Q270: Experiments and Models in Cognitive Science

Apparent Motion Laboratory

When assessing your own subjective experience of apparent motion, not only are the correspondences between elements across frames relevant, but the strength/goodness of motion and its resistance to change should also be considered. In answering the following questions, try exposing yourself (and friends) to the same display a number of times and base your answers on the most frequent interpretation given.

1. By arranging two objects in two frames like

one can see whether A corresponds to C (and B to D) or whether A corresponds to D (and B to C). Using this procedure, you can explore which of two features is more influential in determining apparent motion.

A. Draw two frames that allow you to test whether color or shape is more important for determining motion. Why did you choose the colors and shapes that you did? Describe (draw and label) your displays. Describe your results. Given your results, which dimension is more important?

B. Do the same, pitting color against size.

C. Do the same, pitting size against shape.

D. There is neurophysiological evidence for two separate visual pathways. The Magnocellular pathway is particularly sensitive to motion, sharp edges (altered by size and shape changes), and timing. The parvocellular pathway is particularly sensitive to color (hue), form, and spatial relations. Do your results from A support this distinction? Why or why not?

2. According to the one-to-one mapping constraint, we have a bias to see every object in one frame correspond to one and only one object in the second frame.

A. How does your perception of these frames support the one-to-one mapping constraint?

However, a consideration of the frames

leads one to suspect that there may be more going on than simply a one-to-one mapping constraint in A.

B. Describe what you (and your friends) see in the second situation, and why this suggests a possible alternative account (other than simply a preference for one-to-one-mappings) for your perception of the first situation in Question A. That is, why might people see the motions that they do in Question A even if their perceptual systems did not have a one-to-one mapping constraint? Be as specific about this alternative as possible.

C. Design a display that tests whether there is still a one-to-one mapping constraint without the possible confound described in B, and show a drawing of your frames. Do you still find evidence for a one-to-one mapping constraint? If so, what is your uncontaminated evidence?

3. Consider the Ternus effect:

There are two major interpretation that people give for this display, and you can see them both by changing "Blank Time" from 0 to a somewhat larger number of milliseconds. Describe the two interpretations that you see by describing which dots correspond to each other for each interpretation. Explain why "Blank Time" has the influence that it does.

4. The computational model can also produce either of these interpretations. Report two parameter settings that produce the same two interpretations that you found, and that only differ on one parameter value. Given this simulation outcome, what does the model say is one of the effects that increasing the blank interval between frames has? Does this make sense? Why or why not?

5. Fully describe (include a drawing, model parameters, and timing information) a situation where the simulation makes the wrong prediction (given your own perception) when "Match" is set to 0, but makes the correct prediction if Match > 0. Ideally, setting Match > 0 should be the only way of successfully modeling the situation. What does this suggest about when and why the Match constraint is important? In particular, why is the one-to-one mapping constraint not sufficient?

6. Even though I programmed the model (adapted from models by Dawson and Ullman), I can see that it is far from perfect. Describe two situations where the computer model and the human perceiver do not agree. For each situation, describe the discrepancy between the human and computer. You should make reasonably sure that no parameter setting would allow the model to see what the human sees. In one kind of situation, the human phenomenon is "outside the boundary conditions of the model." This means that the model was not intended to handle this sort of case. An example of this is rotation in 3-D. If people are shown the two frames:

they will often report seeing an entire complex object rotate in depth. The model has no way of seeing such motions and was not intended as a model of 3-D motion, and so there's really no way that the model could possibly get the right answer. A worse problem for a model is when it could potentially account for a phenomenon because the phenomenon deals with the same type of motion that the model deals with, and yet the model still gets the wrong answer. Try to describe an example of this second kind of situation, and also describe, as specifically as possible, how you would modify the model to show human-like behavior.