Using a Narrative Generator to Teach Literary Theory

  1. 1. Peter Havholm

    Department of English - College of Wooster

  2. 2. Larry Stewart

    Department of English - College of Wooster

Work text
This plain text was ingested for the purpose of full-text search, not to preserve original formatting or readability. For the most complete copy, refer to the original conference program.

Using a Narrative Generator to Teach
Literary Theory
The Linear Modeling Kit, an application designed
by the writers, is an environment in which undergraduates can create narrative generators. We use
the LMK in literature courses as a way of helping
students think more deeply and precisely about
narrative and narrative theory. Students are asked
to “program” their own understanding of a narrative theory in the LMK, thus creating a narrative
generator. They then test the powers of their understanding by evaluating the skeletal narratives
produced by their generator.
Using syntactic rules, the LMK can create narrative sequences from text strings grouped in categories, and it is therefore particularly good at
modeling narrative theories influenced by structuralism. (It is a whiz at modeling syntactic structures.) Any discursive analysis of narrative can be
translated into the LMK’s four terms: “category”
(or “element”), “text string,” “link,” and “marker.”
A humorous squib, “Create Your Own Internet
Hype” (quoted in Newsweek from the dance magazine XLRSR), uses a primitive version of such
a process to satirize jargon. The instructions are:
“pick one word from each column [A through C]
to form your very own Net Hype Term.” The
columns (excerpted) are:
interactive multimedia suite
high speed server architecture
networked e-mail engine
revolutionary reality group
visionary protocol site
virtual software agent
To make correct Net Hype terms, one must link
any word from category A to any word from
category B and then link the B word to any word
in category C. Given this syntax (A+B+C) and the
vocabulary as categorized, all of the following are
correct: “interactive multimedia suite,” “revolutionary server agent,” “visionary software architecture,” and “high speed server group.”
The resulting phrases seem semantically correct
because they sound like the kind of thing one hears
in sales presentations and reads in Wired. (The
model’s semantic component is partly a function
of the words selected for the three columns.) That
they carry so little denotative power is the joke. A
simple narrative generator built along these lines
might have for its three categories – with a nod to
Gerald Prince’s Narratology (81–83 – First State,
Modification, and Second State. Into First State,
one might put the strings “Mary was rich” and
“Mary lived in New York.” Into Modification, one
could put “Then disaster struck” and “Then she
changed jobs,” and into Second State, “Then, as a
result, she was poor” and “Then, as a result, she
moved to California.”
We can see at once that the rule A+B+C will not
do, here, because one result would be “Mary was
rich. Then disaster struck. Then, as a result, she
moved to California.” That sequence does not feel
much like a narrative.
The LMK, therefore, allows a user to mark strings
to differentiate them from one another within a
category. In this case, First State strings could be
marked as fitting the “wealth” sequence or the
“move” sequence. “Mary was rich” would be marked “wealth” while “Mary lived in New York”
would be marked “move.” In the “Modification”
category both “Then disaster struck” and “Then
she changed jobs” could be left unmarked because
they will work with either sequence. In the “Second State” category, “Then, as a result, she was
poor” would also be marked “wealth,” and “Then,
as a result, she moved to California” would be
marked “move.” Because the LMK matches marked strings only with other similarly marked
strings or with unmarked strings (like “Then she
changed jobs” in this case), the four possible narratives would then be continuant: 1) “Mary was
rich. Then disaster struck. Then, as a result, she
was poor.” (2) “Mary was rich. Then she changed
jobs. Then, as a result, she was poor.” (3) “Mary
lived in New York. Then she changed jobs. Then,
as a result, she moved to California.” or (4) “Mary
lived in New York. Then disaster struck. Then, as
a result, she moved to California.”
The use of Prince’s technical categories here is to
suggest that, while our students work with more
limited analyses like Vladimir Propp’s of folk
tales or Janice Radway’s of romances, the LMK is
adaptable to global projects like that of Narratology.
In “Story and Theory,” an upper-level course
about approaches to narrative, we have asked students to use the LMK to create models repre-
senting (for example) Propp’s analysis of the Russian folk tale, an assignment which calls for them
to understand his functions and to infer their linking rules in detail.
In response, students follow the procedure suggested above, first translating Propp’s thirty-two
functions into collections of text strings. For example, strings like the following are entered to
represent Propp’s category “Absentation”:
Now it happened that Ivan’s grandmother
was a great lover of sweets, and that she
sometimes went to the next village, where
the baker could make sweets to tickle even
the oldest tooth. She liked to go alone, so she
left Ivan alone in the house.
And Propp’s “Interdiction” includes strings like:
But before she left, his grandmother gave
Ivan a solemn warning: “while I am gone,
do not open the jewelry box on my dresser.”
Before they left, Peter and Ruth were told by
their father, “Stay in the house, whatever you
do, because it is easy to get lost wandering
Once the strings have been entered, students establish the syntactic rules suggested by Propp’s
analysis. For example, it is clear in Morphology
of the Folktale that, while neither the function
“violation” nor the function “absentation” will
occur in every tale, when they do, “violation” is
always preceded (and never followed) by “absentation.” Such syntactic rules may be translated into
the LMK’s links, which set the possible combinations of categories. For example, in the Internet
Hype Generator above, A is linked to B (but not
to C or to itself), and B is linked to C (but not to
A or to itself). That means that only ABC sequences are possible, while ACB, AABBCC, BCA,
ACB, etc., are not.
With enough strings like the examples above –
appropriately marked and their categories linked
in the ways Propp’s analysis suggests, an LMK
generator can produce stories like the one quoted
in part below, from a generator created by Ben
Wachs, a senior at Wooster this year:
There was a rumor among the peasants that if Ivan
were to ever stay in one place for longer than three
months, the world would end. Certainly Ivan wasn’t taking any chances, and had decided to take to
the road to see what new sights he could see. As
Ivan looked at the forest around him he was amazed to see that many of the animals looked sick,
and many were dead! “Oh Ivan,” a flock of geese
cried to him, “there is a terrible plague over the
forest. If you do not help us, we are doomed!” “A
plague!” Ivan gasped. “How terrible!” Ivan had a
great love for the animals of the forest, and did not
wish any of them to come to harm. “Don’t worry,
my friend, I will not sleep until I have found a
cure.” Ivan thought about how to begin to cure the
forest animals. “I have heard,” he said, “that there
are many magical herbs inside the forest. Perhaps
if I travel south, deeper into it, I shall find some.”
So saying, he began his trek. After a walk too long
to be talked about, Ivan came to a clearing where
a huge gryphon sat, tethered to a rock.” If you want
to petition the magic gnome for help,” the gryphon
yawned, “you must do a task for me. Otherwise,
I’ll never let you pass.” “Oh?” said Ivan. “I should
trust a man-eating creature chained up to a rock
like a criminal? Hah. Let me pass. . ..”
The generator that produced this tale can produce
millions of others, each distinct from this one. Like
any storyteller, this generator repeats itself on
occasion – reusing characters, tropes, and plot
devices – but for all practical purposes, it never
tells the same tale twice. The difference between
having students read and write about the Morphology and having them make a story generator
based on it is that, unlike essays, generators either
work or they don’t. They use Propp’s functions or
they do not, and they produce tales that sound like
folk tales or they do not. In our experience, students can tell when their generators are not producing valid tales – and therefore thrash their way
through to success without benefit of a teacher’s
advice. Moreover, a generator can measure the
power of the analysis it models. That is, if some
or all of the generator’s products do not “sound
like” folk tales, students may find that (1) they
have not accurately modeled the analysis, (2) the
LMK is not capable of reproducing the analysis
accurately, or (3) the analysis itself does not account for enough elements of folk tales. In Propp’s
case, for example, there is nothing in the Morphology analogous to the LMK’s markers (like those
for “wealth” and “move” in the kernel narrative
example above). Yet our students have found it
impossible to create a good generator without
using markers, and they commonly mark strings
for character and for situation. This is a sign that
Propp’s rejection of “theme” as a means of classification (7–11) and his dissociation of function
from character (20–21) reduce his system’s ability
to account for all that occurs in a tale. We use the
LMK to allow students to model the consequences
of analytic ideas. While modeling an analysis,
students encounter literary critical texts in an unusually direct and deep way. They do so because
they must wrest from the critic’s discursive analysis the formal rules that the LMK requires. Then,
the model’s operation provides a concrete vantage-point from which students can look back at the
original critical texts. Finally, a continuing question the process raises is the degree to which
principles in any analysis can be formalized into
rules without crippling distortion. Modeling requires that students infer the consequences of a set
of analytic terms and test their inferences – in a
way that reading, writing, and discussion do not –
and this is its primary benefit. A second advantage
is that making a generator, in our experience,
invites students to be inventive. They enjoy creating a variety of text strings that will accurately
represent abstract categories like “Interdiction”
and “Violation,” and they take pleasure in developing a narrative syntax that can link a variety of
possible pieces in many different coherent ways.
They are proud of their generators’ power. The
work, in other words, is good work – engaging in
its own right and good for the critical thinking
required by translating a critical analysis into
terms that can be represented by a computer.
Works Cited
Prince, Gerald. Narratology. Amsterdam: Mouton, 1982.
Propp, Vladimir. Morphology of the Folktale.
Trans. Laurence Scott. Austin: U of Texas P,
Radway, Janice A. Reading the Romance: Women, Patriarchy, and Popular Literature. Chapel Hill: U of North Carolina P, 1991.

If this content appears in violation of your intellectual property rights, or you see errors or omissions, please reach out to Scott B. Weingart to discuss removing or amending the materials.

Conference Info

In review


Hosted at University of Bergen

Bergen, Norway

June 25, 1996 - June 29, 1996

147 works by 190 authors indexed

Scott Weingart has print abstract book that needs to be scanned; certain abstracts also available on dh-abstracts github page. (

Conference website:

Series: ACH/ICCH (16), ALLC/EADH (23), ACH/ALLC (8)

Organizers: ACH, ALLC