| This dissertation presents a novel method of sentence generation, drawing on the insight from Cognitive Semantics (Talmy, 2000a,b) that the effect of uttering a sentence is to evoke a Cognitive Representation in the mind of the listener. Under the assumption that this Cognitive Representation is also present in the speaker and defines (part of) the speaker's communicative intention, sentence generation is seen as the process of transforming Cognitive Representations into sentences. Starting with a brief exposition of the Cognitive Representations of motion events, this dissertation demonstrates how Cognitive Representations can be rendered as graphs, the graphs can be analyzed (parsed) with a graph grammar, and the parse term obtained from the graph can be transformed into the linguist's traditional representation of a sentence, a syntactic tree.;The process of generation is presented in two phases: graph parsing and parse transformation. The graph parsing phase itself has two sub-phases. An initial descriptive phase motivates graphs as a means of depicting Cognitive Representations and tours the development of graph grammars from string grammars. The following analytic phase introduces flowgraphs, a formalism closely related to graphs. It is demonstrated that from any graph a unique flowgraph (up to isomorphism) can be obtained. The analytic phase finishes with a detailed exhibition of an algorithm for parsing flowgraphs based on the work of Lutz (1996).;The parse transformation phase shows first how the important details of a parse can be recorded in a parse term while also discarding irrelevant details of the parse. The transformation of the parse term into a syntactic tree is then manifested with a top-down deterministic macro tree transducer (Engelfriet and Vogler, 1985), and the generative powers of macro tree transducers is briefly explored. |
