We describe a new approach for rescoring speech lattices — with long-span language models or wide-context acoustic models — that does not entail computationally intensive lattice expansion or limited rescoring of only an N-best list. We view the set of word-sequences in a lattice as a discrete space equipped with the edit-distance metric, and develop a hill climbing technique to start with, say, the 1-best hypothesis under the lattice-generating model(s) and iteratively search a local neighborhood for the highest-scoring hypothesis under the rescoring model(s); such neighborhoods are efficiently constructed via finite state techniques. We demonstrate empirically that to achieve the same reduction in error rate using a better estimated, higher order language model, our technique evaluates fewer utterance-length hypotheses than conventional N-best rescoring by two orders of magnitude. For the same number of hypotheses evaluated, our technique results in a significantly lower erro...