Mini reviewMirror neurons: Functions, mechanisms and models
Highlights
► In the literature various functions and supporting mechanisms are attributed to MNs. ► Many of the functions are not observed in monkeys: look for evolutionary explanation. ► The distinction between a brain function and mechanism must be made clear. ► Computational models can be used in clarifying mechanisms that can support MNs.
Introduction
Mirror neurons for manipulation fire both when the animal manipulates an object in a specific way and when it sees another animal (or the experimenter) perform an action that is more or less similar. Such neurons were originally found in macaque monkeys, in the ventral premotor cortex, area F5 [21], [32], [76], and later also in the inferior parietal lobule [28], [78]. Not all visuomotor neurons in F5 show the mirror property. There are F5 visuomotor neurons that selectively discharge to the visual presentation of a given object, which also discharge selectively during grasping of that object [61]. These neurons are called canonical neurons, and are believed to play a crucial role in transforming visual appearance of objects into motor plans for interacting with them [77]. Area F5 also includes auditory mirror neurons [54] that respond not only to the view but also to the sound of actions with typical sounds (e.g. breaking a peanut, tearing paper). Although classically, the actions associated with mirror neurons in the monkey are considered to be transitive, i.e. the action has to be directed to an object, which may be removed from the view of the monkey before hand-contact [e.g. 95], more recent data indicates that transitivity may not be a prerequisite for mirror-like activity (see also Section 7).
Recent neuroimaging findings indicate that the adult human brain is endowed with a system for matching the observation and execution of actions which might be homologous to the macaque mirror neuron system [10], [73], [77]. In spite of the growing number of human brain imaging data related to posited ‘human mirror systems’ [6], [13], [25], [43], [44], [83], the experimental data on mirror neurons are available mainly for monkeys as systematic recordings using electrophysiology cannot be used to investigate the human brain. Therefore it must be emphasized that in humans a mirror system refers to a brain region (or set of brain regions) that becomes active for both observation and execution of a class of actions.
In the literature a set of functions is attributed to monkey and human putative mirror neuron system, and several terms are used to describe mechanisms underlying these functions. However, many of these functions are observed in human but not in monkeys, thus suggesting evolution within the mirror systems or within the wider networks of which they are part. Among these are imitation [13], [58], action understanding [e.g. 95], intention attribution [43] and (evolution of) language [74]. Recently, reviews and meta-analyses that are critical of the claimed mirror neuron functions have started to appear, in particular with the focus on the ambiguity of the terminology used to describe mirror neuron functions such as direct matching and motor resonance [17], [19], [23], [92]. In part such failures to unambiguously describe mirror neuron function, follow from ignoring the distinction between a brain function and a possible mechanism (Box 1). In the majority of mirror neuron literature, functions associated with a mirror system in humans are attributed to “direct matching” or “motor resonance” and sometimes with “motor simulation” as a mechanism to underlie action/intention understanding [34], [78] and theory of mind [33] without either a precise definition of such a “mechanism” nor a clear account of how it contributes to the observed function. It is simply assumed that mirror systems are involved in this property. However, we know (by extrapolation from the macaque) that a mirror system will contain many types of neurons other than mirror neurons. Thus, when a brain imaging study reports increased activity in a mirror system for some task relative to a control it is a mistake (all too frequent in the literature) to assume that activity in mirror neurons underlies the activation – in some cases, it will be, but by no means in all.
Our task in this article is to make the case for the use of computational models, whether in terms of neural networks or higher level constructs such as control systems, in clarifying mechanisms that are sufficient to explain observed phenomena involving mirror systems. The mechanisms we demonstrate might not be the same with those employed by the brain, but their very precision sets the stage for experiments more precise than those guided by claims like “a motor resonance supports the observed function.” To be concrete, we will be interested with these questions:
- •
Is the postulated mechanism sound from a computational point of view?
- •
To what extent can a brain implement the postulated mechanism?
- •
What are the additional mechanisms needed on top of the claimed mechanisms to yield the posited functions?
Section snippets
Direct matching
To evaluate the direct matching or motor resonance hypothesis, we must first clarify what is matched (decoded)? and how is the matched action encoded by the mirror neurons (MNs).
Action understanding and direct matching
Mirror neurons, when initially discovered in macaques, were thought to be involved in action recognition [30], [32], [76]. Although the term “action understanding” was often used, the exact meaning of “understanding” as used is not clear. In fact, the neurophysiological data simply show that a mirror neuron fires both when the monkey executes a certain action and when he observes more or less congruent actions. We think at minimum “understanding” includes the ability of an organism to
But MNs are active during execution of the monkey's own actions
Our conclusion from the above discussion is that it is not meaningful to focus on MNs in isolation but their (possible) role in an understanding network must be the target for research, for which the current anatomical and behavioral data falls short of giving a full picture. However, even this broader discussion relates – as does almost all the extant literature on mirror neurons – to the activity of MNs during observation of another's actions, what about the activity of MNs during the
Simulation theory and direct matching
We return to the emphasis on interpreting MNs in the context of observing the actions of others. Gallese and Goldman [33] suggest that the purpose of MNs is to enable an organism to detect certain mental states of observed conspecifics via mental simulation. According to this view, mirror neurons could be the precursor of mind-reading ability, being compatible with the simulation theory hypothesis according to which mental states of others are represented by representing their states in terms
Distinguishing computational and conceptual modeling
In this section we emphasize the difference of conceptual and computational models. At one extreme the statement “an observed action is represented in motor terms” may define a conceptual model but in developing a computational model, a set of strict and unambiguous specifications must be followed. In particular, the notions of what is matched and how it is represented as in Section 2 must be precisely defined. Furthermore the observation must be also explained in the model specification: is it
Internal models and mirror neurons
Although we have indicated that direct matching (corresponding to inverse modeling at some control level) is the prime focus of the neuroscientific community for mirror neurons, there are a few proposals that involve MNs in forward modeling [13], [44]. The idea being the mirror code is used to generate a visual prediction in the Superior Temporal Sulcus (STS) where neurons have been found with selectivity for biological movement (e.g. of arms, whole body) for comparison. Miall [58] suggested
Development of mirror neurons
Despite the growing number of reports on adult mirror neuron systems, data on development of the mirror system for both human and other primates is scarce. It is known that the macaque mirror neuron system is adaptive: responses to novel actions can be acquired through repeated experience. For example, some mirror neurons respond to both execution and viewing of ripping a sheet of paper, which is not in the ecological repertoire of wild monkeys [54]. Furthermore, as we have seen, some mirror
Evolution of the language-ready brain
The mirror system hypothesis (MSH) of the evolution of the language-ready brain has developed over the years [2], [3], [74] to embrace a wide body of data. This is not the place to review those developments or the details of the current version of the model. Instead, we simply summarize the stages in biological evolution of the brain that the hypothesis posits, noting that it progresses from a mirror system for manual actions posited for our last common ancestor with monkeys 25 mya by a posited
Discussion
Our discussion so far indicates that the key to understanding the function of MNs and the mechanisms that facilitate that function is governed by our knowledge on the coding of MNs and how evolution changed and augmented MNs.
What do MNs encode (during action execution) and decode (during action observation)? As brain imaging cannot offer much help in this endeavor due to the gross temporal and/or spatial smearing of the neural activity; we hold that most direct information can be obtained via
Acknowledgement
This material is based in part on work supported by the National Science Foundation under Grant No. 0924674 (Michael A. Arbib, Principal Investigator). We thank Prof. Akira Murata for his useful comments and pointers on the anatomical connections of area F5. Author MK was supported by the Strategic Research Program for Brain Sciences of Japanese MEXT.
References (99)
- et al.
Learning human arm movements by imitation: evaluation of a biologically inspired connectionist architecture
Robotics and Autonomous Systems
(2001) - et al.
Sensorimotor learning configures the human mirror system
Current Biology
(2007) - et al.
From self-observation to imitation: visuomotor association on a robotic hand
Brain Research Bulletin
(2008) Human cortex: reflections of mirror neurons
Current Biology
(2008)- et al.
A mirror up to nature
Current Biology
(2008) - et al.
Modeling parietal-premotor interactions in primate control of grasping
Neural Networks
(1998) - et al.
Mirror neurons and the simulation theory of mind-reading
Trends in Cognitive Sciences
(1998) - et al.
A unifying view of the basis of social cognition
Trends in Cognitive Sciences
(2004) Where do mirror neurons come from?
Neuroscience and Biobehavioral Reviews
(2010)- et al.
Dorsal and ventral streams: a framework for understanding aspects of the functional anatomy of language
Cognition
(2004)