| |
|
Faculty
Advisor
|
TBD |
|
|
|
Machine Translation is still an open problem in Artificial Intelligence (AI), with important applications in practice. We will explore the idea of translating text from one language to another by having people play a game. The crux of this project relies on the observation that a version of "Multilingual Pictionary" can be used to partially translate text from one language to another. Multilingual Pictionary is the following game: Two players who speak a different language play Pictionary, with one of the players getting a word or phrase in his/her native language, and having to sketch a picture so that the other player guesses the word or phrase in their own native language (different from the drawer's native language). By having players play multilingual Pictionary, we can obtain valuable translation data from one language to another. The most striking part of this approach is that translation is done by two people playing a game, none of whom knows both languages and none of whom feels like they are doing translation. We will explore this idea and the related idea of using Pictionary to solve the word-sense disambiguation problem.
The REU student will help to build prototype games, as well as think about the different components the game should have. The student will learn about open problems in Artificial Intelligence and why current techniques don't work well to solve them. The student will also learn how to design interactive systems that entice people to contribute to the solution of the problems at hand, the importance of usability, and how to design games for which adversaries cannot cheat. The most important prerequisite is strong experience in programming Java.
|
|
CENTER
Carnegie
Mellon 
Aladdin
Back
to the REU page