Conceptualising and modelling hierarchical classification from a behavioural perspective
Slattery, Brian
Slattery, Brian
Loading...
Repository DOI
Publication Date
2013-10-03
Type
Thesis
Downloads
Citation
Abstract
This thesis involved conceptualising and modelling hierarchical classification from a behavioural perspective. Classification refers to grouping stimuli according to their physical or functional similarity (Barnes-Holmes, Hayes, Dymond, & O'Hora, 2001). Hierarchical classification is a particular type of classification whereby classes themselves are classified as members of other classes. For example, a 'Springer spaniel' is a member of the class of 'Dog', and the class of 'Dog' is a member of the class of 'Mammal'. According to the mainstream psychological literature, hierarchical classification is characterised by three features: transitive class containment, asymmetric class containment, and unilateral property induction. Transitive class containment refers to including a particular class as a member of another class. For example a 'Dog' is a member of the class of 'Mammal'. Asymmetric class containment refers to the containment by a higher (superordinate) class of other (subordinate) classes. For example, the class of 'Mammal' includes the class of 'Dog'. Finally, unilateral property induction refers to the asymmetrical transfer of functions from superordinate classes to subordinate classes in a hierarchical network. For instance, the class of 'Mammal' gives birth to live young, therefore, all mammals, including dogs and Springer spaniels give birth to live young. However, the functions of subordinate classes (e.g., Springer spaniel and dog) do not transfer up to superordinate classes (e.g., mammal). Griffee and Dougher (2002) developed the first model of hierarchical classification from a behavioural perspective. Their model was based on contextual control over responding to the formal features of the to-be-categorised stimuli. However, Griffee and Dougher (2002) did not test their model for the emergence of any of the three features that characterise hierarchical classification. The aim of Study 1 was to replicate and extend the Griffee and Dougher (2002) model of hierarchical classification by testing for transitive class containment, and provide remediation training should responding consistent with this feature be found absent. Specifically, Experiment 1 replicated key components of the Griffee and Dougher (2002) model and then included a test for responding consistent with transitive class containment. Only two out of five participants showed the expected pattern. Experiment 2 tested whether repeated exposures to the Experiment 1 protocol would give rise to the expected pattern more reliably. None of three novel participants demonstrated the pattern. In Experiment 3, physically similar stimuli used in Experiments 1 and 2 were replaced across testing cycles by arbitrary stimuli. Transitive class containment consistent responding was observed in all three novel participants. Studies 2 and 3, attempted to address more complex forms of hierarchical classification were the stimuli share no formal similarities. The aim of Study 2 (Experiments 4 and 5) was to model hierarchical classification in terms of contextually controlled bifurcated equivalence classes. Specifically, in Experiment 4 participants were trained and tested for the formation of two 12-member equivalence classes in one context (superordinate), two 6-member classes in a second context (intermediate), and four 3-member classes in a third context (subordinate). Participants were then tested for responding consistent with transitive class containment. Overall, six out of the seven participants established bifurcated equivalence classes under contextual control and five of the six participants who went on to transitive class containment testing, demonstrated responding consistent with this feature. Experiment 5 investigated whether an equivalence-based account of hierarchical classification would produce a transfer of functions in line with unilateral property induction. Participants were first trained and tested for the formation of two 7-member equivalence classes (i.e., H1-A1, B1, C1, D1, E1, F1; H2- A2, B2, C2, D2, E2, F2) in one context (superordinate), and then trained and tested for the formation of four 4-member equivalence classes (i.e., H1.1- A1.1 [A1], B1.1 [B1], C1.1 [C1]; H1.2- A1.2 [D1], B1.2 [E1], C1.2 [F1]; H2.1-A2.1 [A2], B2.1 [B2], C2.1 [C2]; H2.2- A2.2 [D2], B2.2 [E2], C2.2 [F2]) in a second context (subordinate). Stimulus functions were then established for one subordinate class member (H1.1) and one superordinate class member (H2) and participants were then tested for a transfer of functions in line with unilateral property induction. Overall, both participants demonstrated the formation of contextually controlled bifurcated equivalence classes. However, both participants demonstrated a bidirectional transfer of functions, which is not in line with unilateral property induction. The aim of Study 3 was to model hierarchical classification as contextually controlled generalised relational responding or relational framing. In Experiment 6, a training procedure involving non-arbitrarily related multi-dimensional stimuli was used to establish two arbitrary shapes as contextual cues for 'member of' and 'includes' relational responding respectively. Subsequently those cues were used to establish a network of arbitrary stimuli in particular hierarchical relations with each other and then test for derivation of further untrained hierarchical relations as well as for transformation of functions. Resultant patterns of relational framing for all four participants showed properties of transitive class containment, asymmetrical class containment and unilateral property induction, consistent with conceptions of hierarchical classification as described within mainstream cognitive developmental literature. Experiment 7 extended the basic model by using more naturalistic stimuli and providing a better controlled test of transformation of functions. Overall, six out ouf nine participants passed non-arbitrary pretraining, the remaining participants demonstrated transitive class containment and asymmetric class containment and five out of the six participants demonstrated unilateral property induction. Limitations of all three studies and directions for future research are discussed.
Funder
Publisher
Publisher DOI
Rights
Attribution-NonCommercial-NoDerivs 3.0 Ireland