2008 Poster Sessions : Automatic Sentence Simplification and Online Word Games

Student Name : David Vickrey
Advisor : Daphne Koller
Research Areas: Artificial Intelligence
Abstract
In this poster I will present two projects for better understanding natural language. The first is a model we are developing for simplifying sentences. For example, the sentence "I didn't have a chance to eat" would be simplified to "I ate." Although we lose some information, this simplified sentence is generally much easier to process. We apply the sentence simplification algorithm to the task of semantic role labeling, where we try to label the arguments of each verb (e.g., the "Eater" in our example sentence is "I"). Sentence simplification is particularly helpful when we only have a small amount of labeled data The second project is online word games for collecting data about natural language. We have implemented two games. In the first game, a pair of players are given the same set of categories (e.g., "Types of vehicle") and an initial letter (e.g. "C"). They then type answers (e.g., "car") and get points whenever they match. The second game is similar, but instead shows a single word at a time and allows the players to type anything they want (again
with the goal of matching their co-player). To date, the games have been online for almost a year, and we have collected around half a million submitted data instances. Preliminary data analysis has shown that we can extract interesting linguistic resources from this collected data (e.g., lists of vehicles).

Bio
David Vickrey is a Ph.D. student working with Prof. Daphne Koller in the Stanford Artificial Intelligence Lab (SAIL). His work focuses on machine learning methods for natural language processing. Particular areas of interest include learning to rewrite and paraphrase sentences; and learning and using relationships between words. Prior to starting his Ph.D., he graduated from Stanford with a double major in Math and Computer Science. His senior year, he received the Terman Engineering Scholastic Award for placing in the top five percent of the graduating class, and placed in the Top 50 in the national Putnam Mathematical Competition.