Aurora's word prediction also supports grammatical rules, although we do not recommend their use for most users. Below you can see the rules in action, they know for example that people do not often use two inquisitives in a row like "what what" or "who who." You can see when you type the first "w" you get words like "who," "what," "where," "when," and "why." After you select one of these words, and type the "w" for would, the grammatical rules cause Aurora Prediction not to show those inquisitives again.
We have done testing with grammatical rules in our products and in others, and we have found that they are generally more of a hindrance than a benefit. The problem lies in the English language itself which is full of exceptions and special cases. Current grammatical systems are simply unable to adequately describe such a complex language.
What we have found is that in all the systems we have tested, when grammatical rules are used, the word prediction programs often will not offer you the word you are typing in the prediction window to help you finish it, even when the word you are typing is grammatically correct (despite what the rules may believe.) This results in your having to type out the entire word. The problem is that the grammatical rules think you are making a mistake so they don't help you finish the word you are typing, often giving you a completely empty word prediction window. Unfortunately, with current grammatical rule systems, this happens more often than it should.
At the moment, we believe our statistical word prediction algorithms are much more effective than grammatical rules at offering users the correct words since they are based on a large sample of the English language rather than a set of general rules.
Despite their problems, we think grammatical rules have great potential to help our users so we continue to work at improving them.