17
Voted
ACL
2008
14 years 4 months ago
2008
Frequency counts from very large corpora, such as the Web 1T dataset, have recently become available for language modeling. Omission of low frequency n-gram counts is a practical ...