In an approach to text prediction, a computer detects an initialization of a text-based communication between a first user and a second user. The computer determines whether a predictive text dictionary associated with the second user exists, wherein the predictive text dictionary is based, at least in part, on a history of communication between the first user and the second user. Responsive to determining a predictive text dictionary associated with the second user exists, the computer receives the text-based communication. The computer determines whether one or more words are detected in the text-based communication that do not exist in the predictive text dictionary associated with the second user. Responsive to determining one or more words are detected that do not exist in the predictive text dictionary associated with the second user, the computer adds the one or more new words to the predictive text dictionary associated with the second user.