Output details
11 - Computer Science and Informatics
University of St Andrews
Prefix Probability for Probabilistic Synchronous Context-Free Grammars
<22>Probabilistic synchronous context-free grammars are a central concept in modern statistical machine translation. Whereas prefix probabilities have obvious applications for pruning the search space, and thereby speeding up parsing or translation, there had been no attempts to use them for synchronous context-free grammars, as the effective computation seemed too complicated. This paper introduces an entirely new way of looking at the problem, which in addition carries over to several related formalisms. The crux of the solution is in reducing the problem to a number of much simpler problems, so that effective computation becomes straightforward.