MoL-2016-02: A decompositional analysis of discourse relations

MoL-2016-02: Nisidis, Nikolas (2016) A decompositional analysis of discourse relations. [Report]

Preview	Text (Full Text) MoL-2016-02.text.pdf Download (584kB) | Preview
	Text (Abstract) MoL-2016-02.abstract.txt Download (1kB)

Abstract

Current approaches to discourse coherence hypothesize a number of discourse relations which are used to link sentential units to one another. These relations, also called “coherence” or “rhetorical” relations, are defined either semantically in terms of their truth-conditional content or pragmatically in terms of speaker intention. Our enterprise consists in analyzing the constituting features of discourse relations and, more specifically, in investigating the inferences involved when a speaker or a hearer links two parts of a text. The framework we use to carry out our analysis is based on two main assumptions: agents reason in a Bayesian fashion and discourse is characterized by some sense of topicality. For that reason, we use a version of Update Semantics and combine it with Causal Probabilities along with Argumentation and Questions Under Discussion. Causal Probabilities allow us to model causal and identity inferences the interlocutors make as well as probabilistic inferences having to do with modalities, whereas Questions Under Discussion provide an elegant way to model topicality. We analyze the 32 discourse relations of Rhetorical Structure Theory due to the latter’s longevity as a discourse structure theory and its application in a variety of linguistic fields. The main result of our analysis is that disycourse relations can be reduced to specific inferences and there is no need to hypothesize any primitive relations constituting these relations. Given the nature of the inferences, it is also possible that a Bayesian interpreter be able to automatically assign a discourse relation just by answering certain questions pertinent to discourse structure.

Item Type:	Report
Report Nr:	MoL-2016-02
Series Name:	Master of Logic Thesis (MoL) Series
Year:	2016
Uncontrolled Keywords:	Logic, Language
Date Deposited:	12 Oct 2016 14:39
Last Modified:	12 Oct 2016 14:39
URI:	https://eprints.illc.uva.nl/id/eprint/974

Actions (login required)

View Item