Indeed, I was thinking that generating a chording dictionary based on an optimal number of bits of information that would reduce the error rate on typos could potentially speed up overall throughput.