How from action-mirroring to intention-ascription?

This paper is devoted to an assessment of the three-step model offered by Gallese and colleagues in support of the thesis that the function of the mirror mechanism is to mindread an agent's intention. The first step of the model is the acceptance of the direct-matching model of action understanding. The second step is the endorsement of a different model of mirror neuron activity, i.e. the model of chains of logically related mirror neurons (or motor chains) whose application to action-mirroring is supposed to show that the mirror mechanism enables an observer to predict the goal of the agent's forthcoming action. The third step is the endorsement of the 'deflationary' account of intention-ascription according to which to ascribe an intention to an agent is to predict the goal of the agent's forthcoming action. I argue that each step of the model faces insuperable objections.