Representing Meaning in Multilink: Semantic Spreading of Activation Using Ratings and Associations
Keywords
No Thumbnail Available
Authors
Issue Date
2020-02-14
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
The Multilink project aims to unify and formalize cognitive theories of language processing
in a symbolic localist connectionist model for word retrieval. Already achieving high levels
of correlation to empirical data on several linguistic tasks, Multilink has shown its
suitability in simulating human performance on such tasks. An aspect of Multilink that is
as of yet largely unexplored is its implementation of word meaning. Specifically of interest
is the interaction between different concepts, and how they influence word-level language
processes. In this thesis, human ratings and associations are implemented in Multilink, and
used to simulate empirically collected semantic priming data. It was found that the general
priming effect trend can consistently be reproduced using both ratings and associations in
Multilink. Item-wise correlations are low, however, which may indicate that the data is too
noisy to consistently correlate with. It may also indicate that ratings and associations are
not optimal as measures of semantic spreading of activation. In general, associations
appear to perform slightly better than ratings. For future research regarding semantic
representations in Multilink, this thesis recommends ensuring that orthographic word
features are not dominated by semantic features. Also recommended is a closer
investigation into the causation of item-level differences.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen