The human body moves in rhythmic (repetitive) patterns, not only when dancing, but also during walking (step by step), breathing (in and out), and when moving our eyes, the latter being a fundamental constituent of visual cognition. The interplay between rhythm and cognition has mainly been investigated in one direction: how we perceive and produce rhythm, usually in the domain of music psychology (e.g., (31, 54, 78)). However, we here address the more general idea that cognition itself may be rhythmic in nature, especially when associated bodily systems can be regarded as inherently rhythm-based (such as eye movements in the context of visual cognition). The last decades have witnessed growing evidence for an embodied view of cognition (e.g., (8)), a view that also entails the idea that cognitive processes are essentially determined by associated bodily systems (75).

The idea that rhythms can shape cognitive processing has been discussed extensively in the domain of neuroscience. Brain activation contains rhythmic oscillations. Such synchronized activation plays a major role for cognitive processes involved in memory representations and attentional selection (for a review see (14)). Moreover, intrinsic brain rhythms shape visual (5, 24), as well as auditory (28), perception via changes in the excitability of local neuronal ensembles (44). Interestingly, brain rhythms can also be affected by perceptual rhythms. Neural oscillations can be shifted upon perceptual stimulation, an effect called entrainment (44, 43; for a review see (6)). In addition, a single stimulation in one modality (e.g., auditory) can reset the oscillatory phase for processing stimuli from another modality (e.g., visual), demonstrating cross-modal interactions in perception (16, 17, 48). Different from entrainment, auditory-driven phase resets do not need a sequence of auditory stimuli.

There are at least two findings indicating the relevance of brain rhythms for the saccade system. Saccade generation was aligned to visual perceptual oscillations (4, 30). Saccadic reaction times to a visual stimulus onset were reduced by a preceding sound, resulting in crossmodal phase reset (12). That is, attentional dynamics do affect motor behavior.

In the field of visual cognition, classical theories focus on stimulus-dependent processes. However, in recent years the idea of a close coupling of perception and action has transferred to the visual cognition domain, and an “active vision” perspective has been proposed (18), which highlights the role of eye movements as an essential component shaping the visual cognition machinery. Thus, typical everyday visual cognition tasks like reading or searching/scanning for an object have been successfully studied using eye tracking techniques (e.g., 57, 58).

In this context, two diverging views regarding eye movement control have been proposed, namely a “sequential processing” view (e.g., in reading, the EZ-Reader model, see (59, 60)) and a “modulated pulse” view (e.g., in reading, the SWIFT model, see (13); for scene viewing, the CRISP model, see (53); for sequential scanning, see (71)). According to the former, eye movements are driven by an essentially sequential cognitive process aimed at processing stimuli, for example, comprehending words (in the case of reading) or perceiving and categorizing objects (in the case of visual search). Corresponding models therefore assume that eye movements are triggered by a certain stage/level of cognitive stimulus processing (e.g., whether a certain word promises to be decoded successfully, see (60)). Thus, eye movements are triggered only when a certain level of stimulus processing has been reached (“direct control”, (51)).

However, sequential processing models do not adequately capture one crucial aspect of the eye movement system, namely the fact that the eyes “never rest”, but always move. In particular, saccades are continually generated every 250 to 300 ms (57), regardless of any cognitive processing demands. This rhythmic behavior is fundamentally different from other effector systems (e.g., those involving arms, feet, vocal utterances etc.), which usually rest when no movement is required. This special characteristic of the eyes is captured more convincingly in models assuming an autonomous pulse that triggers eye movements (sometimes referred to as “indirect control”, see (32)). Here, cognitive control is regarded as a potential modulator of this pulse (“mixed control”, see (25, 71)). For instance, the pulse is assumed to be slowed down when a difficult word needs to be processed (see (77, 13)). More specifically, prolonged fixation durations are assumed to occur either due to a decreased rate at which the saccade timer accumulates activation until a “move” threshold is reached, or due to a cancellation of an ongoing saccade program (13, 53). There is evidence for “mixed control” in a variety of tasks, e.g. reading (61, 62), scene viewing (27), and visual search (45). The “mixed control” account is also supported by the observation of two distinct saccade populations, of which only one is affected by task settings (52). Whether processing facilitation can also yield a decrease of fixation duration is currently under debate (26, 73). Taken together, this “modulated pulse” view of eye movement control seems especially suited to capture the rhythmic nature of eye movement control, which – when taking the embodiment perspective of cognition into account – in turn should shape (visual) cognition. The “modulated pulse” view is also highly compatible with the neuropsychological evidence discussed above: a saccade timer might relate to ongoing brain oscillations (e.g., (4, 12, 30)).

Based on the assumption of rhythmic cognition, we were looking for a method to test the idea of a pulse underlying both visual cognition and oculomotor action in a straightforward but simple way. Based on previous findings of inter-sensory crosstalk and cross-modal attention (e.g., (67)), we reasoned that one promising way to manipulate the putative pulse of visual cognition would be to utilize an external (auditory) pacemaker, that is, a very simple musical stimulus introducing a rhythmic auditory beat. We assumed that rhythmic patterns on the irrelevant (auditory) channel should modulate the pulse in the task-relevant visual processing channel connected to the generation of eye movements. For example, in “modulated pulse” models of saccade generation, the saccade timer is characterized by a random walk to the threshold of saccade initiation. The tempo of musical beats might affect the transition rate of this random walk and alter fixation durations accordingly.

Interestingly, previous studies in the domain of music psychology have already used research designs in which music was presented in addition to, for example, a relevant reading or search/scanning task, but with a completely different research focus, namely to address the question of whether background music (and which type of music) affects performance in the primary task (e.g., scanning: (10, 37); reading: (2, 22, 68)). For example, literature on reading performance showed impaired comprehension in conjunction with task-irrelevant background music (2, 26, 38, 41, 68; but see (42), for a beneficial effect). It has been assumed that processing music requires cognitive resources which conflict with the activation of working memory representations (e.g., (65)). This might also relate to the finding that background music increased fixation durations during scene viewing (66; but see (21)). In line with this argument, background music also impairs performance in a wide range of memory tasks (7). Interestingly, faster music impaired reading comprehension even more than slower music (68). This effect might be due to higher information load with faster music, because more musical events per time unit must be processed. Alternatively, arousal might contribute to the results. Faster music is associated with higher perceived arousal, and high arousal music might impair reading comprehension (11), as well as memory performance (7). There is one study showing faster reading with fast, compared to slow, background music, using a stop-watch to measure paragraph reading on mobile computers in a cafeteria setting (40). This study indicates that indeed eye-movement control might be affected by the tempo of background music.

Corresponding studies using visual scanning tasks have provided mixed results so far as well. Whereas instrumental and lyrical music can speed up performance when presented simultaneously with a scanning task (10), participants took longer to complete a visual search task after exposure to slow versus fast music (37). Taken together, these previous studies on the effects of background music on reading and visual search yield evidence for the possibility of inter-modal (auditory on visual) crosstalk.

In the present study, we focused on two types of visual cognition tasks which are relevant for everyday behavior: (a) text reading, and (b) sequential scanning. Those tasks differ in several aspects. For example, cognitive processing load during reading for comprehension is relatively high, since it involves many memory-based and linguistic processing demands (from syntax to semantics) to extract the meaning of the text (e.g., (33, 34, 35)). In contrast, cognitive processing load (in terms of memory-based and linguistic processing) is particularly low in our utilized sequential scanning task, since the decision of which item should be processed next is predetermined in a simple, task-inherent manner. Specifically, the search array consisted of Landolt-like rings, and the side of the opening of each ring indicated the direction of which object to scan next (“follow the gaps in the Landolt-rings”) until participants found a closed target ring (see (70)).

The two tasks differ also in a second aspect: the underlying oculomotor (instead of cognitive) control demands. Eye-movement control during reading is known to be a highly-learned process characterized by largely autonomous scanning strategies, as indicated by studies showing similar oculomotor patterns when normal text is exchanged by z-strings which lack any semantic content (29, 72). In contrast, selection of the next saccade target in our sequential scanning task is not highly learned and automatized. Instead, it rather depends on moment-to-moment decision-making that is based on meticulous visual attention processing involving detection of the gap location within the stimulus in order to program the appropriate next saccade (70)). Taken together, the two tasks thus differ on two levels of control demands, namely high-level cognitive demands (text reading more demanding than sequential scanning), and lower-level oculomotor demands (sequential scanning more demanding than highly trained and automatized text reading). While, during reading, the ongoing goal of the reader is text comprehension, during scanning, participants should mainly intend to determine the direction of the following saccade.

Our general hypothesis was as follows: if visual cognition is inherently rhythmic in nature (as assumed on the basis of “modulated pulse” accounts of eye movement control in reading and search), it should be possible to influence these processes by employing an external auditory beat of varying tempo. This influence on processing rhythms should become observable in terms of corresponding shifts in temporal eye movement parameters (i.e., a faster beat should yield shorter temporal oculomotor parameters), which in turn reflect temporal characteristics of the underlying cognitive processes.

Based on this general hypothesis, we reasoned that two outcomes are conceivable, depending on how the external pulse is processed by participants:

(a) processing of the auditory beat might consume central (comparatively high-level) processing resources. Then, adding this auditory stimulus might lead to a resource conflict resulting in general slowing of the reading or scanning task. However, the scanning task has a greater chance to be modulated by the auditory beat. Since text reading for comprehension is assumed to be more cognitively demanding than sequential scanning, the auditory beat should be more effective in the latter, because central resources in reading are consumed by the primary task (reading comprehension) and thus no resources are available for processing of the irrelevant auditory stimulus. Additionally, if the auditory beat influences oculomotor control through a more high-level cognitive processing route, one should expect to see effects especially in those oculomotor parameters that are known to be determined by more high-level, cognitive processing. For example, in reading, the central cognitively relevant unit is the word, that is, cognitive oculomotor control (in terms of the decision “where” and “when” to go next) is basically word centered (or object-centered in the case of visual object search, see (58)). Therefore, the assumption of object-based processing predicts effects on gaze durations or total reading times (which are determined by cognitive decisions based on successful word/object decoding), rather than on basic fixation durations. We will refer to this reasoning as the high-level cognitive load account.

(b) Processing of the auditory beat might operate on a lower (less cognitive) control level more specifically devoted to basic oculomotor control. On such a basic oculomotor control level, we reasoned that text reading relies on largely automatized oculomotor control routines (i.e., there is relatively low oculomotor control demand), whereas the present sequential scanning task is considerably less trained and associated with oculomotor control decisions from stimulus to stimulus. This greater demand on oculomotor control decisions might prevent any influence of the auditory beat in the sequential scanning task. Instead, oculomotor control during text reading should be affected. Given that this presumable crosstalk operates on a relatively low level of basic oculomotor control, one would expect more basic oculomotor control parameters to be affected (i.e., basic fixation durations instead of gaze durations). This is also plausible since the above cited “modulated pulse” models of oculomotor control are devoted to explaining the control of basic fixation durations. We will refer to this possibility as the oculomotor control load account.

There is one important caveat regarding the predictions for temporal parameters that needs to be considered. Previous literature on text reading and sequential scanning has already provided a nearly exhaustive picture of relevant variables determining oculomotor parameters, which together explain a remarkable portion of oculomotor processing variability (e.g., (58), for the case of reading). As a consequence, there is little room left for remaining variables (including external pacemakers) to affect these parameters. Thus, even though we consider the present hypotheses as highly relevant on a theoretical level (as outlined above), it is clear from the start that any potential effects should be very small (i.e., in the range of milliseconds). To make the observation of such small effects more likely, we decided to have participants read through not just one, but many text passages (and to scan through several displays, respectively) in order to maximize the reliability of individual performance estimates. Additionally, we utilized not just two but four different tempi for the auditory beat to further minimize the probability of observing a Type I error, i.e. a false positive result.

In order to implement a more natural experimental situation, we did not use a basic metronome as an external auditory stimulus, but instead composed a melodic-rhythmic pattern (resembling trance-like music with a non-ambiguous, continuous pulse). In this way, our study is also open to interpretation in the context of more applied research questions (i.e., regarding the influence of background music and its tempo on reading and scanning performance, see above for a brief literature review).

The tempi differed strongly, that is between 80 to 140 beats per minutes (bpm), which translates to inter-beat-intervals of 750 to 429 ms or beat frequencies of 1.33 to 2.33 Hz (when regarding the quarter notes as beat throughout, see Table 1). However, fixation durations for silent reading are usually about 225 ms and for scene viewing about 330 ms (see (57)), which translates to frequencies of 3–4 Hz. Importantly, then, the pulse of saccade generation and the beats of our stimuli were on very different time scales. Nevertheless, we expected some (albeit small) modulation effects. Note that our study does not touch on entrainment effects in the narrow sense, which in this case would be reflected in saccades locking onto the beat. First, the range of tempi makes beat entrainment unlikely. Eye movement behavior will unlikely be forced to slow down towards a rate of 1.33 to 2.33 Hz, as given by the quarter beat. However, entrainment to eighths or sixteenths is thinkable but would require a rather complex processing of the simple stimuli. Second, the auditory beats presented in our study were not synchronized in any way to the eye movement recording, rendering a direct relational entrainment analysis impossible. Instead, we simply compared temporal eye movement parameters (e.g., fixation durations) across four different auditory beat conditions to uncover a systematic effect of an external pulse.

Note. Inter-beat-intervals and beat frequencies are presented using either the quarter as a reference, which would most likely be perceived as the basis for an intermediate tempo. However, individual beat perception is difficult to predict. When tempo decreases, the perceived beat reference might shift to the eighth (or to the halves for increasing tempo).

We first compared fixation durations between tasks in the silent condition. Mean fixation durations in the reading task, M = 211 ms (SD = 23), matched what is known for typical silent reading (57). In the scanning task, mean fixation durations were somewhat longer, M = 308 ms (SD = 34). These strong baseline differences, t(38) = 17.38, p < .001, η² = .888, in addition to the very different underlying task demands discussed above, further strengthen the validity of our decision to separately analyze the reading and the scanning tasks. Results for all dependent variables can be found in Table 2 (reading) and 3 (scanning).

The present study addresses the general question of whether visual cognition is inherently rhythmic in nature by measuring the potential influence of an external auditory beat, with varying tempo, on temporal eye movement parameters. Specifically, we reasoned that a faster beat should yield shorter temporal oculomotor parameters, which in turn should reflect temporal characteristics of the underlying cognitive processes. We measured eye-movements in two exemplary visual cognition tasks: text reading and sequential scanning. These tasks were either completed in silence or with an auditory beat (simple, electronic music) at four different tempi.

The most important result is that the tempo of the beat significantly affected basic fixation durations in the reading task. Higher musical tempo resulted in shorter mean fixation durations. While the effect was notably small, several aspects strengthen the reliability of the effect. First, it should be noted that we already expected any effect to be small, since eye movements in reading are known to be strongly determined by automatized routines and linguistic processing, leaving only small room for further external determinants. Second, the effect consists of a highly systematic (monotonous) pattern in the expected direction across all four tempo conditions. The implementation of four (instead of only two) tempo conditions makes it unlikely that such an effect represents a random false positive. Finally, the effect was successfully replicated in an independent parallel study involving comparable demands ((47); see below for further details). On the backdrop of our theoretical reasoning outlined in the introduction, this central result is of high theoretical significance.

Specifically, the fact that the Bayes factor analyses provided substantial evidence against an influence of the auditory beat on oculomotor control (mean fixation durations) in sequential scanning but substantial evidence for a reliable effect during text reading can be regarded as evidence for an oculomotor control load account and against a higher-level cognitive load account regarding the underlying level of processing for the auditory stimulus. While oculomotor control in the present sequential scanning task necessitates moment-by-moment decisions on where to move next, oculomotor control in reading relies on highly automatized motor routines, thus leaving more room (in terms of motor control resources) for the processing of the auditory stimulus. If processing of the beat had occurred on a higher cognitive level, one would have expected a strong influence of the beat tempo in the scanning task, but no substantial influence in reading (a pattern we clearly did not observe).

Our present interpretation in favor of the oculomotor control load account is further corroborated by the observation that gaze durations (as opposed to basic fixation durations) were not affected by beat tempo during reading. This is in line with “modulated pulse” models of temporal oculomotor control (e.g., (13)), which also assume that an autonomous rhythmic timer determines basic fixation durations to a large degree. Most likely, the external auditory pulse affected the speed of the autonomous saccade timer (rhythmic crosstalk between auditory input and oculomotor output). In contrast, a cognitive account of the effects of the auditory stimulus would have predicted an effect on temporal parameters that reflect more cognitively driven decisions (e.g., gaze durations reflecting the decision of when to fixate the next word, see (57)). However, we did not find any significant effect on gaze durations, which speaks against a higher-level cognitive load account of the present data.

Furthermore, the higher-level cognitive load account would have predicted a general detrimental effect of the processing demand of the additional auditory stimulus on primary task performance. However, this was not observed in either of the two tasks. While sequential scanning was completely unaffected by the presence of music, total reading times and reading completion times even decreased when the auditory stimulus was present, while text comprehension was not hampered. Several explanations for increased reading speed in the presence of music appear conceivable. For example, music might have increased motivation to complete the task based on positive emotion induction. Or, given its rhythmic nature, the music might have helped participants to stay focused. In addition, the music conditions likely increased arousal in comparison to a silent control, since increased emotional arousal is known to be generally associated with speed (e.g., (3, 15, 23)). Higher arousal might have resulted in the overall effect of speeding up in the mere presence of music; however, it is unclear, why this would affect reading but not scanning.

One difference between tasks is the involvement of articulatory processes in reading (33), as opposed to sequential scanning (where participants might only articulate self-instructions such as “up”/”right” etc.). One might think that when articulating the text silently, participants might match the tempo between articulation (stressed/unstressed syllables) and the auditory beat. In that case, however, the observed effect should have been much more pronounced (similar to an entrainment hypothesis, see Table 1), which renders the assumption of a strategy to match articulation speed with the auditory beat unlikely.

It is important to note that the two tasks, sequential scanning and reading, differed substantially regarding both the underlying processing requirements and the actual oculomotor characteristics, which prohibits any direct statistical comparison between tasks. For example, fixation durations in the scanning task were much longer, and thus the frequency of the oculomotor pulse was actually closer to the chosen frequencies of the auditory beat, but still differed strongly. Future studies on the influence of beat tempo on reading or scanning, particularly when testing entrainment in the sense of typical corresponding studies (see Introduction), should test the effect of pacemaker tempi higher than those utilized in the present study, which were clearly slower than the oculomotor rhythms. Here, we were less interested in such direct entrainment, but rather in the ecologically more prominent situation of background music during visual cognition. Hence the chosen tempi spanned from very slow to rather fast music.

On a more theoretical level, the fact that our results favor an oculomotor control load account rather than a higher-level cognitive load account also has implications for the underlying cognitive architecture. Previous theoretical frameworks regarding the structural layout of the cognitive system can be divided into models assuming one common central resource for cognitive processing (e.g., (39)) and models assuming a more modular layout (19) involving distinct processing resource pools for the different processing modules (e.g., (74)). A single central module or resource account would have predicted costs associated with additional processing demands, which we did not observe. Thus, our results rather indicate the presence of separate processing modules with rather independent resource pools, one for central, higher-level cognitive processing (linguistic processing, comprehension etc.), and one for more peripheral (and more low-level) oculomotor control, a level on which auditory beat processing appears to operate. However, it appears likely that, despite some degree of modularity, there is still room for inter-modular crosstalk to occur (see (36)), which explains the possibility of an influence of the auditory beat on the oculomotor control system.

Interestingly, the reliability of our main finding is further corroborated by a similarly small but significant effect of musical tempo on fixation durations in free scene viewing as reported recently (47). Note, however, that this study had a somewhat different theoretical focus, comparing the effect of auditory beats between musicians and laymen, as well as between two different musical styles (funk, techno). Only two different tempi were chosen (about 102 and 144 bpm) and task affordances (e.g., differences in involvement of cognitive or oculomotor control processes) were not manipulated. While tempo had an effect similar to the one we observed in our reading task, musical expertise and musical style did not. Despite the differences in theoretical focus, this study therefore nicely confirms our present results and interpretation when assuming that free scene viewing can also rely more on automatized oculomotor scanning routines than the scanning task used in our present study, which necessitates moment-to-moment decisions regarding saccade targets based on the identity of the currently fixated object.

From a more practical viewpoint, our study also speaks to the issue of effects of background music on cognitive task processing. In contrast to other studies (2, 20, 68), we did not find a detrimental effect of music on reading comprehension. Our participants’ comprehension performance was unaffected by the auditory stimulus. In this context, it is important to note that we used very simple musical stimuli, specifically composed for our study. Other studies applied classical or popular music. Such music differs from our stimuli in terms of its complexity and familiarity. For example, music samples taken from top-ten lists might trigger individual memories, associations, and stronger emotional states, potentially conflicting with the reading comprehension task at hand.

Our main finding motivates further research on the relation between auditory rhythms and saccade generation. Even though effects are small and sometimes cannot be observed (see (21), in this special issue of “Music and Eye-Tracking”), they are of high theoretical relevance for understanding crossmodal perception. There are several options for future studies, for example, using experimental designs similar to studies on crossmodal resets in neuroscience (e.g., (12)) and analyzing periodicities for saccade generation (see for example (1)), or using tasks that are suited to uncover different saccade populations (with or without direct control), such as the stimulus-onset delay task (see (52)).

In sum, we demonstrated a theoretically important tempo effect of an external auditory beat on basic eye-movement control in reading. The results are interpreted in favor of modulatory processes that affect the speed parameter of the saccade timer, supporting the assumption of inherently rhythmic underlying processing in exemplary visual control tasks. The present results add to the growing evidence for an embodied view of cognition (e.g., (8)), a view that also entails the idea that cognitive processes are essentially determined by associated bodily systems (75). Thus, tasks requiring rhythmic oculomotor behavior appear to rely on corresponding rhythmic processing strategies, as evidenced by the influence of the task-irrelevant, external pacemaker tempo.

The author(s) declare(s) that the contents of the article are in agreement with the ethics described in http://biblio.unibe.ch/portale/elibrary/BOP/jemr/ethics.html and that there is no conflict of interest regarding the publication of this paper.

We thank Lena Plückebaum and Ruth Julier for data collection. Special thanks to Maria Heuring for her support with the data preprocessing. Special thanks as well to Fabian Greb for generating the musical stimulus. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Correspondence concerning this article should be addressed to Elke.Lange@aesthetics.mpg.de