Research
My overall research interest is to understand how pattern recognition and machine learning methods can enable new user interfaces that are more efficient, fluid and fun to use.
Research interests
- human-computer interaction
- pattern recognition
- information visualization
- end-user programming
Intelligent text entry methods
I have a long-standing research interest in efficient text entry. Efficient text entry may sound like an unusual research interest. After all, the desktop keyboard is relatively efficient. 
My Ph.D. However, there are many situations when the desktop keyboard is less than ideal. For example, in various usage scenarios in mobile text entry lend themselves toward different solutions. For example, if you are sitting in a lecture hall you would desire an absolutely quiet text entry method. However, if you are walking towards your airport gate you may want to use a speech recognizer.
While mobile text entry dominates contemporary efforts, intelligent text entry methods may be necessary in other scenarios as well. Esoteric devices, such as wall-sized displays and eye-trackers, also require efficient text entry solutions.
Text entry methods can be made intelligent by realizing that human languages are highly redundant. The trick is to find out exactly how to take maximum advantage of this redundancy. When optimizing a user interface it is easy to forget that humans aren't machines. For example, text entry methods that are information-efficient in theory need not necessarily provide high entry rates in practice. To create truly efficient intelligent text entry methods (and user interfaces in general) you need to have at least a reasonable amount of understanding of how humans process information and articulate their intentions. In practice, this means intelligent text entry research is highly interdisciplinary and involves, for example, experimental psychology, computer science, and engineering-driven design.
I have been involved in developing completely novel text entry methods, such as ShapeWriter and the "elastic stylus keyboard" (with Shumin Zhai). The former is probably the fastest pen/finger-based text entry method in the world. At Cambridge I have been involved in the development of Parakeet (with Keith Vertanen), which is a speech recognition system for mobile touch-screen devices. Using Parakeet we conducted the first user study with true continuous speech recognition on the mobile device. 
Parakeet. It appears mobile speech recognition can be very fast if the system implements a good error correction interface (errors are unavoidable in speech recognition).
When I am not trying to invent new better systems I study aspects of existing ones. For example, I have studied the text entry performance of handwriting recognition, which, amazingly, no one appears to have done properly before. Instead, researchers have been citing (often via secondary or even tertiary sources) an old inappropriate study from 1967. It turns out that modern handwriting recognition enables the same text entry rates as the standard QWERTY software keyboard. This finding contradicts the literature, which has stated that handwriting recognition is much slower.
I view my work as being part experimentally hypothesis-driven and part creative/engineering-driven. You cannot hope to make breakthroughs without having some intuition of what works and what doesn't.
ShapeWriter
ShapeWriter is probably the biggest research result I have been part of so far. It was previously known as Hybrid Shorthand Keyboard (HSK) early on in my final thesis work in 2001-2002, and as SHARK and SHARK2 in research publications and press articles earlier than 2007. 
With ShapeWriter users write text by sliding a pen or finger over a touch-screen software keyboard. All words in a large lexicon (tens of thousands of words) form shapes on the software keyboard.
Each individual shape is created by tracing out an individual word in the lexicon on the keyboard. For example, the word "the" is traced out on the keyboard by connecting the centre point of the "t" key to the centre point of the "h" key and the centre point of the "e" key. These two inter-connected lines form the shape that represents the word "the". The system then finds the user's intended word by pattern matching the user's input gesture against all these shapes formed by all the words in the lexicon.
Unlike many other intelligent text entry methods ShapeWriter enables a smooth transition from novice to expert user. Novice users trace out words by sliding from letter-to-letter (analogous to "hunt and peck" with a traditional desktop keyboard). After a while the shapes of common words build up in users' motor memory. From that point on, users can quickly recall the shapes for their intended words by not looking much at the keyboard. 
ShapeWriter in action. © 2005 Handelsblatt. This is what enabled ShapeWriter to deliver a peak entry-rate of 99 wpm in a controlled user study (mean entry rate was of course lower, around 45 wpm). ShapeWriter is the earliest text entry method I am aware of that represents words as geometrical shapes defined according to a keyboard topology (first publication is my final thesis in 2002).
ShapeWriter has received considerable attention in international press, such as Time, New York Times, BBC News, Die Zeit and many others (currently there are well over 100 press articles in English, French, German, Italian, Japanese, Russian, Vietnamese, Danish, Norwegian, Finnish, Swedish and other languages). To my knowledge ShapeWriter was first reported in the international press in San Jose Mercury News in April 2003. ShapeWriter is currently commercialized by ShapeWriter, Inc. - a technology start-up I co-founded in 2007.