesions
Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997
Hazeltine et al. - Timing mechanisms
disrupt learning-dependent timing of conditioned eyelid responses
1. Neurosci. 13, 1708-1718
38 Houk, J.C. and Barto, A.G. (1992) Distributed sensorimotor learning, in
Tutorials in Motor Behavior /I (Stelmach, G.E. and Requin. 1.. eds). pp.
71-100, Elsevier
39 Thatch. W.T. (1996) On the specific role of the cerebellum in motor
learning and cognition: clues from PET activation and lesion studies in
man Behav. Brain SC;. 19. 501-502
40 Euonomano, D.V. and Mauk, M.D. (1994) Neural network model of the
cerebellum Neural Computat. 6, 3555
41 Buonomano, D.V. and Merzenich, M.M. (1995) Temporal information
transformed into a spatial code by a neural network with realistic
properties Science 267. 1028-1030
42 Allen, G. et a/. (1997) Attention activation of the cerebellum
independent of motor involvment Science 275, 1940-1943
43 Courschesne, E. and Allen, G. Prediction and preparation, fundamental
functions of the cerebellum Learn. Mem. (in press)
44 Keating. J.G. and Thach. W.T. (1995) Nonclock behavior of inferior
olive neurons: interspike interval of Purkinje cell complex spike
discharge in the awake behaving monkey is random J. Neurophysiol.
73, 132%1340
45 Helmuth. L.L. and Ivry. R.B. (1996) When two hands are better than
one: reduced timing variability during bimanual movements 1. Exp.
Psycho/. Hum. Percept. Perform. 22,278-293
Semantic networks: visualizations of knowledge
Roger T. Hartley and John A. Barnden
n conferences and in the literature,
complexity theory in computer science. Many ex
written, and yet it is our belief that none of them
b& between their use as a formal scheme for knowledge
use as an informal tool for thinking. In our
. i.._ . ( ^
T he history of the development of semantic networks is
well known (for an introduction to semantic networks, see
Box 1). Both Sowa’ and Lehmann’ have expounded in ex-
cellent scholarly fashion as to their origins in the study of
language. Their later development as a tool for representing
knowledge is also well knowr?, as is their role in building
computerized inference systemPO. Indeed, the triad of in-
telligent thought, logic and language will never be far from
our discussion. From all these sources we learn that seman-
tic networks have three main attributes:
(1) They originate in the conceptual analysis of language.
(2) They can have an expressiveness equivalent to first-
order logic, at least (although many do not).
(3) They can support inference through an interpreter
that manipulates internal representations.
Many people would go further and say that semantic net-
works are indistinguishable from formal logic representa-
tions”. However, there is something missing here. The visual
aspect of the semantic network idea is clearly important. As
Sowa says: ‘network notations are easy for people to read”
and this pragmatic aspect ofthe formalism cannot be ignored.
According to Sowa: ‘graphs...can keep all the information
about an entity at a single node and show related information
by arcs connected directly to that node”. In contrast, in
symbolic logic notations: ‘the scattering of information not
only destroys the readability of the formula, but also obscures
the semantic structure of the sentence from which the formula
was derived. So the battle is joined! The visual aspects of the
semantic network notation are preferred (at least by Sowa)
over the arcane, but more traditional notation of symbolic
logic. Interestingly, this traditional notion was invented by
C.S. Peirce before he abandoned it in favor of a diagram-
matic form’a.
In this paper, we hope to show that this argument is only
one component of a larger, more complex one involving the
nature of semantics; we will also show how different notations
can lead to different systems with different pragmatic uses.
Meaning
The design and use of a knowledge representation revolves
around the business of meaning. Actually, one spin-off from
studies in natural language provides a good start, namely,
the meaning triangle of Ogden and Richards”. The triangle
relates objects in the real world, concepts that correspond to
Copyright 0 1997, Elsevier Science Ltd. All rights reserved. 1364-6613/97/$17.00 PII: 51364-6613(97)01057-7
Trends in Cognitive Sciences - Vol. 1. No. 5, August 1997