Computational research is exploring more complex emotion-related phenomena, but lacks models of emotion that accommodate them naturally. A powerful approach assumes that the directly experienced phenomena point to ‘engines of emotion’, which have ongoing functions that we only partially register. Theory has proposed five broad functions: evaluating situations; preparing us to act accordingly, at multiple levels; ensuring that we learn from significant situations; interrupting conscious processes when necessary; and aligning us with other people. Emotional feelings inform conscious awareness of what they are doing, and emotion words split the space of their activity into discrete regions. The natural goal for computation is not to duplicate those forms: it is to describe what the ‘engines’ are doing.