In this thesis, we aim to analyze the computational aspects of discourse annotation. Discourse is not only a concatenation of sentences / in fact the totality of discourse is more than the sum total of the sentences that constitute it. The property that differentiates discourse from a set of arbitrary sentences is defined as coherence. Coherence is established by the relations between the parts of discourse. We have a lexicalized approach to discourse, therefore in this study, discourse relations are considered to be set up by lexical items called discourse connectives. Systematic analysis of coherence requires an annotated corpus in which coherence relations are encoded. We developed an annotation environment to be used in an ongoing discourse level annotation project which aims to generate a theory-neutral source of coherence relations. We followed a data-driven methodology in design of the data structure employed in the annotation software. For this reason, we examined the predicate-argument structure of connectives. This analysis shows that stand-off annotation technique is more suitable than an inline method for such an annotation environment. This thesis also include a brief discussion on the formal implications of coherence relation constructions.

A text is a linguistic structure that is more than a random collection of sentences. A text is cohesive (Halliday &amp / Hasan, 1976) and coherent (Mann &amp / Thompson, 1987, 1988). Mainly ignored in the field of linguistics until recently, the text and the discourse structure have been inquired from various points of view (Asher, 1993 / Asher &amp / Lascarides, 1998 / Grosz &amp / Sidner, 1986 / Mann &amp / Thompson, 1987, 1988 / Webber, 2004). D-LTAG is a discourse grammar work that extends a lexicalized sentence level grammar LTAG (Joshi, 1987) to low-level discourse (Webber, 2004 / Webber &amp / Joshi, 1998). In this framework, discourse connectives such as coordinating conjunctions, subordinating conjunctions, parallel connectives and discourse adverbials are predicates of discourse structure that take text spans that can be interpreted as abstract objects (Asher, 1993). Turkish has a flexible word order in comparison to languages like English. In English, the discourse adverbials are noted for their ability to occupy positions unavailable to other discourse connectives. In Turkish, word order of other discourse connectives, coordinators and subordinators are not expected to be as restricted. This thesis examines the connective position, argument order and the information structure of five Turkish discourse connectives in their eleven uses. The analyses show that the examined features of discourse connectives are related to the syntactic group the connective belongs to. Discourse connectives of the same syntactic groups exploit similar connective position and argument order possibilities, and they tend to be included in similar information units.