The predictive autocomplete function on our phones is both very stupid in human terms, and very smart. The latter is because of a simple programming routine, the former because human language and thought are incredibly complex and occasionally bizarre.
For example, if I type in…
…anyone who’s spoken English from birth will KNOW that the extension of “smaller than a bread___” is “breadbox!” But there’s a lot of nuance and context in there, along with about a hundred zillion colloquialisms, not to mention the possibility of “close but no cigar” usage by someone who doesn’t know the exact phrase (“Excuse me, while I kiss this guy”) and people who are deliberately mixing things up (“it’s not brain science”). This is why IBM took years and hundreds of millions of dollars to create Watson.
On the other hand, it’s relatively easy to watch for words which are not in the lexicon but are used repeatedly. This lets your app “learn” and add initials, proper nouns, names, and so on to its stored dictionary.
So after a few days now, I get…
…”Trumpencritter” from “Trum.”
That’s going to save a LOT of time over the next few months.