Skip to main content

Table 4 Perplexity on 70K word as test from Arabic Open Corpus using different smoothing techniques against proposed algorithm

From: Enhancing recurrent neural network-based language models by word tokenization

Model Perplexity Entropy reduction (%)
GT5 113.473
KN3 99.1785 2.85
KN5 98.9021 2.9
Basic RNN 70.58 10.04
Proposed model 68.42 10.69
  1. Italic values represent proposed model perplexity and entropy reduction against different smoothing techniques