Repository logo
 

Peak Shift in Pigeon and Human Categorisation


Type

Thesis

Change log

Authors

Aitken, Michael R. F. 

Abstract

In a series of experiments, both pigeon and human subjects were trained to categorise two groups of confusable stimuli, with each category being made up of distortions of a ‘Prototype’. Once the subjects had successfully learned to categorise the training stimuli, they were tested on their responding to a variety of previously unseen stimuli: these were distortions of the Prototypes towards (‘Closer’ exemplars), or away from (‘Further’ exemplars), the other category, and the Prototypes themselves. Pigeons responded more to positive Further exemplars that were close to the Prototype than to the Prototype itself, or to exemplars even further away from the category boundary. This result is an example of the peak shift (Hanson, 1959), and can be explained by interacting excitatory and inhibitory generalisation gradients (Spence 1937).

When the pigeons were autoshaped using stimuli from the positive category before learning the categorisation, they failed to show a peak shift; greatest response rates on test were elicited by the positive Prototype. This result could be explained by the interaction of the autoshaping producing a ‘prototype effect’, i.e. a generalisation gradient with a maximum at the Prototype of the positive category, which masks the development of the peak shift. Further experiments showed that a similar abolition of the peak shift occurred when the pigeons were given prior experience of the negative category in an extra-dimensional discrimination designed to produce an inhibitory analogue of the prototype effect.

A connectionist model of categorisation learning is presented, based on representation of the stimuli as sets of independent features. Simulations conducted using this model showed that, with few assumptions, such an analysis was capable of accounting for all the results found with pigeon subjects, some of which present a problem for alternative instance theories of categorisation (e.g. Pearce, 1984).

Human subjects also categorised the Further exemplars better than the Prototypes, but did not show a peak shift. Performance increased with greater distance from the category boundary, consistent with subjects having abstracted and applied a cognitive strategy. When trained in an incidental learning paradigm, designed to minimise the opportunity for using such a strategy, subjects showed evidence of learning without any knowledge of the categorisation ‘rule’. The performance in this ‘implicit’ task had some similarities to the results of the studies with pigeon subjects, suggestive of a peak shift. These results indicate that similar associative processes may underlie categorisation in both humans and non-humans, although higher-level ‘symbolic’ processes may control human performance in laboratory studies.

Description

Date

Advisors

Keywords

Qualification

Doctor of Philosophy (PhD)

Awarding Institution

University of Cambridge