Page 226 - Foundations of Cognitive Psychology : Core Readings

The Auditory Scene 231

movement seems to occur in depth, in a movement slanting away from the ob-
server, the visual system allows more time for the object to move through the
third dimension than it would have if it had appeared to be moving only in the
8
horizontal plane. This happens despite the fact that although a slanting-away
motion would traverse more three-dimensional space, it produces the same
displacement of an object’s image as a horizontal motion does on the retina of
an observer. Therefore Ko ¨rte’slaw appliestorealdistanceinthe worldand not
to retinal distance, and therefore can best be understood as a sophisticated part
of scene analysis.
Another example of a discovery that was guided by the assumption that the
rules of apparent motion exist to group glimpses of real scenes was made by
Michael Mills and myself. We worked with an animation sequence in which a
9
shape disappeared from one part of a drawing and appeared in another. This
change was seen as motion only if the shape was seen as representing the out-
line of a ‘‘figure’’ both before and after the disappearance. If the observer was
induced to see it as ‘‘ground’’ (the shape of an empty space between forms)
before it disappeared, and as ‘‘figure’’ (the shape of an actual figure) when it
reappeared, the displacement was not seen as motion but as an appearance
from nowhere of the figure.
Neither is the auditory streaming effect simply a laboratory curiosity. It is an
oblique glimpse of a scene-analysis process doing the best it can in a situation
in which the clues to the structure of the scene are very impoverished.
In general, all the Gestalt principles of grouping can be interpreted as rules
for scene analysis. We can see this, for example, in the case of the principle of
grouping by similarity. Consider the block-recognition problem shown earlier
in figure 9.2 where the problem was to determine which areas of the drawing
represented parts of the same block. Because this drawing is not very repre-
sentative of the problem of scene analysis as we face it in everyday life, let us
imagine it transformed into a real scene. In the natural world visible surfaces
have brightness, color, and texture. It would be a good rule of thumb to prefer
to group surfaces that were similar in appearance to one another on these
dimensions. This would not always work, but if this principle were given a
vote, along with a set of other rules of thumb, it is clear that it would contribute
in a positive way to getting the right answer.
In the case of sound, the considerations are the same. If in a mixture of
sounds we are able to detect moments of sound that strongly resemble one
another, they should be grouped together as probably coming from the same
happening. Furthermore, the closer in time two sounds that resemble each
other occur, the more likely it is that they have originated with the same event.
Both of these statements follow from the idea that events in the world tend to
have some persistence. They do not change instantly or haphazardly. It seems
likely that the auditory system, evolving as it has in such a world, has devel-
oped principles for ‘‘betting’’ on which parts of a sequence of sensory inputs
have arisen from the same source. Such betting principles could take advantage
of properties of sounds that had a reasonably high probability of indicating
that the sounds had a common origin. Viewed from this perspective, the
Gestalt principles are seen to be principles of scene analysis that will generally
contribute to a correct decomposition of the mixture of effects that reaches our