Kneser-Ney smoothing
At ICASSP-95 in Detroit (9-12 May 1995) Reinhard Kneser of Philips' research laboratories in Aachen and Hermann Ney of Aachen proposed backing-off distributions for n-gram language models optimised for exactly the cases where the longer sequence had not been seen. Perplexity fell by about 10%, and the word error rate in speech recognition by 5%.
Why it matters
N-gram language models for speech recognition and translation gained a way of estimating the unseen that later proved the best of those compared. In Chen and Goodman's explanation of 1998, a word's probability at the lower level should depend not on how often the word occurs but on how many different words it follows.
Backing-off means falling back to a shorter history when the longer n-gram is absent from the training data. Kneser and Ney derived distributions for this step in two ways, which, per the abstract, are quite different from the usual ones. Chen and Goodman's example: Francisco is frequent but occurs only after San, so after an unfamiliar word it is unlikely, however common it is overall. What the record does not claim. The full text was not read: IEEE Xplore shows only the abstract, which is where the 10% and 5% come from, and the record does not know which corpora they were measured on. The "how many different preceding words" mechanism is described in Chen and Goodman's words, not the paper's own.