Modeling the speed of text entry with a word prediction interface

This study analyzes user performance of text entry tasks with word prediction by applying modeling techniques developed in the field of human-computer interaction. Fourteen subjects transcribed text with and without a word prediction feature for seven test sessions. Eight subjects were able-bodied and used mouthstick typing, while six subjects bad high-level spinal cord injuries and used their usual method of keyboard access. Use of word prediction decreased text generation rate for the spinal cord injured subjects and only modestly enhanced it for the able-bodied subjects. This suggests that the cognitive cost of using word prediction had a major impact on the performance of these subjects. Performance was analyzed in more detail by deriving subjects´ times for keypress and list search actions during word prediction use. All subjects had slower keypress times during word prediction use as compared to letters-only typing, and spinal cord injured subjects had much slower list search times than able-bodied subjects. These parameter values were used in a two-parameter model to simulate subjects´ word entry times during word prediction use, with an average model error of 16%. These simulation results are an encouraging first step toward demonstrating the ability of analytical models to represent user performance with word prediction