Toward Expert Typing in ACT-R


This paper describes an effort to integrate the TYPIST theory of expert transcription typing into the ACT-R cognitive architecture. Our goal is to strike a reasonable balance between a match to the highly accurate predictions of TYPIST and the architectural constraints imposed by ACT-R. The model we have built provides good predictions of human performance on most basic typing phenomena, though less accurately than TYPIST. We present the design of the model, a description of software to support model execution and experimentation, and the results of performance tests comparing the model's predictions with human typing data in the literature.