Sequence Modeling for Intelligent Typing Assistant with Bangla and English Keyboard

Typing Assistant is a helping tool that autocompletes writing, both by completing current words and suggesting relevant word sequences. Smart typing assistant is the one that can suggest more than just the current word. Based on the context, it is able to suggest the next few words that are relevant. In this work, we propose such a model using the recent advancement of the recurrent neural network. It is a context-aware model that can predict the next few words conditioning on the context user is writing. This is a general purpose solution to typing assistant system. Our work stands out from other research in two cases. First, rather than using only a temperature variable to introduce randomness in the generation and choosing only one character, we used beam search to generate multiple words suggestion. Second, we used encoder-decoder architecture to generate more than the one-word suggestion. Often more than one word of this suggestion is relevant enough to choose and that makes the typing more efficient. We tried it for both Bangla and English. In both cases, it gave impressive suggestions during typing.