Abstract
A common idea in models of action representation is that actions are represented in terms of their perceptual effects (see e.g., Prinz, 1997; Hommel et al., 2001; Sahin et al., 2007; Umiltà et al., 2008; Hommel, 2013). In this paper we extend existing models of effect-based action representations to account for a novel distinction. Some actions bring about effects that are independent events in their own right: for instance, if John smashes a cup, he brings about the event of the cup smashing. Other actions do not bring about such effects. For instance, if John grabs a cup, this action does not cause the cup to “do” anything: a grab action has well-defined perceptual effects, but these are not registered by the perceptual system that detects independent events involving external objects in the world. In our model, effect-based actions are implemented in several distinct neural circuits, which are organized into a hierarchy based on the complexity of their associated perceptual effects. The circuit at the top of this hierarchy is responsible for actions that bring about independently perceivable events. This circuit receives input from the perceptual module that recognizes arbitrary events taking place in the world, and learns movements that reliably cause such events. We assess our model against existing experimental observations about effect-based motor representations, and make some novel experimental predictions. We also consider the possibility that the “causative actions” circuit in our model can be identified with a motor pathway reported in other work, specializing in “functional” actions on manipulable tools (Bub et al., 2008; Binkofski and Buxbaum, 2013).
Highlights
A common idea in models of action representation is that an agent’s actions are encoded in a way which makes reference to the sensory effects they bring about
An important possibility to consider is that the distinction we propose between networks for causative and non-causative actions coincides with the distinction made in earlier work between a dorso-dorsal “reach/grasp” pathway and a ventrodorsal “use” pathway
The reach network performs well after training; its performance is described in Lee-Hand et al (2012); we discuss the performance of the simple action and causative action networks after training
Summary
A common idea in models of action representation is that an agent’s actions are encoded in a way which makes reference to the sensory effects they bring about. The key idea uniting these models is that motor programs are not defined purely within the motor domain: their neural representation includes a representation of the effects they are expected to have on the world, as apprehended by the perceptual system This idea has been supported in a variety of experiments, and modeled computationally in a number of different ways, as we will summarize below. If John smashes a cup, he brings about the event of the cup smashing This is an event which in other circumstances could happen independently of any action of John’s: it involves the cup changing state in a certain way. On the other hand if John touches or grasps a cup, he does not bring about any event involving the cup that is independent of his own action. “to grasp a cup” doesn’t mean “to cause the cup” to do anything
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have