Rhythm as a Cognitive Skill:
Temporal Processing Deficits in Autism

Judith Sheridan and J. Devin McAuley
School of Psychology
University of Queensland
Brisbane, Queensland 4072
Australia
devin@psy.uq.edu.au

Abstract

Timing accuracy and variability of children with Autistic Spectrum Disorder (ASD), age-matched controls, and college-age controls were evaluated in a rhythmic tapping task. Participants synchronized hand taps with a rhythmic series of tones, then continued that pattern without the tones. In addition to developmental factors, young children with ASD (uncomplicated by intellectual disability) were more variable than age-matched controls. Overall tapping variance was decomposed into clock and motor-delay estimates based on a two-stage model of timing (Wing and Kristofferson, 1973). The worse performance of the ASD group can be attributed to an increased clock variance relative to age-matched controls. No differences in motor-delay estimates were observed between the children with and without ASD. These results provide support for a cascade model of ASD (Courchesne, 1989) linking cerebellar dysfunction in ASD with a primary deficit in attentional timing.

Introduction

Autistic Spectrum Disorder (ASD) is a developmental disorder characterised by three essential domains of dysfunction, impairments in reciprocal social interaction, impairments in verbal and nonverbal communication and imaginative activity, and a restricted repertoire of behaviours and interests (American Psychiatric Association (APA), 1994). It is generally agreed that the clinical manifestation of ASD reflects an underlying neurological dysfunction (Gillberg and Coleman, 1992; Goodman, 1989) and there have been a number of attempts to integrate clinical observations with current neurological knowledge (Bachevalier, 1994; Courchesne, 1989; Damasi and Maurer, 1978; Maurer and Damasio, 1982; Ornitz, 1983; 1989; Waterhouse, Fein, and Modahl, 1996). The single most consistent finding of neuropathology and radiological-scan studies in autism has been pathologic features of the cerebellum (Courchesne et al., 1987; Courchesne, Townsend, and Saitoh, 1994; Hashimoto et al., 1989; Pr! ior et al., 1984).

The cerebellum has been established as essential for regulating many aspects of motor control including movement timing, coordination, muscle tone, and posture (Lechtenberg, 1988; Thatch, Goodkin and Keating, 1992). In contributing to skilled motor performance, there is evidence to suggest that a key contribution of the cerebellum is to act as a central timing mechanism or ``clock'' for both the perception and production of temporal intervals (Braitenberg, 1967; Fahle and Braitenberg, 1984; Keele et al., 1985; Ivry and Keele, 1989). Ivry and Keele (1989) compared the timing ability of patients with Parkinson's disease, focal cerebellar lesions, degenerative cerebellar disease, peripheral neuropathy, and cortical lesions to age-matched controls using a time-discrimination task and a rhythmic tapping task. For the tapping task, subjects synchronized finger-taps with a rhythmic series of pacing tones and then continued tapping at the same rate in the absence of the tones. Tap! ping performance was measured by examining accuracy and variability of subject's inter-response-intervals (IRIs) during the continuation phase of the task. IRI variability was separated into clock and motor-delay components using the two-stage timing model proposed by Wing and Kristofferson (1973). Wing and Kristofferson assume that variability in rhythmic tapping is due to two independent sources: clock variability and motor-delay variability. The model posits a central clock timer (which for Ivry and Keele is localized in the cerebellum) with a mean clock-interval specified by the task. Following each clock interval C_i, a motor response is generated with a variable delay. By assuming independence between the motor delay (D_i) and clock interval (C_i), the predicted inter-response-interval is given by

IRI_i = C_i + D_i - D_i-1

with the variance of the inter-response-intervals determined by the sum of the clock variance plus twice the motor-delay variance:

²_IRI = ²_C + 2²_D

The clock and motor variances are determined from the observation that an independent motor delay predicts a negative correlation (bounded between 0.0 and -0.5) for successive IRIs (see Wing and Kristofferson (1973) for more details). The magnitude of the negative lag-one covariance provides an estimate of motor-delay variance, which (given the IRI variance) determines the estimate of clock variance.

By localizing the clock in the cerebellum, Ivry and Keele predicted that patients with cerebellar damage should exhibit increased clock variance relative to the non-cerebellar patients and the normal controls.

Rhythmic tapping studies focussing on IRI variability of normal subjects provide broad support for the Wing and Kristofferson model; most of the reported lag-one correlations are between 0.0 and -0.5, permitting a direct decomposition of tapping variability into clock and motor components (Wing and Kristofferson, 1973; Wing, Gentner, and Church, 1989). However, with the different patient groups, Ivry and Keele found violations of the model (i.e., positive lag-one correlations) corresponding to negative motor-delay estimates for 21% of the tapping trials. When the groups were examined separately, the peripheral neuropathy group was found to rarely produce violations of the model (providing evidence of large motor-delays), whereas the Parkinsonian, cortical, and cerebellar groups violated the model on 18%, 14.3% and 26% of the trials, respectively. Consistent with the hypothesis of a cerebellar clock, they reported that the patients with cerebellar damage showed reliab! ly increased clock variance compared with the Parkinsonian group, peripheral neuropathy group and the control group. In addition, the cerebellar group was the only group to show a deficit in the perception of duration relative to the controls.

Leiner, Leiner and Dow (1986, 1989) hypothesized that the cerebellum could contribute to skilled mental performance in the same way that it contributes to skilled motor performance--by coordinating the timing of cognitive and motor processes (e.g., in verbal fluency and temporal sequencing). This hypothesis was based on the observation that the cerebellum is connected (via the modulating output of the Purkinje cells) with almost every brain system including the limbic system in the cerebro-ponto-cerebello-thalamo-cortical loop \cite{Diamond:85,Leiner:86}. Support for a general role of the cerebellum in cognitive functioning has been provided by several separate sources. Using a positron emission tomography (PET) study of the ``articulatory loop'', Paulescue, Frith, and Grackowich (1993) reported that during sub-vocal rehearsal tasks the supplementary motor area (SMA), cerebellum, and sensori-motor areas were activated even when there was no overt speech, suggesting that the! se s tructures (including the cerebellum) are engaged as part of the language planning process. In a detailed neuropsychological evaluation on a patient with idiopathic cerebellar degenerative disorder (with relatively little cerebral atrophy), Akshoomoff et al., (1992) reported significant impairments in verbal associative learning and visuo-spatial skills. In a study of attentional switching, Akshoomoff and Courchesne (1992) demonstrated that patients with damage to the cerebellum were significantly impaired in their ability to respond rapidly with shifts of attention between sensory modalities, but are unimpaired in their ability to maintain attentional focus. They concluded that the neo-cerebellum is necessary for rapid voluntary shifts in attention between sensory modalities.

Given the experimental evidence highlighting the importance of the cerebellum for both skilled mental and motor performance, Courchesne and colleagues have proposed a cascade model of ASD based on cerebellar dysfunction in infancy (Courchesne, 1989; Courchesne, Townsend, and Saitoh, 1994; Akshoomoff, 1992). Courchesne and colleagues have argued that cerebellar dysfunction in infancy leads to a fundamental deficit in attentional timing and that delays in attentional switching between sensory modalities lead to an impairment in the autistic child's ability to coordinate shifts of attention with salient changes in the environment. As a result, the knowledge that an autistic child develops about the world consists of disconnected fragments of (gestural, facial, vocal, and emotional) information, giving rise to the major characteristics of autism including impaired social skills.

In support of the model, Courchesne et al. (1994) showed that autistic individuals (similar to patients with cerebellar damage) have more difficulty switching attention between auditory and visual modalities than do normal controls. However, Ivry and Keele's conclusion that the cerebellum functions as a timing mechanism for both the perception and production of duration suggests an obvious (untested) corollary to Courchesne's hypothesis that cerebellar dysfunction in ASD leads to a primary deficit in attentional timing: Autistic individuals should also exhibit deficits in the perception and production of duration. If autistic individuals exhibit timing deficits, then to what extent can those deficits be accounted for by overall intellectual functioning and to what degree can those deficits be given a developmental explanation? The present study was designed to address this question by comparing the timing ability of children with ASD (between the ages of 7 and 15) with age-! matched and college-aged control groups using a simple rhythmic tapping task.

Method

Participants. Children between the ages of seven and fifteen with a prior diagnosis of ASD were recruited primarily via the Asperger's Syndrome Support Network in Queensland, Australia. Asperger's Syndrome represents a variant of ASD characterized by preserved language and less severe disability than is seen in Kanner's autism. An independent diagnosis by a medical practitioner or registered clinical psychologist was required for inclusion in the ASD group. Age-matched controls were recruited via school newsletters, local newspapers, or families with a child with ASD already participating in the study.

In order to assess IQ, each child was administered the WISC-III (the third edition of the Weschler Intelligence Scale for Children; Weschler, 1949). The WISC-III is an individually administered clinical instrument for assessing the cognitive abilities of children (aged 6 years through 16 years and 11 months) which yields three composite scores: Verbal IQ (VIQ), Performance IQ (PIQ), and Full Scale IQ (FSIQ). The distributions of VIQ, PIQ, and FSIQ scores have a mean of 100 and standard deviation of 15; About two-thirds of all children obtain scores between 85 and 115, and about 95\% score within the 70 - 130 range. Significantly sub-average intellectual functioning is defined as an IQ score of about 70 or below (DSM-IV, American Psychiatric Association, 1994).

Table 1: Mean verbal, performance, and full-scale IQ scores on the WISC-III for the five groups of children.

Fourteen children with ASD met an inclusion criterion of Full Scale and Verbal IQ scores above 75 (see Table 1). This group was divided by age into two sub-groups: ages 7-10 (n = 11) and ages 12-15 (n = 3). Four children with ASD who did not meet the inclusion criterion were assigned to a low-functioning group (Low IQ) and their data was treated separately. For the ASD (7-10) and (12-15) groups, the mean scores were all near 100, indicating that both of these groups were functioning in the average range of intellectual ability; any observed deficits in rhythmic tapping ability for these groups were unlikely to be due to an effect of impaired intellectual functioning (e.g., difficulty in understanding the task requirements).

Children were excluded from the control group if they had been diagnosed previously with a developmental, learning or behavioural disorder; had a prior diagnosis of epilepsy or head injury; or had a verbal or full-scale IQ score on the WISC of 75 or below. Eleven children without ASD met the inclusion criteria. These children divided by age into two subgroups: ages 7-10 (n = 6) and ages 12-15 (n = 5). For the control (7-10) and (12-15) groups, the WISC scores represented the high-average range of intellectual functioning (about 75th percentile), indicating a sampling bias (probably due to the volunteer nature of the recruitment of controls).

There were 18 adult-age controls that participated in the study for partial credit in a first-year psychology course at the University of Queensland.

Stimuli. The stimulus for the uni-manual task consisted of an isochronous series of 1100-Hertz 50-ms tones with a fixed-inter-onset interval of 600 ms.

Equipment. The test rhythm was delivered over small speakers (placed directly in front of the subjects) generated on a PC computer which also controlled all aspects of stimulus presentation and response collection. Speakers were used in preference to headphones as many children with ASD show tactile hypersensitivity particularly for the face and scalp areas. The experimental task used a response board with which subject's left- and right-hand tapping responses were recorded on left- and right-hand copper response plates. Each participant wore a loose fitting wrist band (with copper insert) that was attached by flexible insulated wire to the tapping device and which signalled a response by completing an electrical circuit when the subject touched either plate. Press and release times were recorded to the nearest ms.

Procedure. Children in both the ASD and control groups performed a set of tapping tasks using the isochronous rhythm described above. For the presented rhythm (a trial), the participant's task was to tap in synchrony with the tones as accurately as possible. After 40 pattern cycles, the tones stopped and the participant was instructed to continue tapping the rhythm at the same rate. The continuation phase also lasted 40 pattern cycles. Tapping responses were collected for a single-hand condition in which the participant tapped with either the left or right hand. Both hands were always tested for each subject, but the order was counterbalanced between subjects.

Prior to the collection of tapping responses, each participant (tested separately) was seated in front of the response board and allowed to adjust the copper plates and the board so that they were able to tap comfortably. The children were especially encouraged to ask questions about the equipment and the procedure until they seemed comfortable with the environment. Participants were then told that they would hear some tones. This was followed by a demonstration of the tested rhythm and the instructions to tap in time with the tones. The tested rhythm was then played again and the participants were encouraged to tap on the response board in synchrony with the tones. When it was clear to the experimenter that the participant understood and was able to produce the response requirements of task , they were told that after a period of time the tones would stop, but that they should continue tapping as if the tones were still playing. To start a trial, participants were instruct! ed t o begin tapping with the tones when they were ready, followed by the experimenter initiating the computer generation of the test rhythm. A second set of trials was run after the completion of the first.

Rhythmic Tapping Performance

The analysis of the tapping data was based on the inter-response-intervals (IRIs) between successive taps (presses) on the copper response plates and the asynchrony between the onset of the pacing tones and the synchronization response of the subject. Some initial data screening was necessary. IRIs less than 100 ms were treated as unintended, and added to the following IRI (i.e., the unintended response, essentially a ``finger bounce'', was removed from the response stream). The mean IRI scores revealed a slow linear drift in rate that occurred for all of the subjects during the continuation phase. Before calculating IRI variances, the linear trend in each trial was removed. All statistical analyses reported below were sensitive to this slow drift in rate. Omnibus significance tests were replaced with a series of planned comparisons between the different groups examining both developmental and ASD effects. One-tailed t-tests were used in the comparisons of IRI variances, gi! ven the hypotheses that children with ASD should be more variable in their tapping than children without ASD, and that younger children should be more variable in their tapping than older children and adults. Two-tailed t-tests were used to compare mean IRI and asynchrony (in the absence of directional hypotheses).

Group Comparisons

The group means for each of the control and ASD groups for the isochronous rhythm during single-hand performance are reported in Table 2 (synchronization) and Table 3 (continuation). No significant differences were observed between the two hands, and thus the tapping data for the dominant and non-dominant conditions have been pooled.

Table 2: Group means during synchronization phase of tapping task: inter-response-interval and asynchrony.

A clear developmental pattern was observed when the performance of the groups was compared. Overall, the college-age controls were least variable in their tapping. The controls (ages 12-15) were reliably more variable in the production of the target interval compared with the college-age controls during both synchronization (t(21) = 4.34, p < 0.001) and continuation (t(21) = 2.47, p < 0.01) Similarly, the 7-10 controls were more variable in their tapping during synchronization than the 12-15 controls (t(9) = 2.67, p < 0.01); however, during continuation, no significant difference in tapping variability was found between the 7-10 and 12-15 groups (t(9) = 0.96, p 0.1).

All the groups showed a decrease in their ability to maintain the target interval during the continuation phase of the tapping task. Consistent with previous studies of rhythmic tapping, subjects showed a tendency to speed up (drift). When comparing the different groups, a developmental pattern of drift was observed; the college-age controls maintained a mean IRI close to the target (IRI mean = 583.29), whereas the mean IRI for the 12-15 and 7-10 groups (combined) was significantly shorter (IRI mean = 570.9).

When comparing the ASD and control groups, the children with ASD were found to be more variable in the production of the target interval than the children without ASD. Although there was a consistent trend of reduced tapping performance in ASD compared with aged-matched controls, the planned comparisons revealed that the only marginally reliable difference in tapping variability occured between the 7-10 groups during the continuation phase (t(15) = 1.55, p < 0.1). The effect of ASD on the ability to maintain tapping speed during continuation was similar to the developmental effect, with the 7-10 ASD group showing the greatest overall increase in tapping speed (IRI mean = 542.79). A comparison between the mean IRI during continuation for the 7-10 ASD group and the 7-10 control group revealed a significant difference (t(15) = 2.07, p < 0.05). The overall weaker ASD effects suggest that larger group sizes are needed to separate the ASD and developmental effects.

Not surprisingly, children with ASD and intellectual disability were less accurate and more variable in their tapping compared with all of the other groups. In a planned comparison between the low-IQ ASD and 7-10 ASD groups, a significant difference was found only in IRI variability during synchronization (t(13) = 1.84, p < 0.05). The worse performance of the low-IQ ASD group highlights the importance of controlling IQ when assessing timing ability.

Consistent with previous studies of synchronization, mean asynchronies for all five groups were negative (Mates, 1994). Mates provided an explanation for negative asynchronies based on the coincidence of stimulus and feedback information and the minimization of subjective synchronization error. However, in the present study, no consistent pattern of asynchrony differences was observed between groups for the isochronous rhythm and was not considered further for the purposes of this paper.

Decomposition of Tapping Variability

Data was analyzed using the Wing and Kristofferson (W&K) model to decompose tapping variability (during continuation) into clock and motor-delay estimates. This decomposition procedure yielded negative motor-delays on some trials, indicating that the independence assumptions of the W&K model had been violated. Negative motor delays (positive lag-one correlation of the IRIs), occurred primarily for the ASD groups, with the ASD groups producing violations on 28.5% of all trials, compared with only 9% violations for the age-matched controls. The largest percentage of violations (43%) occurred for the low-IQ ASD group, whereas the college-age controls produced no violations.

Table 3: Group means during continuation phase of tapping task: clock and motor-delay estimates were obtained by decomposing IRI variance into clock and motor components using the Wing and Kristofferson model.

In order to minimize the effect of the negative motor delays on the computation of clock variance, trials for which model violations occurred were eliminating from the decomposition. Table 3 reports the mean clock and motor estimates for the six groups. The main question concerned whether differences in overall variability (reported in the previous section) were due to clock difference, motor-delay differences, or both. For the reported developmental effects, the differences in variability were due to both clock and motor-delay differences. As the age of participants increased, both clock and motor estimates decreased. However, for the reported ASD effects, the larger variability of the 7-10 ASD and Low IQ ASD groups compared with the 7-10 control groups was primarily due to increased clock variance. Age-matched motor-delay estimates for children with and without ASD were similar.

Conclusions

The strongest effects in this study can be accounted for by developmental factors. Adults are less variable and more accurate in rhythmic tapping than the 7-10 or 12-15 controls. Similarly, 12-15 controls are less variable than 7-10 controls. When the differences between the timing ability of children with and without ASD are examined, the primary finding is that the 7-10 ASD group is more variable during continuation and less able to maintain a target tapping speed than the aged-matched controls. The reduced timing ability of the 7-10 ASD group is primarily due to an increased clock variance. In contrast, developmental effects are due to a combination of motor and clock factors.

The differences in timing ability of children with and without ASD reported in this paper are not strong. This may be because of small group size, or that the task was too simple to isolate consistent differences. Unpublished data from the same participants using more complex rhythmic tasks suggest a greater differentiation of timing abilities.

The findings of the present study with ASD subjects parallel those of Ivry and Keele with cerebellar patients. A consistent finding with the ASD group is that they exhibit a large percentage of violations of the W&K model. In Ivry and Keele's study, cerebellar patients also produced the greatest percentage of violations of the model. Following adjustment for those violations, the cerebellar patients demonstrated increased clock variance in comparison with the other patient groups, as did the ASD groups in the present study. This provides additional support for the Ivry and Keele's hypothesis that timing deficits occur as a consequence of cerebellar damage. A broader issue is raised by the consistent finding of violations of the W&K model in both studies. Although the W&K can consistently account for the performance of neurologically intact individuals during the continuation phase of rhythmic tapping tasks, it is an an open-looped timing model that can not accou! nt f or synchronization performance (Jones, 1976; Mates, 1994; Large, 1994; McAuley, 1996). In addition, it consistently has trouble accounting for the data generated by neurologically compromised groups, limiting the extent to which it can provide an adequate explanation of the cognitive (and ultimately) neural mechanisms underpinning timing. Nonetheless, given Ivry and Keele's assumption of a cerebellar clock, the increased clock variance observed for the ASD group provides some additional support for Courchesne's cascade model of ASD linking neuro-developmental dysfunction of the cerebellum with deficits in timing.

References

Click here for the list of references and a postscript copy of the paper which appears in the Proceedings of the Fourth Australiasian Cognitive Science Conference.