SDRT

Segmented Discourse Representation Theory (SDRT) is a formal framework for discourse semantics and coherence in natural language. It extends Discourse Representation Theory (DRT) by providing a structured representation of how sentences and passages relate within a larger text or dialogue.

The central idea of SDRT is to partition a discourse into segments, each with its own discourse representation structure (DRS). These segments are connected by coherence relations such as Narration, Elaboration, Explanation, Background, and Contrast. The overall discourse is modeled as a graph in which nodes correspond to DRSs and edges encode coherence relations, enabling the representation of non-local dependencies and cross-sentence reference that are difficult to capture in linear, tree-like accounts.

SDRT handles phenomena such as anaphora and presupposition across segments, allowing referents introduced in one part of the discourse to be accessible to later segments. It supports the dynamic updating of the discourse model as new information is introduced and provides a compositional account in which each segment contributes to the global meaning in a way governed by its attached coherence relation. The approach aims to produce a coherent, truth-conditional interpretation of multi-sentence discourse, given an appropriate context.

Applications and impact: SDRT has been influential in formal semantics and computational linguistics for analyzing discourse structure, coherence relations, and cross-sentential reference. It is often discussed alongside other theories of discourse coherence such as Rhetorical Structure Theory (RST) and various DRT-based approaches. The framework supports discourse parsing, interpretation of long-distance dependencies, and the development of natural language understanding systems that require robust modeling of discourse-level meaning.