The advent of graphical interfaces such as Windows and the MacOS was rightly hailed as an advance in usability, but as anyone who ever used Unix or pre-Windows DOS will know they could make some tasks more cumbersome and less natural to perform. You could, for instance, tell DOS with a single command to list to a file all filenames in a particular folder containing the letters xkz; try doing that in Windows.

One of the least-appreciated aspects of speech-recognition programs is that they have brought back the command-line in a new form. Instead of writing near-English commands, with obscure and daunting modifiers, you can tell your program verbally what to do - which is far more natural than clicking a mouse.

They show that commands can make things easier even for that majority of people who would run a mile at being asked to cope with anything that looks like a line of code. Take, for instance, the task of setting up four-column-by--row table in Word 2003. First you need to know the difference between Draw Table and Insert on the menu; then you have to complete a dialogue box. With a spoken command you need to know nothing about the menu structure; you just say something like: Insert 10 by 4 table.

Natural-language processing has come a long way since DOS days so it should be possible for commands to be closer to the way we speak. There are problems, such as possible security breaches or data loss stemming from ambiquities. Firefox is making a start with its about-to-launch Ubiquity command interface. It looks like a project well worth watching.