Research Repository

Computational Linguistic Models of Mental Spaces

O'Reilly, Cliff (2016) Computational Linguistic Models of Mental Spaces. Masters thesis, University of Essex.

[img]
Preview
Text (Dissertation)
Essex_CliffOReilly_Dissertation_Oct2014_Final_20160527.pdf
Available under License Creative Commons Attribution.

Download (1MB) | Preview

Abstract

In this report we describe a computational linguistic model of mental spaces. We take theories from cognitive science as inspiration and, using the FrameNet database, construct a model upon which we execute a number of experiments. Our underlying assumption is that, in order to develop computer systems that have near human capacities for natural language processing, those systems will need to model cognitive processes. Gilles Fauconnier's theory of Mental Spaces provides a detailed background of partitioned semantic relations. These relationships can be constrained by Frames and Scripts. We use pre-existing computer tools to develop a model that mimics this framework. Fauconnier's and Turner's work on Conceptual Integration and current theories of dynamic systems are further inspiration for a model of conceptual integration using Latent Dirichlet Allocation, a topic modelling algorithm. We choose three experiments with which to validate the usefulness of this approach. Our fi�rst experiment investigates text classi�cation using the Full Text corpus within FrameNet. Our second experiment uses the corpora supplied for the SemEval Textual Semantic Similarity Task in order to validate the hypothesis that mental space networks are related to semantic similarity. The third experiment in this report investigates the Blending model and the hypothesis that this is related to the style of the document text. The results for these experiments were mixed. We are pleased with some high Micro F1 scores (0.9), but disappointed that overall the results are not conclusive. We describe the analysis of the outcomes and also the drawbacks of our methods. Finally we explain our thoughts on how these models could be improved and extended by learning lessons from our work and also including other work and approaches.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Computational Linguistics, NLP
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions: Faculty of Science and Health > Computer Science and Electronic Engineering, School of
Depositing User: Cliff O'Reilly
Date Deposited: 25 Nov 2016 15:08
Last Modified: 25 Nov 2016 15:10
URI: http://repository.essex.ac.uk/id/eprint/18085

Actions (login required)

View Item View Item