Imagine you go to your favorite ice cream shop to get an ice cream cone. When you arrive, you learn that the delivery truck did not come in this morning, and the shop is out of sugar cones. You are a little irritated, because you were really craving ice cream on a cone, but you simply thank the server for the information, and decide to go to another shop that is nearby.
Most of us handle encounters like that described above frequently and fairly easily. For many individuals with autism, however, an unexpected event like that would be quite stressful and difficult to overcome. Everyday life consists of many complex social situations within which a wide variety of obstacles may arise, and each of these obstacles can be overcome in a number of ways. We are interested in using human computation to develop models of social scripts, which are scripts that people naturally develop that help them know how to navigate everyday situations. We elicit three types of data from the crowd; steps to complete the task in question, obstacles that may arise, and solutions to those obstacles. The data collected will be analyzed to develop a model that shows probabilistically how events follow each other and all the ways in which an everyday experience can unfold.
A 6-phase process will be used to enable the collection and classification of this data (left). Our goal is to empower parents and other caregivers to create instructional modules for their children and students. We will use the models to assist them in the authoring process. They will be asked to specify the location and task or activity they would like to create a story about (e.g. attending a concert at a theatre), where they would like the story to start (e.g. at home) and where they would like the story to end (e.g. when the show ends). We will automate the transitions between phases and iterations using Turkit .
To explore the validity of this approach, we conducted 2 studies. These indicated that rich models can be created using data collected via this approach. For more information please see our paper: http://www.cc.gatech.edu/~fatima/Scripts.pdf
The social world that most of us navigate effortlessly can prove to be a perplexing and disconcerting place for individuals with autism. Our approach could enable the creation of models for a plethora of complex and interesting social scenarios, possible obstacles that may arise in those scenarios, and potential solutions to those obstacles. In turn these models will help parents to create and share instructional modules representing a variety of complex social situations. We believe that human input is the natural way to build these models, and in so doing create valuable assistance for those trying to navigate the intricacies of a social life.
Fatima Boujarwah is pursuing her PhD in Computer Science at the Georgia Institute of Technology. This research is part of her thesis work, and is being done under the advisement of Dr. Gregory D. Abowd, and Dr. Rosa I. Arriaga.
 Little, G., Chilton, L. B., Goldman, M., and Miller, R. C. Turkit: Tools for Iterative Tasks on Mechanical Turk. In KDD-HCOMP ’09 (2009).