The dichotomy of scaled connectionist compositionality
: (S)RAAM - trick or treat?

  • John C. Flackett

    Student thesis: Doctoral Thesis


    The use of compositional recursive structures form the bedrock of many application areas. Such structures allow data to be stored and manipulated. Within artificial intelligence (AI) symbolism has traditionally provided the mechanisms to create and deal with these recursive types of data representations. In contrast, connectionism has a much shorter history in providing techniques to encode the recursive structures necessary for tasks such as natural language processing (NLP). However, even though such connectionist techniques now exist they have not been shown to scale adequately for use in real-world problem domains (such as parsing). Indeed, although connectionist techniques have been employed in systems that begin to deal with real-world data, the representation of structured data within such systems is still left to symbolic modules. To address this, an existing [cutting edge] hybrid (connectionist/symbolic) parsing architecture has been modified to make use of Simplified Recursive Auto-Associative Memory ((S)RAAM). (S)RAAM is employed to encode parsed sentences (from Lancaster Parsed corpus) in recursive connectionist representations. These representations act as the final output from the parsing architecture and maintain the parse state during processing. The use of (S)RAAM removes the need for a symbolic stack mechanism in the parser and allows the first in-depth study of the technique using corpus data. This thesis begins by examining the properties of (S)RAAM to show that the compositional representations created exhibit properties that may be utilised by other connectionist techniques.. The parsing framework (outlined above) is then presented and this framework allows the scaling of (S)RAAM to corpus data. Conclusions are then drawn about (S)RAAM's suitability as a vehicle for continued research into recursive connectionist representations with (S)RAAM showing promising capabilities in the learning phase, but limited generalisation properties.
    Date of Award2005
    Original languageEnglish
    Awarding Institution
    • Nottingham Trent University

    Cite this