obstructed views of trees.
The 40 other slides, the "negative instances," contained no trees. The 80 slides had been picked from a large collection assembled for several category-learning experiments, not necessarily involving trees. Positive instances sometimes contained trees only incidentally or in the distance. An attempt was made to span a range of difficulty in the visibility of trees, and also to make the positive and negative instances comparable except for the presence or absence of trees, at least to a human observer. The pigeons saw the slides projected for about 45 sac on a small screen next to the standard pigeon response-key. Brief food reinforcement come intermittently for packing if the slide on the screen was positive and not at all if it was negative.
Since the pigeons saw the same 80 slides daily, more discrimination may not seem like convincing evidence of an over-arching "tree" category. The pigeons could have been learning 80 distinct patterns, 40 of which happen to be positive and 40 negative. We would like to know if this is whit the pigeon, with its relatively small nervous system, is limited to. An alternative is that the pigeon, like a human observer, sees each slide as representing objects In a three-dimensional space. The patterns of light, shade, and color would then be seen as aspects of a finite collection of objects, and the pigeon's task is to respond if there is a tree among them. The speed of discrimination suggests the latter interpretation. The slowest of the four pigeons in the experiment was discriminating significantly by the fifth session, after having seen the 80 slides only four times. The other three pigeons were discriminating significantly by the second session, having seen the slides only once before. Occasionally, in experiments like this, pigeons seem to got the idea before "first session is over and start discriminating significantly during the first rotation of the carousel slide tray.
To say the pigeon gets an idea is to say it has an over-arching principle of classification. Once a category forms, the pigeon knows, as it were, that certain slides are positive, even if they have never been associated with reinforcement. In this sense, categorization transcends more induction or, for that matter, more conditioning. But, to be useful, a category has to be at least nearly right. Were the pigeons really looking for trees, or is there some other category or perhaps a set of categories that overlaps with the experimenter's notion of trees? This proves to be a hard (perhaps unanswerable) question, but we can chip away at it. We can test to see if the pigeons' principle of classification applies to now instances the way our category of tree does.
After the pigeons were stably discriminating between the original 40 positive and 40 negative instances, they were tested with now slides. Five positives and five negatives were replaced with five now slides containing trees and five now slides without them, interspersed among the 70 remaining original slides. Then, a few sessions later, another ten now slides were substituted for ten of the original set. Would the pigeons correctly classify new positives and negatives on first presentation? All the pigeons generalized, which means they were sorting slides according to criteria that at least approximated trees for a large proportion of the instances examined. In terms of rates of pecking, the now positives were ranked as high as or higher than the training stimuli, as If they were