TU Darmstadt / ULB / TUbiblio

Automatic Identification of Novel Metaphoric Expressions

Do Dinh, Erik-Lân (2013):
Automatic Identification of Novel Metaphoric Expressions.
Darmstadt, Technische Universität, [Online-Edition: https://download.hrz.tu-darmstadt.de/media/FB20/Dekanat/Publ...],
[Diploma Thesis or Magisterarbeit]

Abstract

Manually annotating novel metaphors in philosophical and historical texts is a difficult and time-consuming task, so any automated method would be welcome. In this work, we implement such a method for novel metaphor identification. The approach uses Gaussian mixture models, the parameters of which are estimated by expectation maximization. We seek to improve the results of this baseline by incorporating selectional preferences, whose violation can be indicative for metaphorical use of a term or phrase. As we intend to find only novel metaphorical expressions, we further refine the results by employing a large diachronic n-gram corpus. We find that incorporating selectional preferences and additional n-gram novelty filtering significantly improves on the baseline results. As an example application, the resulting classifications will then be presented in a web-based tool.

Item Type: Diploma Thesis or Magisterarbeit
Erschienen: 2013
Creators: Do Dinh, Erik-Lân
Title: Automatic Identification of Novel Metaphoric Expressions
Language: English
Abstract:

Manually annotating novel metaphors in philosophical and historical texts is a difficult and time-consuming task, so any automated method would be welcome. In this work, we implement such a method for novel metaphor identification. The approach uses Gaussian mixture models, the parameters of which are estimated by expectation maximization. We seek to improve the results of this baseline by incorporating selectional preferences, whose violation can be indicative for metaphorical use of a term or phrase. As we intend to find only novel metaphorical expressions, we further refine the results by employing a large diachronic n-gram corpus. We find that incorporating selectional preferences and additional n-gram novelty filtering significantly improves on the baseline results. As an example application, the resulting classifications will then be presented in a web-based tool.

Place of Publication: Darmstadt
Uncontrolled Keywords: UKP_a_LangTech4eHum
Divisions: 20 Department of Computer Science
20 Department of Computer Science > Ubiquitous Knowledge Processing
Event Location: Darmstadt, Germany
Date Deposited: 31 Dec 2016 14:29
Official URL: https://download.hrz.tu-darmstadt.de/media/FB20/Dekanat/Publ...
Identification Number: TUD-CS-2013-0216
Export:
Suche nach Titel in: TUfind oder in Google

Optionen (nur für Redakteure)

View Item View Item