People engage in a variety of rhythmic motor behaviors, including walking and running, which are defined in part by a regular period or rate at which the rhythm recurs. A wide range of movement rates are available, but the range is always limited and in general there is an optimal rate that an individual gravitates toward for any given behavior. These optimal rates are estimated using the spontaneous production rate (SPR), which is the rate at which one produces sequences of sounds in the absence of external cues. We address rhythmic behaviors that are used to communicate via audition: music and speech. These behaviors are associated with different cognitive domains, referring to a class of information that may be associated with a distinct cognitive network, possibly associated with distinct modules (Fodor, 1983). Music and language may constitute separate domains according to some claims (e.g., Patel, 2008; Peretz and Coltheart, 2003). With respect to the rate at which people produce music and speech, two questions emerge: First, is there any association between an individual’s optimal rate for speech and their optimal rate for music production, and—if so—the second question involves the basis on which that association is made.

The second question is of particular interest given that the production of music and language can vary both in terms of the actions used to generate auditory feedback, and in terms of the effector system used to produce the actions. We define effector systems as a collection of muscles and joints, such as finger or tongue movements, that function in a coordinated way in order to execute the goals of an action, forming synergies (Bernstein, 1967). The present study addresses the role of these factors in the Spontaneous Production rates (SPRs) at which people produce rhythmic sequences like speech or music.

We focus primarily on how individual differences are associated across music and speech production tasks, and not on the average rates at which each task is produced. Recent evidence suggests that music and language from around the world differ with respect to the rate at which they are produced, with considerably faster rates for the rhythms formed by successive spoken syllables than by the onsets of musical notes in a melody (Ding et al., 2017; Ozaki et al., 2024). The fact that these differences exist does not necessarily indicate that different timing systems contribute to an individual’s optimal rates of production across domains. It is well known that different gait patterns within species, including horses (Hoyt and Taylor, 1981) and humans (Kung et al., 2018) are associated with different optimal rates, for example with faster optimal rates for running and slower optimal rates for walking. These differences can be attributed to changes in the attractor landscape within a single system when a control pattern (here, rate) varies, and do not require any attribution to the use of different timing systems for the different coordination patterns (e.g., Kelso, 1995). In a similar way, it is possible that SPRs for speech and music production may vary in overall rate while being based on the same timing system. Given this flexibility, we focus on whether an individual who chooses a relatively “fast” rate for one task (compared to other individuals) also chooses a “fast” rate on a different task, even if the two rates differ considerably from each other. If so, we may say that this individual timekeeper engages a “fast” timekeeper during both motor tasks, while varying rates flexibly from instance to instance in response to other contextual factors.

Results from previous studies offer mixed evidence concerning whether SPRs are correlated across music and language production. Whereas Pfordresher et al. (2021) found no significant correlation between SPRs for speech production and the production of melodies on a keyboard, Coleman et al. (2024) found significant associations between production of speech and production of sung melodies. These results are consistent with the view that SPRs are based on the effector system used, given that the correlated SPRs in Coleman et al. were found when speech and music were produced via the same system. Similarly, Engler et al. (2024) found that tapping rates were more strongly correlated when participants tapped under conditions of varying auditory feedback (taps producing a melody, a single tone, or silent) in contrast to correlations between tapping rates with either hand clapping or walking. Constraints on optimal rates due to effector system may reflect the biomechanics associated with effector systems that function synergistically when one engages in rhythmic movements. The integrated movements of limbs that vary in mass, flexibility and muscular dexterity significantly influence the optimal rate of rhythmic movements. These effects have been modeled in past research by varying masses of pendula that are moved rhythmically (e.g., Turvey et al., 1986). These studies did not, however, manipulate the presence or absence of auditory feedback, an important dimension of timing in music and speech.

Other results suggest that auditory feedback plays a role in governing optimal rates. Tranchant et al. (2022) found that the spontaneous rate of each participant’s productions of musical sequences using finger movements were not significantly correlated with their spontaneous motor tempo (fingered tapping at consistent rates in the absence of auditory feedback; Drake et al., 2000; McAuley et al., 2006), whereas other studies have found significant correlations for similar conditions (Engler et al., 2024; Pfordresher et al., 2021). Engler et al. (2024) found stronger correlations between tasks with shared auditory information—tapping to produce different melodies, or rhythmic actions that generated no sound (silent tapping versus hand clapping)—than across tasks that generate different kinds of auditory patterns. These results suggest that auditory information associated with produced actions—here the simple presence or absence of auditory feedback—may determine optimal production rates.

Other theoretical approaches associate spontaneous production rates with information based on principles of information transmission, possibly leading to differences in production rates across music and speech (Assaneo and Poeppel, 2018; Doelling et al., 2019). Thus, differences in SPRs may result for actions that lead to music or speech, independent of effector system.

The present study tests three hypotheses concerning the basis on which timekeepers may or may not be distinguished. The first (H1) is that spontaneous rates are constrained by the biomechanics of the effector system used to generate a rhythm. Some aforementioned results support H1: Whereas Coleman et al. (2024) found significantly correlated SPRs across speaking and singing, which share the same (vocal) effector system, Pfordresher et al. (2021) found no significant association (and a near zero correlation) for SPRs from speaking and playing melodies on a piano. The current study tests whether this dissociation is replicated within a single study.

The second hypothesis (H2) is that SPRs are influenced by the presence of auditory feedback generated from actions. Recent studies provide mixed evidence. Whereas Tranchant et al. (2022) found no association between silent tapping and a condition in which tapping led to a sequence of melody pitches, Pfordresher et al. (2021) found a significant but modest correlation between silent tapping and SPRs for piano production. As shown in Table 1, H2 predicts strong correlations between spontaneous rates among sung melodies, speech, and tapped melodies, for which feedback is present, but weak or non-significant correlations between any of these conditions and silent tapping.

The third hypothesis (H3) is that SPRs are based on the acoustic patterning of domain-specific information that is produced. This hypothesis predicts that SPRs will be correlated within domain (i.e., across production of music via singing and on a keyboard), but not across domains (i.e., music versus speech production). Although the results of Pfordresher et al. (2021) are consistent with H3, the results of Coleman et al. (2024) are not. However, neither of these studies included comparisons within a domain across different effector systems. A recent study that examined synchronization that crossed effector system (hand claps versus articulation) with the domain associated with the auditory target (syllable versus tone), was consistent with the idea that timekeeping is independent of effector and domain for accurate synchronizers (Mares et al., 2023). The current study uses a comparable design and can address whether this generality extends to SPRs.

We report an experiment that tested these hypotheses by having participants generate SPRs for short memorized sequences. Participants included those with and without musical training. Tasks that could be produced easily (without formal training) were presented to each participant in four conditions: Speaking short phrases, singing familiar melodies on the syllable ‘da’ (Sung Melody), producing familiar melodies by tapping on a keyboard (Tapped Melody), or Silent Tapping. Participants were given different melodies for the Sung Melody and Tapped Melody conditions to avoid carryover effects across conditions. Table 1 illustrates the relationship between these experimental conditions and the three hypotheses. Whereas Speech and Sung Melody conditions both involve the vocal effector system, Tapped Melody and Silent Tapping conditions both involve the hand and finger effector system. H1 predicts shared SPRs within the verbal and within the tapping conditions, but not across these systems. The Silent Tapping condition was used to test H2: If the presence of auditory feedback determines SPRs, then production rates in this condition should not correlate with the Tapped Melody condition. H3 predicts a significant association between Sung and Tapped Melody conditions, but no correlation between these conditions and the Speech condition.

Figure 2 show the distribution of mean IOIs across participants. Bars in each panel represent the mean IOI across trials, and error bars reflect variability across trials for that participant and task. Participants are ordered in each panel according to their rank order for Sung Melody IOIs, and axis scales vary to show the range of IOIs in the task (e.g., note that smaller Y-axis range for Sung Melody trials than music trials). These plots verify that individuals varied significantly from each other on each task, thus validating the use of correlation to assess the degree of covariation across tasks.

Figure 3 displays associations between SPRs across tasks. Each point in each plot represents the joint mean SPR for two conditions. Because each condition participates in three different bivariate correlations, we adopted a Bonferroni correction and considered correlations at p < 0.017 to be statistically significant. Pearson correlations are shown in the leftmost column of Table 2. All correlations that compared conditions in which actions led to auditory feedback (SPR tasks), yielded significant correlations. By contrast, all correlations with Silent Tapping (bottom row of Figure 3) did not yield significant correlations (p > 0.10 for each).

We also analyzed the results shown in Figure 3 using the Bayes Factor criterion (Rouder et al., 2009). Using the Bayesian correlation function in JASP version 0.17.2.1 (JASP Team, 2023), the Bayes Factor criterion (BF) tests the likelihood of the alternative hypothesis (significant correlation, BF₁₀) over the null (no correlation) or support for the null over the alternative (BF₀₁). Because only positive associations of SPRs across tasks are interpretable, we treated all tests as one-tailed. BF values, shown in the middle and right columns of Table 2, indicated overwhelming support of the alternative hypothesis (i.e., a significant correlation) for correlations of Sung Melody with both Tapped Melody and Speech, and the association between Speech and Tapped Melody SPRs yielded strong support for the alternative hypothesis. By contrast, all associations involving Silent Taping yielded strong support for the null hypothesis. These results were consistent with those found for frequentist (Pearson) correlations, while providing additional insights on whether results support the null hypothesis.

We also analyzed whether this pattern of results differed by language experience or musical background. The current sample included a sizable proportion of multilingual participants and included predominantly nonmusicians. Although the vast majority of multilinguals in our sample were L1 English speakers, it is possible that experience learning the sound patterns of other languages may influence speech SPRs. Table 3 shows the across-task correlations for all participants in comparison with multilingual and monolingual participants. The same basic pattern is found in all groups, although the correlation between Tapped Melodies and Speech (more modest in size for the entire group than the other significant correlations) was non-significant for the two smaller subsets (despite being of higher magnitude for multilingual participants than for the whole sample). We take this to be simply a byproduct of sample size. We also analyzed correlations for those participants who were categorized as non-musicians, defined as having fewer than 6 years of formal training on any musical instrument or singing. These correlations also yielded the same pattern as found for the entire sample. We did not run correlations on participants defined as musicians given how few participants were categorized as such (n = 10).

We analyzed correlations of the Silent Tapping condition with other conditions based on its relative position within the order of tasks. One complication with interpreting this condition as representing the absence of feedback is that in certain situations participants may imagine auditory feedback resulting from their taps (Brown and Palmer, 2012, 2013; Highben and Palmer, 2004). This is particularly likely for participants who experience the second silent tapping trial after the Melody Tapping condition, which is similar to Silent Tapping with respect to the effector system. This happens in the second of two orders that a participant could experience (see Methods). Thus, we re-analyzed all correlations of Silent Tapping with the other conditions, separately for the first and second trials of Silent Tapping (trial 1 always began the session), and further separating correlations with trial 2 for participants who experienced trial Order 1 (Silent Tapping precedes Melody Tapping) versus those who experienced trial Order 2 (Silent Tapping follows Melody Tapping). Spontaneous tapping rates for the second trial of Silent Tapping did correlate significantly with Melody Tapping for participants who experienced trial Order 2, r(27) = 0.35, p = 0.03, suggesting that auditory imagery associated with Melody Tapping may have carried over to the later Silent Tapping trials in these cases. No other correlations of any SPR trials with Silent Tapping were significant (all p > 0.10).

We next addressed whether SPRs are reliable across trials within a given task (Figure 4). It is possible, for instance, that the lack of correlation for the Silent Tapping conditions in most cases may reflect generally unreliable production rates for this task. We assessed reliability by averaging IOIs across the first trial for each item in a task (for Silent Tapping, this is just the mean produced rate for a single trial) and comparing this with the comparable mean for the final trial (trial 3 for production of Speech, Sung Melody, or Tapped Melody; trial 2 for Silent Tapping). All correlations were strong, positive, and statistically significant, indicating high reliability. High reliability between Silent Tapping trial 1 (before any auditory feedback trials) and trial 2 (in middle of study) indicates that any effects of auditory imagery did not significantly modulate the pattern of individual differences in tapping rate for this task. The plot in Figure 4D removes one participant from the Tapped Melody condition whose mean IOI for trial 1 (1.98) was highly discrepant from trial 3 (0.35), in contrast to other participants. When this participant was included, the correlation was much smaller but still significant at p < 0.05, r(56) = 0.25, p = 0.027.

Although our main focus is on correlations across tasks with respect to individual differences in SPR, we also analyzed mean SPR across tasks to assess whether the present data replicate results found in other studies. The mean IOI across all trials and items for each participant and task were analyzed via a within-participants single-factor ANOVA, which yielded a strong and significant effect of task, F(3, 171) = 113.43, p < 0.001, η p 2 = 0.67. Means and distributions are shown in Figure 5. Post-hoc tests (Bonferroni corrected) revealed significant contrasts across all pairs of conditions. Mean IOIs thus proceeded in an ordered manner with Speech (M = 0.223, SE = 0.005) < Sung Melody (M = 0.339, SE = 0.008) < Tapped Melody (M = 0.387, SE = 0.009) < Silent Tapping (M = 0.763, SE = 0.042). This reflects what has been reported elsewhere (Coleman et al., 2024; Pfordresher et al., 2021). It is worth noting that none of the conditions approximates the 0.6 sec IOI rate first reported by Fraisse (1982), and often repeated in the literature (e.g., Honing, 2013), as shown by the light gray line. The correlations across tasks that we report thus were found in the context of mean differences in production rate that are typical for the tasks we used.

We used a set of tasks to assess whether individual differences in spontaneous production rates (SPRs) reflect shared timekeeping mechanisms based on the effector system used to generate rhythmic actions and/or the auditory feedback generated by those actions. We tested three hypotheses concerning the basis for the selection of an optimal production rate, observed via SPRs: basis on shared effector systems (H1), basis on the presence or absence of auditory feedback (H2), or basis on domain-specific differences in the structuring of information within a domain (H3). Table 4 provides a qualitative summary of how well the pattern of correlations across tasks fit these hypotheses, using both positive (significant correlation) and negative (nonsignificant correlation) evidence. Results clearly support the auditory feedback hypothesis over the other hypotheses. As shown in Table 4, all significant and non-significant correlations are consistent with this hypothesis, and with the results of Tranchant et al. (2022). By contrast, the pattern of results is only consistent with other hypotheses at rates suggesting chance performance.

The primary finding of this experiment (H2) was that spontaneous rates vary across individuals in similar ways for tasks in which production generates auditory feedback (such as the SPR, spontaneous production rate), but not for tasks in which production does not generate auditory feedback (such as the SMT, spontaneous motor tempo). These findings are consistent with Tranchant et al. (2022) who speculated that differences in SPR and SMT values within individuals may be related to the fact that the auditory feedback present in SPR tasks provides error correction for timing. Another possibility is that SPR tasks are used to generate auditory patterns that may convey meaningful information to a perceiver. It is possible that the engagement of motor systems in tasks that contain a goal to create perceptual patterns may be less strongly associated with biomechanical constraints than tasks that do not contain perceptual goals.

The findings reported here differ from two other studies that we know of that reported positive correlations between SMT and SPR tasks. Engler et al. (2024) reported positive correlations between conditions similar to our Tapped Melody condition and silent tapping condition, and Pfordresher et al. (2021) reported a significant correlation between spontaneous rates of production of melodies on a piano and silent tapping. We suggest that these differences reflect sensitivity of SMT tasks to carryover effects from condition orders. Similar to what we reported in the present Results section, Pfordresher et al. (2021) found that the significant correlations across conditions varied depending on whether SMT was measured before or after the piano production task, with a significant correlation found for SPR and SMT only when SMT was measured after SPR. Although Engler et al. (2024) did not report such an analysis, they note that their SMT condition never occurred at the start of the session, but instead always occurred at a position intermingled with several other conditions (including SPR). Thus, we propose that other positive correlations between SMT and SPR may reflect a carryover of auditory imagery and/or motor imagery commonly seen in musicians (Highben and Palmer, 2004). An important implication of these results is that the use of SMT as a generalizable measure of internal timekeeping must be handled with care, given its potential vulnerability to order effects.

The current findings are also consistent with a timekeeping mechanism that is domain- and effector-general (contrary to H1 and H3), consistent with cognitive clock-counter models of timekeeping (e.g., Killeen and Weiss, 1987; Wing and Kristofferson, 1973). The correlation reported here between SPRs arising in Sung Melody and Speech production is consistent with the results of Coleman et al. (2024) and suggests domain-general timekeeping when sequences are produced with the same (vocal) effector system. Correlations of these tasks with the Tapped Melody condition suggest timekeeping that generalizes across effectors both within domain (Tapped with Sung Melody) and even across domains (Tapped Melody with Speech). The significant (although modest compared to others) correlation between Tapped Melody and Speech production was not consistent with the null (near zero) correlation between piano and speech production reported in Pfordresher et al. (2021) and deserves consideration. Some possible explanations for these different results include the use of familiar tunes that were associated with lyrics (used in the current study but not the previous study), the complexity of the fingered motor task (low in the current study—single-finger tapping—relative to the previous study), and degree of musical training (untrained participants in the current study, pianists in the previous study).

In sum, the present findings suggest that optimal rates may be defined by the incorporation of sensorimotor plans to form action-sound contingencies, based on the integration of motor actions with auditory feedback (cf. Tranchant et al., 2022). This conclusion is supported both by positive evidence (correlations across SPR tasks) as well as by Bayesian statistics that suggest support for the null hypothesis with respect to correlations across SPR and SMT tasks. The fact that music can be produced by different effector systems (including vocal and arm movements) makes this domain uniquely suitable for decoupling auditory from effector-based information.