Timing intrinsically involves prediction. Determining when to act upon a future event requires the ability to predict it. For instance, ensemble music performance requires precise estimation of the passage of time in order to synchronize and coordinate sounds to re-produce the musical structure.
A central idea in the predictive coding account of cognition is that prior knowledge is used to guide sensory interpretations and action decisions. Identifying the periodicity of an event in the world is typically an ill-posed problem: How does the agent know beforehand what constitutes the signals that indicate a period? To infer the beat in a complex musical piece, or when a quail will reappear from behind a bush, are underspecified problems in the sensory signal. In both cases, prior experience appears necessary to play to the beat or to catch the quail.
Another key idea in the predictive coding framework is information compression. Representing music or other temporally structured events as cycles reduces the entropy in the signal and allows for more efficient storage. Action can serve to further bootstrap timing. For instance, humans spontaneously tap along with their hands or feet to music (Brown Reference Brown2003) and entrain their movements to other people's movements (Demos et al. Reference Demos, Chaffin, Begosh, Daniels and Marsh2012; Merker et al. Reference Merker, Madison and Eckerdal2009). Just like active interactions with an object improve perception (Harman et al. Reference Harman, Humphrey and Goodale1999), timed activities have been shown to improve the reliability of temporal perception (Grahn & McAuley Reference Grahn and McAuley2009; Phillips-Silver & Trainor Reference Phillips-Silver and Trainor2007). A benefit of having induced the rhythm is that violations of rhythm are easier to detect (Ladinig et al. Reference Ladinig, Honing, Haden and Winkler2009).
Bayesian inference of timing requires temporal uncertainties to be represented. The nature of the timing signal remains open to debate. One candidate is trace strength that decays with time (Buhusi & Meck Reference Buhusi and Meck2005). A function of decay, trace strength conveys information about the time since it occurred. Another time signal candidate is populations of oscillating neurons. Timing could then be established by coincidence detection in the oscillating network (Matell & Meck Reference Matell and Meck2004; Miall Reference Miall1989). Regardless of the signal format, its representation is noisy and its uncertainty should reasonably increase with timing over long durations. Indeed, human temporal perception and production do deteriorate monotonically with time scale (Buhusi & Meck Reference Buhusi and Meck2005). Exactly how the human system deals with temporal signal uncertainty remains an open question.
A key notion in the target article is the hierarchical division of labor from bottom sensory to top associative cortical control. For timing, the scaling of time appears as a likely attribute to stretch across such a hierarchical structure. Millisecond control of motor timing cannot feasibly be carried out directly by the prefrontal cortical regions involved in working memory, due to transfer speed, and the accumulated signal error that such an extensive chain of transmission would involve. Instead, millisecond control might be represented closer to the action output (e.g., cortical effector representation and the cerebellum) and involve a more direct pathway between sensory input and motor output. In contrast, when observation and action become more detached in time, the window of opportunity for planning opens up, involving more prefrontal processing.
Consistently, many studies support the view that there is a distinction in neural representation, for example, above and below about one second (Gooch et al. Reference Gooch, Wiener, Hamilton and Coslett2001; Lewis & Miall Reference Lewis and Miall2003; Madison Reference Madison2001). Furthermore, time representation for sub-second intervals appears at least to some extent to be sensory specific (Morrone et al. Reference Morrone, Ross and Burr2005; Nagarajan et al. Reference Nagarajan, Blake, Wright, Byl and Merzenich1998), and under some conditions even limited to spatial locations (Burr et al. Reference Burr, Tozzi and Morrone2007; Johnston Reference Johnston, Arnold and Nishida2006). Additionally, there appear to be breakpoints in interval discrimination such that there are scalar properties in timing performance for intervals above about one second, but nonlinear relationships between time and perception below one second (Karmarkar & Buonomano Reference Karmarkar and Buonomano2007; Rammsayer Reference Rammsayer1999) – further supporting the notion that longer time intervals are controlled by different brain regions from those involved in sub-second timing. Also, with longer time periods under consideration, a larger part of the prefrontal cortex gets activated (Lewis & Miall Reference Lewis and Miall2006; Simons et al. Reference Simons, Schölvinck, Gilbert, Frith and Burgess2006). This timing-related frontal lobe network is also largely overlapping with those employed by working memory and executive control processes (Jahanshahi et al. Reference Jahanshahi, Dirnberger, Fuller and Frith2000; Owen et al. Reference Owen, McMillan, Laird and Bullmore2005), suggesting that timing constitutes a general cognitive control problem at longer time durations. The hierarchical organization from accurate and dedicated timing devices at sensory levels and less accurate but flexible timing at longer time frames in the prefrontal cortex might be accounted for by signal averaging in the time domain from sensory to frontal cortical regions (Harrison et al. Reference Harrison, Bestmann, Rosa, Penny and Green2011). Harrison and colleagues suggested that decay rate is faster close to the sensory input level and slower at later stages in the visual hierarchy, thus allowing for a differentiation across time scale and brain region. Taken together, there is abundant support for the differentiation of brain regions involved in timing at different time scales.
Communication of temporal information across the levels of the outlined timing hierarchy is currently rather unclear. Intuitively, the more temporally extended control processes associated with prefrontal working memory processes might still influence control at shorter time frames without interfering in direct control, such as in initiation of a drumming exercise, without employing moment to moment volitional control of the individual beats. Recent findings from our research group suggest that executive functions are indirectly related to motor timing via, for example, effector coordination (Holm et al., in press). Furthermore, there is a well-established yet poorly specified relationship between intelligence and simple motor timing (Galton Reference Galton1883; Madison et al. Reference Madison, Forsman, Blom, Karabanov and Ullén2009). More research is clearly needed to identify how high-level temporal expectations might influence brief interval timing. Another important question is how the brain identifies the time scales from noisy input and learns how to treat those signals. The predictive account of cognition seems like a useful theoretical framework for understanding timing, and the Bayesian formalism is a promising tool to investigate and explain its operation.
Timing intrinsically involves prediction. Determining when to act upon a future event requires the ability to predict it. For instance, ensemble music performance requires precise estimation of the passage of time in order to synchronize and coordinate sounds to re-produce the musical structure.
A central idea in the predictive coding account of cognition is that prior knowledge is used to guide sensory interpretations and action decisions. Identifying the periodicity of an event in the world is typically an ill-posed problem: How does the agent know beforehand what constitutes the signals that indicate a period? To infer the beat in a complex musical piece, or when a quail will reappear from behind a bush, are underspecified problems in the sensory signal. In both cases, prior experience appears necessary to play to the beat or to catch the quail.
Another key idea in the predictive coding framework is information compression. Representing music or other temporally structured events as cycles reduces the entropy in the signal and allows for more efficient storage. Action can serve to further bootstrap timing. For instance, humans spontaneously tap along with their hands or feet to music (Brown Reference Brown2003) and entrain their movements to other people's movements (Demos et al. Reference Demos, Chaffin, Begosh, Daniels and Marsh2012; Merker et al. Reference Merker, Madison and Eckerdal2009). Just like active interactions with an object improve perception (Harman et al. Reference Harman, Humphrey and Goodale1999), timed activities have been shown to improve the reliability of temporal perception (Grahn & McAuley Reference Grahn and McAuley2009; Phillips-Silver & Trainor Reference Phillips-Silver and Trainor2007). A benefit of having induced the rhythm is that violations of rhythm are easier to detect (Ladinig et al. Reference Ladinig, Honing, Haden and Winkler2009).
Bayesian inference of timing requires temporal uncertainties to be represented. The nature of the timing signal remains open to debate. One candidate is trace strength that decays with time (Buhusi & Meck Reference Buhusi and Meck2005). A function of decay, trace strength conveys information about the time since it occurred. Another time signal candidate is populations of oscillating neurons. Timing could then be established by coincidence detection in the oscillating network (Matell & Meck Reference Matell and Meck2004; Miall Reference Miall1989). Regardless of the signal format, its representation is noisy and its uncertainty should reasonably increase with timing over long durations. Indeed, human temporal perception and production do deteriorate monotonically with time scale (Buhusi & Meck Reference Buhusi and Meck2005). Exactly how the human system deals with temporal signal uncertainty remains an open question.
A key notion in the target article is the hierarchical division of labor from bottom sensory to top associative cortical control. For timing, the scaling of time appears as a likely attribute to stretch across such a hierarchical structure. Millisecond control of motor timing cannot feasibly be carried out directly by the prefrontal cortical regions involved in working memory, due to transfer speed, and the accumulated signal error that such an extensive chain of transmission would involve. Instead, millisecond control might be represented closer to the action output (e.g., cortical effector representation and the cerebellum) and involve a more direct pathway between sensory input and motor output. In contrast, when observation and action become more detached in time, the window of opportunity for planning opens up, involving more prefrontal processing.
Consistently, many studies support the view that there is a distinction in neural representation, for example, above and below about one second (Gooch et al. Reference Gooch, Wiener, Hamilton and Coslett2001; Lewis & Miall Reference Lewis and Miall2003; Madison Reference Madison2001). Furthermore, time representation for sub-second intervals appears at least to some extent to be sensory specific (Morrone et al. Reference Morrone, Ross and Burr2005; Nagarajan et al. Reference Nagarajan, Blake, Wright, Byl and Merzenich1998), and under some conditions even limited to spatial locations (Burr et al. Reference Burr, Tozzi and Morrone2007; Johnston Reference Johnston, Arnold and Nishida2006). Additionally, there appear to be breakpoints in interval discrimination such that there are scalar properties in timing performance for intervals above about one second, but nonlinear relationships between time and perception below one second (Karmarkar & Buonomano Reference Karmarkar and Buonomano2007; Rammsayer Reference Rammsayer1999) – further supporting the notion that longer time intervals are controlled by different brain regions from those involved in sub-second timing. Also, with longer time periods under consideration, a larger part of the prefrontal cortex gets activated (Lewis & Miall Reference Lewis and Miall2006; Simons et al. Reference Simons, Schölvinck, Gilbert, Frith and Burgess2006). This timing-related frontal lobe network is also largely overlapping with those employed by working memory and executive control processes (Jahanshahi et al. Reference Jahanshahi, Dirnberger, Fuller and Frith2000; Owen et al. Reference Owen, McMillan, Laird and Bullmore2005), suggesting that timing constitutes a general cognitive control problem at longer time durations. The hierarchical organization from accurate and dedicated timing devices at sensory levels and less accurate but flexible timing at longer time frames in the prefrontal cortex might be accounted for by signal averaging in the time domain from sensory to frontal cortical regions (Harrison et al. Reference Harrison, Bestmann, Rosa, Penny and Green2011). Harrison and colleagues suggested that decay rate is faster close to the sensory input level and slower at later stages in the visual hierarchy, thus allowing for a differentiation across time scale and brain region. Taken together, there is abundant support for the differentiation of brain regions involved in timing at different time scales.
Communication of temporal information across the levels of the outlined timing hierarchy is currently rather unclear. Intuitively, the more temporally extended control processes associated with prefrontal working memory processes might still influence control at shorter time frames without interfering in direct control, such as in initiation of a drumming exercise, without employing moment to moment volitional control of the individual beats. Recent findings from our research group suggest that executive functions are indirectly related to motor timing via, for example, effector coordination (Holm et al., in press). Furthermore, there is a well-established yet poorly specified relationship between intelligence and simple motor timing (Galton Reference Galton1883; Madison et al. Reference Madison, Forsman, Blom, Karabanov and Ullén2009). More research is clearly needed to identify how high-level temporal expectations might influence brief interval timing. Another important question is how the brain identifies the time scales from noisy input and learns how to treat those signals. The predictive account of cognition seems like a useful theoretical framework for understanding timing, and the Bayesian formalism is a promising tool to investigate and explain its operation.