Text prediction techniques based on the study of constraints and their applications for intelligent virtual keyboards in learning systems

Abstract

This paper aims to observe how one can apply fixed and probabilistic constraints on information sources and to emphasize the benefits obtained from this process for learning systems. Constraints applied may be both fixed, that are initially established with rigorous formulation, and probabilistic, in which case the process becomes an adaptive one, using statistical properties of the input symbols. Using fixed constraints, one can determine the fixed probabilities of appearance for letters or words and using probabilistic constraints one can determine the relationships between letters or words in a row by building lists of probabilities containing conditions allowed for transitions that are based on certain letters prefix or reference to a particular word. In order to illustrate the usefulness and applicability of these constraints predictive techniques of text insertion will be presented, as well as how they are implemented in applications such as predictive keyboard for smartphones or portable devices.