Information Technology Reference
In-Depth Information
and “extra-structural,” but in fact mean rather “captured by the grammar” and
“not captured by the grammar.” If the “syntactic” structure of a portion of a
text makes a particular rule applicable, then the relationship of this compo-
nent to others is gleaned from the annotation to the rule. Unfortunately, the
“syntax” given in story grammars doesn't rule out many constructions; while
the “semantic” annotations are inadequately defined. This deficiency leaves the
grammars open to wishful parsing and generation, a serious flaw which propo-
nents of story grammars were unable to overcome. A major part of this work is
a rigorous, formal framework used in relating story components to one another
(Goldman & Lang 1993; Lang 1997).
Following Rumelhart's “Notes on a Schema for Stories,” other researchers
expanded on Rumelhart's grammar (Bower 1976; Frisch & Perlis 1981; Johnson
& Mandler 1980; Mandler & Johnson 1977; Stein & Glenn 1979) while others
attacked the foundations of the possibility of a “grammar for stories” (Black &
Bower 1980; Black & Wilensky 1979; Garnham 1983; Wilensky 1982). Eventu-
ally, the story grammars project was abandoned as unsuccessful, largely due to
the crude state of formal techniques available at the time, but also due to the
excessive demands made of story grammars as a cognitive mechanism.
Story Generation by Author Modeling
Around the same time as Rumelhart's seminal paper on schemas for stories,
Meehan published his dissertation on story generation (Meehan 1976). His
system, Tale-Spin, inspired work in story generation from the perspective of
author modeling, that is, by modeling the cognitive processes of a human au-
thor of stories. Turner's Minstrel (Turner & Dyer 1985; Turner 1990, 1991a,
b) and the system described by Okada and Endo (Okada & Endo 1992) are
representative samples of author-modeling systems for stories.
Meehan's Tale-Spin is a simulation of a forest world, producing natural lan-
guage output describing the interactions of characters pursuing goals such as
eating and drinking in a context where duplicity and hostility occur along with
honesty and friendliness. Although Tale-Spin provides access to the meanings
(conceptual dependency forms, in this case) from which the natural language
text is constructed, the model by which the meanings themselves are gener-
ated is left implicit; and the relationships among the components of a story are
deeply entwined in the procedures which drive the simulation.
Michael Lebowitz develops a model of story telling based upon an extensi-
ble library of plot fragments (Lebowitz 1985). These plot fragments serve the
goals of the author, which may be nonsensical from the point of view of the
