Predictionary

When reading particularly elevated or technical prose, I sometimes run into words I can't define. When this happens, I try to glean the word's definition from context, and try to remember to look it up in the dictionary later.

I never, ever remember to look up definitions later. The predictionary solves this problem before I even start reading the book.

The predictionary, on first activation, quizzes its user on a list of, say, fifty words. Based on the user's responses, the predictionary is given a rough estimate of his vocabulary level. When the user prepares to read a book which may contain challenging vocabulary, he feeds its etext into the predictionary, which returns a list of all words it calculates as likely to give the user trouble, with their definitions.

The predictionary also serves as a traditional dictionary; the user's searches, along with his choice of reading material, provide additional feedback that allows the device to continuously refine its estimate of the user's vocabulary level.

Networked predictionaries share data on commonly-requested definitions, correlating data in a helpful manner ("users that know `apogee' are likely to also know `nadir'; users that don't know `gravitas' will likely need help with `non sequitur'").