ACL
2000
13 years 7 months ago
2000
We propose a distribution-based pruning of n-gram backoff language models. Instead of the conventional approach of pruning n-grams that are infrequent in training data, we prune n...