Embodiments relate to systems and methods for predictive text entry for small touch-screen devices. A two-state input interface (116) of a device (102) can include a seek area (120), which displays a subset of letters representing the full range of available letters. The user can select a first letter in that subset. A prediction engine (160) can generate a next most-likely letter (162) based on the currently inputted letter, using look-up tables or other techniques based on letter sequence probabilities. The user can respond with a right or left-swipe gesture, causing the prediction engine (160) to seek the most-likely next letter upstream or downstream in the alphabet. The user can also respond by touching the seek area causing the prediction engine (160) to seek the next predicted letter closest to the touch point. The prediction engine (160) can also generate a predicted completed word, for convenient user selection.