Narrative Prose Generation

Abstract

Story generation is experiencing a revival, despite disappointing preliminary results from the preceding three decades. One of the principle reasons for previous inadequacies was the low level of writing quality, which resulted from the excessive focus of story grammars on plot design. Although these systems leveraged narrative theory via corpora analyses, they failed to thoroughly extend those analyses to all relevant linguistic levels. The end result was narratives that were recognizable as stories, but whose prose quality was unsatisfactory. However, the blame for poor writing quality cannot be laid squarely at the feet of story grammars, as natural language generation has to-date not fielded systems capable of faithfully reproducing either the variety or complexity of naturally occurring stories. This paper presents the A UTHOR architecture for accomplishing precisely that task, the S TORYB OOK implementation of a narrative prose generator, and a brief description of a formal evaluation of the stories it produces.