Predictive Disambiguation

Predictive disambiguation is the class of text-entry techniques in which each user input event is intentionally ambiguous (one keystroke covers several possible letters, one swipe covers many possible paths) and software resolves the ambiguity into a most-likely word using a dictionary and a language model. T9, swipe-to-type keyboards, and many switch-scanning and gaze-typing systems all rely on predictive disambiguation. Because each keystroke carries more information under disambiguation, the approach is particularly valuable for users whose per-keystroke cost is high — users of switch input, eye-gaze keyboards, AAC devices, or mobile keypads with few keys. Modern predictive-disambiguation systems use neural language models that rank candidates by semantic and syntactic context in addition to raw word frequency.