The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

PhD Thesis #8609

Title:Interpreting Systematic Grammar as a Computational Representation: a Problem Solving Approach to Text Generation.
Date: 1986
Abstract:A new approach to automatic natural-language text generation is described. The approach exploits artificial intelligence (AI) problem-solving techniques and explicitly represents the grammars using an established linguistic formalism. It is demonstrated that AI problem solving and Halliday's theory of systemic grammar share some fundamental properties, and that this has resulted in an equivalence between the representations found in these two fields. The equivalent representations mean that a systemic grammar can be translated trivially and automatically into AI knowledge-representation languages and then used as a linguistic knowledge base. This knowledge base can be put at the disposal of a powerful, general-purpose problem solver which applied goal-directed knowledge-based techniques to selectively and efficiently generate text. A significant linguistic characteristic of this approach is that it allows the incorporation of the socio-semantic level of systemic theory exactly as described in Halliday's recent writings. This is also of major computational significance because the socio-semantic level acts as highly-compiled knowledge that guides the grammitical problem solving. The approach thus exploits the state-of-the-art computational techniques while manifesting an established linguistic theory. The approach is ideal for generating explanations in expert systems because the same problem solver that applies the expert knowledge to problems can also apply the linguistic knowledge in the grammar to text-generation problems. This not only supplies a powerful and efficient mechanism for the text-generation problem, but also greatly simplifies the system as a whole. Although any one of several AI knowledge-representation languages could have been used to represent systemic grammars, production rules were chosen for this project. Since systemic grammars can be trivially translated into production rule form, a formalization of systemic grammars

[Search These Pages] [DAI Home Page] [Comment]