Dynamic Location Updates: The Impact of Motion on Discourse Structure

This dissertation presents an analysis of the language of motion in discourse and a computational model that captures the observed phenomena. This model, called Dynamic Location Update (DLU), is motivated by the ultimate goal of tracking moving objects through time and space. Similar computational models have proved necessary for the analysis of temporal information. Given this model, a framework that is based within it is described and its potential implementation into an existing resource for semantic analysis, Boxer, is discussed.;The analysis begins with the inspection of two corpora. Three full length texts that are particularly rich with spatiotemporal information are closely examined, which results in a distinction between explicit and implicit motion that is crucial in the final model. A collection of discourse segment pairs that involve explicit motion predicates is also analyzed. As a result of these analyses, the belief that locations are dynamically updated in discourse is confirmed. The first steps towards defining the DLU model are then made with the introduction of several variables that are used to capture both explicit and implicit motion.;Given these observations, several phenomena are defined and requirements for any framework that uses the DLU model to account for the phenomena are designated. At this point in the presentation, the observed behaviors are accounted for by the DLU theory, but there is no specific framework that implements this theory. The second half of the dissertation presents one such framework.;This framework builds on Dynamic Interval Temporal Logic (DITL). DITL for Discourse (DITL-D) consists of axioms that work from the DITL expressions for the given discourse segments as well as the Discourse Representation Theory (DRT) and Segmented Discourse Representation Theory (SDRT) treatments of the discourse. Rhetorical relationships are crucial to the DLU phenomena. The second component of the framework, MotionDRT, reflects the variable resolutions that DITL-D mandates. Together, DITL-D and MotionDRT account for most of the DLU phenomena and all of the stated requirements of the DLU theory. Finally, speculation on how this framework could be implemented into Boxer is presented.