The zero-frequency problem: Estimating the probabilities of novel events in adaptive text compression

By Ian H. Witten, T. Bell. 1991

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Learning API Usages from Bytecode: A Statistical Approach (2015, methodology): “We also implemented Witten-Bell smoothing [43] technique for this model.”
Two decades of statistical language modeling: where do we go from here? (2000, methodology): “These include discounting the ML estimates [14, 15 ], recursively backing off to lower order -grams [16, 17, 18], and linearly interpolating -grams of different order [19].”
On Prediction Using Variable Order Markov Models (2004, background): “As noted by Witten and Bell (1991) there is no principled justification for any of the various ppm escape mechanisms.”
Recovering Traceability Links between Code and Documentation (2002, unknown): “whenever any word wk is not present in the document Di. This problem, known as the zero-frequency problem [ 55 ], can be avoided using different approaches (see [21]).”
IRSTLM: an open source toolkit for handling large scale language models (2008, methodology): “In the current version, a standard interpolation scheme is considered in combination with two well-established smoothing techniques, namely the Witten-Bell method [10] and a simplified version of the improved Kneser-Ney method [8].”
Where should the bugs be fixed? More accurate information retrieval-based bug localization based on bug reports (2012, background): “SUM smoothes the probability distributions by assigning non-zero probabilities to the unseen words [17, 36].”
Dialog Act Modeling for Conversational Speech (1998, methodology): “We trained standard backoff N-gram models (Katz 1987), using the frequency smoothing approach of Witten & Bell (1991).”