![]() | ![]() |
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Introduction Our Members Current INCITE Publications INCITE Projects Virtual Brown Bags The Hanging Milestones Project Grad Tea Ideas ![]() |
INCITE ProjectsApril Fool's Proposal
Title: Reclaiming Joyce: Solving the NLP Problem to Understand the Writings of James Joyce Idris H. Hsi INCITE Lab, College of Computing Georgia Institute of Technology Date: April 31st Time: 1:00 pm Location: KACTB 243 Committee Members: R. Daneel Olivaw Larry P. Waterhouse R. Chandra Richard Daystrom James Joyce (External) Abstract: The works of James Joyce, specifically Ulysses and Finnegan's Wake, have been notorious for their impenetrability due to his experimental use of language and complex narrative structures. To make Joyce more accessible to non-English Literature majors and to reduce the number of people who have to lie about having read him, we propose a translator for recompiling his text into intelligible English using a novel machine learning method for solving partially observable markov decision processes. Our application, working through a context-aware, self-adaptive user interface will return normal English phrases providing a learning scaffold for the user. In previous work, we developed Phonix, an application that uses a combination of techniques from data mining, data clustering, and stochastic search trees in conjunction with WordNet, a dictionary based on semantic networks. By using WordNet as the domain model for the English language, our program can determine meanings for any given piece of text. Our case studies reveal it to have a 62% success rate at interpreting text in works such as Dr. Seuss's "Green Eggs and Ham" and Margaret Wise Brown's "Goodnight, Moon" and a 27% success rate at interpreting the nonsense words in Seuss's works such as 'oobleck' and 'zamp'. We do believe that this algorithm can eventually read Joyce with the secondary result of solving the NLP problem but our current implementation of the compiler is intractable (equivalent to 3-SAT). To overcome this problem,we propose to write the compiler in L-STAR, a Turing-complete language guaranteeing program termination. L-STAR has the added benefit of translating any program equivalent to 3-SAT into a polynomial time solution. To guarantee run-time efficiency, we will implement our final application on a grid-based network of quantum computers. We hope that this line of inquiry will lead to publishable results. Last modified 20 April 2006 at 7:44 am by Huan |