We studied the contribution of action–effect cue integration to the perception of action and its sensory outcome in intentional binding. According to the cue integration hypothesis, the compression in the perceived temporal interval between a voluntary action and its sensory consequence results from using both events as temporally informative cues. The time estimates are based on a weighted average of the two events, in which the weight of each cue is determined by its relative reliability. As predicted by cue integration, our data show that reducing the reliability of the sensory outcome results in a smaller shift in the perceived time of action toward its outcome (reduced action binding). The cue integration hypothesis also predicted an increase in the shift of the perception of the sensory outcome toward the action with increasing uncertainty (i.e., increased tone binding). However, the results of additional analyses point to a separate mechanism involved in tone binding.
Action binding has been described in terms of a postdictive or inferential process, as it occurs even when the action is not strongly predictive of a tone, as long as the tone event occurs (Moore and Haggard 2008). Our results suggest that this postdictive process could be mediated by a cue integration mechanism. On the other hand, a predictive process has also been proposed to support action binding, as when the action is highly predictive of a tone, action binding occurs even in trials in which tones are absent (Moore and Haggard 2008). Consistent with this notion, even in our high uncertainty condition, in which tones were at each individual subject’s perceptual threshold, action binding measures were significantly above zero (data not shown). The association between the action and its outcome has been suggested to explain the predictive component of action binding (Moore et al. 2009).
Our finding that action binding is supported by cue integration is consistent with a previous study, suggesting that the estimation of time of movement depends on cue integration (Lau et al. 2007). This integration combines information about the time of action with other sources, as in the sensory outcome of an action. Information about the time of one’s own voluntary action could draw upon proprioceptive as well as internal volitional signals, such as an ‘efference copy’ of motor commands (Von Holst 1954) or components of the readiness potential (Lau et al. 2007). When sensory uncertainty is high or in the absence of sensory feedback, the perception of action relies more on these internal representations, thereby reducing action binding.
The cue integration framework has been successfully used to explain many perceptual phenomena in the sensorimotor system (e.g., Ernst and Banks 2002) and has been suggested to support the sense of agency (Moore and Fletcher 2012). Particularly, the integration of internal, volitional signals with external sensory cues can help dealing with uncertainty in the attribution of agency. Therefore, alongside the well-described action–effect association mechanism (see above), this integration could be another mechanism that links agency and intentional binding and reflects the volitional components that are captured in binding. For example, abnormal agency in disease states or under experimental interventions could arise from impairments in the internal volitional signals that normally contribute to the experience of agency. In turn, these impaired signals can lead to distinct changes in intentional binding, resulting from abnormal weighting of the action and outcome events. Future studies can thus apply the cue integration approach to explain abnormalities in action binding in terms of volitional deficits.
If cue integration can account for action binding effects, can it also explain tone binding? Our data suggest not. Tone binding was enhanced with increasing tone uncertainty, which at first glance is consistent with cue integration. However, this effect is attributable to increasing perceptual shifts in the baseline condition (tone only) as a function of tone intensity, rather than changes in the operant conditions, in which both action and tone events occurred together. Moreover, the changes in action and tone binding that resulted from sensory uncertainty were not correlated, suggesting different underlying mechanisms. Crucially, the prediction that integrating cues reduces performance variability was not satisfied for tone binding: Variability in estimation errors was significantly increased when two cues were provided in the operant tone condition, compared to baseline tone condition. These results show that the perceptual changes we observed in tone binding are likely to be driven by an alternative mechanism.
What might the alternative mechanism for tone binding be? Tone binding has been recently suggested to be mediated by a ‘pre-activation’ mechanism (Waszak et al. 2012). According to this account, the neural representation of a predicted sensory event, such as a sensory outcome following a voluntary action, is activated prior to its occurrence. Because of this ramped predictive activity, when the predicted sensory outcome occurs, it reaches perceptual threshold faster than when it is not predicted. Consequently, estimation errors are smaller in the operant tone condition than they are in the baseline tone condition, leading to tone binding. Our results suggest that this pre-activation mechanism can better account for the changes in tone binding under different levels of uncertainty. We found increased tone binding under high uncertainty, resulting from increased estimation errors in baseline tone condition. As tone intensity was reduced against a background noise for increasing sensory uncertainty, more time would be required for the tones to reach the perceptual threshold for detection. This additional time would be reflected in the increased estimation errors in the baseline condition. By contrast, in the operant condition, the learned action–effect association could diminish these differences in perceptual latencies. In other words, for pre-activated tone representations, the differences in intensities could be negligible, resulting in the lack of differences in estimation errors that was observed in the operant condition across uncertainties. Our results thus support the hypothesis that tone binding results from changes in perceptual latencies, driven by a predictive pre-activation mechanism (Waszak et al. 2012).
These results add to the growing evidence that different mechanisms underlie action and tone binding (Waszak et al. 2012). For example, whereas establishing a specific action–effect association is required for action binding (e.g., Moore and Haggard 2008), a more general association is sufficient for tone binding to occur (Desantis et al. 2012). In addition, experimental interventions can specifically affect tone binding without changing action binding. Such interventions include transcranial magnetic stimulation of the pre-supplementary motor area (Moore et al. 2010a) and manipulation of subjects’ causal beliefs (Desantis et al. 2011). Our results further support this notion: While action–effect cue integration is the most plausible explanation for the effect of uncertainty on action binding, differences in tone binding could be better accounted for by changes in perceptual latencies. Nevertheless, some mechanisms underlying action and tone binding may be shared. For example, the learned action–effect association can contribute to action binding through a prediction mechanism (Moore and Haggard 2008). Similarly, a prediction mechanism could be implemented for the tone to reach the perceptual threshold more rapidly and thereby lead to tone binding (Waszak et al. 2012).
Preliminary studies of the functional anatomy of binding (Moore et al. 2010a) have motivated the study of volitional disorders in patients with neurological and psychiatric illnesses, as well as healthy adults (e.g., Moore et al. 2010b, 2011). Often, action and tone binding measures have been added together (i.e., action binding plus the negative of tone binding) to generate an ‘overall binding’ measure. This measure has been used as a single metric of agency for comparing groups or measuring the effects of experimental interventions. However, if, as our data suggest, action and tone binding have different underlying contributory mechanisms, then disease states or interventions may have differential effects on these two forms of binding. Not only that our data indicate these two measures can be partially independent, but we also show that under some circumstances action and tone binding can be inversely related: Recall that high sensory uncertainty led to a reduction in action binding, while tone binding was increased. We therefore suggest that future studies should consider action and tone binding separately, rather than summing up these measures for studying volition.
The current study also has several limitations. First, our study draws upon the principles of cue integration, but does not apply computational techniques to model the data. Formal modeling of individual subject data would require many more trials for each condition per subject, which were not obtained here. Moreover, statistically optimal cue integration has been classically described for integrating multiple sources of information about one sensory event or object. Although action and tone events are synthesized in binding, it is possible that some of the principles of cue integration may not apply for the binding task, such as the statistical optimality. Second, we did not use a continuous variation of uncertainty, which would allow us to examine the psychometric properties of sensory uncertainty as a continuous effect. Such continuous variation was opted instead for the greater power conferred by the ordinal uncertainties. Third, our study only varies uncertainty in perception of action outcome tone and does not alter uncertainty in perception of time of action for fully covering the contribution of action–effect cue integration to binding. Manipulating temporal uncertainty of action would be experimentally more challenging. One solution could be to study clinical populations, such as patients with movement disorders, in which there is uncertainty over actions.
In conclusion, our results suggest that cue integration between action and effect contributes to the intentional binding effect for actions. By contrast, cue integration did not account for the observed changes in tone binding. This supports the notion that action and tone binding are driven by distinct underlying mechanisms. Our data support the use of intentional binding in the investigation of the mechanisms of volition, but suggest that action and tone binding should be considered separately in future studies.