Figure 1.

(A) Task as decision. The current model (deduced from the literature), where an NLP task (eg, classification, entity recognition) serves a discrete function that relates to a modeling objective, but is not explicitly designed to assist clinician decision-making. Dotted lines represent permeable boundaries; solid lines impermeable. (B) Task as need. The alternative model proposed by this paper, where one or more NLP tasks are directly designed to interact and contribute to providing evidence to support clinical decision-making. Dotted lines represent permeable boundaries.