Experimental BrainResearch

Exp Brain Res (1988) 71:116-128

9 Springer-Verlag 1988

Trajectory control in targeted force impulses V. Gradual specification of response amplitude W. Hening*, M. Favilla**, and C. Ghez Center for Neurobiologyand Behavior,New York State PsychiatricInstitute, Collegeof Physiciansand Surgeons, ColumbiaUniversity, New York, NY 10032, USA

Summary. This study was undertaken in order to determine the time course of the process by which information derived from a visual target is used to accurately set the amplitude of a simple motor response. We refer to this process as response specification. Separate auditory and visual cues were given to the subjects in order to independently control the moment of response initiation and the time available for processing amplitude information from the target. Six subjects initiated impulses of isometric force in synchrony with the last of predictable series of regular tones. Response amplitudes were to match one of three visual target steps occurring at random times between 0 and 400 ms before the response-synchronizing tone. Using these separate auditory and visual cues, we were able to systematically vary the time interval between target presentation and response onset, termed here Stimulus-Response or S-R interval. Target steps were presented in blocks of either predictable (simple condition) or unpredictable (choice condition) amplitudes. The peak forces and the peaks of their time derivatives were analyzed to determine how subjects achieved accuracy under the different conditions and at different S-R intervals. The trajectories of responses produced in the simple condition were independent of the S-R interval. In contrast, when targets were presented in unpredictable order, the distribution of the peak forces of the subjects' responses depended on the S-R interval. At short SR intervals (< 125 ms), subjects made responses whose peak forces were distributed around the center of the range of target steps. These responses formed Present addresses: *Department of Neurology, LyonsVeterans AdministrationMedicalCenter, Lyons,NJ 07939, USA ** Istituto di FisiologiaUmana, Universit~di Ferrara, V. Fossato di Mortara, 64B, Ferrara, Italy Offprint requests to: C. Ghez, Center for Neurobiology and Behavior, 722 W. 168th Street, New York, NY 10032, USA

a unimodal, but broad distribution which was independent of actual target amplitude. With increasing S-R interval (> 125 ms), the distributions of peak forces gradually shifted toward the correct target amplitudes, with the means reaching the appropriate amplitudes at S-R intervals of 250-400 ms. At S-R intervals comparable to a reaction time, the range of peak forces was constricted to a similar extent as previously observed in a reaction time task (Hening et al. 1988). We found that the gradual improvement of accuracy was not achieved through changes in trajectory control: at all S-R intervals, subjects utilized a pulse-height control policy (Gordon and Ghez 1987a). Different peak forces were achieved by varying the rate of rise of force, while force rise time was held relatively invariant. We did find, however, that within the constraints imposed by rise time regulation, compensatory adjustments to the force trajectories (Gordon and Ghez 1987b) were greatest during the period of specification. We conclude that (1) subjects can initiate their responses independent of the degree of specification achieved and that the normal process of specification of amplitude begins earlier and continues longer than the latency of responses in a reaction time task; (2) before target presentation, subjects prepare a default response whose amplitude is biased by prior experience with the targets presented in the task. We hypothesize that the central mechanisms that specify response amplitude do so by a progressive adjustment of the default parameters.

Key words: Humans - Trajectory control - Isometric force - Motor program - Reaction time - Information processing

117

Introduction

In the preceding study (Hening et al. 1988), we have shown that when subjects initiate responses as soon as possible to unpredictable targets, the scaling of their responses shows a systematic distortion. Responses to targets requiring small force changes are hypermetric while those to large force targets are hypometric. This constriction in the range of response amplitudes was shown to reflect a systematic bias toward the center of the range of targets or toward the most probable target. This range effect (Poulton 1975, 1981) was present in all subjects, independent of their level of practice. However, the effect almost disappeared completely when experienced subjects were allowed to delay responding until they felt optimally prepared to respond accurately. This dependence of scaling on the urgency of responding suggested that the process of specification extends over a time interval that exceeds a minimal reaction time and that when an urgent response is required subjects can respond before specification is complete. The biases of responses produced as soon as possible also suggested that amplitude specification derives from a common default value that is prepared in advance of target presentation. The present study was undertaken in order to test the hypotheses that (1) specification of response amplitude is an extended process beginning earlier and terminating later than a reaction time, and (2) that responses of different sizes are derived from a common default. To address these questions it was necessary to develop an experimental procedure in which the processes responsible for triggering a response could be dissociated from the processing required to specify the amplitude of their trajectories. We achieved this dissociation by having subjects synchronize their responses with a tone occurring at a predictable time. A visual target provided information only about the amplitude of the required response. By providing these separate auditory and visual cues and varying the time interval between them, we were able to interrupt the process of specification at any time after target presentation. This allowed us to examine the trajectories of responses produced at varied times following target presentation and thus to analyze the time course of amplitude specification. We refer to this paradigm as the timed response paradigm. A paradigm somewhat similar to ours was used by Schouten and Bekker (1967) to examine speed-accuracy tradeoffs in a button pressing task. In this study, we have examined the time course of amplitude specification in responses made to target shifts of three different sizes. The three targets

were presented in either predictable (simple condition) or unpredictable (choice condition) order. The distributions of response amplitudes made at intervals following the visual stimulus that were too short to be influenced by the target were analyzed to determine whether, in the absence of specific information, subjects guess among the different response alternatives or whether they use a single default. We also sought to determine whether specification is a gradual process or one where subjects abruptly switch from a default to the correct amplitude. Finally, we examined the relations between trajectory parameters to determine whether, in the course of specification, subjects alter the pulse height control policy which they normally use to accurately scale force impulses (Gordon and Ghez 1987a; Hening et al., 1988). Lacking sufficient information to correctly predict the required peak force, subjects might produce responses in which the initial portions of the trajectories were unscaled. The peak force might then be determined by such modifications to the later trajectories as variations in force rise times (Cordo 1987) and compensatory adjustments for initial errors (Gordon and Ghez 1987b). Preliminary accounts of this work have appeared as an abstract (Hening and Ghez 1984) and as part of a short communication (Ghez et al. 1988).

Methods Subjects were 6 neurologically normal adults (ages 31 to 42, 4 men and 2 women). All were given at least one session of training with the present paradigm prior to testing, in order to familiarize them with the task. The subjects also had at least two sessions of previous experience in the production of aimed isometric force trajectories produced at the elbow (for description of task, see Ghez and Gordon 1987; Gordon and Ghez 1987a). Two subjects, $2 and $5, participated in the preceding study of force impulses produced at the metacarpophalangeal joint of the index finger (Hening et al. 1988). The apparatus used to conduct these studies was described in detail in an earlier report in this series (Ghez and Gordon 1987). In the preceding study (Hening et al. 1988), we examined isometric flexion responses produced by the index finger. However, since we subsequently made further progress in characterizing the trajectories of isometric force impulses produced at the elbow (Ghez and Gordon 1987; Gordon and Ghez 1987a and b), the present study examines elbow responses. In brief, subjects sat with their right shoulder abducted to 70 ~ elbow flexed to 90~ and the arm and wrist immobilized with a series of padded but rigid metal restraints. A system of strain gauges measured the forces applied to the wrist cuff by elbow flexors and extensors. Subjects viewed two oscilloscopes. On the upper oscilloscope, a target level and the force registered by the strain gauge were displayed at a fast sweep speed and appeared as two horizontal lines. The force line moved up and down with flexor and extensor forces, respectively. As in the earlier papers of this series, the required response was a single uncorrected impulse of force with as brief a rise time as possible. Subjects were instructed to allow the force to return passively to

118

4, ,

!

4, . . . .

!

Tone # 1

2

4, . . . .

!

4, . . . .

i

3

Target -.~

. . . .

4

i

' ' 0.1 s

i~/~'.~ ~

Force

~,4

~ Tone # 4

U//////A

i

i Timing i error S-R interval t,

S-T interval

( randomized ) Fig. 1. Response specification task. Four successive tones of increasing pitch (above) allow subject to synchronize response initiation with the last tone. The target level (hatched area) shifts to one of three possible target levels (interrupted hatched zones). Blow-up below shows time relations and a negative timing error is depicted (see Methods for details)

baseline after it had reached its peak. A storage oscilloscope, placed below the monitor, allowed subjects to check the timing and the trajectories of their responses.

from zero to 400 ms before the fourth tone. For two subjects ($2 and $14), however, the target step was also presented at a fixed ST interval of 350 ms in some of the blocks of trials. In the course of a single trial, subjects would first align their force with the target for 4 s. Then, as shown schematically in Fig. 1, the sequence of ascending tones was initiated. Between the third and fourth tones, the target was stepped to a new level. Subjects initiated the required force impulse at the time of the fourth tone, using whatever information they could from the target. As shown in the lower part of the figure, we calculated two different intervals for each response. The difference between the time of response onset (measured as the first change in dF/dt from zero) and that of the tone with which it was to be synchronized was taken as the timing error. A positive timing error indicates that response onset occurred later than the synchronizing tone. The interval between the step change in the target level and the time of onset of the subsequent response was defined as the Stimulus-Response or S-R interval. It represents the time which the subject had available to incorporate amplitude information from the target into their responses. The S-R interval differs from a conventional reaction time in that response onset is controlled by the experimenter and, therefore, does not depend upon a decision by the subject of when to initiate responses. Testing sessions consisted of 6 to 10 blocks of 60 trials. The three target amplitudes were presented in both predictable (simple) and unpredictable (choice) blocks. Since behavior in the simple blocks was found to be quite consistent and relatively independent of S-R interval (Fig. 3), the majority of blocks were in the choice condition. Typically, sessions consisted of one block for practice, two simple blocks, and a variable number of choice blocks. Choice blocks were presented later to minimize performance deficits due to lack of intra-session practice. The simple blocks consisted of the three target steps presented in a fixed order (X, Y, Z, X, Y, Z, X . . . ) while choice blocks consisted of the targets presented in a pseudo-randomized order. We presented the simple targets not in blocks of uniform amplitude, as done in the preceding paper (Hening et al. 1988), but in a cyclical series to limit the distinction between simple and choice to that of predictability alone. In the more usual simple block, the subject can readily correct amplitude errors from one trial to the next, since the required response remains constant. The present procedure forces the subject to learn responses to a range of targets.

Data acquisition and analysis Experimental procedure: the timed response paradigm We obtained independent experimental control of response initiation and response amplitude by providing separate auditory and visual cues (Fig. 1). Response timing was controlled by training subjects to initiate the force impulses in synchrony with the last in a series of four predictable 20 ms tones presented through earphones. Tone intensity was adjusted to a comfortable level, approximately 40 to 50 dB above hearing threshold. To facilitate correct timing, successive tones in the series had a progressively higher pitch (increasing from 625 to 1300 Hz) and were separated by constant intervals of 500 ms. The cue signalling the desired response amplitude was a step change in the target level displayed on the monitor oscilloscope. Subjects were instructed to just match this target level with the peak of a correctly timed force impulse. The amplitudes of the target steps were in ratio of i : 3 : 5 and the largest force required was chosen at a level which did not produce significant fatigue. This level varied from 30 to 50% of the maximal isometric force the subject could produce. The time interval between the occurrence of the target step and the fourth tone (Stimulus-Tone or S-T interval) was varied unpredictably

Experiments were controlled and data collected and analyzed using a small general purpose computer (PDP 11/23, Digital Equipment Corporation). The time of target presentation and its amplitude were collected with the subjects' force response. Additionally, electromyographic (EMG) activity of biceps and triceps muscles was recorded using surface electrodes (Boston Elbow Myoelectrodes, Liberty Mutual). Automatic computer programs marked the force trajectories and reduced the trial data to arrays of variables which described trajectory parameters (amplitude and timing of the peaks of force and of its first two time derivatives, timing error, and S-R interval). Trials with responses showing more than one peak in the trajectory of the first time derivative of force (dF/dt) were assumed to show evidence of voluntary corrections and were rejected. Trials initiated in the wrong direction were also rejected. The percentage of rejected trials varied between 5% and 20% in different subjects. The arrays of trajectory variables were then subjected to statistical analyses, including multiple regression analysis. Data from individual sessions were first analyzed separately. Then, in order to achieve the highest resolution in determining the time course over which responses were specified, data from up to four experimental

119 250" .... : :...

..

Simple

9. . . . . . .

Choioe

0

15C

"6 d Z

Responses to targets of predictable amplitude are specified in advance of target presentation

:..=

50

i

-150

-0.02 (r = 0.08) to -0.28 (r = 0.53). This dependence suggested that, in the face of uncertainty, the subjects briefly withheld responding in hope of obtaining more information about the target.

i

I

i

0

=

150

Timing Error (ms) Fig. 2. Distributions of timing errors in simple and choice conditions (simple, N = 259; choice, N = 1097)

sessions of each subject were grouped. Curves were fit to bivariate plots of the data using a non-parametric least squares estimation procedure referred to as LOWESS (Cleveland 1979; Chambers et al. 1983), or Locally Weighted Scatterplot Smoother. This procedure employs an algorithm which is conceptually similar to a quadratic moving-average smoothing performed on time series data. It also includes a "robustness" criterion which gives less weight to outlying data points. In order to estimate the dispersion of points around the mean in each local region, another LOWESS curve was fitted to the absolute values of the residuals. The value of this curve at each point was then added to and subtracted from the mean at each point to provide a measure of dispersion analogous to the standard deviation (Cleveland 1979; Chambers et al. 1983).

Results

Response timing All subjects learned to produce smooth force impulses that were well synchronized to the last tone after one or two sessions of practice. Figure 2 shows the timing errors made by one subject in the course of three sessions. Responses in both simple (solid line) and choice (dotted line) conditions were, on average, initiated slightly before the tone. For the trials illustrated here, the mean timing errors were - 5 ms, +34 ms (mean, + s.d.) in the simple condition and - 1 1 ms, +_34 ms in the choice condition. In all subjects, the dispersion in response timing did not differ significantly between simple and choice conditions. However, in the choice condition, all six subjects showed a systematic dependence of timing error on the S-T interval: responses with the shortest S-T intervals had a more positive timing error than responses with long S-T intervals. This relationship was statistically significant in 5 of the subjects (p < 0.05). The relationship between S-T interval and timing error was linear, but the slope of the relationship was quite low (average, -0.13) and ranged from

When subjects knew the required response amplitude in advance of target presentation, both the amplitude and shape of their responses remained independent of the S-R interval. These findings are illustrated for a typical subject ($2) in Fig. 3. In part A, the peak force of each response is plotted as a function of the S-R interval. Empty circles indicate responses to the targets requiring responses of small and large size, while responses to the middle target are indicated by 'x' symbols. The LOWESS lines (see Methods) fitted through each cloud of points can be seen to closely approximate the horizontal line corresponding to each target. The dispersion in response amplitudes was also similar at all S-R intervals. Figure 3B compares the trajectories of responses initiated at S-R intervals shorter and longer than 150 ms (vertical dashed line in Part A), a value corresponding approximately to the subject's simple reaction time. Both groups of force impulses have the features of force impulses which we have described in previous papers of this series (Gordon and Ghez 1987a): different amplitudes of force are achieved by a proportional scaling of trajectories with the same shape and rise time. These findings indicate that when the required amplitude of the response was known in advance, the subject produced a predictive and appropriate response whose trajectory was not substantially altered during the S-R interval by information derived from the actual target.

Specification of response amplitude to unpredictable targets is gradual and proceeds from a default value When subjects did not know the required size of the response in advance, the distributions of response amplitudes varied systematically with the S-R interval. Figure 4 illustrates the changes in the peak forces and trajectories of the responses to each target produced by subject $2. The amplitudes of individual responses to the large, medium and small targets are plotted separately against S-R interval in the three parts of Fig. 4A together with the LOWESS lines fitted to the data. LOWESS lines and associated average variability curves for each target amplitude are superimposed in Fig. 4B. AT short S-R intervals

120

A

a

b o

o o oo q o oo I~ | ...........o......~..~ % ~ o _ ~

/

- ~ i ~

I ~~' .............."6 o o

o

o

o

..................~ ....

.......................................................... ~ o x

o

~

x

~

o

' 8 ~ ..................................................s ~

z

~

o

o

•

.....~ " ~

x x

~

x

.................... . . . . . . . oO~.

o

o

o

x~.,~. " ' ~ i ~

,.-..:........ x ~ . . . .

............................

• ~• .............................................................. ,~ ~ ~..;._....._~_.~ ~~ ~ . . ~ . -*,~7,',~2.~'#~ - ~ ................ (1) C1

_

I

•

o

o I

2

5

o~

~

1

1

"

x

~

•

"""'"0-~'~'""~"~"~,~,~.

-

/

o

~

J

o

/

o~

I"~...-..-~....~.....

~

oo

~

o

~. -

-o

9,

9 ............ :-.

~

---,,,

.................... o

-

.....

I

I

.

S-R

B

.

(ms)

IntervQI

a

b

............. T": :

/, 75N

i'//--X,

.............. ",.

! ,,

,

"'.,

............... i;e....:'::.'2.:..~._.,..:,i

.,. --......

I

'

o

i

'

,IkTk./ A,,'

oa L%]M7,

ms

I

0

6o'

26o ms

Fig. 3A, B. Response specification in simple condition. A Individual peak forcesfrom subject $2 are plotted against S-R interval. Open circles: responses to large and small targets (25 and 125 N); crosses: responses to the middle target size (75 N). Thick horizontal lines indicate target levels. The responses to each of the three targets are fit by LOWESS (solid lines) together with flanking lines indicating dispersion (see Methods). B Averaged force trajectories of responses to three targets. (a) Trials initiated at S-R intervals shorter than 150 ms. (b) Trials initiated at S-R intervals greater than 150 ms. Force trajectories are synchronized on force onset. Number of trials by target size: small, 131; medium 139; large, 136

( < 125 ms), the amplitudes of responses to all three targets overlap and are centered around the middle sized target. As S-R interval increases beyond 125 ms, the responses to the large and small targets diverge from those to the middle-sized target and their amplitudes gradually approach the correct values. In parallel with the progressively more accurate

scaling of the responses to the three targets, Fig. 4A and B also shows a progressive reduction in the dispersion of response amplitude. The results shown in Fig. 4A, B support the hypothesis that the accurate matching of peak force to a visual target shift, evident at S-R intervals greater than 250 ms, results from a gradual and extended process of specification that proceeds from a common default. Alternative explanations for our data, however, had to be 'excluded. First, responses made at short S-R intervals might have resulted from random guesses among the three targets rather than a true common default with a somewhat more variable amplitude. Such guesses might be more often correct when responses were initiated at longer S-R intervals after the target shift. Second, the specification might appear as a gradual process if an abrupt shift from an unspecified to a fully specified state occurred at different S-R intervals on different trials. T h e s e possibilities can, however, be excluded by examining the distributions of response amplitudes in successive S-R intervals. Figure 4C shows the amplitude distributions of the responses to the different targets for three intervals corresponding roughly to the periods before ( < 125 ms), during (125-250 ms) and following ( > 250 ms) specification. In the earliest time slice, the peak forces of the responses to each target are distributed around modal values which are similar for each target. In each case, there is a single mode to the distribution. Accordingly, the distribution of all responses produced in the earliest interval (dashed histogram in left middle row) is unimodal and is similar in shape to the distribution of the responses to the middle-sized target (solid histogram). None of the distributions are trimodal as would be expected if the subjects were making random guesses among three alternative amplitudes. In the later intervals, the distributions of responses to the large and small targets are skewed and their modes shift progressively toward the appropriate targets (dotted horizontal lines). We do not observe the bimodal distribution which would be predicted in the middle interval for a mix of unspecified and specified responses. Similar findings were obtained in all subjects and confirm the conclusion that response amplitude is specified gradually and proceeds from a default value which is independent of the target presented on that trial. Subjects might also have achieved increasing accuracy by selectively modifying responses that all had the same initial trajectory. For example, response amplitude could be adjusted by feedback mechanisms through control of force rise time. In this case, the subject would be utilizing a pulse width control system (Bahill et al. 1975), rather than the

121

o

C.

A

125 ~

'

~

b

c

7..........-.....

..................... ]

I ~l ..............................

--I 1 ....................... T-I

.............

"

Z5

1:5 I

A

z

]

9 o

., 9

J

..

oo

9

i

-7

i

z@ ~

2S~,

o,

LI

o

75 u_

.................

0~. : % : - _ 4 . .

.'o

. o l l, i r a . o, : 9. . :2 9

o 9 9

o

..~

9

o

F

o_

o

.

U

I

/

o o

o

oo

L__ 75 ~ L ; :

k

o

25

~L

L

I

I

,.~ ...............................................

+i .................................

j

o,+,,A-.~._._

I 160

360 460 I N T E R V A L (ms)

S-R

B

0

260

o

b

c

2 0 (60)

I

I

1 ...........:.:..~i!i !.

0

i

D

I

125

i

~

f 20

0

i

20

i

a

]

I s

b

c

125N ..................................

''4." ..............

...............

"

"'i

/ (9 0

t.

75

~

I=J

25

a60

e60

a6o

46o

!

o

10o

200 ms

0

I

100

200 ms

o

10o

200 ms

S-R INTERVAL (ms)

Fig. 4A-D. Response specification in choice condition. A Individual peak forces are plotted against S-R interval for responses to the small (25 N, below), middle (75 N, center), and large (125 N, above) targets. Lines fitted by LOWESS are shown for each target size (solid lines). B LOWESS lines for the three target sizes are combined with flanking lines indicating dispersion. For Parts C and D: short (a, < 125 ms), intermediate (b, between 125 and 250 ms) and long S-R (c, > 250 ms) intervals. C Distribution of peak forces within time intervals. Histogram bin width is 12.5 Newtons. Scales for the number of trials in each bin are shown below each interval. In interval 'a' for the middle target size, the distribution of responses to all targets in the same time interval (dashed line) is superimposed and its scale is given in parenthesis below. D Average force trajectories of responses to each target within each time interval. Averages synchronized on force onset. Overall number of trials by target size: small, 364; medium, 369; large, 364

A

C 90

9

9 9

;'5-

9

9

9

b

be

9

_0

o ~

.................

_~I, 9

45-

......i

,~...." : ~ . .

30-

9

o9 9

iiiiii ii

i

....... i ........ '

i

Ooo

..5...........

i

75

6o

~"2.1"~...."9

9

45

o

e 09

._. ~ F ~ , .

30

~

:

~

9

9

9

15-

:: :-~ o9

.

~ e~176 ..~ 9 ." ~

9. ~a, . ~ . 2 , ; ,

99 v

[:

i

o~ o 9 o9 9

15

u.

.........

9 "..'"~K,P~.. ,#,."

60-

"0

c

9

e~

9

o

.

~,,

9

.~o

.

.... i i, ........ ......

~

9

W 0.

75 1 60

i_

9 9

]

"e. 9

/

I.

,o,, ?

",hJ~,9 [---

l

L___

I

I ' . . ~ ' ~ ~

". 9

J

9

r- ....... "7J :-~ . . . . . . . . . . . . . . . . . . . . . . .

15 i" "9 9 qT, ..',~,.~P,t~- 9 16o S-R

B

2%0

360

0i

460

lJ

0I.....20

20J (60)

7

7I -? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

I

t

1

20!

!

INTERVAL (ms)

a

b

c

9o I

,

,

I ?5 4

I I

I I.......

o

D

b

c

/".

F- I

'.,2 i

i00

S-R

200

300

INTERVAL ( m s )

400

O

i

ZOO

200 ms

0

100

a00 ms

, o

V zoo

coo ms

Fig. 5A-D. Specification of peak dZF/dt2 in the choice condition. Figure arranged as Fig. 4 and derived from the same data. A Individual peak dZF/dt2 of responses are plotted against S-R interval for responses to the small (25 N, below), middle (75 N, center), and large (125 N, above) targets. Lines fitted by LOWESS are shown for each target size (solid lines). B LOWESS lines for the three target sizes combined with flanking lines indicating dispersion. C, D As in Fig. 4, trials have been divided into short, intermediate, and long time intervals. C Histograms of peak d2F/dt2 within the respective intervals of responses aimed at each target. Histogram bins are 7.5 kN/s 2. Scales for the number of trials in each bin are shown below each interval. In interval 'a' for the middle target size, the distribution of all trials in the same time interval (dashed line) is superimposed and its scale is given in parenthesis below. D Trajectories of d2F/dt~ for responses within each interval have been averaged for the three target sizes

123

pulse height control seen here in simple responses. This was not the case, however. The correct specification of response amplitude reflected a progressive improvement in specification of the initial trajectory parameters themselves. Figure 5 shows the changes in d2F/dt 2 of the same response illustrated in Fig. 4. The peak values of dZF/dt2 corresponding to target forces of different amplitudes show the same trends as the peak forces seen in Fig. 4 and are also gradually specified (Fig. 5A-C). Figures 4D and 5D show ensemble averages of the responses produced in the S-R intervals depicted in parts B and C. Both the force and d2F/dt 2 trajectories can be observed to have the same configuration before, during and following specification (parts a, b and c in Figs. 4D and 5D). Force rise time remains essentially independent of amplitude at all S-R intervals while response amplitude is adjusted by variation in the rate of change of force, as described in our earlier reports (Gordon and Ghez 1987a).

A

100

'0 80 O)

E

g

60

o o


g

0

l

|

I

I

100

200

300

400

B lOO

"13 80 ID E 60 o o < 40

co

Time course of specification and response updating

9~

20

g

0

o~ >

In order to ascertain more quantitatively the time course of amplitude specification and to compare different subjects, we grouped responses into discrete latency intervals. We then performed linear regressions within each interval between the target amplitudes of individual trials and the peaks of force and of its second time derivative, d2F/dt 2. We calculated the regression coefficients (r) and their squares (r2). The squared correlation coefficients indicate the proportion of variance in the dependent variables (peak force and peak d2F/dt 2) accounted for by the target amplitude. In accurate responses, a large proportion of the variance in both dependent variables is accounted for by the target (see also Gordon and Ghez 1987a, b). We have, therefore, used the percentage of variance (r 2 x 100) in peak force and peak d2F/dt 2 that could be accounted for by target amplitude in each interval as a measure of the degree of specification achieved. Values calculated for the simple and choice responses illustrated in Figs. 3 through 5 are plotted against the midpoint of each interval in Fig. 6A, B. In the simple condition (Fig. 6A), target amplitude accounts for 90 to 95% of the variance in peak force (filled squares) and about 80% of the variance in peak d2F/dt 2 (open squares) and varies minimally with S-R interval. This degree of specification is similar to that which we have previously reported for responses made without time constraints (Gordon and Ghez 1987a). In the choice condition (Fig. 6B), at short S-R intervals (less than 100 ms), the target accounts for little or none of the

100

C

13

i

i

!

200

300

40O

40

30

E D o o o


g

20

10

0

'

0

i

i

1

100

200

300

S-R Interval

400

(ms)

Fig. 6A-C. Specification of amplitude and compensatory adjustments to trajectory. Individual trials from Subject $2 (see Figs. 2 through 5) have been separated into intervals of 80 ms in A, 20 ms in B and C. A, B The percentage of variance in the peaks (r 2) accounted for by target amplitude are plotted against the midpoint of each interval in the simple, A, and choice, B, conditions (see text for explanation). C The difference between the percent of variance in peak force accounted for by the target and the corresponding value for peak d2F/dt2 (open triangles) and the percent of additional variance accounted for when target amplitude is added as a second independent variable to a multiple regression of peak force on peak d2F/dt2 (filled triangles) are plotted against the midpoint of the intervals. Total number of trials: simple, 406; choice, 1097. Within the individual intervals, there are 34 to 151 trials in the simple condition and 54 to 158 trials for the intervals from 80 to 300 ms in the choice condition

124

variance in peak force (filled squares). It first accounts for a significant proportion of variance at S-R intervals between 100 and 120 ms. This percentage increases progressively until it reaches asymptotic values after 260 ms. Thereafter the percent of the variance in peak force accounted for by the target is similar to that of the simple responses. The time course of specification of the peak d2F/dt 2 (open squares) parallels that of the peak force, with a lag of roughly 20 ms. Therefore, target information influences the peak force of responses which are initiated before there is any specification of the early portion of trajectory, as measured by the peak d2F/dt 2. This finding supports the conclusion of our earlier study (Gordon and Ghez 1987b) that trajectories can be modified in the time interval between peak d2F/dt 2 and peak force. To examine further whether there are modifications to trajectory in the interval between peak d2F/dt 2 and peak force, we have plotted (Fig. 6C, open triangles) the difference between the percent of variance in peak force and peak d2F/dt2 accounted for by the target (i.e. the difference between the two sets of values shown in Fig. 6B). The difference is maximal at intervals of 130 to 170 ms. In an earlier study of this series, we used a multiple regression analysis to determine whether there were compensatory adjustments to trajectory (Gordon and Ghez 1987b). We have repeated that analysis here within each of the intervals. Peak force was the dependent variable and peak d2F/dt2 and target amplitude were entered sequentially as independent variables. The percent of additional variance accounted for when target amplitude is added as a second independent variable to the regression is plotted against the midpoint of the time intervals (Fig. 6C, filled triangles). This percentage is highest at S-R intervals between 130 and 190 ms. Thus, both measures of the contribution of compensatory adjustments to the trajectories reach maximum values in the same S-R intervals. Similar findings were obtained in other subjects and are shown in Fig. 7. Individual trials from all 6 subjects have been separated into 80 ms intervals. The amount of variance in peak force (Part A) and peak d2F/dt 2 (Part B) accounted for by the target amplitude is again plotted against the midpoint of each interval. All subjects show the same general tendency for increasing specification with S-R interval as shown for subject $2 in Fig. 6. However, individual subjects differ both in the amount of time required for accurate specification (compare $2 with $5) as well as in the degree of accuracy of fully specified responses (i.e. the amount of variance accounted for at long S-R intervals). The differences

A lOO"- 80 "10 r

o-a 60 O

g

40

20 0 0

i

i

i

100

200

300

400

100

200

300

400

,

,

,

100

200

300

B lOO-

-o 8 0 -

== oo 60 oQ 4o "~ 20 0--

c4o] 0

t elO-I

o ~ . 1 0l[

i

0

400

S-R Interval (ms)

Fig. 7A-C. Specification of amplitude and compensatory adjustments to trajectory. A, B The responses from all six subjects have been grouped into 80 ms intervals and plotted as in Fig. 6A, B. The percents of variance (r 2) in peak force (A) and in peak d2F/dt2 (B) accounted for by the target force have been plotted against the midpoints of the intervals. C Difference between the percent of variance in peak force accounted for by the target and the corresponding value for peak d2F/dt2 is plotted against the midpoints of the intervals. Total number of trials: $2: [3, 1097; $3: D, 418; $5: !1,, 156; S12: [], 343; S13: ~ , 217; S14: B, 340

between the variance in peak force accounted for by the target and the corresponding value for peak d2F/ dt 2 are plotted in Part C. All subjects show the greatest difference during the time when specification is increasing rather than before or after. This

125 100

Discussion

-o 80

E o 60 < 9 40 r 20 0

!

!

100

200

300

400

S - R I n t e r v a l (ms)

Fig. 8. Influence of prior trial target upon peak force in the choice condition. Trials from three subjects grouped into 80 ms intervals. Percent of variance in peak force accounted for by the target presented on prior trial is plotted against the midpoints of intervals. $5, I1~; $12, []; $14, []

result indicates that one mechanism used by subjects to improve accuracy in the face of an insufficient time to process target information is the compensatory modification of the later phases of responses.

Influence of previous target on default response In parallel with the progressive improvement in response scaling with longer S-R intervals, all subjects showed a progressive reduction in the variability of response amplitude. This suggests that, in the absence of information about the current target, the amplitude of the default response is specified with less precision than that of responses matched to known targets. In three of the six subjects, however, this increase in variability of the default in part reflected an influence of the target that had been presented on the preceding trial. This effect is shown in Fig. 8 where individual trials from the three subjects have been grouped into S-R intervals of 80 ms. The percent of variance in peak force accounted for by the target amplitude of the prior trial is plotted in the center of each of these ranges of S-R intervals. In all 3 subjects, the dependence of peak force on the target presented on the previous trial is significant at short S-R intervals where it explains up to 50% of the variance in peak force. At these intervals (see Fig. 7), there was no significant relationship between the peak force and the amplitude of the current trial target. This dependence of peak force on the previous target decreased in the subsequent time interval, as the dependence on the current target became significant.

In this study, we have examined how subjects specify the amplitude of a simple motor response using information derived from a visual target. In order to assess the time course of specification, it was necessary for response initiation to be independent of decisions based upon the subjects' perception of the target. To achieve this, we provided subjects with a predictable auditory cue with which to synchronize their responses. By this stratagem, it was possible to transform the stimulus-response interval, normally a dependent variable in reaction time experiments, into an independent variable under experimental control. This has allowed us to examine the degree of specification at time intervals that were both shorter and longer than the reaction times reported in the preceding paper. Our timed response paradigm has thus permitted us to assess how the subject prepares to produce responses to targets of either predictable or unpredictable amplitudes and to determine the rate at which response specification occurs. Schouten and Bekker (1967), using a somewhat similar paradigm to examine the speed-accuracy relationship, found that error rates rose rapidly when responses were made at latencies briefer than the subjects' reaction times. However, those authors did not examine trajectories and, therefore, only limited conclusions could be drawn about how apparently random performance was transformed into accurate responding. They also used a somewhat different synchronizing procedure than ours, with much briefer intervals between tones (80 vs. 500 ms), which appears to have been less effective in controlling latency. Our paradigm also resembles forced choice procedures used in sensory psychophysics. The findings reported here have important implications for understanding the processes that lead to the production of accurate trajectories in our task. First, prior to target presentation, subjects prepare a default response whose amplitude is dependent upon the targets previously presented. Second, the specification of response amplitude is an extended process. It is first evident in responses initiated about 100 ms after target presentation. Specification of the responses improves progressively over the ensuing 150 to 300 ms. During this period, the default amplitude is progressively adjusted to conform with the amplitude of the target. Third, the processes responsible for response initiation can be engaged independently of those responsible for the specification of response amplitude. In the absence of information concerning the appropriate target, a response with default parameters is produced. When response onset occurs while specification is underway, but not

126 complete, partially specified responses are produced. Fourth, throughout the course of specification, response amplitude is governed by the same overall pulse height control policy (Ghez et al. 1983; Gordon and Ghez 1987a) that is used to produce responses of variable amplitude in the absence of time constraints. In both tasks, adjustments (Gordon and Ghez 1987b) are made to the trajectory to compensate for initial errors. The trajectories of responses initiated before specification is complete are thus updated by further processing to improve their accuracy.

different subjects typically became manifest in responses between 100 and 150 ms. This interval is shorter than the mean latency for responses to similar stimuli in reaction time paradigms. F o r instance, we found longer mean latencies in the preceding study of isometric finger flexions (Hening et al. 1988) as well as longer latencies in data obtained at the elbow (Favilla et al. 1987). The process of specification was complete when the S-R interval exceeded 250 to 400 ms.

Independence of triggering and specification The default response At very short S-R intervals, the amplitudes of responses to all three targets were distributed around the same modal value, which was close to that of the middle target force. Therefore, subjects must have prepared a single default response, rather than selecting randomly from among responses to the three targets. Moreover, the amplitude of the default can be assumed to reflect the subjects' preceding experience in the task. In the preceding paper we showed that incompletely specified responses were biased toward the center of the target range and depended on the range of target amplitudes and the probability of target occurrence (Hening et al. 1988). These factors are likely to operate by influencing the default amplitude (Favilla et al. 1986). In addition, in three subjects, we have shown a significant dependence of peak force upon the amplitude of the target presented in the preceding trial. The greater dispersion of both default responses and partially specified responses produced early in the course of specification parallels the greater dispersion of reaction time responses produced as soon as possible in the choice condition (Hening et al. 1988; Fig. 10B).

Gradual specification of response amplitude The smooth relationship between peak force and S-R interval (Fig. 4A, B), where no discontinuities can be seen, and the unimodal distributions of the response amplitudes to single targets (Fig. 4C) demonstrate that specification is a gradual and continuous process. During the course of specification the distributions of responses to large and small targets were never bimodal (Figs. 4C, 5C), as would be predicted if subjects shifted abruptly, but at different times from trial to trial, from an unspecified to a fully specified response. The process of specification in the

The constriction of range and increased dispersion of responses produced as soon as possible in the finger is similar to that obtained at the elbow in the present paradigm at comparable time intervals after target presentation (compare Figs. 4C and 5C here with Figs. 4 and 10 in the companion study, Hening et al. 1988). Moreover, data obtained in a reaction time task requiring the same elbow responses to the same stimuli indicate that, at comparable S-R intervals, the degree of specification was similar for the two tasks (Favilla et al. 1987). It follows from these findings that, in the production of a targeted isometric response, the triggering of response initiation and specification of response amplitude can be carried out by independent mechanisms. The minimal amount of time needed to trigger the initiation of a response is shorter than the time needed to fully specify its amplitude. Response initiation need not, as suggested by certain theoretical formulations, occur only after the completion of obligatory sequential processing stages (Donders 1869; Welford 1952, 1967, 1980; Sternberg 1969; Taylor 1976; Sanders 1980; Miller 1982; Gottsdanker and Shragg 1985; Kantowitz 1985). Our results indicate that the degradation of accuracy that occurs in the reaction time task arises because, when responses are initiated as soon as possible, they are only partially specified. Although response updating does occur, it is not sufficient to fully correct initial errors. Our observations also provide evidence of interactions between the processes leading to response initiation and those subserving amplitude specification. We found that, in the choice condition, there was a systematic dependence of the timing error on the time between target stimulus and the tone. Thus, the need to specify response amplitude interferes with the subjects' ability to correctly synchronize their responses with the tone. Schouten and Bekker (1967) reported a similar finding using a paradigm which also used tones to control the initiation of responses to two unpredictable stimuli.

127

Trajectory control during response specification Throughout the course of specification, response trajectories remained governed by a pulse height control policy which we have found to determine responses made both with (Hening et al., 1988) and without reaction time constraints (Gordon and Ghez 1987a). Thus, the amplitudes of the default, partially specified, and fully specified force responses were largely predicted by the rate of change of force (d2F/dt2) while the rise times of force were largely independent of amplitude. Our subjects did not vary response amplitude by producing a common initial trajectory and then modulating rise time in order for the peak force to match the different targets (Cordo 1987). In these and our earlier studies (Ghez and Gordon 1987; Gordon and Ghez 1987a, b), we have attempted to preclude the use of such a strategy by having our subjects produce impulses rather than step changes in force and by insisting that they not make voluntary corrections to their trajectories. In addition, our subjects were provided with extensive training during which they learned to scale their responses to the different targets. Less practiced subjects and those making step changes of force might choose to use a strategy involving primarily corrective modifications of force rise time which would be carried out after response onset. As we have previously demonstrated, in targeted force impulses produced without temporal constraints (Gordon and Ghez 1987b), compensatory adjustments were made to the trajectories of responses at all S-R intervals. These compensatory adjustments for initial errors occur despite the brief nature of the responses we have studied (< 110 ms on average) and the smooth monotonic contours of the force trajectories. We have suggested that such corrective adjustments are analogous to ones documented in other contexts (e.g. Higgins and Angel 1970) and are probably dependent upon internal feedback mechanisms. It is significant, however, that in the present experiments such adjustments were maximal during the time interval when the degree of specification was changing most rapidly. In this circumstance, these adjustments are likely to reflect the subject's further processing of target information resulting in updating of the trajectory. It is interesting to note that, as in our previously reported data (Gordon and Ghez 1987b), changes in accuracy and differences in amount of corrective adjustments occurred in the course of specification without changes in response duration (i.e. in force rise time) which could be expected according to Fitts' law (Fitts 1954; Keele and Posner 1968). The present

results are rather in agreement with the kind of speed-accuracy tradeoff in which reaction time is included in the calculation of response duration (Pew 1969; Swensson 1972; Pachella 1974; Wickelgren 1977). The ability of subjects to produce incompletely specified responses indicates the need for caution in using reaction time as a measure of the duration of completed sensorimotor processing under different conditions (Miller 1982). Our data indicate that a detailed analysis of response trajectories is also necessary in order to dissect the component processes responsible for the achievement of accuracy. In summary, our results indicate that accurate responding to unpredictable targets depends on at least three classes of mechanisms. First, in advance of responding, subjects develop a task specific transform which enables them to accurately match a range of targets (Ghez 1979; Gordon and Ghez 1987a). This transform is based on a pulse-height control policy whereby a stereotyped response trajectory is scaled to the desired peak force (Gordon and Ghez 1987a). Second, by mechanisms which we have collectively termed response specification, the subject uses sensory information to convert a default response into one that is accurately matched to the specific target. Specification results in a progressive change of the amplitude of responses, while duration remains relatively fixed. Third, during the course of specification, compensatory adjustments allow for the use of late-arriving information to update the terminal portions of the trajectory (Gordon and Ghez 1987b).

Acknowledgements. We are indebted to Dr. James Gordon for valuable discussionsin the course of this study and for reviewing the manuscript. We are grateful to Robert Woolleyfor assistance in preparation of the figures. Supported by NIH grant NS 22715, W.H. supported in part by the Veteran's AdministrationMedical Research Service. References

Bahill AT, Clark MR, Stark L (1975) The main sequence: a tool for studyingeye movements. Math Biosci 24:191-204 Chambers JM, Cleveland WS, Kleiner B, Tukey PA (1983) Graphical methods for data analysis. DuxburyPress, Boston Cleveland WS (1979) Robust locally weighted regression and smoothing scatterplots. J Am Stat Assoc 74:829-836 Cordo PF (1987) Mechanisms controlling accurate changes in elbow torque in humans. J Neurosci 7:432-442 Donders FC (1869) On the speed of mental processes. Translated and reprinted in Acta Psychol30:412-431 Favilla M, Hening W, Ghez C (1985) Human tracking performance: parallel specification of amplitude and direction. Neuroscience Abstr 11:72 Favilla M, Hening W, Ghez C (1987) Independent specificationof response features in a reaction time paradigm. Neuroscience Abstr 13:13

128 FaviUa M, Hening W, Gordon J, Ghez C (1986) Preparatory set: independent specification of direction and size in the preparation of responses to a range of targets. Neuroscience Abstr 12: 972 Fitts PM (!954)The information capacity of the human motor system in controlling the amplitude of movement. J Exp Psychol 47:381-391 Ghez C (1979) Contributions of central programs to rapid limb movement in the cat. In: Asanuma H, Wilson V (eds) Integration in the nervous system. Igaku-Shoin, Tokyo, pp 305-319 Ghez C, Gordon J (1987) Trajectory control in targeted force impulses. I. Roles of opposing muscles. Exp Brain Res 67: 225-240 Ghez C, Hening W, Favilla M (1988) Parallel specification of response amplitude in human tracking performance. Behav Brain Res (in press) Ghez C, Vicario D, Martin JH, Yumiya H (1983) Sensorimotor processing of targeted movements in motor cortex. In: Desmedt JE (ed) Motor control mechanisms in health and disease. Raven Press, New York, pp 61-92 Gordon J, Ghez C (1984) EMG patterns in antagonist muscles during isometric contraction in man: relations to response dynamics. Exp Brain Res 55:167-171 Gordon J, Gtiez C (1985) Mechanisms contributing to accuracy in aimed force impulses: rise time regulation and corrective adjustments. Soc Neurosci Abstr 11:75 Gordon J, Ghez C (1987a) Trajectory control in targeted force impulses. II. Pulse height control. Exp Brain Res 67:241-251 Gordon J, Ghez C (1987b) Trajectory control in targeted force impulses. III. Compensatory adjustments for initial errors. Exp Brain Res 67:252-269 Gottsdanker R, Shragg GP (1985) Verification of Donders' subtraction method. J Exp Psychol (Hum Percept Perform) 11:765-776 Hening W, Ghez C (1984) Processes underlying the initiation and specification of responses in a human tracking task. Neuroscience Abstr 10:801 Hening W, Vicario D, Ghez C (1988) Trajectory control in targeted force impulses. IV. Influences of choice, prior experience and urgency. Exp Brain Res 71:103-115 Higgins JR, Angel RW (1970) Correction of tracking errors without sensory feedback. J Exp Psychol 84:412-416

Kantowitz BH (1985) Channels and stages in human information processing: a limited analysis of theory and methodology. J Math Psychol 29:135-174 Keele SW, Posner MI (1968) Processing of visual feedback in rapid movements. J Exp Psychol 77:155-158 Miller J (1982) Discrete versus continuous stage models of human information processing: in search of partial output. J Exp Psychol (Hum Percept Perform) 8:273-296 Pachella RG (1974) The interpretation of reaction time in information processing research. In: Kantowitz B (ed) Human information processing. Erlbaum, Hillside, New Jersey, pp 41-82 Pew RW (1969) The speed-accuracy operating characteristic. Acta Psychol 30:16-26 Poulton EC (1975) Range effects in experiments on people. Am J Psychol 88:3-32 Poulton EC (1981) Human manual control. In: Brooks VB (ed) Handbook of physiology, Sec 1. The nervous system, Vol 2. Motor control. American Physiological Society, Bethesda, pp 1337-1389 Sanders AF (1980) Stage analysis of reaction processes. In: Stelmach GE, Requin J (eds) Tutorials in motor behavior. Amsterdam, North Holland, pp 331-354 Schouten JF, Bekker JAM (1967) Reaction time and accuracy. Acta Psychol 27:143-153 Sternberg S (1969) The discovery of processing stages. Extensions of Donders' method. Acta Psychol 30:276-315 Swensson RG (1972) The elusive tradeoff: speed versus accuracy in visual discrimination tasks. Percept Psychophys 12:16-32 Taylor DA (1976) Stage analysis of reaction time. Psychol Bull 83: 161-191 Welford AT (1952) The "psychological refractory period" and the timing of high-speed performance - a review and a theory. Br J Psychol 43:2-19 Welford AT (1967) Single-channel operation in the brain. Acta Psychol 27:5-22 Welford AT (1980) Reaction times. Academic Press, New York Wickelgren WA (1977) Speed-accuracy tradeoff and information processing dynamics. Acta Psychol 41:67-85

Received July 1, 1987 / Accepted December 11, 1987