Skip to main content
Fig. 4 | Human-centric Computing and Information Sciences

Fig. 4

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

Fig. 4

The proposed recurrent neural network-based language model architecture with input layer segmented into three components: the prefix, the stem and the suffix. Work in this paper suggests splitting the input layer of the recurrent neural network-based language models to contains three parts of the word; prefix part, the word stem part and the word suffix part, where each part of the word is presented to the network using a 1-of-n encoding.

Back to article page