Fig. 2
From: Enhancing recurrent neural network-based language models by word tokenization
Effect of corpus size on the network input layer neurons number. Recurrent neural network-based language models input layer size depends on the vocabulary words number, training time and memory cost increased remarkably as the number of neurons in input and output layers becomes big number. This figure illustrates the relationship between corpus size and the recurrent neural network input layer size, as shown as we add more words to the corpus it will cause vocabulary expansion and as a result recurrent neural network input layer size will increase which will downgrade the network performance.