Computational Narratology and Thematics

Stories’ themes, or thematic concepts, are what readers think that the stories are about. Stories’ theses, or thematic statements, are what readers think that authors expressed about these themes.

Computationally modeling stories’ plots, characters, themes, and theses can enable the creation and revision of powerful datasets with which to train and evaluate artificial-intelligence systems. With such computational models stored alongside stories’ texts, artificial-intelligence systems could be trained to perform data-driven story generation, comprehension, and evaluation tasks.

Computational Modeling

Stories’ themes and theses can be modeled using data structures that include properties having text-string values with which to express noun concepts, noun phrases, (ethical) questions, moral messages, proverbs, adages, or other content.

Stories’ plots, characters, themes, and theses can be represented using knowledge graphs. Annotations can be used to relate excerpts or selections of stories’ texts to these data, to nodes, statements, and/or subgraphs. Selections or excerpts of stories’ texts can provide evidence for interpretive arguments and reasoning in support of interpretive claims about stories’ plots, characters, themes, and theses.

Stories and their subparts can have multiple valid interpretations. Computational models of stories interpretations and interpretive arguments could involve hierarchical and recursive data structures (Le, 2002). Stories’ outline nodes could provide interpretive data, arguments drawing evidence from stories’ corresponding text content and, perhaps, other source materials, while also describing relationships to containing and contained parts as well as their claims, properties, and text contents.

Question-answer pairs could be included in or, interestingly, derived from computational models of stories’ plots, characters, themes, and theses. Included and/or derived question-answer pairs can enhance the training and evaluation of artificial-intelligence systems.

Story Search

With stories’ themes and theses computationally modeled, new techniques for indexing and searching for stories would become possible. Stories’ themes and theses could, for instance, arise during conversational story searches.

Story Recommendation

Artificial-intelligence systems could utilize recommender systems for parables and adages, or for stories’ themes and theses, more broadly, to suggest themes and theses for individual readers or described audiences.

Once themes and theses were thusly recommended, matching stories could be searched for and retrieved for readers or story-generation components could create new stories.

Story Generation

Interestingly, when stories’ themes and theses are among the input ingredients, more purposeful creative writing can be achieved by artificial-intelligence systems performing story-generation tasks and by human authors using these tools.

Story Comprehension

One view of story comprehension is that artificial-intelligence systems have successfully comprehended a story when they can correctly answer reading-comprehension questions about that story.

Beyond comprehending stories, however, might artificial-intelligence systems, one day, be able to “reverse engineer” stories’ texts into story outlines, into computational models of stories’ plots, characters, themes, and theses, or into other “project files” for writing tools?

Story Evaluation

By developing and using rule-based requirements about their desired stories, authors could use evaluation components to receive useful informational messages, warnings, and errors pertaining to their story revisions.

By developing and using preferences about their desired stories, authors could use evaluation components to receive multi-objective, multi-criteria costs and scores to be minimized and maximized, respectively.

Components can also be envisioned for predicting described audiences’ responses to story revisions. Modeled simulations, predictions, and other forecast data could be presented to authors in a variety of ways.

With new and emerging techniques for evaluating stories, agentic generation-evaluation-revision loops could be greatly enhanced, these loops being useful to contemporary approaches to generating stories with artificial-intelligence tools.

Bibliography

Le, Elisabeth. “Themes and hierarchical structure of written texts.” In Thematics: Interdisciplinary Studies, pp. 171-187. John Benjamins Publishing Company, 2002.


by

Tags: