Connectionist and Process Modelling of Long-Term Sequence: The Integration of Relative Judgements, Representation and Learning

Doing Harris, Kristina Mary (1995) Connectionist and Process Modelling of Long-Term Sequence: The Integration of Relative Judgements, Representation and Learning. PhD thesis, University of Glasgow.

Full text available as:
[img]
Preview
PDF
Download (7MB) | Preview

Abstract

A large amount of psychological research is devoted to the representation of sequences. It is a fundamental upon which most of the processes of cognition are based. Despite the amount of research into sequencing, there has been relatively little investigation of the types of representations generated in response to sequential information. These representations must allow operations to be performed on individual elements, as well as operations between and among elements. This thesis begins by describing the effects found when subjects are asked to make relative order judgements using sequences which are in long-term storage (e.g. the alphabet or number series). These effects are then used to examine some of the theories and models which have been developed, with a view toward generating a general purpose mechanism with the ability to model all of the different effects found with different types of stimuli. In the course of developing the new model, the neuropsychological findings in the area are examined in Chapter Two. Deficit studies and neuropsychological investigations are able to isolate which aspects of a task appear to be processed in different structures. If a patient loses the ability to perform one aspect of a task but not another, trying to model both of these aspects in one network may be counter-productive. The construction of a new model is begun in Chapter Three. This model is developed in the PDP environment as it offers the ability to change (learn) as a result of experience, and demands a more thorough definition of the mechanisms operating within the network. Chapter Four details a formal definition of the Serial Order Network (SON) model outlined in Chapter Three, including a section devoted to relative order judgements, called the Response Generation Network (RGN) and undertakes a comparison between the SON/RGN and Poltrock's (1990) random walk model described in Chapter One. A review of some of the sequence learning networks developed is undertaken in Chapter Five. This review is used to choose sequence learning networks, which may be used to learn the type of representation needed. These sequence learning networks are investigated in Chapter Six for their ability to learn the sequence incrementally. It is determined that not one of these networks is appropriate. Thus, in Chapter Seven a recitation mechanism is added directly to the representation in the SON. The resulting system is investigated, and it is determined that the system's success in recitation is not dependent on an idiosyncratic setting of the parameters in the network. The definition of the SON and complementary recitation network is not sufficient. The resulting mechanisms should also be compatible with the developmental literature for both children learning sequences for the first time, and adults learning novel sequences. A review of this literature is conducted in Chapter Eight. It is also necessary to explain how this SON representation can be developed, and how the model can be used to explain the seeming hierarchic nature of some sequences. In Chapter Nine a mechanism designed to mimic a hierarchical structure for the representation of a sequence is developed. A learning mechanism is defined for the resulting system. This system is then investigated for its ability to recite both hierarchic and non-hierarchic sequences, and to generate the relative order and developmental effects referred to in Chapters One and Nine. The model developed in this thesis is the only model existent which is able to explain sequence learning, representation and relative order effects, and represents an advance in the approach to modelling sequence information.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Additional Information: Adviser: Tony Sanford
Keywords: Developmental psychology, Cognitive psychology
Date of Award: 1995
Depositing User: Enlighten Team
Unique ID: glathesis:1995-74441
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 13 Nov 2019 15:58
Last Modified: 13 Nov 2019 15:58
URI: http://theses.gla.ac.uk/id/eprint/74441

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year