Archive for September, 2002

Dasher – a new input method

Monday, September 2nd, 2002

This week’s Economist reports on a new innovative input method invented by a group at Cambridge. Dasher is something you really have to see to understand, but once you do, its effectiveness hits you immediately, especially for mobile devices.

The Paradox of Mobile Devices
The problem lies in the paradox of mobile device design. This paradox arise from two competing optimizations: you need to make the device large enough to house a readable screen and a reliable text input system, but you also want to make the device small enough to carry around effortlessly.

Jeff Hawkins’ Palm solved the problem with his invention of Grafitti; RIM’s Blackberry device innovated with its thumbboard, a mini QWERTY keyboard that added little size to the pager.

The thumbboard seems to be the future of mobile devices, at least in the near future. Handspring has moved in its direction in its treo devices, and has completely ditched Grafitti. But what if the space occupied by this thumbboard could be used for display?

Dasher may be on to something. It’s first novel feature is that a keyboard isn’t needed at all. Dasher uses a language model to predict the most probable next letter you’re about to enter, based on what letters you’ve already entered. And by moving a cursor with the mouse, you guide Dasher through the choices.

Yeah, it’s hard to describe. But it’s easy to understand once you’ve seen it.

The Dasher project includes Windows, Linux, and Pocket PC versions of its open software. I intend to try the Pocket PC version and see if it’s a viable alternative for text entry.

Besides the benefits, there are some very interesting aspects of Dasher that certainly demand further investigation:

(1) Claude Shannon’s observation that the entropy of English is about 1 bit per character. Coupled with Dr. MacKay’s observation that the keyboard, with 80 keys, is capable of log2(80) = 6.3 bits per gesture, and therefore is 6 times inefficient. A better system would help limit choice based on the range of valid possibilities given the language – and Dasher does this with its use of a predictive system.

(2) Compression as a way of prediciting text input. Dasher uses PPM5D+, a variant of PPM, or prediction by partial match.

I hope to see some improvements in Dasher’s capabilities to make it even more usable, things like:

punctuation and capitalization, and a better integration with other applications (currently it’s a standalone app). But these are usability, not core innovation. Dasher is cool and fun.