Exposure to delayed visual feedback of the hand changes motor-sensory synchrony perception

We examined whether the brain can adapt to temporal delays between a self-initiated action and the naturalistic visual feedback of that action. During an exposure phase, participants tapped with their index finger while seeing their own hand in real time (~0 ms delay) or delayed at 40, 80, or 120 ms. Following exposure, participants were tested with a simultaneity judgment (SJ) task in which they judged whether the video of their hand was synchronous or asynchronous with respect to their finger taps. The locations of the seen and the real hand were either different (Experiment 1) or aligned (Experiment 2). In both cases, the point of subjective simultaneity (PSS) was uniformly shifted in the direction of the exposure lags while sensitivity to visual-motor asynchrony decreased with longer exposure delays. These findings demonstrate that the brain is quite flexible in adjusting the timing relation between a motor action and the otherwise naturalistic visual feedback that this action engenders.


Introduction
While interacting with the external world, our brain is provided with a massive amount of information that is processed by diVerent neural mechanisms. For example, the clapping of the hands is initiated by the motor system, but subsequently, visual, auditory, and tactile information is provided and processed. Despite that motor planning and the subsequent motor execution vary in time for diVerent actions, and diVerent sensory modalities process information from a common event at diVerent speeds (Fain 2003), people still experience the sensory consequences of their actions as causal and as simultaneous with the actions. Though, in artiWcial situations like a teleconference or a telesurgery in which the auditory and/or visual information can be noticeably delayed, people may experience incoherence between their motor actions and the sensory consequences. Here, we examined to which extent the brain can adapt to these temporal disturbances of motor-sensory events by exposing participants to delayed, but otherwise naturalistic visual feedback of their own body movements. The question was whether participants would recalibrate their sense of motor-visual synchrony of what, at Wrst sight, would seem to be a rigid timing relation.
Initial studies on temporal recalibration have demonstrated that participants do indeed adapt to small asynchronies between artiWcial audio-visual stimuli like Xashes and beeps (Fujisaki et al. 2004;Vroomen et al. 2004). For example, participants who were exposed to a Xash occurring »100 ms after a beep were in simultaneous test trials more likely to say that the Xash occurred before the beep. Ever since, this so-called cross-modal temporal recalibration eVect (i.e. TRE) has been reported many times in diVerent situations and in diVerent modalities (see Vroomen and Keetels 2010 for reviews; Keetels and Vroomen 2012). In principle, though, there may be nothing implausible about small temporal delays between the senses, given that we are exposed to visual and auditory stimuli in all sorts of temporal conWgurations due to diVerences in air-transduction and neural processing times (Fain 2003). The order of motor actions and their sensory feedback, though, is much more constrained because sensory feedback is normally expected to occur only after motor actions are initiated. As an example, you will only hear yourself speaking after you actually spoke the utterance. Yet, Xexibility to motor-sensory timing relationships seems to be necessary, at least to some extent, because conduction times in sensory and motor pathways change due to, for example, diVerences in retinal response times in diVerent lighting conditions (Purpura et al. 1990), attention to speciWc modalities (i.e. prior entry: attention towards a speciWc modality can speed up processing time in the attended modality ;Titchener 1908) or, on a longer time scale, due to limb growth (Campbell et al. 1981) or muscle decay.
For the visual-motor domain, several authors have reported that temporal recalibration does indeed occur and that it can actually change the causal relationship between a motor action and the associated artiWcial sensory feedback (i.e. like the movement of a cursor or a visual Xash; Stetson et al. 2006;Cunningham et al. 2001;Sugano et al. 2010;Pesavento and Schlag 2006;Heron et al. 2009;Stekelenburg et al. 2011). Stetson et al. (2006), for example, exposed participants to a Wxed delay between a self-initiated key-press and a subsequently delivered Xash and reported that Xashes appearing at unexpectedly short delays were often perceived as occurring before the motor action, thus demonstrating a reversal of 'cause-before-eVect'. Sugano et al. (2010) also examined motor-sensory recalibration and explored whether it is the sensory or motor event that is shifted in time. Participants were exposed to a 150 ms lag between a Wnger tap and a Xash or tone pip, and immediately thereafter, they performed a temporal order judgment (TOJ) task about the tap-feedback test stimulus. The modality of the feedback stimulus was either the same as the adapted one (within modal) or diVerent (cross-modal). The results showed that the point of subjective simultaneity (PSS) was uniformly shifted in the direction of the exposed lag within and across modalities, indicating that the TRE of sensor-motor events is mainly caused by a shift in the motor component (though, see Stekelenburg et al. 2011 for eVects in the visual ERPs).
At this stage, though, motor-visual temporal recalibration has only been studied with artiWcial sensory feedback, but not with naturalistic feedback. There are several reasons why the causal relationship between a motor action and the resulting natural sensory feedback might be diVerent when compared with artiWcial sensory feedback. Most importantly, delays to natural feedback are far less variable than delays to artiWcial feedback. For example, when watching your own hand reaching for an object, it is only the relatively stable visual sensory transduction time that might cause a delayed perception of the limb's movement. For artiWcial feedback, though, delays can be quite large and variable due to for example, technical constraints or operating system delays. For example, the time that passes between a Wnger touching a keyboard and a letter appearing on a computer screen diVers substantially because the sensitivity of keys on diVerent keyboards varies widely. Another potentially relevant factor is that there may be more intentional binding between a voluntary movement and natural rather than artiWcial feedback. One consequence of intentional binding is that the interval between a voluntary action and its outcome is perceived to be shorter than the interval between a physically similar involuntary movement (Engbert et al. 2008;Haggard et al. 2002b;Moore et al. 2009;Buehner and Humphreys 2009). Intentional binding thus can diminish the perceived lag between an action and its outcome, and this may prevent temporal recalibration to occur simply because no timing error is detected.
One study that hints for an adaptation eVect to delayed naturalistic auditory feedback was performed by Katz and Lackener (1977). In their study, participants were exposed to motor-sensory temporal disturbances by delaying auditory feedback (DAF; Lee 1950aLee , b, 1951 while performing speech tasks (i.e. reading word lists, short sentences, and a prose passage). The authors found that DAF had a disturbing eVect directly after introducing the delay (i.e. high error rate and slow speech), but after being exposed to DAF for a few minutes, the disturbing eVect declined. The authors concluded that participants had adapted to DAF in speech, and the reduction of errors might be due to the recalibration of motor-sensory timing. Alternatively, though, participants might also have learned to ignore the disturbing sounds. Given that the authors did not directly measure perceived motor-sensory timing, it remains unclear whether motorsensory timing was recalibrated after exposure to DAF.
In the present study, we used Wnger taps as a voluntary action while the visual feedback consisted of an online video recording of that movement from the perspective of the actor, played back via a monitor with an adjustable delay. In Experiment 1, the monitor was placed in front of the participant at »25 cm from the hand, which implied that the location and orientation of the seen and real hand did not match. In Experiment 2, we further increased the naturalness of the situation by aligning the orientation and location of the virtual and real hand. A single trial consisted of a short-exposure phase immediately followed by a test phase. During exposure, participants tapped their index Wnger while seeing the movement of the Wnger via the monitor (see Fig. 1a, b for Experiment 1 and Experiment 2, respectively). The normal synchrony between the motor action (M) and the accompanying visual information (V) was modiWed by delays in the video (varying from »0 to »120 ms; the shortest delays are hardly noticeable, the longer ones are). In the test phase, the delay of the video was again changed (varying from »0 to »160 ms), and participants decided whether the video was in-or out-of-synchrony with their Wnger taps (a simultaneity judgment task, SJ). 1 We expected that motor-visual temporal recalibration would manifest itself as a shift in the point at which maximum simultaneity (the point of subjective simultaneity, PSS) between the motor act and the visual feedback was perceived.

Participants
Twenty-nine students (eight male, mean age: 21.4, 25 righthanded) from Tilburg University participated in return for course credits after giving informed consent. All had normal hearing and normal or corrected-to-normal seeing.

Stimuli and apparatus
Participants sat at a desk in a dimly lit and soundproof booth. Their right hand was positioned in front of them on a custom made touch pad (8 £ 8 cm). A video camera (Conrad; KC-3800; CCD-Camera; PAL with 400 TV-lines at 50 Hz) was positioned on the table to record the participant's Wnger movements (see Fig. 1a). The participant's hand, touch pad, and camera were covered by a black cloth in order to prevent them from seeing their own real-time hand movements. Participants were instructed to rest their consists of an exposure phase and test phase, in which participants tapped 10 and 5 times, respectively, while the visual feedback of their Wnger movements was artiWcially delayed and played back on a monitor in front of them. Participants judged whether the video of the last 5 taps was synchronous or asynchronous (SJ task) relative to their movements (a) head on a chin rest and to look at a RGB monitor (PHI-LIPS, CM8533/00G) at about 45 cm viewing distance on which the real-time or delayed videos of their own hand movements were displayed. The video was delayed by dedicated hardware (Ovation Systems Ltd.; DelayLine; customized unit: up to 320 ms total delay, in 20 ms steps) connected between the video camera and the monitor. The intrinsic delay of this device was »3 ms. 2 To ensure that all participants received equal amounts of exposure and that they tapped at an equal speed, we presented an auditory train of click sounds that served as a pacing signal (ISI 750 ms; each click 10 ms duration; presented via headphones at 70 dB(A)). White noise was continuously presented via two speakers on each side of the monitor at 59 dB(A) to mask sounds produced by the taps.

Design
Two factors were varied: the exposure delay (0, 40, 80, or 120 ms video delay), and the delay of the test stimulus (0, 40, 80, 120, or 160 ms video delay). Half of the participants were tested with the 0 and 80 ms exposure delays (N = 15), and the other half were tested with 40 and 120 ms exposure delays (N = 14). The two exposure delays were split into two consecutive test sessions with at least a 1-h pause in between. The order of the exposure delays was counterbalanced across participants. The video delay in the test phase varied randomly. The whole experiment consisted of six blocks of 20 trials (three blocks per exposure delay), resulting in a total of 12 repetitions of each combination of test delay and exposure delay. Each block of 20 trials took about 8 min with a short break after each block. To acquaint participants with the procedures, experimental trials were preceded by a practice session of 6 trials with a 0 ms exposure delay.

Procedure
A single trial consisted of a short-exposure phase followed by a test phase (see Fig. 1c). During the exposure phase, participants were presented a sequence of 12 auditory clicks as a pacing signal and were instructed to tap along with the last 10 clicks on the touch pad with the index of their right hand. At the start of the exposure phase, the monitor was switched on, and depending on the particular exposure delay, a real-time video (0 ms video delay) or a delayed video (40, 80, or 120 ms video delay) of the participant's hand was displayed. After the exposure phase, the monitor then switched oV for 1.5 s (dark screen) after which the test phase started. During the test, participants were presented seven pacing clicks (750 ISI), and they tapped their right index Wnger along with the last Wve clicks. At the start of the test phase, the monitor was switched on again, and depending on the test delay, a realtime video (0 ms video delay) or a delayed video (40, 80, 120, or 160 ms video delay) of the participant's actual Wnger movements was shown. After the last tap, the screen turned black, and a pure tone (70 dB(A), 440 Hz, 500 ms) was presented that indicated that participants had to decide whether the video that was seen during the test phase was synchronous with their Wnger movements or asynchronous. Participants made an unspeeded response by pressing one of the two buttons on a response box with their left hand. After the response, the exposure phase of the next trial started.

Results
Trials of the training session were excluded from further analyses. The proportion of 'synchronous' responses was calculated for each participant and for each combination of exposure delay (0, 40, 80, 120 ms) and test delay (0, 40, 80, 120, 160 ms). For each of the obtained distributions of responses, an individually determined psychometric function was calculated by Wtting the Gaussian normal distribution [exp(¡ (x ¡ Mean)^2/(2*SD^2))] over the data using the 'fminsearch' function in Matlab 7.9 (MathWorks, Natick, MA, USA; the multidimensional unconstrained nonlinear minimization method). The average goodness of the Wt was R 2 = 0.95. The mean of the resulting distribution is the point where simultaneity is maximal (the PSS), and the standard deviation (SD) is a measure of the sensitivity. Figure 2 displays an example of two representative participants and the averaged normal distributions. Table 1 shows the average PSSs and SDs at each exposure delay.
A mixed-linear model analysis with exposure delay (0, 40, 80, 120 ms) as a Wxed factor was performed on both the PSSs and the SDs. For both measures, a signiWcant eVect of exposure delay was found: PSS: F(3,58) = 6.92, p = <.001, SD: F(3,58) = 3.72, p = <.05. The eVect of exposure delay on the PSS demonstrates that participants adjusted the perception of synchrony, as the most synchronous test stimulus gradually shifted with exposure delay: The PSSs were 23.4, 27.5, 35.1, and 45.2 ms after exposure delays of 0, 40, 80, and 120 ms, respectively. The eVect of exposure delay on the SDs shows that sensitivity to motor-sensory asynchronies gradually declined when the exposure delay increased (SD: 57.1, 61.2, 62.7, and 74.3 ms after exposure delays of 0, 40, 80, and 120 ms, respectively). The best Wtting linear function through the PSS values was as follows: PSS = (0.18* exposure delay) + 22.0 ms, which thus indicates that the PSS shifted about 18 % of the exposure delay. A similar linear function Wtted over the standard deviation (SD = 0.13* exposure delay + 55.9), indicates that sensitivity declined with about 13 % of the exposure delay.

Discussion
The most important result of Experiment 1 is that the PSSs were shifted in the direction of the exposure delays. This shows that there is Xexibility in motor-sensory timing, even with naturalistic feedback whose timing relation with the motor event is usually rather Wxed. We also found that sensitivity to visual-motor asynchrony (SDs) declined with longer delays. Although PSS shifts are a common Wnding in temporal recalibration studies (for reviews, see Vroomen and Keetels 2010;Keetels and Vroomen 2012), the decrease in sensitivity has only been reported in some studies (Navarra et al. 2005;Vatakis et al. 2008Vatakis et al. , 2007. One concept that might have aVected the PSSs and SDs in the present study is the 'feeling of agency'. Because the motor system includes speciWc mechanisms for predicting the sensory consequences of our own actions Blakemore et al. 2000), it might be that body actions were less strongly felt as 'owned' when participants saw their own hand movements at a diVerent location and orientation than expected. In Experiment 1, the virtual hand was located »25 cm further than the real hand, and this might have led to a reduced feeling of agency. A reduced feeling of agency, in turn, might cause a reduction in the number of 'synchronous' responses. To further examine this, we replicated Experiment 1 with the locations of the real and virtual hand spatially aligned.

Method
Stimuli and design were the same as in Experiment 1 with the following exceptions. Thirty-two new participants (eight male, mean age: 19.9, 29 right-handed) from Tilburg University were tested (N = 17 for 0 and 80 ms exposure delays, and N = 15 for 40 and 120 ms exposure delays). They sat at a desk with their right hand holding a wooden block (7 £ 7 £ 4 cm) in the middle of the table. A doublesided mirror was installed above the participant's hand (see Fig. 1b for a schematic overview of the apparatus set-up). The video camera projected the image of the participant's  hand via the backside of the mirror. The angle of the mirror and camera were adjusted in such a way that the seen orientation and location of the arm and hand matched the real one. A black cloth was attached to the front edge of the mirror to prevent the participant from seeing their own realtime wrist and hand movements.

Results
PSSs and SDs were calculated as in Experiment 1 (see Fig. 2

Between experiments analysis
A mixed-linear model analysis with exposure delay (0, 40, 80, 120 ms) and Experiment (1 vs. 2) as Wxed factors was performed on both the PSSs and the SDs. A signiWcant overall eVect of exposure delay was found on PSSs (F(3,122) = 21.14, p = <.001) and SDs (F(3,122) = 7.25, p = <.001). There was no diVerence in the PSSs between experiments (F(1,122) < 1), but SDs were larger (a 6.0 ms overall diVerence) when the location of the real hand and its visual feedback were misaligned (Experiment 1) rather than aligned (Experiment 2; F(1,122) = 6.83, p = <.01). For both measures, no interaction was found between experiment and exposure delay (PSS: F(3,122) < 1; and SD: F(3,122) = 1.15, p = .33), indicating that the spatial separation between the seen and real location of the hand did not aVect the pattern of temporal recalibration. To further analyse the data, we examined the proportion of synchronous responses at the SOA that-under normal circumstances-should be the most naturalistic, namely the 0-ms SOA. If lag adaptation entails a true shift of the PSS rather only than a widening of the window of acceptable asynchronies in the direction of the adapted lag (see Yarrow et al. 2011a), one may expect the 0-ms SOA test stimulus to become 'unnatural' after exposure to large delays because observers may start to experience 'consequencebefore-cause'. To test this, we ran a separate analysis on the number of synchronous responses at 0 ms SOA (the leftmost portion of the raw data in Fig. 2) with exposure delay (0, 40, 80, 120 ms) as a Wxed factor. This analysis showed that for the 0-ms test trials, the proportion 'synchronous' responses declined after exposure to longer delays (average proportion 'synchronous' responses were .98, .96, .91, .81 for 0, 40, 80, and 120 ms exposure delays, respectively, F(3,122) = 9.1, p = <.001), thus conWrming that there was a true shift in the PSS.

General discussion
We examined the eVect of exposure to delayed naturalistic visual feedback on the perception of motor-visual synchrony. Participants tapped their Wnger on a surface while viewing a video of their taps at 0, 40, 80, or 120 ms delays. Following a short-exposure phase, perception of motorvisual synchrony was tested by asking participants to judge whether the video of their Wnger was delayed relative to their taps. The results demonstrated that exposure to delayed visual feedback shifted the point of maximal perceived synchrony. This shift was not aVected by whether the location of the seen hand and real hand was spatially aligned or misaligned. This suggests that observers are quite Xexible in what constitutes motor-visual synchrony. This is remarkable given that the timing of self-initiated actions and their natural visual feedback is usually Wxed.
The Wnding that the PSS was shifted after exposure to naturalistic delayed feedback concurs with the previous Wndings of temporal recalibration in both the intersensory (see for reviews: Vroomen and Keetels 2010;Keetels and Vroomen 2012) and the motor-sensory domain (Stetson et al. 2006;Sugano et al. 2010;Heron et al. 2009) in which so far only artiWcial stimuli have been used. Apparently, even though we are less frequently exposed to delays of naturalistic visual feedback, the brain is still able to adapt to these changes in timing (see also Kopinska and Harris 2004 who showed that the brain corrects for changes in visual timing, which result from wearing light-attenuating dark glasses). This Wts with the idea that the brain needs to be Xexible, to some extent, in order to correct for (small) changes in the motor and the visual system like limb growth, muscle decay, or diseases like Parkinson that aVect motor control. Presumably, sensory signals from a motor event (tactile/kinaesthetic information) and the visual feedback are remapped to preserve their perceived simultaneity because events appearing at a constant delay after motor actions are interpreted as consequences of those actions. The brain then recalibrates the timing relation so that it is consistent with the prior expectation that sensory feedback follows motor actions without noticeable delay.
Several mechanisms have been proposed for how temporal recalibration between the senses might be envisaged (see Vroomen and Keetels 2010 for a review on this topic). One possibility that Wts with the observation that there was a shift of the PSS in the direction of the adapted lag is that the criterion for simultaneity between the corresponding modalities is adjusted in accord with the previous experience (a Bayesian approach; Ernst and BulthoV 2004). Alternatively, though, it may also be that one modality (vision, touch, or motor information) is 'shifted' towards the other because the sensory threshold for stimulus detection in the adapted modality is adjusted (Navarra et al. 2009). For example, participants may have adopted a more conservative criterion for detecting when they moved their Wnger or touched the surface, or they may have adopted a more lenient criterion for detecting when they actually saw the visual motion of the Wnger. Yet another possibility is that the 'window of temporal integration' was widened due to the asynchronous exposure. This idea of widening of the temporal window is in line with the Wnding that SDs became larger (i.e. sensitivity became worse) when the exposure delays increased. Note, though, that this widening of the temporal window cannot account for the whole pattern of results because physically synchronous test stimuli (0-ms SOA) were perceived as less synchronous (not more) after exposure to long video delays. Note that these options do not mutually exclude each other, and several of them remain open. The current data do indicate, though, that for motor-visual recalibration there may not be a critical distinction between naturalistic and artiWcial feedback because both kinds of feedback induce a shift in the PSS. Further research is necessary, though, to compare them in a more direct manner.
It is noteworthy that there was a decrement in sensitivity, because this has not always been reported. Navarra et al. (2005) and Vatakis et al. (2008) tested audio-visual temporal recalibration using audio-visual speech stimuli and also reported a decline in sensitivity. Their observers had to monitor a continuous speech stream for target words that were presented either in synchrony with the video of a speaker or with a delayed video. Concurrently with the speech-monitoring task, participants performed an audiovisual temporal order judgement task (TOJ; Navarra et al. 2005;Vatakis et al. 2007) or an SJ task (Vatakis et al. 2008) on simple Xashes and noise bursts that were overlaid on the video. The results showed that participants became in both tasks less sensitive to audio-visual temporal asynchrony, but without a shift in the PSS, if exposed to desynchronized rather than synchronized audio-visual speech. Similar eVects (a decrease in sensitivity after exposure to audiovisual asynchrony) were found when participants were exposed to audio-visual music stimuli. This led the authors to conclude that the window of temporal integration was widened because of asynchronous exposure (see also Navarra et al. 2007;and Winter et al. 2008 for eVects on sensitivity after adaptation to asynchronous audio-tactile and motor-tactile stimuli, respectively). The authors argued that this widening may reXect an initial stage of recalibration in which a more lenient criterion is adopted for simultaneity. With prolonged exposure, participants may then ultimately shift the PSS. Alternatively, though, it might also be that participants became confused by the non-matching exposure stimuli in the background, and as a result made more errors, thus aVecting the JND and not the PSS. Note that in the present study, the reduction in sensitivity after exposure to asynchronous videos was less likely caused by 'confusion' as such, because the exposure and test phase were clearly separated in time rather than that they were overlapping. However, given that the eVect on the JND has most frequently been demonstrated in studies using naturalistic stimuli (i.e. hand motion, speech, music), it remains for future studies to examine whether the nature of the adapting stimuli (naturalistic vs. artiWcial) is critical for a widening of the temporal window.
An additional process that might be involved in the present study relates to the concept of intentional binding (Engbert et al. 2008;Haggard et al. 2002a, b;Moore et al. 2009;Blakemore et al. 2000;Engbert and Wohlschlager 2007;Cravo et al. 2009;Buehner and Humphreys 2009). It is commonly thought that intentional actions and their resulting eVects are perceived as temporally attracted towards each other. In our study, it is therefore conceivable that the visual feedback was perceived as a consequence of a participant's voluntary actions, and intentional binding may have contributed to a reduction of the perceived temporal delay. As a consequence, intentional binding may have led to an overall large proportion of 'simultaneous' responses, and sensitivity may for that reason be low. It should be noted, though, that from this perspective it is not clear how exposure to increasing delays-that presumably lead to less binding-could worsen sensitivity on test trials, because less binding should improve, not hamper sensitivity. In a similar vein, we observed that when the seen and felt location of the hands were aligned rather than misaligned-the latter evoking less binding-participants were in fact more (not less) sensitive to temporal asynchronies. Intentional binding thus cannot account for the diVerences in sensitivity when hands were aligned or when participants were exposed to large delays.
It is also of interest to note that equal shifts in the PSS were obtained when the virtual and real hand were aligned or misaligned. This Wts an earlier study (Keetels and Vroomen 2007) in which participants were exposed to an asynchronous train of spatially matching or mismatching auditoryvisual stimulus pairs. Following exposure to a Wxed lag, participants were tested on an auditory-visual TOJ task. Temporal recalibration manifested itself as a shift of the PSS in the direction of the adapted auditory-visual lag. As observed here, this shift was equally big for spatially matching and mismatching sound/light pairs. Apparently, a spatial mismatch between otherwise corresponding inputs does not disrupt temporal recalibration. This is not to say, though, that spatial information is completely irrelevant for setting up the initial correspondence between inputs. As an example, in a study by Yarrow et al. (2011b), spatial correspondence resolved temporal ambiguity. Their participants were exposed to a train of sounds and lights whose locations alternated between left and right (e.g. left-sound, leftlight, right-sound, right -light). The time between each stimulus was equal (200 ms), and in the absence of spatial information, it would be unclear whether sounds were actually leading or lagging the lights. The results though showed that temporal recalibration was obtained as implied by spatial grouping (in the previous example, sound-leading). This Wnding demonstrates that-in the absence of other cues-spatial information can help in the segmentation of the audio-visual scene.
A natural action like a Wnger tap is a complex stimulus, and this raises the question what the actual cues are in the signal that observers use as a timing marker for the action and for the visual feedback of that action. A Wnger tap might be decomposed into an intention to make a movement, followed by the actual motor command and an eVerent copy of that command. While the Wnger is moving, the perceiver also receives proprioceptive feedback about the Wnger movement and the position of the joints, and tactile feedback at the moment that the Wnger touches an object. Conceivably, the visual feedback of the Wnger tap also contains several markers like the onset of the motion, the motion itself, and the oVset of the visual motion when the Wnger touches the object. In the current situation, it is diYcult to pinpoint which of these cues observers actually used to estimate the timing of the action and the feedback thereof. This requires further studies in which one could, for example, examine temporal recalibration with passive motion (to remove the intentional component), Wnger tapping without the touch on a surface (to remove the tactile feedback), or a restriction of the visual feedback to the onset or the oVset of the Wnger tap.
Another relevant aspect of our study is that we asked participants to tap in synchrony with an auditory pacer (a click) presented at a constant rate (ISI = 750 ms). These isochronous clicks at a relatively slow rate ensured that consecutive Wnger taps were temporally distinct so that the timing relation with the video remained unambiguous (at faster rates, Wnger taps might overlap with the delayed video of the previous taps in which case the temporal relation would be lost). A typical outcome of auditory synchronization task is that Wnger taps precede the sound by about 20-60 ms (a 'negative asynchrony'), most likely because the central representation of the tactile feedback of the Wnger is synchronized with the auditory code that represents the click. Because processing times are diVerent for these two modalities, the tap has to lead the sound (for reviews, see Repp 2005;Aschersleben 2002). It remains for future studies to examine to what extent the presence of the auditory pacer was of help (e.g. serving as a timing anchor) for establishing the timing relation between the Wnger tap and its visual feedback.
Another question is to know what the maximum delay is to which the brain can still adapt. Of relevance is a study by Heron et al. (2009) that tested a substantially wider distribution of delays than used here (i.e. 50, 100, 200, 400, and 800 ms). In this study, participants pressed a mouse button Wve times. During the Wrst four taps, an auditory, tactile, or visual stimulus was presented with a Wxed delay ('exposure phase'), and with the Wfth mouse click, a sensory stimulus was presented with a variable delay (test stimulus). Participants judged whether the Wnal stimulus appeared before or after the button click. Their results showed that the PSS was shifted as a function of exposure delays up to 200 ms, but then the eVect declined for the 400 and 800 ms delays. The authors suggested that at the two largest delays, there was no 'feeling of agency', which then reduced recalibration. In the light of these Wndings, it is reasonable to assume that the 120-ms temporal delays used in the present study did not exceed the boundary for motor-visual unity. Further research is required, though, to examine whether the same criterion of agency holds for natural feedback and for other modalities like audition, as it may well be that for natural feedback (as in audio-visual speech) the criterion may wider.
Another relevant feature of our study is that we only used a relatively short-exposure phase of ten Wnger taps to induce temporal recalibration, while others have commonly used a single and longer exposure period of, for example, 3 min followed by short 'top-up' exposures (see Di Luca et al. 2009;Hanson et al. 2008;Vroomen et al. 2004;Vroomen 2007, 2008;Fujisaki et al. 2004). In a recent study, Wozny and Shams (2010) also reported a fast recalibration process in the Weld of spatial recalibration (see also Frissen et al. 2012). In this study, it was shown that recalibration of sound location by a displaced visual stimulus did occur after just a single-exposure trial. This Wnding suggests that the modiWcation of a sensory map may not necessarily require the accumulation of a substantial amount of evidence, but that it operates in a fast and continuous fashion. This Wts with the observation that we obtained temporal recalibration after only a limited amount of exposure, though further testing is needed to examine the exact time course and whether there is accumulation across trials.
A related question is the extent to which temporal recalibration dissipates. Machulla et al. (2010) explored this by testing whether the strength of temporal recalibration decays over time or whether it declines due the presentation of new stimuli in the test phase. Their data showed that recalibration did not dissipate over time provided that no new sensory information was presented. Only when information was presented that diVered from the stimuli used during adaptation, temporal recalibration diminished. Although the set-up of our study is not suitable for making strong claims about the time course of temporal recalibration, it seems safe to conclude that for the duration that the test phase lasted (i.e. Wve taps in total), and it did not completely undo the recalibration that was built-up during the relatively short-exposure phase (i.e. 10 taps).
To conclude, then, our results demonstrate that participants can adapt to a delay in the naturalistic visual feedback of a self-initiated motor action. This is remarkable, because the timing of visual feedback is in a natural situation usually without delay. While being exposed to delayed visual feedback, participants most likely adjust the point of subjective simultaneity (and sensitivity) of motor-visual synchrony, though other mechanisms may also be at play. Further research is required to examine the time course and the extent to which other modalities than vision can adjust the timing of naturalistic feedback.