How a computer should talk to people

I would now like to summarize briefly what has been discussed. I begand by stating that one cannot produce good messages by just letting them grow like weeds during program development, then rewording them after the fact to make them read well. My experience and reflection have convinced me that, to create messages that work for their intended readers, we must do a number of things: Set human goals for messages. We should first commit ourselves to certain goals. A major goal is to be tolerant of “user errors.” We have to decide how much error-correction the program will do. By anticipating the kinds of mistakes humans will make, we can identify the messages needed to account for those mistakes, as well as for error-free operation.