Opponent Modeling in Scrabble

Abstract

Computers have already eclipsed the level of human play in competitive Scrabble, but there remains room for improvement. In particular, there is much to be gained by incorporating information about the opponent's tiles into the decision-making process. In this work, we quantify the value of knowing what letters the opponent has. We use observations from previous plays to predict what tiles our opponent may hold and then use this information to guide our play. Our model of the opponent, based on Bayes' theorem, sacrifices accuracy for simplicity and ease of computation. But even with this simplified model, we show significant improvement in play over an existing Scrabble program. These empirical results suggest that this simple approximation may serve as a suitable substitute for the intractable partially observable Markov decision process. Although this work focuses on computer-vs-computer Scrabble play, the tools developed can be of great use in training humans to play against other humans.