Five Challenges for Intelligent Text Entry Methods

ing feature of civilization. The continual improvements and refinements of writing systems throughout history are a testament to their importance. In the last 10,000 years, the media of writing systems have changed from cave walls, rune stones, and sand to papyrus, wax tablets, parchment, paper, and—in our age—networked computers. Carvings, pieces of scribble, and elaborate texts have been generated from tools ranging from stones, iron rods, and fingers to styli, quills, pens, typewriters, and keyboards. In parallel the very heart of writing systems has also changed and adapted. The original logographic writing systems, such as the hieroglyphs, have throughout the course of Western civilization gradually been replaced by refined alphabetic systems. Between the 16th and 19th centuries the need for fast and accurate transcriptions of proceedings in courts and parliaments motivated research into various shorthand systems, culminating in the Pitman and Gregg shorthand writing systems that are still with us today. This article is about the latest generation of writing systems— intelligent text entry methods. Such text entry methods use AI techniques to improve our writing systems. The high-performing speech and handwriting recognition systems we have available today would be impossible without decades of research and optimization. However, there is more to these systems than engineering. I explain why research on intelligent text entry methods is important in the next section. Then I sample selected text entry methods and set them into their historical context. I argue that for text entry methods to make a difference in society, we need to leave the narrow objective functions of high entry rates and low error rates. To that end I raise five challenges Articles