Next word prediction for phonetic typing by grouping language models

In this paper, we present a language model based framework for instant messaging, that can predict probable next word given a set of current words. Our goal is to facilitate the task of instant messaging by suggesting relevant words to the user. Generally, at the time of sending personal messages, a user follows a specific style of communication with a specific group of people. This phenomenon is much more evident in other languages apart from English. For example, in Bengali language, there are three counterparts of you, that is used to address the second person in English. Considering this fact, there are at least three styles of writing texts in Bengali language: informal, semi-formal and formal. Therefore, it is quite necessary to generate next words based on the linguistic style adopted by a user, when sending messages to a specific set of people. In this paper, we develop a solution to this issue by adopting different language models when exchanging messages with different groups of people. Our method clusters language models based on user interactions, and we show the effectiveness of our method using a popular metric hit ratio. This model can be widely adapted for predicting next words in smart-phone devices and expedite the communication between users, specifically at the time of phonetic typing.