A probabilistic line breaking algorithm

Abstract

We showh ow a probabilistic interpretation of an ill defined problem, the problem of finding line breaks in a paragraph, can lead to an efficient newalgorit hm that performs well. The graphical model that results from the probabilistic interpretation has the advantage that it is easy to tune due to the probabilistic approach. Furthermore, the algorithm optimizes the probability a break up is acceptable over the whole paragraph, it does not show threshold effects and it allows for easy incorporation of subtle typographical rules. Thanks to the architecture of the Bayesian network, the algorithm is linear in the number of characters in a paragraph. Empirical evidence suggests that this algorithm performs closer to results published through desk top publishing than a number of existing systems.