news and views

Visual perception

was found to have systematic inaccuracies.

Where are the things we see?

sparsely furnished visual world: there was

Cai et a/.2 exposed their subjects to a

nothing visible within 200 milliseconds of

Joel M. Miller and Christopher Bockisch

the subjects' saccades to a target, except for a

vernier probe composed of a vertical pair of

ur visual perception of the direction

dots (which were visible from the start of the

of objects seems to be based on sever

trial), and a briefly flashed dot (which was

al inaccurate visuomotor processes,

vertically

ment of the eye to the target, at various hori

issue. Even a simple sideways glance to grasp

zontal locations in the neighbourhood of the

a coffee mug involves visuomotor complexi

saccade target, and subjects judged whether

ties that we have only begun to unravel. Sup

the flashed dot was collinear with the flank

pose you flick your eyes to the mug: that is,

ing dots. An accurate collinearity judgement

you prepare (during a 200-millisecond

might have been made from retinal-image

latency period) and execute (another 50 mil

information alone, during the brief period

liseconds) a 'saccadic' eye movement. You

when flashed and flanking dots were simul

note the location of the mug and the orienta

taneously shown. Strikingly, in the 100-mil

tion of its handle, and look back, as you begin

lisecond period before a saccade, flashes at all

reaching for the mug-the whole process

horizontal locations were mislocalized in the

takes, perhaps, half a second. If you don't

direction of the saccade, by about one-quar

find yourself grabbing at air or mopping up

spilt coffee then, supposing you moved your

centred between the flanking

dots). The flashed dot preceded the move

which are the subject of two studies by Ross et

a/.1 and Cai etal.2on pages598 and601 of this

ter the magnitude of the saccade. So visual

Figure I Detail of 'Portraits in an Office (New

direction is influenced by an extraretinal sig

nal, even when more accurate, purely retinal,

arm without continuous visual guidance,

Orleans)' by Edgar Degas, with a hypothetical

you have accurately assigned the informa

scanpath (red) showing six fixations joined by

information is available. But there is a hint in

tion from your visual 'snapshot' of the mug

saccades. a-f indicate the predicted retinal

the study of Cai et al. that the retinal infor

to its proper relative location in space.

content of these fixations. Based on information

mation is also involved, because other stud

within the images, it seems possible that the

ies8 involving only extraretinal information

if our eyes did not move. An image falling on

the specialized central fovea of the retina

visual system could properly relate the

find even larger rnislocalizations.

newspaper in b with its reader in c, and perhaps

Ross et al. 1 asked subjects to make sac

would then indicate an object lying straight

the cotton sample in d with the man to whom it

cades from a fixation point to a target spot in

ahead. An image falling to the left of fovea

is being presented in e. But information about

a dimly lit environment, and to judge the

would correspond to an object to the right,

eye position is probably required to ascertain

horizontal position of a single vernier-probe

and so on. But because only the fovea sup

that the man in f is watching the presentation,

flash relative to a ruler that was shown to

ports

and that the man in a is self-absorbed.

them afterwards. In a second experiment,

with saccades (brief, rapid movements),

cessing times imply that post-saccadic snap

flashed early in the saccadic latency period

which quickly bring successive images to the

fovea. However, a given retinal locus must

shots are not available for more than 50 mil

(so as to be unaffected by relocalization

liseconds. In contrast, information about

processes), and the other was flashed just

then signify different spatial directions,

self-produced movement could be available

before the saccade (when large localization

depending on the position of the eye. The

for perception of direction even before a

errors are typically found). The authors

Locating visual targets would be simpler

high-acuity

trichromatic

v1s1on,

foveate animals constantly scan visual scenes

one component of a vernier probe was

suddenness of saccades (which reach 400

planned movement is completed. And it

found that visual objects presented in the

per second in about 20 milliseconds) and

would be surprising if saccadic eye move

100-millisecond period around the begin

their frequency (several each second) proba

ments were not accompanied by informa

ning of a saccade are mislocalized by as much

bly imposes an extreme computational bur

tion about 'camera' movement because,

as half the magnitude of the saccade.

den, which can be used to study localization

here, we are the cameramen.

mechanisms.

Ross et al. 1 and Cai et al. 2 have consolidat

Since at least the mid-nineteenth century,

ed previous findings that probes around the

We easily understand the spatial arrange

ments represented in cinematic images.

scientists have thought that the brain uses

fixation spot, and short of the saccade target,

extraretinal

are mislocalized in the direction of the sac

Although we have no direct information ;1bout

about eye position (Fig. 2): knowing how the

cade. Ross et al. also show that probes which

camera movement, patterns of visual flow can

eyes have moved, the brain adds the angle of

are presented beyond the target can be mislo

provide indirect information about continu

eye-rotation to the position on the retina of

calized in the opposite direction to the sac

ous or overlapping camera movements, so that

an object's image, to compute the spatial

cade. So, around the time of a saccade, the

we can understand the visual worlds they rep

direction of the object4'5: Several studies,

retinal direction-map not only translates

resene. Even 'jump cuts' between non-over

placing retinal and extraretinal cues in con

as it must to compensate for the movement

lapping camera positions, which provide the

viewer with no information about movement,'

flict, have shown that rich visual information

of the eye-but also compresses in the

determines perceived changes in location,

vicinity of visual targets. That is, the retinal

do not seem unnatural.

despite the contradictory extraretinal infor

mechanisms, which are usually thoughtto be

(non-retinal)

information

It is possible that our visual world is

mation6. But other experiments have shown

highly accurate, introduce a distortion here.

assembled, at least in part, like a jigsaw puz

the reverse7 Now, Cai et az.l describe a clear

In a third experiment, Ross et al. confirmed

zle, on the basis of visual cues within frag

case in which extraretinal information is

the compression effect by presenting hori

mentary images. But where visual snapshots

dominant, and Ross et al. 1 have mapped the

zontally spaced compound probes around

do not overlap, there is no intrinsic basis for

time course of saccadic relocalization, and

the saccade target, and asking the subjects to

assembling them-that is, jigsaw-puzzle

mechanisms do not work for large saccades

show that there are combined effects of reti

judge them for separateness .

nal and extraretinal information. In both

Although Cai et al. describe clear evidence

(Fig. 1). Even for small saccades, visual-pro-

studies, the measured saccadic relocalization

for extraretinal influence on perceived direc-

550

NATUREIVOL386IJOAPRIL 1997

news and views

flashed

pressed in normal vision9'10 Future studies in

beyond the saccade target, in the direction of

tion

(mislocalization

of

probes

specially contrived visual situations could

the saccades), this is the opposite of the direc

determine how several information sources

tion found by Ross et al. The studies straddle

interact in normal vision. This should allow

an interesting territory in which the visual

improved design of, and performance in,

system responds to both extraretinal and

artificial, high-information-flow environ

purely visual determinants of direction, and

ments, such as aircraft, in which, for exam

subtleties in the task determine the weight

ple, suppressive mechanisms are undesir

that is given to each. So, the study by Cai et al.

able.

shows (and that of Ross et al. is consistent

joel M. Miller and Christopher Bockisch are at the

with) imperfect compensation for eye move

Smith-Kettlewell Eye Research Institute,

ment, based on extraretinal information. The

Webster Street, San Francisco, California 94115,

results from Ross et al. further indicate that

USA.

1. Ross, J., Morrone, M. C. & Burr, D. C. Nature 386, 598-601

(1997).

2. Cai, R. H., Pouget, A. Schlag-Rey, M. & Schlag, ). Nature 386,

'601-604(1997).

3. Gibson, J. J. TI1e Ecological Approach to Vis11al Ptrception 1-332

(Houghton Mifflin, Boston, 1979).

4. von Holst, E. Br. ]. Anim. Behav. 2, 89-94 (1954 ).

5. von Helmholtz, H. Treatise on Physiological Opt ics (Dover, 1925).

.
6

Matin, L., Picoult, E., Stevens. J., Edwards, M. & MacArthur, R.

Sci ence216,198-201(1982).

7. Rine, R. M. & Skavenski, A. A. Vision Res. 37, 77>-787(1997).

8. Honda, H. Vision Res. 33,709-716(1993).

9. Bridgeman, B., Hendry, 0. & Stark, L. Visiotl Res. 15,719-722

2232

(1975).

IO.MacKay, D. M. in Handbook ofSemory Physiology 7, Part 3 (ed.

lung, R.) 307-331 (Springer, Berlin, 1973).

visual objects, or the attention that is directed

to them, can exert an attractive bias on the

Metrology

perceived locations of other objects. In both

Stars and ships and superfluids

cases, the perceived directions of objects

depend on complex interplay between many
sources of flawed information.

C. W. F. Everitt

Because vision is so important for effec

tive behaviour, we would expect visuomotor

two

elegant

experiments,

Schwab,

the helium behaves normally and is dragged

Bruckner and Packard (on page585 of this

viscously like water. For the superfluid, at

issue1) and Avenel and Varoquaux2, have

slow rotations- where the total initial

environment. Distortion of biological sig

measured the rotation of the Earth to 0.5%

angular momentum is less than half him-

nals, and the computational limitations of

and 2% respectively by exploiting the macro

the liquid stops rotating altogether, and the

any single mechanism, may be overcome by

scopic quantum properties of superfluid

bucket correspondingly speeds up. Super

using several mechanisms, to improve over

helium. Given the many current applications

fluid circulating through a thin hollow

all performance in a range of situations. The

of gyroscopes, we may ponder the technical

ring behaves similarly, but one can place a

worst failings of such a 'patchwork' system

future of such exotic rotation sensors- and

diaphragm with a minute hole in it (in prac

can be minimized by still other mechanisms.

their relevance to ancient, still unsettling

tice, of diameter -1 j.Lm) across the section of

For example, the visual instability that would

questions about absolute rotation.

mechanisms to exploit all possible sources of

information about distal objects in a lighted

be expected to result from transient mis

localizations around saccades is simply sup-

Perception:

object is 20"
to the right

Perception

object is 20"

the ring to form a 'weak link' in the super

Macroscopic quantization is a peculiar

phenomenon

in

superconductors

and

flow, producing a hydrodynamical Joseph

son effect. That is the key to the new rotation

superfluids, related to Bose-Einstein con

sensors. By modifying the topology of the

densation but best understood by compar

ring to make the orifice part of a resonant

ing the superconducting case with the de

chamber driven by a vibrating membrane,

Broglie picture of the hydrogen atom. In

one can, as Schwab, Bruckner and Packard

hydrogen, stable electron orbits occur when

remark1, determine "the state of absolute

the wavefunction of the lone electron closes

rotation of the containment vessel".

on itself around the orbit. Fritz London con

But what does that innocent phrase

jectured3 in 1950 that an analogous condi

absolute rotation mean? We are drawn back

tion holds in a superconductor, not for single

to Newton's famous rotating bucket experi

electrons but for the collective behaviour of

ment. Centrifugal force makes a rotating liq

the roughly 101 6 electrons (strictly, electron

uid surface parabolic; so what is that rotation

pairs) that constitute a circulating super

relative to? Not the bucket, for the surface

current.

can clearly be flat or parabolic with identical

This phenomenon has several intriguing

relative rotations. Newton concluded that it

consequences. One is that the magnetic flux

must be rotation with respect to 'absolute

Figure 2 Simple scheme for combining retinal

through a superconductor is quantized in

space'. Enter two centuries later Ernst Mach.

and extraretinal information on direction. The

units of hc/2e (h being Planck's constant,

He argued that inertia originates not from

image of an object located 20 to the right is

the velocity of light and e the charge on the

space but from the matter in space. Accord

projected onto the retina, which signals the

electron). Another, recognized in 1962 by

ing to Mach's principle, if we abolish from

location of the image to a summing junction

Brian Josephson\ is that if one introduces

the Universe all matter except the bucket and

(retinal information). A copy of the command

one or more thin insulating layers (or other

water, the water will stay flat - centrifugal

that determines eye position provides

'weak links') into a superconducting loop,

forces vanish.

to the right

extraretinal information to the summing

there occur interference effects, resembling

Einstein made Mach's principle a found

junction. The algebraic sum () appears at the

optical interference, which can be exploited

ing assumption of General Relativity. He

output, and this is the perceived direction of the

to measure very accurately the magnetic field

was, therefore, vastly annoyed when Kurt

object. If retinal information accurately reflects

through the loop.

Godel, having recovered from destroying

the position of the image, and extraretinal

In superfluid helium it is vorticity rather

mathematics, derived from General Rela

information accurately reflects the position of

than magnetic flux that is quantized, the unit

tivity a most un-Machian model universe7

the eye, then perceived direction is accurate,

of quantization being not hc/2e but him,

Godel's universe is not ours, but that misses

whether the eye is looking at the object (b) or

where m is the mass of the helium atom. An

the point, which is that without special extra

elsewhere (a). But, as Ross et a/.1 and Cai et a/.2

old experiment by G. B. Hess and W. M.

assumptions General Relativity does not

explain inertia. One escape is frame-drag

now show, both sources of information are

Fairbank5'6 illustrates it. They cooled a tiny

inaccurate around the time and destination of

rotating bucket of liquid helium through its

ging. According to General Relativity, rotat

saccades.

superfluid transition (2.16 K). Above 2.16 K,

ing massive bodies like the Earth partially

NATUREIVOL386! IOAPRIL 1997

551