Learning Narrative Intelligence
Narrative intelligence includes the ability to create, tell and understand narratives. It has been proposed as a central component of human intelligence, and hence a central component of any machines aiming to simulate human intelligence or to communicate effectively with humans. A number of computational narrative intelligence systems have emerged during the past few decades. However, most such systems rely on hand-crafted knowledge structures that require extensive expert labor. As a result, these systems are constrained to operate in a few domains where the needed knowledge has been authored.
This project investigates the learning of knowledge structures that can support story generation, storytelling, and story understanding in any domain. The ability to learn supportive knowledge structures, I claim, is an integral part of computational narrative intelligence systems that are flexible enough for the real world. Due to the complexity of narrative structures, learning of supportive knowledge has remained a largely unsolved problem to date.
The proposed system aims to learn complex knowledge structures from an input corpus of simple stories, and subsequently create, tell, and understand stories from the same domain by utilizing the learned structures. This knowledge representation balances the complexity of learning and the richness of narrative applications, so that we can (1) learn the knowledge robustly in the presence of noise, (2) generate a large variety of stories based on one learned representation and (3) understand stories efficiently.
As a result, the system is able to demonstrate narrative intelligence in any domain where a small number (~60) of input stories can be collected. Those stories can be easily collected as writing them does not require training in computer science. One inexpensive way (but not the only way), as I demonstrate, is to crowdsource the stories from Amazon Mechanical Turk. The algorithms are evaluated by checking their results against human intuition. This project is the first step toward extending computational narrative intelligence to the real world, which is both incredibly rich and incredibly noisy.
Download the crowdsourced data set here.
- Boyang Li. Learning Knowledge to Support Domain-Independent Narrative Intelligence. School of Interactive Computing, Georgia Institute of Technology. Ph.D. Dissertation. 2015.
- Boyang Li, Mohini Thakkar, Yijie Wang and Mark O. Riedl. Storytelling with Adjustable Narrator Style and Sentiments. The 7th International Conference on Interactive Digital Storytelling, Singapore, 2014.
- Boyang Li, Mohini Thakkar, Yijie Wang, and Mark O. Riedl. From Data to Storytelling Agents. The 14th International Conference on Intelligent Virtual Agents (IVA 2014). Boston, MA. 2014.
- Boyang Li, Mohini Thakkar, Yijie Wang, and Mark O. Riedl. Data-Driven Alibi Story Telling for Social Believability. Social Believability in Games Workshops , Fort Lauderdale, FL, 2014.
- Boyang Li, Stephen Lee-Urban, George Johnston and Mark O. Riedl. Story Generation with Crowdsourced Plot Graphs. The 27th AAAI Conferece on Artificial Intelligence, Bellevue, Washington, 2013.
- Boyang Li, Stephen Lee-Urban, D. Scott Appling and Mark O. Riedl. Crowdsourcing Narrative Intelligence. First Conference on Advances in Cognitive Systems, Palo Alto, California, 2012.
- Boyang Li, Stephen Lee-Urban, and Mark O. Riedl. Toward Autonomous Crowd-Powered Creation of Interactive Narratives. Fifth Workshop on Intelligent Narrative Technologies, Palo Alto, California, 2012.
- Boyang Li, D. Scott Appling, Stephen Lee-Urban, and Mark O. Riedl. Learning Sociocultural Knowledge via Crowdsourced Examples. Fourth AAAI Workshop on Human Computation, Toronto, Canada, 2012.
- Boyang Li, Stephen Lee-Urban, D. Scott Appling, and Mark O. Riedl. Automatically Learning to Tell Stories about Social Situations from the Crowd. LREC 2012 Workshop on Computational Models of Narrative, Istanbul, Turkey, 2012.
Which Paper Should I Read?
The dissertation contains the most up-to-date information and details that previous papers did not cover. The ICIDS paper describes efforts to tell stories using different narration styles, with potential applications in believable virtual characters. The AAAI paper and the ACS paper describe learning of the plot graph knowledge structure.