Activity 6

In this activity, you will learn about how production systems work by experimenting with a program which implements a simple production system model of meter assignment. As with earlier activities, you are free to collaborate with other students in the class, as long as you let us know.

1. For each of the following periodic patterns, create the pattern (option 0 in the program menu) and run the production system for 60 epochs (select option 1 in the program menu 3 times).
• Pattern A: (1 1 0)
• Pattern B: (1 1 0 0)
• Pattern C: (1 1 1 1 1 0 0 0)
• Pattern D: (1 1 1 0 0 1 0 0)
• Pattern E: (1 0 1 0 1 1 1 0)
• Pattern F: (1 1 1 0 0 1 0)
• Pattern G: a pattern of your own choice with period no greater than 8
Because there is some randomness in the way rules are selected, you may want to run a pattern twice. Then answer the following questions:

• What sort of meter would people assign to the pattern?

That is, where would the perceived beats of different strength underlying the pattern be? For example, for the repeating pattern (1 1 1 0), people would place accents on the first and third beats (the beginning and end of a run), one stronger than the other:

```2 0 1 0 2 0 1 0 ...
```
For the repeating pattern (1 0 1 0 1 1 1 1), people would place accents on every other beat, accenting every fourth beat more and every eighth beat even more, for example:
```3 0 1 0 2 0 1 0 3 0 1 0 2 0 1 0 ...
```
Note that there is often more than one possible meter assignment for a pattern?

• How difficult for people is this pattern compared to the others?

Recall that patterns are difficult to assign meter to when there is a conflict between grouping and meter, that is, when the places that are accented because of how the events are grouped are not equally spaced. For example, the above patterns are relatively easy, but the pattern (1 1 0 1 1 0 1 0) is difficult because the grouping would lead one to accent things in this way:

```1 0 0 1 0 0 1 0
```
or
```0 1 0 0 1 0 1 0
```
But this leaves the distances between the accents unequal.

• How well does the production system assign meter to the pattern? What mistakes does it make?

To assess the performance of the production system, you will need to look at the working memory after you run the system on the pattern. The last line of the "interpreted" working memory shows the strengths for each of the 24 time steps in the sequence. Remember that the pattern is repeated several times in the sequence. If you can, figure out a way to combine the information from the different repetitions of the pattern.

• How well is the supposed difficulty of the pattern reflected in the performance of the production system? That is, does the model have more trouble with patterns which are more difficult for people?

There are two ways in which we might expect to see greater difficulty for the network. First, it may take the network longer to find a suitable meter for a pattern. Second, it may not settle on one stable meter. Instead it may show different beat assignments over different places within the sequence of 24 time steps, or it may vary from one cycle of the program to the next. To look for evidence of both of these, you will need to compare the performance after 20, 40, 60, and 80 cycles for the different patterns.

2. Suggest two ways in which you think the model could be improved so that it comes closer to human performance. These could involve the rules themselves, the way in which rules are selected on each cycle, or the way in which patterns are represented. Be as specific as possible, explaining why you believe these changes would improve the performance.

Last updated: 4 December 1995
URL: http://www.indiana.edu/~gasser/activity6.html