Content area
Abstract
Text messaging is a rapidly growing form of communication that is still in its initial stages of optimization. Many text message users complain of poor text prediction and inadequate dictionary management techniques on their mobile phones. To help improve the user’s experiences when using this method of communication, we have developed an improved dictionary-based method for text input on cell phones using bigram language models. We have collected our own corpus of text messages and formatted them properly in order to train the models. Experiments described in this thesis demonstrate that our system shows significant improvement over current commercially available systems with regards to the average number of key-presses required per character, disambiguation between words having the same numeric codes, and dictionary management, as well as user evaluation.
