3 ms·

Current input methods are based on large corpora and use context to disambiguate. They are surprisingly smart.