www.sciencedirect.com

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1

Available online at

ScienceDirect

Journal homepage: www.elsevier.com/locate/cortex

Research report

The cerebellum and visual perceptual learning:Evidence from a motion extrapolation task

Cristina Deluca a,1, Ashkan Golzar a,b,1, Elisa Santandrea a,1,Emanuele Lo Gerfo a, Jana E�sto�cinov�a a, Giuseppe Moretto c,Antonio Fiaschi a,d, Marta Panzeri e, Caterina Mariotti e, Michele Tinazzi a,d

and Leonardo Chelazzi a,d,*

a Department of Neurological and Movement Sciences, University of Verona, Verona, Italyb Department of Physiology, McGill University, Montreal, Canadac Neurology Unit, Borgo Trento Hospital, Verona, Italyd National Institute of Neuroscience, Verona, Italye Department of Genetics of Neurodegenerative and Metabolic Diseases, IRCCS Foundation Carlo Besta, Milan, Italy

a r t i c l e i n f o

Article history:

Received 27 August 2013

Reviewed 16 February 2014

Revised 9 April 2014

Accepted 26 April 2014

Action editor Branch Coslett

Published online 2 June 2014

Keywords:

Cerebellum

Perceptual learning

Perceptual adaptation

Motion extrapolation

* Corresponding author. Department of NeuVerona, Strada Le Grazie 8, 37134 Verona, It

E-mail address: leonardo.chelazzi@univr.1 The first three authors (C.D., A.G., and E

http://dx.doi.org/10.1016/j.cortex.2014.04.0170010-9452/© 2014 Published by Elsevier Ltd.

a b s t r a c t

Visual perceptual learning is widely assumed to reflect plastic changes occurring along the

cerebro-cortical visual pathways, including at the earliest stages of processing, though

increasing evidence indicates that higher-level brain areas are also involved. Here we

addressed the possibility that the cerebellum plays an important role in visual perceptual

learning. Within the realm of motor control, the cerebellum supports learning of new skills

and recalibration of motor commands when movement execution is consistently per-

turbed (adaptation). Growing evidence indicates that the cerebellum is also involved in

cognition and mediates forms of cognitive learning. Therefore, the obvious question arises

whether the cerebellum might play a similar role in learning and adaptation within the

perceptual domain. We explored a possible deficit in visual perceptual learning (and

adaptation) in patients with cerebellar damage using variants of a novel motion extrapo-

lation, psychophysical paradigm. Compared to their age- and gender-matched controls,

patients with focal damage to the posterior (but not the anterior) cerebellum showed

strongly diminished learning, in terms of both rate and amount of improvement over time.

Consistent with a double-dissociation pattern, patients with focal damage to the anterior

cerebellum instead showed more severe clinical motor deficits, indicative of a distinct role

of the anterior cerebellum in the motor domain. The collected evidence demonstrates that

a pure form of slow-incremental visual perceptual learning is crucially dependent on the

intact cerebellum, bearing the notion that the human cerebellum acts as a learning device

for motor, cognitive and perceptual functions. We interpret the deficit in terms of an

inability to fine-tune predictive models of the incoming flow of visual perceptual input over

rological and Movement Sciences, Section of Physiology and Psychology, University ofaly.it (L. Chelazzi)..S.) contributed equally to this work.

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 53

time. Moreover, our results suggest a strong dissociation between the role of different

portions of the cerebellum in motor versus non-motor functions, with only the posterior

lobe being responsible for learning in the perceptual domain.

© 2014 Published by Elsevier Ltd.

1. Introduction

Perception of the visual world is adjusted (fine-tuned) by

experience e known as visual perceptual learning, which can be

defined as the improvement in detecting and discriminating

low-level or more complex features of the visual input as a

result of extended practice with a specific set of stimuli and

task (Ahissar, Nahum, Nelken, & Hochstein, 2009; Byers &

Serences, 2012; Censor, Sagi, & Cohen, 2012; Dosher & Lu,

2009; Fahle, 2009; Gilbert, Li, & Piech, 2009; Lu, Hua, Huang,

Zhou, & Dosher, 2011; Roelfsema, van Ooyen, & Watanabe,

2010; Sasaki, Nanez, & Watanabe, 2010). Although the under-

standing of perceptual learning phenomena and the under-

lying neuralmechanisms is far from complete, a long tradition

of research in the field has focused on plastic changes occur-

ring along the cerebro-cortical pathways, including at the

earliest stages of processing, though increasing evidence in-

dicates that higher-level brain areas are involved as well

(Ahissar et al., 2009; Byers & Serences, 2012; Fahle, 2009;

Gilbert et al., 2009; Lu et al., 2011; Roelfsema et al., 2010;

Sasaki et al., 2010). So far the possibility that the cerebellum

contributes significantly to perceptual learning has never

been addressed.

It is instead solidly established that the cerebellum is

critical for motor control, supporting smoothly unfolding and

precise movements (Dow & Moruzzi, 1958; Rothwell, 1994,

chap. 10, pp. 387e445). This role is largely mediated by a key

contribution of cerebellar mechanisms to various forms of

motor learning (Blazquez, Hirata, & Highstein, 2004; Doyon,

1997; Imamizu et al., 2000; Ioffe, Chernikova, & Ustinova,

2007; Krakauer & Shadmehr, 2006; Manto et al., 2012; Smith

& Shadmehr, 2005; Thach, 1998) and adaptation (Bastian,

2008; Golla et al., 2008; Manto et al., 2012; Prsa & Thier, 2011;

Tseng, Diedrichsen, Krakauer, Shadmehr, & Bastian, 2007;

Werner, Bock, & Timmann, 2009), which allow the system to

acquire new skills and re-calibrate motor commands when

movements become inaccurate as a result of consistent per-

turbations. The latter notion, initially formulated in theoret-

ical, computational terms on the basis of the available

knowledge at the time (Albus, 1971; Marr, 1969), has subse-

quently been supported by a vast array of converging empir-

ical observations (e.g., Ito, 2006). Moreover, different forms of

plasticity within the cerebellar circuitry have been hypothe-

sized tomediate this role (e.g., Carey, 2011; Gao, van Beugen,&

De Zeeuw, 2012; Lamont & Weber, 2012).

More recently, numerous findings have led to the emerging

notion that the cerebellum is also involved in non-motor

functions (e.g., Bellebaum & Daum, 2011; Bostan, Dum, &

Strick, 2013; E, Chen, Ho, & Desmond, 2012; Ito, 2008; Leiner,

Leiner, & Dow, 1989; Leiner, Leiner, & Dow, 1991;

Schmahmann, 1998; Strick, Dum, & Fiez, 2009; Timmann &

Daum, 2007; Timmann et al., 2010), including cognitive do-

mains e such as language (Durisko & Fiez, 2010; Marvel &

Desmond, 2010; Murdoch, 2010), executive control (Bellebaum

& Daum, 2007), emotion (Strata, Scelfo, & Sacchetti, 2011;

Timmann et al., 2010) and working memory (Ben-Yehudah,

Guediche, & Fiez, 2007; Durisko & Fiez, 2010), sensory/percep-

tual domains (Bastian, 2011; Bhanpuri, Okamura, & Bastian,

2012; Molinari et al., 2008) and time processing (Bueti,

Lasaponara, Cercignani, & Macaluso, 2012; Ivry & Spencer,

2004). This view of a more diverse role of the cerebellum fits

well with the detailed description of widespread and bidirec-

tional cerebro-cerebellar connections,whichengagemotorand

non-motor areas of the cerebral cortex, including frontal, pre-

frontal, parietal and temporal territories (Bostan et al., 2013;

Stoodley, 2012; Strick et al., 2009; Sultan et al., 2012).

Whether or not the cerebellum contributes significantly to

non-motor functions, and especially the very nature of this

putative contribution, is nevertheless still a matter of debate

(e.g., Glickstein, 2007; Glickstein & Doron, 2008). Mainly based

on the complex, often times subtle, and highly variable

pattern of cognitive disturbances which can be detected in

patients suffering from cerebellar damage, the proposal has

been made that the cerebellum contributes to cognitive

functions in a way similar to its role in the motor domain,

allowing well-coordinated and smoothly unfolding, cognitive

processes (Schmahmann, 1991; 1998). This has generated the

idea that damage to the cerebellum leads to what has been

termed “dysmetria of thought” (Schmahmann, 1991; 1998).

Consistent with the view that the cerebellum exerts similar

functions within the motor and the cognitive domain, the

obvious prediction would be that the cerebellum mediates

learning phenomena also within non-motor, cognitive func-

tions, and within perception. Surprisingly, however, the claim

that the human cerebellum plays a key role in learning within

pure forms of perceptual processing has so far never been

made (but see Section 4 for an analysis of related findings). It is

noteworthy that in a number of recent, non-selected and

highly authoritative review articles on perceptual learning,

terms such as “cerebellum” or “cerebellar” do not appear a

single time (Ahissar et al., 2009; Byers& Serences, 2012; Censor

et al., 2012; Fahle, 2009; Gilbert et al., 2009; Lu et al., 2011;

Roelfsema et al., 2010; Sasaki et al., 2010). Likewise, in a

comparable number and type of review articles on cerebellar

function, including in the cognitive domain, the term

“perceptual learning” is again entirely absent (Bastian, 2011;

Ito, 2008; Ramnani, 2006; Schmamann, 2010; Stoodley, 2012;

Strick et al., 2009). Similarly, no mention to a possible role of

the cerebellum in “perceptual learning” can be found in a

comprehensive Special Issue addressing the cerebellar

contribution to non-motor functions, which appeared

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 154

recently in the journal Cortex (Language, cognition, and the

cerebellum: grappling with the enigma, 2010).

To fill this major gap, we devised a series of experiments to

test whether visual perceptual learning is compromised, or

abolished altogether, following cerebellar damage. For this

purpose, variants of a novel visual motion extrapolation para-

digm were developed, which enabled us to assess slow and

cumulative improvement at the task as a result of extended

practice (slow learning), as well as fast recalibration of

perceptual judgments as a result of a perturbing manipulation

(adaptation).We chose this type of task assuming that it would

be exquisitely suited to assess the role of the cerebellum in

creating and refining an internal model of a sensory event that

changes over time. More specifically, the task was designed to

assess theability ofparticipants tofinely evaluate thepatternof

motion of a decelerating visual target and infer the spatio-

temporal unfolding of its trajectory through a predictive pro-

cess (Courchesne & Allen, 1997; Miall, Weir, Wolpert, & Stein,

1993; also see Section 4). Noteworthy, as suggested in a num-

ber of studies (e.g., Sokolov, Ehrenstein, Pavlova, & Cavonius,

1997; Watamaniuk, 2005; Watamaniuk & McKee, 1995), the

ability to extrapolate motion can be ascribed to the same sen-

sory/perceptual mechanisms that support motion perception.

Patients with focal anterior versus posterior cerebellar

damage or with diffuse cerebellar pathology were tested,

along with a suitable group of age- and gender-matched

controls. The collected evidence demonstrates that, as a

group, cerebellar patients show a marked deficit in visual

perceptual learning, likely reflecting an inability to fine-tune

predictive models of the incoming flow of visual perceptual

input over time. Moreover, a striking correlation emerged

between the extent of the learning deficit and the site of the

lesion in focally damaged patients, with strong deficits in

patients with posterior (but not anterior) damage. The latter

finding fits well with data in the literature pointing to a

prevalent role of the posterior cerebellum in non-motor

functions (Salmi et al., 2010; Stoodley & Schmahmann, 2010;

Stoodley, Valera, & Schmahmann, 2012).

2. Materials and methods

2.1. Participants

Fifteen normal subjects, ten patients with cerebellar damage

and as many gender- and age-matched controls were enrolled

in the study. Specifically, seven young normal adults (mean

age ± SD: 24 ± 1.7 years; 6 females) participated in the slow

perceptual learning experiment, and eight (mean age ± SD:

24 ± 4.9 years; 4 females) participated in the adaptation exper-

iment.Theywereall studentsat theUniversity ofVerona, right-

handed and with normal or corrected-to-normal vision.

Additionally, six cerebellar patients (mean age ± SD:

45.5 ± 6.3 years; 3 females) and six gender- and age-matched

control subjects (mean age ± SD: 41.2 ± 8 years) were

recruited for the slow perceptual learning experiment. Finally,

four cerebellar patients (mean age ± SD: 57 ± 16.9 years; 4

males) and four gender- and age-matched control subjects

(mean age ± SD: 56.5 ± 12.4 years) were instead recruited for

the adaptation experiment.

Overall, five patients suffered from focal ischemic cere-

bellar lesions, while five suffered from diffuse cerebellar

degeneration (4 patients: spinocerebellar ataxia type 1e SCA1;

1 patient: type 2, SCA2). Magnetic resonance imaging (MRI)

and clinical records indicating an isolated cerebellar lesion

were considered critical for the inclusion of focal patients,

while MRI indication of no additional extracerebellar pathol-

ogy and a severity of ataxia higher than 5, as measured by

means of the Scale for the Assessment and Rating of Ataxia

(SARA; Schmitz-Hubsch et al., 2006), were considered critical

for the inclusion of patients with cerebellar degeneration.

Focal patients were recruited from the Verona Neurology

Clinics (Department of Neurological and Movement Sciences

and Civil Hospital) and from IRCSS Santa Lucia in Rome, while

patients with cerebellar degeneration were recruited through

the ataxia clinic of the Besta Institute in Milan.

Two additional patients suffering from spinocerebellar

ataxia (SCA1), but at a very early phase of the disease, were

subsequently tested with the slow perceptual learning para-

digm (see Supplementary material). Finally, two additional

patients with focal ischemic cerebellar lesions were enrolled

in a later phase of the study in order to verify the validity of the

conclusions about a crucial role of the posterior cerebellum in

learning within the perceptual domain. Details about the pa-

tients enrolled in the study are provided in Table 1 (for pa-

tients with focal cerebellar damage, see also Fig. 4 for a

detailed description of the size and site of the lesion, including

the identification of the specific cerebellar lobules involved).

Six of the normal subjects, 6 of the cerebellar patients and

their gender- and age-matched control subjects, participating

in either of the two main experiments, performed the control

experiment to assess speed discrimination performance (see

Supplementary material), again as indicated in Table 1.

All participants were naïve as to the purpose of the study

and took part in only one of the two main experiments.

Experimental procedures were designed according to the

Declaration of Helsinki and approved by the local Ethics

Committee. All participants read and signed an informed

written consent. While healthy subjects received a small

monetary compensation for their participation, calculated on

the basis of their actual performance at the task, only a

reimbursement of expenses was given to the patients. We

were concerned that differences in monetary compensation

between patients and healthy participants might be a con-

founding factor. For this reason, at the end of each trial, we

displayed a positive or negative score for correct

versus incorrect responses in order to elicit comparable levels

of motivational engagement across all participants (see Sec-

tion 2.3). Furthermore, we underscore that we found a striking

divergence in learning ability within the patient group

depending on the site of the focal damage (and the stage of the

degenerative disease) (see Section 3 and Supplementary

material). This rules-out any potential influence of monetary

compensation to determine the main results of the study

since all patients were treated equally in this regard.

2.2. Lesion assessment

For all the focal patients enrolled in the study, we performed a

volumetric assessment of lesion size. Lesions were drawn by

Table 1 e Detailed information about the individual cerebellar patients enrolled in the study. In addition to demographicdetails (sex and age), the type of cerebellar pathology is reported together with information about either the severity of thedisease or the site of the lesion, and the SARA score (Schmitz-Hubsch et al., 2006). The rightmost column reports theexperiment (or experiments) in which the patient was enrolled.

Patient ID Sex Age Cerebellarpathology

Severity of thedisease

Site of the lesion SARA Experiment

B.D. F 40 Focal damage Posterior, bilateral 2 Slow learninga

G.C. M 53 Focal damage Posterior, left 3 Slow learning

M.F. F 49 Focal damage Anterior, right 10.5 Slow learning

D.A. F 46 Focal damage Anterior, right 2 Slow learning

C.L. M 45 Focal damage Posterior, left 4 Adaptationa

G.Z. M 53 Focal damage Anterior, right 3 Slow learningc

Z.G. M 37 Focal damage Anterior, right

Posterior, bilateral

3 Slow learningc

R.M. M 49 SCA1 Medium 14.5 Slow learninga

M.R. M 36 SCA1 Medium-severe 10.5 Slow learninga

N.D. M 42 SCA1 Medium-severe 13 Adaptationa

L.B.G. M 64 SCA1 Medium-severe 15 Adaptationa

Z.R. M 78 SCA2 Medium-severe 12 Adaptation

M.C. F 52 SCA1 Mild 5 Slow learningb

C.B. F 34 SCA1 Pre-symptomatic 2 Slow learningb

a Participation to the control experiment to assess sensitivity in speed discrimination.b The two additional patients at a very early stage of the degenerative disease who were tested with the slow learning paradigm in a second

phase of the study.c The two additional patients with a focal cerebellar lesion who were enrolled in a late phase of the study to confirm the main results from the

original group of participants.

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 55

an examiner on the T1-weighted template MRI scan from the

Montreal Neurological Institute provided with MRIcron soft-

ware (Rorden & Brett, 2000; available at http://www.mricro.

com/mricron). Lesion volumes in cubic centimeters (cc) were

estimated by superimposing each patient's lesion onto the

standard brain.

In addition, the location of cerebellar lesions in all focal

patients was analyzed in order to precisely identify the

damaged lobules. Specifically, we established the involve-

ment of specific lobules according to Schmahmann's atlas

(Schmahmann, MacMore, & Vangel, 2009) by comparing the

patients' individual images of the cerebellum in the axial

plane with equivalent sections in the atlas (from superior to

inferior sections). According to the atlas, we identified the

primary fissure that demarcates the anterior lobe from lobule

VI, and the superior posterior fissure that separates lobule VI

from lobule VII in the posterior lobe. Based on these land-

marks, we identified which specific lobules were damaged at

the level of both the anterior and the posterior lobe.

Finally, the location of cerebellar damage in focal patients

was also analyzed in order to ascertain the degree of

involvement of the deep cerebellar nuclei. To this aim, the

location of tissue damage was determined with reference to

theMRI Atlas of theHumanCerebellar Nuclei (Dimitrova et al.,

2002). In most cases, 9-to-10 images of the patients' cere-bellum in the axial plane from superior to inferior were

available for comparison with equivalent sections in the atlas.

The axial sections that contain the cerebellar nuclei in the

atlas are sections z_25 to z_43. The involvement of the deep

cerebellar nuclei in each patientwas examined by comparison

of the critical horizontal sections (z_25ez_43) in the MRI atlas

with the corresponding levels on the patient MRI to determine

whether tissue damage included the expected location of the

nuclei.

2.3. Experimental procedure

All experimental protocols were developed and controlled

with the MATLAB software by means of the Psychophysics

Toolbox extension 2.54 (Brainard, 1997; Pelli, 1997). Partici-

pants sat in a quiet and dimly lit room, at a viewing distance of

70 cm from a computer monitor. A chinrest was used to

minimize head movements, and maintenance of central fix-

ation was required.

2.3.1. Slow perceptual learningA novel motion extrapolation task was devised in order to

measure slow and cumulative improvement in visual

perceptual judgment as a result of extended practice. On each

trial the partially occluded trajectory of a moving target was

shown. Its complete decelerating trajectory was to be inter-

nally extrapolated to indicate where the target had virtually

stopped within a structured visual display (Fig. 1A; stimulus).

The latter depicted the upper half of a circular surface con-

taining six alternating, black and white, concentric curved

sectors (the distance from fixation to the external border of

the outermost sector was 11.5�). A green fixation cross

(.65� � .65�) was presented at the center of the entire imagi-

nary circumference, i.e., at the bottom of the actual display

(Fig. 1A). The moving target consisted of a circular gray

element with a radius of .15� of visual angle; its luminance

was set to a value corresponding to themean luminance of the

display, in order to avoid the occurrence of the so-called

“footstep illusion”, according to which the perceived velocity

of a moving stimulus increases as the contrast of the stimulus

increases (Anstis, 2001). Starting position of the moving

stimulus was randomly chosen within an area of about 1�

around the fixation cross. Direction of straight motion ranged

from �35 to þ35� around the upward direction. Six ranges of

Fig. 1 e A novel paradigm to assess slow visual perceptual learning. (A) Sequence of events in an example trial of the motion

extrapolation task. On each trial, the partially occluded trajectory of a decelerating stimulus (here enlarged and indicated by

an orange arrow for illustration purposes) was displayed. Its complete trajectory was to be internally extrapolated to then

indicate where the target had virtually stopped within the structured visual display. Following a non-speeded verbal

judgment, a positive (or negative) score was assigned for correct (or wrong) responses. Finally, feedback was provided (see

Section 2). (B) Slow perceptual learning in normal observers. The average learning curve (mean and SEM in dark and light

shades of green, respectively) was constructed by means of the dynamic analysis of learning (Smith et al., 2004) (see Section

2). Ordinate is probability of a correct response and abscissa is trial number. Baseline performance (first 25 trials) is also

reported (black square). (C) Rate of learning. Learning curves (mean) were constructed by smoothing data from sequential

trials performed by each subject with a boxcar filter of a fixed size (100 trials), shifted by one single trial at the time, and then

by averaging data across subjects (see Section 2). Ordinate is proportion of correct responses. Conventions as in panel (B). (D)

Learning trial. Results from a series of one-tailed, one-sample t-tests comparing baseline performance with each

subsequent point in the learning curve depicted in panel (C) (see Section 2). Ordinate is the p value and abscissa is trial

number. The dashed line indicates p value ¼ .05.

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 156

initial velocity of the moving stimulus were used, between

4.76 and 11.5�/sec, and they were selected to be within the

interval of optimal sensitivity for human observers (De Bruyn

& Orban, 1988; McKee & Nakayama, 1984). As soon as the

stimulus started moving, it also started to decelerate with a

deceleration constant of 5.7�/sec2 and it virtually stopped

within one of the six sectors. The pattern of initial decelera-

tion of themoving target and its starting position provided the

critical information to the subject to infer the end point of the

trajectory on each trial.

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 57

Crucially, the stimulus trajectory was partially occluded

and the range of visibility of the trajectory was set for each

individual subject prior to the start of the experiment by

means of a staircase procedure (Levitt, 1971; see

Supplementarymaterial) in order to adjust the subjects' initialperformance to chance level. Based on the actual range of

visibilities used in the experiment, the average range of ec-

centricities at which the stimulus was visible spanned be-

tween 1.9 and 8.6� of visual angle.The stimulus display lasted 2500msec, afterwhich subjects

were required to perform a non-speeded verbal judgment,

reporting the color of the sector in which the extrapolated

trajectory came to its end (binary decision). Following the

report, a positive (þ2 points) or negative (�1 point) score was

assigned for correct and wrong responses, respectively. Next,

an updated cumulative score was shown for 500 msec and

consisted in a green (after correct responses) or red (after er-

rors) number, accompanied by an arrow pointing upwards or

downwards respectively (Fig. 1A; reward).

Finally, feedbackwas provided, corresponding either to the

re-play of the entire unoccluded trajectory of the target (in 75%

of the trials; lasting 2500msec) or to its end point (in 25% of the

trials; lasting 1000 msec). In either case, the end point of the

trajectory was signaled by a change in the color of the stim-

ulus from gray to green (Fig. 1A; feedback). Importantly, cen-

tral fixation was required both during the stimulus display

and during the feedback display. Participants were expected

to actively use feedback to create and update an internal

model of the constant deceleration pattern of the moving

stimulus. Crucially, based on a comparison between the

extrapolated trajectory of the moving stimulus and the actual

end point of the trajectory, as provided by feedback, subjects

should manage to progressively refine their internal model, in

turn becoming more proficient at the task.

Each participant completed at least one preliminary session

(typically two) in which a staircase procedure (Levitt, 1971) was

applied to adjust initial performance at 50% (see Fig. S1 and

Supplementary material), followed by at least eight experi-

mental sessions on consecutive days (except weekends), each

comprising 388 trials. Trial order was randomized in order to

avoid that the sequence of initial velocities of the moving

stimulus could yield any systematic impact on performance,

since it is known that motion perception is affected by previ-

ouslyviewedvelocities (Makin, Poliakoff,Chen,&Stewart, 2008).

In addition, each control subject performed the task with iden-

tical experimental conditions (same number of daily sessions,

same sequence of trials, adjustment of the initial performance

atchance leveland,whenpossible, thesamefractionofvisibility

of the moving trajectory) to those applied for the matched

patient.

2.3.2. Fast perceptual adaptation taskAvariantof themotionextrapolation taskused for theprevious

experiment was developed following the logic of typicalmotor

adaptation tasks, in which good performance at a specific

paradigm is challenged by introducing a perturbation that

immediately results in a significant drop in performance.

Adaptation is then assessed as the rate and amount of per-

formance recovery following the initial drop. Task settings and

the temporal sequence of events were identical to those

previously described. Again, the range of visibilities of the

trajectory was set for each individual subject prior to the start

of the experiment by means of a staircase procedure (Levitt,

1971; see Supplementary material), but in this case perfor-

mance was adjusted to an initial level of 85% of correct re-

sponses. Accuracy was then consolidated during one or two

experimental sessions (basal sessions; on consecutive days) in

which the task was identical to what previously described. In

other words, during basal sessions, subjects were expected to

consolidate an internalmodel of thedeceleratingmotionof the

stimulus.Aperturbationwas then introducedat the startof the

following session of the experiment, run on the subsequent

day, and this caused a suddenand robust drop in performance.

The perturbation consisted in a form of false feedback (Choi &

Watanabe, 2012), andwassimilar in logic to that appliedwithin

the context of saccadic adaptation paradigms (McLaughlin,

1967), where fast adaptation is triggered by the systematic

misplacement (jump) of the saccadic target by a certain dis-

tance during the execution of a targeted saccade, such that a

mismatch is detected between the intended saccade landing

position and the feedback obtained after saccade termination.

Specifically, in this kind of paradigm, a peripheral saccade

target is initially presented and then shifted to a new position

while the subject has already initiated a saccade towards it, so

that the subject is not aware of the target jump. In an initial

phase, the apparent error in the landing point of the saccades

drives a corrective saccade to foveate the target correctly.

Subsequently, based on the detected error, the subject gradu-

ally learns to adapt the amplitude of the initial saccade in order

to correctly foveate the shifted target (McLaughlin, 1967).

Notably, adaptation is typicallymeasuredwithin ~30e60 trials

(Deubel, Wolf, & Hauske, 1986; Frens & van Opstal, 1994; Hopp

& Fuchs, 2004). Following the same logic, during the critical

session of our fast perceptual learning task, feedback provided

to the subjects after completion of each trial was not veridical;

namely, the feedback was provided as if the moving stimulus

was decelerating at a lower rate (5.0�/sec2) than during the

actual task (5.7�/sec2), so that its trajectory extended further

(Fig. 5A). Subjects were not alerted of any modification to the

task when the perturbation was introduced. Moreover, both

during thefinal basal sessionandduring theperturbedsession,

on 100% of the trials feedback consisted of the end point of the

trajectory, since the re-play of the entire unoccluded trajectory

could allow subjects to become quickly aware of the pertur-

bation. To reiterate, in the critical (perturbed) session, the

applied deceleration constant was 5.7�/sec2 during the stim-

ulus display and 5�/sec2 during the feedback display. The

change in deceleration rate during feedback was such that,

given the same rangeof startingpositions and initial velocities,

the target stopped within the same sector as before the

perturbation on ~50%of the trials andwithin the sector farther

awayon the remaining trials (Fig. 5A). Followingan initial fall in

performancedue to theperturbation,weexpectedparticipants

to update their internal model rather rapidly, thus recovering

at least partly the pre-perturbation level of accuracy.

2.4. Statistical analyses

For all statistical analyses performed in relation to the main

experiments, data collected from trials in which the trajectory

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 158

of the moving target reached its end point in the first (most

proximal) or in the last (most distal) sector of the structured

visual display (Fig. 1A) were not included. The rationale for

excluding them was that they constituted an especially easy

condition, most likely because the boundary position of those

two sectors represented an additional clue for correct judg-

ment, in turn allowing subjects to virtually exclude one of the

possible responses. As assessed in normal subjects partici-

pating to the slow learning experiment, the average perfor-

mance in such easy conditions was well above chance already

in the first experimental session (79.3% ± 4 SEM), and this

obviously reduced the opportunity for further improvement of

performance. Nonetheless, accuracy improved slightly along

the experiment, reaching an accuracy of 84.7% ± 2.1 SEM on

average, which corresponded to a marginally significant

improvement (p ¼ .069; one-tailed, paired t-test).

2.4.1. Learning curves and learning (or adaptation) trialcalculationIn order to examine the temporal dynamics of learning in the

slow learning paradigm, we first concatenated data collected

across all experimental sessions. Analyses were performed on

a total of 2244 trials per subject, corresponding to the mini-

mum number of trials performed by all single participants.

Learning curves for a given group of subjectswere constructed

by smoothing data from sequential trials performed by each

subject in the group with a boxcar filter of a fixed size (100

trials), shifted by one single trial at the time, and then by

averaging data across subjects. Therefore, each point in the

curve represents the average proportion of correct responses

calculated in a sequence of consecutive trials. In order to

describe learning curves with a single statistical measure

reflecting speed of learning, we estimated the learning trial as

the initial trial of the epoch in which performance became e

and then remained, significantly better than baseline (see

Seitz, Kim, & Shams, 2006). This was estimated by means of a

running one-tailed, one-sample t-test comparing each sub-

sequent epoch along the learning curve with the average

baseline performance, calculated over the first 25 trials. When

we studied learning in pairs of patients, we performed each t-

test by using data from two subsequent epochs along the

learning curve in order to increase sample size.

Learning curves for the adaptation task were constructed

by applying the above procedure for the perturbed session,

and a filter size of 30 trials was used in this case. The adapta-

tion trialwas estimated as the initial trial of the epoch inwhich

performance of the subjects became e and then remained,

significantly better than the baseline in the perturbed session.

Again, this was estimated by means of a running one-tailed,

one-sample t-test comparing subsequent epochs along the

learning curve with the average baseline performance,

calculated here over the first 10 trials of the perturbed session.

2.4.2. Dynamic learning curves and learning scoreAs a further measure of learning efficiency, we calculated the

global amount of improvement as a function of practice in

single participants. Learning curves were constructed ac-

cording to the method developed by Smith et al. (2004),

namely the dynamic analysis of learning. This method pro-

vides not only the average performance level along the

experiment, but also confidence intervals, thus allowing an

instantaneous estimate of the probability that a correct

response will be produced. For each point along the trial

sequence, the lower boundary of the confidence interval was

then compared with the initial probability of delivering a

correct response, which was at chance level (50%), and we

thus computed the probability of responding correctly as a

function of trial number (Smith et al., 2004). In order to obtain

a measure of the global amount of learning for each individ-

ual, while discounting local fluctuations in performance that

are common in perceptual learning (Mednick, Arman, &

Boynton, 2005; Mednick et al., 2002; Seitz et al., 2006), we

gave each participant a numeric score e the learning score,

which was calculated as the integral over the entire experi-

ment of the differences between the probability of responding

correctly in each consecutive trial and chance level.

3. Results

3.1. A novel paradigm to assess slow visual perceptuallearning

The paradigm we devised required participants to internally

extrapolate the complete trajectory of a decelerating moving

target after its trajectory became suddenly invisible (Fig. 1A;

see Section 2.3) and to verbally report where the target had

virtually stopped within the structured visual display. Partic-

ipants were expected to slowly refine an internal model of the

stimulus motion, aided by comparing their report with the

feedback provided at the end of each trial. Improvement in

performance was measured as a function of practice.

Seven young healthy participants (see Section 2.1) showed

robust improvement in performance along the entire experi-

ment, as appreciable from their average dynamic learning

curve (Fig. 1B), thus indicating that the paradigm was suitable

to characterize slow perceptual learning. Success rate rose

from around chance level in the initial phase of the experi-

ment (51%± 3 SEM) to 76% (±4 SEM) towards the ende a highly

significant improvement (p ¼ .000436, one-tailed, paired t-

test). We then computed average success rate within 9 sub-

sequent epochs (the first comprised the initial 25 trials, cor-

responding to baseline performance, whereas epoch 2e9 each

included subsequent 277 trials). A one-way analysis of vari-

ance (ANOVA) revealed a highly significant performance

improvement across epochs [F(8,54) ¼ 5.02, p ¼ .0001]. We also

assessed the amount of practice that was needed for our

participants to achieve a level of performance reliably and

stably superior to that at the start of training. We plotted the

moving-average level of performance along the experiment

(Fig. 1C), and compared performance in each subsequent 100-

trial epoch with baseline performance computed across the

first 25 trials (Fig. 1D). To remind the reader, we defined the

learning trial as the initial trial of the epoch in which perfor-

mance became e and then remained, significantly better than

baseline (see Section 2.4). As a result, our group of young

healthy participants showed reliable and stable improvement

in performance at trial #341. Clearly, our subjects were able to

refine their judgment of the initial pattern of motion of the

moving target as a result of extended practice and became

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 59

increasingly more proficient at extrapolating its entire tra-

jectory in order to determine the trajectory end point. Might

this type of learning be impaired following damage to the

cerebellum? Moreover, might any observable deficit be spe-

cifically linked to damage to a specific portion of the cere-

bellum? We tested these questions in the subsequent

experiment.

3.2. Slow visual perceptual learning following cerebellardamage

After having established that our paradigm was suitable to

reveal slow perceptual learning in normal individuals, we

assessed whether there was a reliable difference in the rate of

learning between a selected group of 6 patientswith cerebellar

pathology and a group of gender- and age-matched controls

(Table 1; see Section 2.1).

Importantly, the initial performancewas virtually identical

between the two groups (patients: 47% ± 3 SEM; controls:

48% ± 2 SEM; p ¼ .465, two-tailed, paired t-test). Even more

importantly, this was achieved (through the staircase proce-

dure) by exposing subjects in the two groups to highly similar

periods of visibility of the moving target (between 36% and

46% of the trajectory across the patients, and between 32%

and 42% across the controls; p ¼ .3632, two-tailed, paired t-

test), indicating that there was no distinct deficit in speed

discrimination in the patients relative to the controls. This

was further ascertained through a series of control experi-

ments (see Supplementary material).

Next, we compared overall improvement in performance

between patients and controls. As can be appreciated by

inspecting the learning curves of the two groups (Fig. 2A),

there was lesser improvement in the patients relative to the

controls. Average performance computed over the final 150

trials was 58% (±3 SEM) in the patient group and 63% (±3 SEM)

in the control group, showing reliably decreased improvement

in the patients (p ¼ .0213, two-tailed, paired t-test). The dif-

ference was partly obscured by what appears to be a seren-

dipitous drop in performance of the controls in the late phase

of the experiment. Therefore, separately for the two groups,

we computed average success rate within 9 subsequent

epochs (as described for normal subjects). The resulting data

are plotted in Fig. 2B. A two-way ANOVA with Group (patients

vs controls) and Epoch (1e9) as the main factors revealed a

significant effect of Epoch [F(8,90) ¼ 3.7, p ¼ .0009] and Group

[F(1,90) ¼ 18.07, p ¼ .0001], while the interaction was not sig-

nificant [F(8,90) ¼ .34, p ¼ .9472]. Results were fully replicated

in a two-way ANOVA including the factor Trial (performance

at each trial was estimated from the learning curve, by

applying a boxcar filter of 25 trials) and Group [Trial:

F(2219,22,200) ¼ 1.06, p ¼ .0411; Group: F(1,22,200) ¼ 2073.42,

p ¼ 0; Trial � Group: F(2219, 22,200) ¼ .49, p ¼ 1]. Lack of a

significant interaction is accounted for by the rapidly

diverging pattern of performance between the two groups (see

Supplementary material for a detailed discussion of this point

and further analyses in line with this interpretation). More-

over, since initial performance at the task did not differ be-

tween patients and controls, the main effect of the factor

Group can unequivocally be interpreted as an index of a

significant difference in the amount of learning between the

two groups.

As an additional measure of learning, we compared the

two groups in terms of the amount of practice required to

achieve a level of performance reliably and stably above

baseline. To this aim, separately for the two groups, we con-

structed learning curves as the moving-average level of per-

formance along the whole experiment (Fig. 2C) and calculated

the learning trial for each group of participants (see Section

2.4). Whereas control subjects attained reliably above-

baseline performance at trial #568 e not too dissimilar from

the estimate in young adults, cerebellar patients attained a

comparable level of proficiency only at trial #1735 (Fig. 2D),

thus indicating much slower learning in the patients

compared to the controls. To further substantiate these re-

sults while correcting for multiple comparisons, we applied a

permutation test separately on data from each group (see

Supplementary material) in order to estimate the probability

of obtaining the reportedmeasures of learning trial by chance.

For the control group, the probability to obtain a learning trial

equal or lower than #568 in the permuted samples corre-

sponded to p ¼ 0; for the patient group, the probability to

obtain a learning trial equal or lower than #1735 corresponded

to p ¼ .0066.

Overall, the above results indicate that patients suffering

from cerebellar disease show diminished learning in a purely

perceptual task. The deficit does not appear to be dramatic,

however, as some degree of improvement was observed in the

patients as well. There can be several explanations for why

the deficit is not more severe. For example, one could

conjecture that learning may depend on the cerebellum on

normal circumstances, but to a lesser extent it can be sup-

ported by other brain structures when damage to the cere-

bellum occurs. A second possibility, however, is that a partial

deficit in perceptual learning e as manifested by our patients,

may be the result of non-homogenous patterns of learning

across patients (see Fig. S2 for individual learning curves),

with some of them showing little-to-moderate learning deficit

and others showing a more dramatic impairment. To shed

light on this aspect, we looked closely at the performance of

each individual patient and correlated any observed deficit in

learning to the type and site of cerebellar damage in the same

patient.

3.3. Selective contribution of the posterior cerebellum toperceptual learning

Of the 6 patients tested, 2 had circumscribed lesions in the

posterior lobe, 2 in the anterior lobe, and the remaining 2

suffered from diffuse degenerative disease (SCA1). Given the

established notion that the anterior cerebellum is critically

involved in motor functions, including forms of motor

learning (Salmi et al., 2010; Stoodley & Schmahmann, 2010;

Stoodley et al., 2012), and also based on the recent claim

that instead the posterior cerebellum may contribute espe-

cially to cognitive domains (Salmi et al., 2010; Stoodley &

Schmahmann, 2010; Stoodley et al., 2012), we looked sepa-

rately at the pattern of performance of the patients (M.F. and

D.A.; Fig. 3A) with anterior versus those (B.D. and G.C.; Fig. 3B)

with posterior damage (Table 1; see Section 2.1).

Fig. 2 e Slow perceptual learning following cerebellar damage. (A) Average learning curves for controls (in blue) and patients

(in red). Conventions as in Fig. 1B. (B) Reduced learning in patients versus controls. Ordinate is the proportion of correct

responses and abscissa is the succession of epochs in the experiment (the first epoch comprised the initial 25 trials,

whereas epochs 2e9 each included subsequent 277 trials). (C) Slower rate of learning in patients versus controls. Learning

curves (mean) were constructed by applying a running boxcar filter (see Section 2). Conventions as in Fig. 1C. (D) Learning

trial. Results from a series of one-tailed, one-sample t-tests comparing baseline performance with each subsequent point in

the learning curve depicted in panel (C) (see Section 2). Conventions as in Fig. 1D.

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 160

As a first step, we constructed average learning curves for

the two pairs of patientswith focal cerebellar damage (Fig. 3C).

Inspection of the curves reveals a striking divergence in the

pattern of performance between the two pairs. Whereas the

two patients with anterior damage showed clear signs of

learning, qualitatively similar to the controls, this was not the

case for the two patients with a posterior lesion (Fig. 3C). Even

after having executed thousands of trials, the performance of

these patients was indistinguishable from initial perfor-

mance, indicating a complete failure of any learning

mechanism.

We then compared the rate of learning between patients

with anterior versus posterior damage, by estimating the

learning trial e as previously defined (Fig. 3D). Whereas the

learning trial for the two patients with focal anterior cerebellar

damage could be detected at trial #607 e a value highly

comparable to that of the controls (trial #568; see Fig. 2D), it

was not possible to detect a learning trial for the two patients

with focal posterior damage, revealing that their performance

never became reliably and stably better than baseline.

In further support to the above observation, as an index of

learning efficiency, we measured the global amount of

improvement as a function of practice for each of the four

patients with focal damage. For this purpose, we constructed

individual dynamic learning curves (Fig. 4), according to the

method developed by Smith et al. (2004), and calculated a

learning score for each participant, corresponding to the cu-

mulative difference between the probability of responding

correctly at each trial and chance level (see Section 2.4). In

agreementwith qualitative assessment of the learning curves,

we found that the two patients with a focal lesion in the

anterior cerebellum were entirely unimpaired in learning the

Fig. 3 e Reduced learning following focal posterior versus focal anterior damage. (A) MRI images of patients with focal

anterior ischemic cerebellar lesions (M.F. and D.A.) and illustration of the site and size of the lesion. (B) MRI images of

patients with focal posterior ischemic cerebellar damage (B.D. and G.C.) and illustration of the site and size of the lesion. (C)

Rate of learning. Learning curves were separately constructed for patients with focal anterior (left panel, in magenta) and

focal posterior (right panel, in aqua) ischemic lesions. Conventions as in Fig. 1C. (D) Learning trial. Results from a series of

one-tailed, one-sample t-tests comparing baseline performance with subsequent points along the learning curve (see

Section 2). Conventions as in Fig. 1D.

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 61

task, relative to the control group (patient M.F.: learning

score ¼ 48.06; patient D.A.: learning score ¼ 117.97; Controls:

average learning score ¼ 94.85 ± 63.91 SEM). In sharp contrast,

the two patients with focal damage restricted to the posterior

cerebellum showed no sign of learning (patient B.D.: learning

score ¼ �136.57; patient G.U.: learning score ¼ �261.75). In

keeping with the above data, the average learning score of the

former pair (83.01) was not reliably different from that of the

control group (p ¼ .9149, two-tailed, Z-test); in contrast, the

average learning score of the latter pair (�199.16) differed

Fig. 4 e Slow perceptual learning in individual patients with focal cerebellar damage. (A) Dynamic learning curves (Smith

et al., 2004) for single patients with focal anterior lesions. In each panel, for an individual patient, the estimated probability

of a correct response as a function of time is reported in magenta, with the shading representing the 90% confidence

interval, as delimited by the upper and lower 95% confidence bounds. Individual learning scores are reported on each panel

(see Section 2). Detailed description of the size and site of the lesions (including the involvement of specific lobules and of

the deep cerebellar nuclei), together with the results of the clinical assessment, are reported in the table at the center of the

figure. (B) Dynamic learning curves (Smith et al., 2004) for single patients with focal posterior lesions. Here, in each panel,

the estimated probability of a correct response as a function of time is reported in aqua, with the shading again representing

the 90% confidence interval. Conventions as in panel (A).

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 162

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 63

significantly from that of the control group (p ¼ .0079). In sum,

we can cleanly dissociate the contribution of the posterior

versus anterior cerebellum to a form of perceptual learning.

Our data demonstrate that slow, incremental learning in our

motion extrapolation task relies strongly and specifically on

neural circuits located in the posterior cerebellum.

One natural objection to the above conclusion could be that

perhaps the 2 patients with a lesion in the posterior cere-

bellumweremore severely impaired overall, as assessed upon

clinical examination, compared to the patients with an ante-

rior lesion. For example, this might be due to somewhat larger

lesions in the patients with posterior (28.81 cc ± .83 SEM)

versus anterior (13.98 ± 10.01 SEM) damage (see Section 2.2).

However, the above possibility is easily ruled out, since it was

instead the patients with a focal anterior lesion to be sub-

stantially more severely impaired clinically, in spite of the

smaller lesion volume in these patients. Their average SARA

score (Schmitz-Hubsch et al., 2006) was 6.25, and was much

higher than that of the patients with a focal posterior lesion,

2.5 (see Table 1). Importantly, even the patient (M.F.) with the

larger lesion in the anterior cerebellum (23.99 cc) showed fully

preserved learning at our task (Fig. S2; Fig. 4A, left panel). The

above data are fully consistent with a double-dissociation

pattern, where patients with posterior cerebellar damage are

especially impaired in learning our novel motion extrapola-

tion task, whereas patients with anterior cerebellar damage

show more severe clinical signs of motor disturbance and

likely more severe deficits in motor learning, as suggested by

previous work in the literature (Salmi et al., 2010; Stoodley &

Schmahmann, 2010; Stoodley et al., 2012; Werner, Bock,

Gizewski, Schoch, & Timmann, 2010).

One last concern about the above results relates to the

possibility that tissue damage may have extended to the deep

cerebellar nuclei in some of the focal patients, but not in

others, and we addressed this possibility directly (see

Supplementary material). An involvement of the deep cere-

bellar nuclei was ascertained only in one of the patients with

an anterior lesion (M.F.). Crucially, deep nuclei were instead

intact in both patients with posterior damage, thus ruling out

the possibility that the observed deficit in slow perceptual

learning in these patients could at least partly reflect damage

of the cerebellar nuclei (see Supplementary material). There-

fore, the observed deficit in slow visual perceptual learning

can safely be ascribed to the isolated lesion of the posterior

cerebellar cortex and underlying white matter.

The results described above were confirmed in two addi-

tional patients with focal cerebellar damage (Table 1). The first

one, G.Z., had an ischemic lesion restricted to the anterior lobe

(Fig. S3A), and was again entirely unimpaired in learning our

motion extrapolation task (learning score ¼ 97.51; Fig. 4A,

rightmost panel). The second, Z.G., had more widespread

ischemic damage, with the most prominent lesion located in

the posterior cerebellar lobe (Fig. S3B) and a small lesion in the

anterior lobe; Z.G. showed no reliable sign of learning at the

task (learning score: �39.29; Fig. 4B, rightmost panel) as ex-

pected from the damage of the posterior lobe. To summarize

the results, Fig. 4 reports the individual dynamic learning

curves computed for all the focal patients enrolled in the

study, together with their learning scores and a detailed

description of the size and site of the lesion, including the

indication of the damaged lobules and of any involvement of

the deep cerebellar nuclei (see Section 2.2). The figure clearly

illustrates the profound difference between patients with

focal anterior versus posterior cerebellar damage in the ability

to learn our motion extrapolation task: while patients with a

focal lesion to the posterior lobe were clearly impaired at the

task and did not show any reliable sign of learning, patients

with a focal lesion to the anterior lobe were completely un-

impaired, such that their learning abilities were undis-

tinguishable from those of healthy controls.

We also made an attempt to correlate the ability to learn

our perceptual task with the severity and progression of the

disease in degenerative patients. Here we found that a clear

deficit in perceptual learning was only evident in the case of

patients at an advanced stage of the disease, as assessed upon

clinical examination, withmedium-to-severemotor signs and

symptoms. Instead, no clear perceptual learning deficit could

be appreciated in mild or pre-symptomatic patients (Fig. S4;

see Supplementary material).

Many issues and potential confounds need to be properly

addressed before one can firmly establish that the cerebellum

is critically involved in perceptual learning. We made a

considerable effort to address any such problem, and the

relevant evidence is provided in the Supplementary Results. In

particular, by means of a series of control experiments and of

additional analyses (see Supplementary material), we ruled

out the possibility that the observed learning deficit could be

explained in terms of a general cognitive impairment (Fig. S5;

Table S1), or in terms of a speed discrimination deficit in the

cerebellar patients, or finally in relation to the involvement of

ocular tracking in the execution of the motion extrapolation

task (Fig. S6; Table S2).

3.4. Fast perceptual adaptation following cerebellardamage

Having demonstrated that the posterior cerebellum is criti-

cally involved in a form of slow visual perceptual learning,

accruing during the course of many consecutive sessions, we

also wished to test whether the cerebellum might likewise be

engaged when learning occurs on a much shorter time scale,

such as is required during a condition of perceptual adapta-

tion taking place within few tens of trials. For this purpose, in

a subsequent experiment, we devised a variant of the para-

digm that was used previously to measure slow perceptual

learning (see Section 2.3). In this case, on the first day of the

experiment we adjusted initial performance of each partici-

pant through the staircase method to be around 85% correct

(see Supplementary material). Participants then practiced the

task for 1e2 daily sessions in order to consolidate such high

performance level, and this sometimes led to further subtle

improvements. A perturbation, consisting in a form of false

feedback (Choi & Watanabe, 2012) and reminiscent of that

used in typical saccadic adaptation paradigms (Deubel et al.,

1986; Hopp & Fuchs, 2004; McLaughlin, 1967), was introduced

at the start of the subsequent session, on the following day,

and was maintained for the remainder of the experiment.

Specifically, after completion of each trial, the provided

feedback was not veridical in that the target was nowmade to

decelerate at a rate (5�/sec2) that was lower than during the

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 164

actual task (5.7�/sec2), so that its trajectory was extended

further. Subjects were not alerted of any modification to the

task when the perturbation was introduced.

The change in deceleration rate during feedback was such

that, given the same range of starting positions and initial

velocities, the target stopped in the same sector as before in

~50% of the trials and in the sector farther away in the

remaining trials (Fig. 5A). The perturbation caused a sudden

and robust drop in performance. Rate and amount of adap-

tationwere reflected in the time course and degree of recovery

from the initial drop. Following an initial fall in performance,

we expected subjects to update the internal model rather

rapidly, thus recovering at least partly the pre-perturbation

level of accuracy. It is important to note that evaluation of

performance in each trial was done according to the feedback.

Therefore, even when the observer correctly inferred the final

position of the moving target on the basis of its initial speed

and position during the trial, the response could be scored as

Fig. 5 e Fast visual perceptual learning (adaptation) following ce

of the effect of the perturbation introduced in the crucial session

the end point of the trajectory according to the real target decele

feedback, provided as if the target were to move with decreased

perturbation in controls (blue) and patients (red). Ordinate is pr

from the last basal session (1 ¼ trials #1e150; 2 ¼ #151e300) a

included in the analyses. (C) Adaptation in controls. Ordinate is

comprising subsequent 30 trials, except for the first two points,

patients. Conventions as in panel (C).

incorrect in accordance with what was then shown during

feedback in the same trial. For this reason, it should be clear

that a drop in performance following the perturbationmust be

expected because it does not depend on any worsening of the

actual performance, but rather it depends on a form of

misattribution of responses to correct versus incorrect

categories.

As a first step, we validated this new version of the task by

testing young healthy adults and confirmed that the task was

perfectly adequate to reveal a distinct form of fast perceptual

adaptation (Fig. S7; see Supplementary material). The para-

digm was then applied to a selected group of patients with

cerebellar pathology. In this case we were able to recruit a

total of 4 patients and as many age- and gender-matched

controls. Among the patients, 3 were affected by hereditary

cerebellar disease (SCA1 or SCA2), while the fourth patient

suffered from a focal lesion in the posterior cerebellum

(Table 1; Fig. S8A; see Section 2.1).

rebellar damage. (A) Adaptation task. Illustrative examples

of the experiment (see Section 2). The green dots represent

ration (5.7�/sec2), whereas the red dots represent the ‘false’

deceleration (5�/sec2). (B) Drop in performance following the

oportion of correct responses and abscissa is epoch. Data

nd the first 30 trials of the third session (perturbed) were

proportion of correct responses and abscissa is epoch (each

comprising 10 and 20 trials, respectively). (D) Adaptation in

Fig. 6 e Rate of adaptation following cerebellar damage. (A)

Rate of adaptation. Average performance for controls (blue)

and patients (red) along the perturbation session. Learning

curves (mean) were constructed by smoothing data from

sequential trials performed by each subject with a boxcar

filter of a fixed size (30 trials), shifted by one single trial at

the time, and then by averaging data across subjects (see

Section 2). Conventions as in Fig. 1C. (B) Adaptation trial.

Results from a series of one-tailed, one-sample t-tests

comparing baseline performance (first 10 trials) with each

subsequent point along the perturbation session for the

two groups (see Section 2). Conventions as in Fig. 1D. (C)

Between group comparison along the perturbation session.

Average adaptation curves for the patients and the controls

are shown in the left panel. Conventions as in Fig. 1B. The

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 65

Patients as well as their controls showed a significant drop

in performance (Fig. 5B). For the patient group, average suc-

cess rate was 85.7% and 84.7%, respectively, in the two base-

line epochs, before the perturbationwas applied, while it went

down to 50% immediately after introducing the perturbation

(first 30 trials; individual data for the patients are reported in

Fig. S8B). Likewise, for the control group, average success rate

was 81.7% and 90.5%, respectively, in the two baseline epochs,

while it went down to 56.7% following the perturbation. The

drop in performance was confirmed by a two-way ANOVA

with Group (patients vs controls) and Epoch as the main fac-

tors, which revealed a significant main effect of Epoch

[F(2,18) ¼ 32.23, p ¼ 0], whereas neither the effect of Group

[F(1,18) ¼ .55, p ¼ .467] nor the interaction [F(2,18) ¼ .81,

p ¼ .4617] were significant. Planned post-hoc comparisons

confirmed that the performance of the two groups did not

differ at any of the three tested epochs (two-tailed, paired t-

test; Bonferroni corrected). Incidentally, the slight difference

in performance between the two groups during the second

pre-perturbation epoch most likely reflects a deficit in slow

learning in the patients relative to the controls.

Having proven that the perturbation caused a large and

reliable drop in performance in both groups, we then

concentrated our analysis on the question whether there was

a divergence between the two groups in their ability to adapt

to the perturbation. Based on the evidence gathered from

normal individuals that recovery tends to occur rapidly

(Fig. S7), we first focused our attention on the first 150 trials

following onset of the perturbation (individual data for the

patients are reported in Fig. S8B).We subdivided these trials in

six successive epochs of 10 (epoch 1), 20 (epoch 2) and then 30

trials (epochs 3e6). Separately for each group, we performed a

one-way ANOVA with Epoch (1e6) as the main factor: while

the controls showed a significant increase in performance

over time [F(5,18)¼ 6.13, p¼ .0017; Fig. 5C], no reliable recovery

could be assessed for the patients [F(5,18) ¼ .09, p ¼ .9937; Fig

5D]. We then identified the first epoch showing significantly

better performance than that in the first epoch by means of a

one-tailed, paired t-test. With this approach, we found that

whereas control participants significantly improved their

performance (p ¼ .0160, Bonferroni corrected) already at the

third epoch (i.e., trials #31e60; Fig. 5C), cerebellar patients did

not show any sign of reliable recovery along the entire stretch

of 150 trials, thus indicating a severe impairment in this form

of perceptual adaptation (Fig. 5D).

It might still be that patients adapt to the perturbation, but

with a slower time course. To check this possibility, we con-

ducted further analyses by looking at performance along the

entire perturbed session (Fig. 6A) in order to identify the

adaptation trial (see Section 2.4), or the trial where performance

became significantly superior to that measured immediately

after the perturbation, and remained such thereafter.

right panel shows results from a series of two-tailed,

unpaired t-tests comparing each point between the two

curves. The level of accuracy of the two groups was

indistinguishable at the beginning, but it soon diverged

and remained different until the end of the session

(p < .05).

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 166

Whereas the adaptation trial for the controls corresponded to

trial #11, wewere not able to reliably identify an adaptation trial

for the cerebellar patients, thus again indicating a severe

deficit in adaptation (Fig. 6B). Finally, in order to directly

compare the pattern of behavior between the two groups, we

contrasted their performance along the entire perturbed ses-

sion by means of two-tailed unpaired t-tests and we found

that their level of accuracy soon diverged after start of the

session, and then remained mostly different until the end of

the session (Fig. 6C).

On the basis of the converging evidence presented thus far,

we can conclude that cerebellar patients, unlike their controls,

showed a marked deficit in fast perceptual adaptation,

therefore extending our previous observation that cerebellar

damage, especially of the posterior lobe, impairs perceptual

learning.2

4. Discussion

The reported findings shed new light on the role of the cere-

bellum in non-motor domains, as well as on its general role in

brain function, supporting the notion that the human cere-

bellum operates as a learning device (“neuronal learning

machine”; Raymond, Lisberger, & Mauk, 1996) for both motor

and non-motor functions, including perception. Specifically,

the collected evidence clearly demonstrates that cerebellar

damage impairs slow perceptual learning in a visual motion

extrapolation task. Interestingly, the degree of impairment

depends on severity of diffuse degeneration in patients

affected by spinocerebellar ataxia (SCA1), with negligible

deficit at mild-pre-symptomatic stages of the disease. Most

importantly, data collected in focal patients point to a strong

correlation between the learning deficit and the site of the

lesion, with a substantial deficit only in patients with poste-

rior versus anterior lesions. This finding supports a clean

functional dissociation between anterior and posterior re-

gions of the cerebellum, as recently suggested (Ramnani, 2006;

Salmi et al., 2010; Stoodley, 2012; Stoodley & Schmahmann,

2010; Strick et al., 2009; Werner et al., 2010), with the former

mainly concerned with motor control and learning and the

latter mainly concerned with non-motor, cognitive and

perceptual functions, and learning within these functions. Of

course, we realize that this pattern will have to be confirmed

in a larger sample of patients, and we put considerable effort

to do so. However, this was not a trivial task owing to the very

strict inclusion criteria that we adopted, and the fact that our

paradigmswere very demanding for the patients, with several

sessions of testing on separate days and hundreds-to-

thousands of trials in total. We underscore that the exten-

sive testing (including with control tasks) is a key feature of

2 The data from both main experiments were presented at thefollowing international meetings: FENS Forum of EuropeanNeuroscience, Amsterdam, NL (2010); IBRO Word Congress ofNeuroscience, Florence, I (2011). We had already been pursuingpublication of these data for a long time when a paper (Roth et al.,2013) was published, which used a task paradigm highly similarto the one employed here, and reporting results fully compatiblewith those from our second experiment (but see Section 4 formore detailed analysis of that study).

the present work and is instead highly unusual in the relevant

literature, where in essentially all previous studies investi-

gating learning deficits in the motor and non-motor domain

following cerebellar damage, testing typically occurs within a

single experimental session comprising tens-to few hundred

trials (e.g., H€andel, Thier, & Haarmeier, 2009; Rabe et al., 2009;

Roth, Synofzik, & Lindner, 2013; Tseng et al., 2007).

Based on the reported evidence, it will now become espe-

cially important to assess whether other forms of perceptual

learning are also dependent upon an intact cerebellum. On the

one hand one might predict that perceptual learning engages

the cerebellum only to the extent that the perceptual judg-

ment depends on some form of dynamic, spatio-temporal,

predictive computation, as it is the case in our motion

extrapolation task (see also Courchesne & Allen, 1997; Miall

et al., 1993; O'Reilly, Mesulam, & Nobre, 2008). In this view,

the contribution of the cerebellum might instead be marginal

or absent for kinds of perceptual learning tapping static rep-

resentations, as it occurs for instance in orientation discrim-

ination or hyperacuity tasks. However, this does not have to

be the case, and the definite answer will require further

empirical testing.

In a second experiment, we also demonstrated that cere-

bellar patients are impaired in fast perceptual adaptation in

the context of a modified version of our motion extrapolation

paradigm, showing a marked inability in recalibrating their

internal model of motion based on the “perturbed” feedback

signal. A compatible pattern of results was reported in a

recent study by Roth et al. (2013), who elegantly addressed the

potential role of the cerebellum in updating internal pre-

dictions about external sensory events, by implementing a

baseline-recalibration experiment similar to ours. In their

study, cerebellar patients (mostly affected by a degenerative

disease, and a minority affected by acute focal posterior

damage) and matched controls were required to judge the

time of reappearance of a moving visual target, the trajectory

of which was partially occluded. After the target reappeared

on the screen, participants had to indicate whether it reap-

peared “too late” or “too early”, assuming that it maintained

constant speed behind the occluder. No marked divergence

was assessed between patients and controls in their ability to

judge the time of reappearance of the moving target in a

baseline phase, where the discrepancy between the actual

and theoretical reappearance time was varied along a sym-

metrical distribution around coincidence. However, cerebellar

patients were impaired in the subsequent recalibration phase

of the experiment, where participants were expected to

“adapt” to an experimentally added delay in the time of

reappearance of the moving target (i.e., on a fraction of the

total trials the moving visual target reappeared with a delay

with respect to the estimated “correct” reappearance time e

as measured individually by the point of subjective equiva-

lence). Although generally compatible with our adaptation

experiment, the results reported by Roth et al. (2013) generate

some concern, and this is mainly for two reasons. First, no

mention is made in the study of the time course of the

“recalibration” effect, which is a key element to characterize it

as a fast learning phenomenon. In other words, did the reca-

libration effect take some time to occur, or was it instanta-

neous? Second, and most crucially, the reported effects could

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 67

truly reflect a recalibration of sensoryeperceptual predictions,

as argued by the authors, but this is not the only viable

interpretation. For example, we would argue that task per-

formance measured in the recalibration phase might have

been affected by development of a “response” bias, i.e., par-

ticipants learned to respond “too late” more often as the time

of reappearance was systematically delayed in a large fraction

of the total trials, which might be seen as a form of cognitive

associative learning. Obviously, data from our second, adap-

tation experiment do not lend themselves to these critiques.

First, because we report a distinct time course for the adap-

tation effect to take place; second, because in no way can the

adaptation effect be explained as the result of a response bias,

since the perturbation and the resulting adaptation did not

lead to a change in the probability that the moving target

ended its trajectory in black versus white sectors, whereas the

subjects' reports concerned this binary distinction.

Although to our knowledge our paper is the first report of a

deficit in pure forms of visual perceptual learning following

cerebellar damage, a role of the cerebellum in non-motoric

types of learning has been supported by earlier work. This

included demonstrations of a learning deficit in cognitive as-

sociation tasks (Drepper, Timmann, Kolb, & Diener, 1999),

impaired updating of internal predictions about the sensory

consequences of action (Synofzik, Linder, & Thier, 2008), and

impaired learning of abstract rules (Balster & Ramnani, 2011).

Overall, these findings converge to establish the notion that

the cerebellum plays a key role in learning and adaptation

outside the realm of motor control.

Ofmore direct relevance to our work, a few functional brain

imaging studies have reported distinct patterns of cerebellar

activity within the context of perceptual learning paradigms

(Schiltz etal., 1999;Vaina, Belliveau, desRoziers,&Zeffiro, 1998).

Schiltz et al. (1999) performed a thorough positron emission

tomography (PET) study to investigate the brain correlates of

perceptual learning using an orientation discrimination task

and, among other findings, reported reduced levels of activity

within the cerebellum (vermis and anterior lateral region of the

left hemisphere) after relative to before training. Although this

observation would seem compatible with the notion that the

cerebellum contributes critically to perceptual learning mech-

anisms (including for static visual properties), as we arguehere,

it should be noted that the authors did not establish such link

between the observedpattern of cerebellar activity and learning

at the orientation discrimination task. To the contrary, they

suggested that the reduced cerebellar activity following training

was likely related to non-specific aspects of task performance

and learning (“Ourfinding that someof thenon-specific training

effects occur in motor regions, notably the cerebellar vermis,

supports this view”, p. 58). In fact, the observed pattern of

decreased cerebellar activity following training was presented

as a rathermarginal finding by the authors, devoid of anymajor

functional significance for the specific learning process.

Vaina et al. (1998) used functional magnetic resonance im-

aging (fMRI) to explore brain mechanisms underlying percep-

tual learning in a global motion discrimination task. In

addition to a variety of changes at the cortical and subcortical

level, as the putative correlate of learning, they reported

massive activation of the superior posterior lateral cerebellum

during early phases of learning,which thengreatly diminished

when learning approached asymptote. Again, however, the

authors did not link this activity changewithin the cerebellum

to specific components of the perceptual learning mecha-

nisms, notably the refinement of perceptual representations

as a result of practice at the task. Instead, they suggested that

the cerebellum might have been recruited in early phases of

the learning process because of its alleged role in the deploy-

ment of attention, under the assumption that attention to the

motion display was especially important during early phases

of the experiment to then become less important with accu-

mulating practice. Although in principle this might also be a

plausible account for the results reported here ewith some of

the patients being impaired at learning the motion extrapola-

tion task because of an attentional deficit, we find this expla-

nationveryunlikely for anumberof reasons. First, our patients

did not manifest appreciable deficits of attention at the neu-

ropsychological testing (Table S1). Second, our patients,

including the ones more severely impaired in learning, dis-

played levels of performance fully comparable to that of their

matched controls when tested with the motion extrapolation

task prior to any learning (staircase procedure, Fig. S5), as well

as when tested with a challenging speed discrimination task

(see Supplementary material), and we would argue that effi-

cient deployment of attention was needed to perform these

tasks.More generally, the view that cerebellar damage leads to

appreciable deficits of attention is still highly controversial

(e.g., Dimitrov et al., 1996; Ignashchenkova et al., 2009;

Stoodley, 2012; Timmann & Daum, 2010). Finally, one should

also note that our current understanding of the network of

brain areas, comprising both cortical and subcortical struc-

tures, assumed to be responsible for directing attention to the

visual space, does not place much emphasis on the role of the

cerebellum (e.g., Beck&Kastner, 2009; Bisley&Goldberg, 2010;

Corbetta & Shulman, 2002, 2011; Mesulam, 1999; Noudoost,

Chang, Steinmetz, & Moore, 2010; Serences & Yantis, 2006).

For all these reasons combined we find an attentional account

of our findings, and those reported by Vaina et al. (1998), very

unlikely. Instead we suggest that, based on the reciprocal in-

direct connections between portions of the cerebellar cortex

and large extents of the cerebral cortex, including motion

sensitiveareas along thedorsal visual pathway (seebelow), the

cerebellum is critically involved in building and refining in-

ternal predictive models of spatio-temporal patterns of visual

motion information (e.g., see Cerminara, Apps, & Marple-

Horvat, 2009), allowing highly proficient performance at our

task as a result of extended practice.

Incidentally, one should also note that while the orienta-

tion discrimination task employed by Schiltz et al. (1999)

tapped a form of slow-incremental learning, occurring over

the course of many sessions and thousands of trials, this was

not the case in the study of Vaina et al. (1998), where learning

at the global motion discrimination task was characterized

within minutes of performance and tens-to-few hundreds of

trials (fast learning). It is widelymaintained that fast and slow

perceptual learning phenomena represent distinct forms of

learning, presumably tapping different sub-components of

the learning process and perhaps reflecting changes of a

different nature and occurring within different brain circuits

(Censor et al., 2012; Karni & Bertini, 1997). In this regard we

underscore that the results presented here indicate that the

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 168

cerebellum is engaged for visual perceptual learning occurring

on both time scales.

As a final comment, it is also important to underscore that

evidence obtained with functional brain imaging methods is

inherently correlational in nature, and it is not easy to draw

firm conclusions based on this evidence as to whether the

cerebellum plays a necessary role in perceptual learning. This

is not the case in the present study, where we could demon-

strate that patients suffering from cerebellar damage, espe-

cially of the posterior cerebellum, were impaired at refining

their perceptual judgments as a result of practice, in turn

indicating that the cerebellum is indeed crucial for perceptual

learning to take place.

Synofzik et al. (2008), by comparing performance of patients

with cerebellar damage and controls, recently provided a bril-

liantdemonstration that thecerebellumisnecessary for rapidly

updating the internal predictive models about the sensory

consequencesofaction (one's ownarmposition in spaceduring

the execution of a pointingmovement). This is clearly different

from what we claim here, namely that the cerebellum is

involved in forms of slow-cumulative and fast perceptual

learning phenomena that are entirely decoupled from action

planning and execution. In other words, our work demon-

strates for thefirst time that the (posterior) cerebellumprovides

a key contribution to a form of pure perceptual learning.

Our findings can be readily explained by known in-

terconnections between parts of the posterior cerebellar cortex

and large extents of fronto-prefrontal and parietal cortices (via

the dentate nucleus, the thalamus and, reciprocally, the pons)

(Bostan et al., 2013; Glickstein&Doron, 2008; Strick et al., 2009),

includingdorso-lateral prefrontal regionsand territorieswithin

and around the intraparietal sulcuse areaswhose contribution

to perceptual decision-making concerning visual motion has

been decisively established (Gold & Shadlen, 2007; see also

Dosher& Lu, 2009; Law&Gold, 2008). However, one shouldnote

that also crucial nodes along thedorsal visual stream, including

the middle temporal complex (MTþ), are functionally con-

nected with the cerebellum (H€andel et al., 2009; O'Reilly et al.,

2008), and might therefore support learning at our motion

extrapolation task. Elegant work using magnetoencephalogra-

phy (MEG) has recently demonstrated that damage to the cer-

ebellum in humans leads to altered processing of motion

signals in the MT region (H€andel et al., 2009).

The task paradigm that we devised for the present study is

reminiscent of that used some years ago by O'Reilly et al.

(2008). In that study, two versions of a motion extrapolation

task were employed and compared within the context of an

fMRI design. One task required participants to detect de-

viations from the expected direction of a moving target whose

trajectory was transiently occluded and therefore had to be

mentally extrapolated; the other task required them to detect

deviations from the expected velocity of the same target.

Crucially, for our purposes, the latter task, but not the former,

engaged portions of the posterior cerebellum (lobule VII crus I)

bilaterally. According to the authors, involvement of this re-

gion reflected its role in providing the basis for a spatio-

temporal forward model of the motion trajectory. We argue

that it is precisely this model that participants were asked to

refine during our learning tasks. Moreover, in the same study,

analysis of functional connectivity revealed that the same

portions of the cerebellar cortex entertained synergistic in-

teractions with a network of cerebral cortical regions,

including prefrontal and parietal regions (see also Dosher &

Lu, 2009; Law & Gold, 2008), as well as the MTþ complex.

Based on this prior evidence (see also H€andel et al., 2009), we

claim that it is likely the same region of cerebellar cortex,

along with the detected network of interacting cerebral

cortical areas, which form the core circuit underlying the type

of perceptual learning that we have studied here in patients

with cerebellar pathology.

We assume that performance improvement at our slow

perceptual learning task likely rests on a refinement both in

the ability to extract the critical initial parameters of the

decelerating pattern of motion, as well as in the ability to

extend the “obstructed” motion pattern (motion extrapola-

tion), which in turn allows a correct spatio-temporal predic-

tion of the end point of the target trajectory. Importantly,

however, we underscore that a number of studies provide a

conceptualization of motion extrapolation as a function

afforded by the same sensory/perceptual mechanisms that

allow motion perception (e.g., Sokolov et al., 1997;

Watamaniuk, 2005; Watamaniuk & McKee, 1995).

Some years ago, based largely on a comparative evolu-

tionary approach, Paulin (1993) proposed that “the cerebellum

may be better characterized as a tracking system, with an

important role in control and coordination of movements

because of an animal's need to trackmoving objects, to track its

own movements, and to analyze the sensory consequences of

movements in order to control movements” (p. 39). Along the

same lines, he also maintained that cerebellar dysfunction

should lead to “an inability to accurately follow and predict

trajectories of objects moving in the environment” (p. 39). Our

present report is fully consistentwithPaulin'sviewof cerebellar

function (also see Courchesne & Allen, 1997; Miall et al., 1993).

Acknowledgments

We acknowledge generous support from Compagnia di San

Paolo, Turin, Italy, and from the International Brain Research

Organization (Studentship 2009 to A.G.). We wish to thank Pier-

giorgio Strata for his support. We thank Davide Massidda for

precious advice on statistical approaches, Giampaolo Tomelleri

for help in the recruitment of cerebellar patients, and Marco

Veronese for help in preparing some of the figures.We alsowish

to thank thepatients enrolled in the study for their collaboration.

Supplementary material

Supplementary data related to this article can be found at

http://dx.doi.org/10.1016/j.cortex.2014.04.017.

r e f e r e n c e s

Ahissar, M., Nahum, M., Nelken, I., & Hochstein, S. (2009). Reversehierarchies and sensory learning. Philosophical Transactions of

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 69

the Royal Society of London Series B Biological Sciences, 364,285e299.

Albus, J. S. (1971). A theory of cerebellar function. MathematicalBiosciences, 10, 25e61.

Anstis, S. (2001). Footsteps and inchworms: illusions show thatcontrast affects apparent speed. Perception, 30, 785e794.

Balster, J. H., & Ramnani, N. (2011). Cerebellar plasticity and theautomation of first-order rules. Journal of Neuroscience, 31,2305e2312.

Bastian, A. J. (2008). Understanding sensorimotor adaptation andlearning for rehabilitation. Current Opinion in Neurology, 21,628e633.

Bastian, A. J. (2011). Moving, sensing and learning with cerebellardamage. Current Opinion in Neurobiology, 21, 596e601.

Beck, D. M., & Kastner, S. (2009). Top-down and bottom-upmechanisms in biasing competition in the human brain.Vision Research, 49, 1154e1165.

Bellebaum, C., & Daum, I. (2007). Cerebellar involvement inexecutive control. Cerebellum, 6, 184e192.

Bellebaum, C., & Daum, I. (2011). Mechanisms of cerebellarinvolvement in associative learning. Cortex, 47, 128e136.

Ben-Yehudah, G., Guediche, S., & Fiez, J. A. (2007). Cerebellarcontributions to verbal working memory: beyond cognitivetheory. Cerebellum, 6, 193e201.

Bhanpuri, N. H., Okamura, A. M., & Bastian, A. J. (2012). Activeforce perception depends on cerebellar function. Journal ofNeurophysiology, 107, 1612e1620.

Bisley, J. W., & Goldberg, M. E. (2010). Attention, intention, andpriority in the parietal lobe. Annual Review of Neuroscience, 33,1e21.

Blazquez, P. M., Hirata, Y., & Highstein, S. M. (2004). Thevestibulo-ocular reflex as a model system for motor learning:what is the role of the cerebellum? Cerebellum, 3, 188e192.

Bostan, A. C., Dum, R. P., & Strick, P. L. (2013). Cerebellar networkswith the cerebral cortex and basal ganglia. Trends in CognitiveSciences, 17, 241e254.

Brainard, D. H. (1997). The psychophysics toolbox. Spatial Vision,10, 433e436.

Bueti, D., Lasaponara, S., Cercignani, M., & Macaluso, E. (2012).Learning about time: plastic changes and interindividual braindifferences. Neuron, 75, 725e737.

Byers, A., & Serences, J. T. (2012). Exploring the relationshipbetween perceptual learning and top-down attentionalcontrol. Vision Research, 74, 30e39.

Carey, M. R. (2011). Synaptic mechanisms of sensorimotorlearning in the cerebellum. Current Opinion in Neurobiology, 21,609e615.

Censor, N., Sagi, D., & Cohen, L. G. (2012). Common mechanismsof human perceptual and motor learning. Nature Reviews.Neuroscience, 13, 658e664.

Cerminara, N. L., Apps, R., & Marple-Horvat, D. E. (2009). Aninternal model of a moving visual target in the lateralcerebellum. Journal of Physiology, 587(2), 429e442.

Choi, H., & Watanabe, T. (2012). Perceptual learning solelyinduced by feedback. Vision Research, 61, 77e82.

Corbetta, M., & Shulman, G. L. (2002). Control of goal-directed andstimulus-driven attention in the brain. Nature Reviews.Neurosciences, 3, 201e215.

Corbetta, M., & Shulman, G. L. (2011). Spatial neglect andattention networks. Annual Review of Neuroscience, 34, 569e599.

Courchesne, E., & Allen, G. (1997). Prediction and preparation,fundamental functions of the cerebellum. Learning & Memory.,4, 1e35.

De Bruyn, B., & Orban, G. A. (1988). Human velocity and directiondiscrimination measured with random dot patterns. VisionResearch, 28, 1323e1335.

Deubel, H., Wolf, W., & Hauske, G. (1986). Adaptive gain control ofsaccadic eye movements. Human Neurobiology, 5, 245e253.

Dimitrov, M., Grafman, J., Kosseff, P., Wachs, J., Alway, D.,Higgins, J., et al. (1996). Preserved cognitive processes incerebellar degeneration. Behavioural Brain Research, 79,131e135.

Dimitrova, A., Weber, J., Redies, C., Kindsvater, K., Maschke, M.,Kolb, F. P., et al. (2002). MRI atlas of the human cerebellarnuclei. NeuroImage, 17, 240e255.

Dosher, B. A., & Lu, Z. L. (2009). Hebbian reweighting on stablerepresentations in perceptual learning. Learning & Perception, 1,37e58.

Dow, R. S., & Moruzzi, G. (1958). The physiology and pathology of thecerebellum. Minneapolis: University of Minnesota Press.

Doyon, J. (1997). Skill learning. International Review of Neurobiology,41, 273e294.

Drepper, J., Timmann, D., Kolb, F. P., & Diener, H. C. (1999). Non-motor associative learning in patients with isolateddegenerative cerebellar disease. Brain, 122, 87e97.

Durisko, C., & Fiez, J. A. (2010). Functional activation of thecerebellum during working memory and simple speech task.Cortex, 46, 896e906.

E, K. H., Chen, S. H. A., Ho, M. H. R., & Desmond, J. E. (2014). Ameta-analysis of cerebellar contributions to higher cognitionfrom PET and fMRI studies. Human Brain Mapping, 35, 593e615.

Fahle, M. (2009). Perceptual learning and sensomotor flexibility:cortical plasticity under attentional control? PhilosophicalTransactions of the Royal Society of London Series B BiologicalSciences, 364, 313e319.

Frens, M. A., & van Opstal, A. J. (1994). Transfer of short-termadaptation in human saccadic eye movements. ExperimentalBrain Research, 100, 293e306.

Gao, Z., van Beugen, B. J., & De Zeeuw, C. I. (2012). Distributedsynergistic plasticity and cerebellar learning. Nature Reviews.Neuroscience, 13, 619e635.

Gilbert, C. D., Li, W., & Piech, V. (2009). Perceptual learning andadult cortical plasticity. Journal of Physiology, 587, 2743e2751.

Glickstein, M. (2007). What does the cerebellum really do? CurrentBiology, 17, R824eR827.

Glickstein, M., & Doron, K. (2008). Cerebellum: connections andfunctions. Cerebellum, 7, 589e594.

Gold, J. I., & Shadlen, M. N. (2007). The neural basis of decisionmaking. Annual Review of Neuroscience, 30, 535e574.

Golla, H., Tziridis, K., Haarmeier, T., Catz, N., Barash, S., &Thier, P. (2008). Reduced saccadic resilience and impairedsaccadic adaptation due to cerebellar disease. European Journalof Neuroscience, 27, 132e144.

H€andel, B., Thier, P., & Haarmeier, T. (2009). Visual motionperception deficits due to cerebellar lesions are paralleled byspecific changes in cerebro-cortical activity. Journal ofNeuroscience, 29, 15126e15133.

Hopp, J. J., & Fuchs, A. F. (2004). The characteristics and neuronalsubstrate of saccadic eye movement plasticity. Progress inNeurobiology, 72, 27e53.

Ignashchenkova, A., Dash, S., Dicke, P. W., Haarmeier, T.,Glickstein, M., & Thier, P. (2009). Normal spatial attention butimpaired saccades and visual motion perception after lesionsof the monkey cerebellum. Journal of Neurophysiology, 102,3156e3168.

Imamizu, H., Miyauchi, S., Tamada, T., Sasaki, Y., Takino, R.,Putz, B., et al. (2000). Human cerebellar activity reflecting anacquired internal model of a new tool. Nature, 403, 192e195.

Ioffe, M. E., Chernikova, L. A., & Ustinova, K. I. (2007). Role ofcerebellum in learning postural tasks. Cerebellum, 6, 87e94.

Ito, M. (2006). Cerebellar circuitry as a neuronal machine. Progressin Neurobiology, 78, 272e303.

Ito, M. (2008). Control of mental activities by internal models inthe cerebellum. Nature Reviews. Neuroscience, 9, 304e313.

Ivry, R. B., & Spencer, R. M. (2004). The neural representation oftime. Current Opinion in Neurobiology, 14, 225e232.

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 170

Karni, A., & Bertini, G. (1997). Learning perceptual skills:behavioral probes into adult cortical plasticity. Current Opinionin Neurobiology, 7, 530e535.

Krakauer, J. W., & Shadmehr, R. (2006). Consolidation of motormemory. Trends in Neurosciences, 29, 58e64.

Lamont, M. G., & Weber, J. T. (2012). The role of calcium insynaptic plasticity and motor learning in the cerebellar cortex.Neuroscience & Biobehavioral Reviews, 36, 1153e1162.

Language cognition and the cerebellum: grappling with theenigma [Special issue]. Alan Beaton & Peter Marïen (GuestEds.). (2010). Cortex, 46, 811e946.

Law, C. T., & Gold, J. I. (2008). Neural correlates of perceptuallearning in a sensory-motor, but not a sensory, cortical area.Nature Neuroscience, 11, 505e513.

Leiner, H. C., Leiner, A. L., & Dow, R. S. (1989). Reappraising thecerebellum: what does the hindbrain contribute to theforebrain? Behavioral Neuroscience, 103, 998e1008.

Leiner, H. C., Leiner, A. L., & Dow, R. S. (1991). The human cerebro-cerebellar system: its computing, cognitive, and languageskills. Behavioral Brain Research, 44, 113e128.

Levitt, H. (1971). Transformed upedown methods inpsychoacoustics. Journal of the Acoustical Society of America, 49,467þ.

Lu, Z. L., Hua, T., Huang, C. B., Zhou, Y., & Dosher, B. A. (2011).Visual perceptual learning. Neurobiology of Learning andMemory, 95, 145e151.

Makin, A. D., Poliakoff, E., Chen, J., & Stewart, A. J. (2008). Theeffect of previously viewed velocities on motion extrapolation.Vision Research, 48, 1884e1893.

Manto, M., Bower, J. M., Conforto, A. B., Delgado-García, J. M., daGuarda, S. N., Gerwig, M., et al. (2012). Consensus paper: rolesof the cerebellum in motor controldthe diversity of ideas oncerebellar involvement in movement. Cerebellum, 11, 457e487.

Marr, D. (1969). A theory of cerebellar cortex. Journal of Physiology,202, 437e470.

Marvel, C. L., & Desmond, J. E. (2010). The contributions ofcerebro-cerebellar circuitry to executive verbal workingmemory. Cortex, 46, 880e895.

McKee, S. P., & Nakayama, K. (1984). The detection of motion inthe peripheral visual field. Vision Research, 24, 25e32.

McLaughlin, S. C. (1967). Parametric adjustment in saccadic eyemovements. Perception & Psychophysics, 2, 359e362.

Mednick, S. C., Arman, A. C., & Boynton, G. M. (2005). The timecourse and specificity of perceptual deterioration. Proceedingsof the National Academy of Sciences of the United States of America,102, 3881e3885.

Mednick, S. C., Nakayama, K., Cantero, J. L., Atienza,M., Levin, A. A.,Pathak, N., et al. (2002). The restorative effect of naps onperceptual deterioration. Nature Neuroscience, 5, 677e681.

Mesulam, M. M. (1999). Spatial attention and neglect: parietal,frontal and cingulate contributions to the mentalrepresentation and attentional targeting of salientextrapersonal events. Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences, 354, 1325e1346.

Miall, R. C., Weir, D. J., Wolpert, D. M., & Stein, J. F. (1993). Is thecerebellum a smith predictor? Journal of Motor Behavior, 25,203e216.

Molinari, M., Chiricozzi, F. R., Clausi, S., Tedesco, A. M., De Lisa, M.,& Leggio, M. G. (2008). Cerebellum and detection of sequences,from perception to cognition. Cerebellum, 7, 611e615.

Murdoch, B. E. (2010). The cerebellum and language: historicalperspective and review. Cortex, 46, 858e868.

Noudoost, B., Chang, M. H., Steinmetz, N. A., & Moore, T. (2010).Top-down control of visual attention. Current Opinion inNeurobiology, 20, 183e190.

O'Reilly, J. X., Mesulam, M. M., & Nobre, A. C. (2008). Thecerebellum predicts the timing of perceptual events. Journal ofNeuroscience, 28, 2252e2260.

Paulin, M. G. (1993). The role of the cerebellum in motor controland perception. Brain Behavior and Evolution., 41, 39e50.

Pelli, D. G. (1997). The VideoToolbox software for visualpsychophysics: transforming numbers into movies. SpatialVision, 10, 437e442.

Prsa, M., & Thier, P. (2011). The role of the cerebellum in saccadicadaptation as a window into neural mechanisms of motorlearning. European Journal of Neuroscience, 33, 2114e2128.

Rabe, K., Livne, O., Gizewski, E. R., Aurich, V., Beck, A.,Timmann, D., et al. (2009). Adaptation to visuomotor rotationand force field perturbation is correlated to different brainareas in patients with cerebellar degeneration. Journal ofNeurophysiology, 101, 1961e1971.

Ramnani, N. (2006). The primate cortico-cerebellar system:anatomy and function. Nature Reviews. Neurosciences, 7,511e522.

Raymond, J. L., Lisberger, S. G., & Mauk, M. D. (1996). Thecerebellum: a neuronal learning machine? Science, 272,1126e1131.

Roelfsema, P. R., van Ooyen, A., & Watanabe, T. (2010). Perceptuallearning rules based on reinforcers and attention. Trends inCognitive Sciences, 14, 64e71.

Rorden, C., & Brett, M. (2000). Stereotaxic display of brain lesions.Behavioural Neurology, 12, 191e200.

Roth, M. J., Synofzik, M., & Lindner, A. (2013). The cerebellumoptimizes perceptual predictions about external sensoryevents. Current Biology, 23, 930e935.

Rothwell, J. (1994). The cerebellum. In Control of human voluntarymovement (2nd ed.). Chapman & Hall.

Salmi, J., Pallesen, K. J., Neuvonen, T., Brattico, E., Korvenoja, A.,Salonen, O., et al. (2010). Cognitive and motor loops of thehuman cerebro-cerebellar system. Journal of CognitiveNeuroscience, 22, 2663e2676.

Sasaki, Y., Nanez, J. E., & Watanabe, T. (2010). Advances in visualperceptual learning and plasticity. Nature Reviews.Neurosciences, 11, 53e60.

Schiltz, C., Bodart, J. M., Dubois, S., Dejardin, S., Michel, C.,Roucoux, A., et al. (1999). Neuronal mechanisms of perceptuallearning: changes in human brain activity with training inorientation discrimination. NeuroImage, 9, 46e62.

Schmahmann, J. D. (1991). An emerging concept: the cerebellarcontribution to higher function. Archives of Neurology, 48,1178e1187.

Schmahmann, J. D. (1998). Dysmetria of thought: clinicalconsequences of cerebellar dysfunction on cognition andaffect. Trends in Cognitive Sciences, 2, 362e371.

Schmamann, J. D. (2010). The role of the cerebellum in cognitionand emotion: personal reflections since 1982 on the dysmetriaof thought hypothesis, and its historical evolution from theoryto therapy. Neuropsychology Review, 20, 236e260.

Schmahmann, J. D., MacMore, J., & Vangel, M. (2009). Cerebellarstroke without motor deficit: clinical evidence for motor andnon-motor domains within the human cerebellum.Neuroscience, 162, 852e861.

Schmitz-Hubsch, T., du Montcel, S. T., Baliko, L., Berciano, J.,Boesch, S., Depondt, C., et al. (2006). Scale for the assessmentand rating of ataxia: development of a new clinical scale.Neurology, 66, 1717e1720.

Seitz, A. R., Kim, R., & Shams, L. (2006). Sound facilitates visuallearning. Current Biology, 16, 1422e1427.

Serences, J. T., & Yantis, S. (2006). Selective visual attentionand perceptual coherence. Trends in Cognitive Sciences, 10,38e45.

Smith, A. C., Frank, L. M., Wirth, S., Yanike, M., Hu, D., Kubota, Y.,et al. (2004). Dynamic analysis of learning in behavioralexperiments. Journal of Neuroscience, 24, 447e461.

Smith, M. A., & Shadmehr, R. (2005). Intact ability to learn internalmodels of arm dynamics in Huntington's disease but not

c o r t e x 5 8 ( 2 0 1 4 ) 5 2e7 1 71

cerebellar degeneration. Journal of Neurophysiology, 93,2809e2821.

Sokolov, A. N., Ehrenstein, W. H., Pavlova, M. A., & Cavonius, C. R.(1997). Motion extrapolation and velocity transposition.Perception, 26, 875e889.

Stoodley, C. J. (2012). The cerebellum and cognition: evidencefrom functional imaging studies. Cerebellum, 11, 352e365.

Stoodley, C. J., & Schmahmann, J. D. (2010). Evidence fortopographic organization in the cerebellum of motor controlversus cognitive and affective processing. Cortex, 46, 831e844.

Stoodley, C. J., Valera, E. M., & Schmahmann, J. D. (2012).Functional topography of the cerebellum for motor andcognitive tasks: an fMRI study. NeuroImage, 59, 1560e1570.

Strata, P., Scelfo, B., & Sacchetti, B. (2011). Involvement ofcerebellum in emotional behavior. Physiological Research, 60,S39eS48.

Strick, P. L., Dum, R. P., & Fiez, J. A. (2009). Cerebellum andnonmotor function. Annual Review of Neuroscience, 32, 413e434.

Sultan, F., Augath, M., Hamodeh, S., Murayama, Y.,Oeltermann, A., Rauch, A., et al. (2012). Unravelling cerebellarpathways with high temporal precision targeting motor andextensive sensory and parietal networks. NatureCommunications, 3, 924.

Synofzik, M., Lindner, A., & Thier, P. (2008). The cerebellumupdates predictions about the visual consequences of one'sbehavior. Current Biology, 18, 814e818.

Thach, W. T. (1998). A role for the cerebellum in learningmovement coordination. Neurobiology of Learning and Memory,70, 177e188.

Timmann, D., & Daum, I. (2007). Cerebellar contributions tocognitive functions: a progress report after two decades ofresearch. Cerebellum, 6, 159e162.

Timmann, D., & Daum, I. (2010). How consistent are cognitiveimpairments in patients with cerebellar disorders? BehaviouralNeurology, 23, 81e100.

Timmann, D., Drepper, J., Frings, M., Maschke, M., Richter, S.,Gerwig, M., et al. (2010). The human cerebellum contributes tomotor, emotional and cognitive associative learning. A review.Cortex, 46, 845e857.

Tseng, Y. W., Diedrichsen, J., Krakauer, J. W., Shadmehr, R., &Bastian, A. J. (2007). Sensory prediction errors drivecerebellum-dependent adaptation of reaching. Journal ofNeurophysiology, 98, 54e62.

Vaina, L. M., Belliveau, J. W., des Roziers, E. B., & Zeffiro, T. A.(1998). Neural systems underlying learning and representationof global motion. Proceedings of the National Academy of Sciencesof the United States of America, 95, 12657e12662.

Watamaniuk, S. N. (2005). The predictive power of trajectorymotion. Vision Research, 45, 2993e3003.

Watamaniuk, S. N. J., & McKee, S. P. (1995). ‘Seeing’ motionbehind occluders. Nature, 377, 729e730.

Werner, S., Bock, O., Gizewski, E. R., Schoch, B., & Timmann, D.(2010). Visuomotor adaptive improvement and aftereffects areimpaired differentially following cerebellar lesions in SCA andPICA territory. Experimental Brain Research, 201, 429e439.

Werner, S., Bock, O., & Timmann, D. (2009). The effect ofcerebellar cortical degeneration on adaptive plasticity andmovement control. Experimental Brain Research, 193, 189e196.