Effects of coda consonant on syllable-initial Voice Onset Time (VOT) in English and French: The Syllable-Body Stiffness Hypothesis

Abstract

The aspiration of syllable-onset voiceless stops in English, indexed as positive Voice Onset Time (VOT), is shorter before a voiceless consonant in the syllable coda than before a voiced one. Vowel duration is also known to be shorter before voiceless codas. We propose a unitary framework in which both patterns emerge from a cohesive change in onset and vowel articulatory stiffness modulated by coda constriction. In Experiment 1, onset VOT was measured from acoustic signals in American English monosyllabic words with a mildly constricted coda (/ɹ/) vs. a highly constricted coda (voiceless stop), as in par vs. pat. In Experiment 2, the effect was investigated for the first time in French mono- and disyllables containing the rhotic /ʁ/ vs. a voiceless stop, as in part /paʁ/ vs. patte /pat/ and patrie /patʁi/ vs. pâté /pate/. VOT in English was significantly longer before coda /ɹ/ than coda voiceless stop. In contrast, no significant effect was observed in French, the French rhotic patterning as a voiceless obstruent. Our English results suggest that voiceless obstruents, which have higher stiffness than voiced ones, are anticipated with a stiffness increase beginning in the onset, making stiffness an important aspect of the syllable’s role in motor control. We propose the Syllable-Body Stiffness Hypothesis to account for both the VOT and vowel duration differences: Stiffness increases in the syllable body (onset and vowel together) to anticipate the intraoral pressure buildup and articulatory force of voiceless codas, producing faster articulations and shorter acoustic durations, including shorter onset aspiration.

1. Introduction

Voice Onset Time (VOT) is a major phonetic correlate of stop voicing distinctions. VOT measures the time interval from the stop release to the onset of vocal fold vibration in stop-vowel sequences (Lisker & Abramson, 1964), found for many languages. The voiceless aspirated stops of English have long positive VOT, while the voiceless unaspirated stops of languages such as French have short positive VOT (Cho et al., 2019). One main goal of VOT research has been to understand why VOT varies depending on surrounding sounds (Klatt, 1975; Nearey & Rochet, 1994). However, one contextual effect remains understudied: the effect of coda consonant on onset VOT. In the present study, we investigate how coda consonants influence the aspiration of voiceless stops in the preceding onset, indexed as onset positive VOT.

Four previously published studies have reported longer onset aspiration when the coda is a voiced stop, a nasal, or a liquid, compared to a voiceless stop (Choi et al., 2016; Mielke & Nielsen, 2018; Port & Rotunno, 1979; Weismer, 1979). The proposed explanations are incompatible and incomplete (see below). Understanding why onset aspiration varies with the coda may shed light on the long-standing issue of the syllable as a speech production unit (Kozhevnikov & Chistovich, 1965; Rousselot, 1897; Stetson, 1928).

This coda-to-onset effect should be studied with reference to the better-known influence of final coda voicing on vowel duration. Changes in vowel duration due to coda voicing were already noted in early studies of German and Spanish (Meyer, 1904; Navarro Tomás, 1916), and later in multiple studies of English (e.g., Chen, 1970; House & Fairbanks, 1953; Raphael, 1975). It has been found in most languages; it is smaller than the English effect in some languages (French, Russian, Norwegian, Korean, and Spanish; Chen 1970: 138, Fig. 2) and very small or absent in others (Polish; Coretta, 2019; Keating, 1980). It would be parsimonious to have a unified explanation for coda-driven duration effects in both the onset and the vowel.

Fig. 2.

A schematic representation of articulatory gestures and audio signal for the syllable /pa/. The aspiration interval, measured as VOT, encompasses the acoustic consequences of supralaryngeal release and vocal fold adduction. In this model, both laryngeal and supralaryngeal gestures are controlled by a stiffness parameter k. The Syllable-Body-Stiffness Hypothesis proposes that the stiffness of both gestures is modulated by upcoming stiffness demands, thereby influencing movement duration and, in turn, aspiration in the syllable onset.

To further explore the coda-to-onset VOT effect, we partially replicated Mielke and Nielsen (2018), the only report to date of onset aspiration being longer before coda liquids /ɹ, l/ compared to coda voiceless obstruents. Confirming the effect of coda rhotics on onset positive VOT in American English (English; Experiment 1) will establish a comparison test in Parisian French (French; Experiment 2).

To explain the effects of voiceless coda together, we propose a unified account for both the onset and vowel, the Syllable-Body Stiffness Hypothesis: Depending on its voicing, constriction, and intraoral pressure, the coda consonant requires a more or less forceful closure gesture; the stiffness of this gesture modulates the stiffness in the preceding onset and vowel (the syllable body), influencing their articulatory and acoustic durations. We next lay the foundation for our hypothesis by reviewing how syllable structure, articulatory dynamics, and intraoral pressure influence the production of different consonants.

1.1. Background on syllable structure: Rhyme and body in phonetics and phonology

The role of the syllable in phonetic variation has been studied for over a century (Kozhevnikov & Chistovich, 1965; Krakow, 1999; Rousselot, 1897; Stetson, 1928). The syllable position of a sound strongly influences the temporal organization of its articulatory gestures and its acoustic structure (Sproat & Fujimura, 1993), as well as its perception (Whalen, 1989). In dynamical approaches to speech, such as Task Dynamics and Articulatory Phonology, syllable variation stems from gestures being differently synchronized—or “coupled”—depending on their syllable position: In a CVC syllable, the onset and vowel, being synchronized in an in-phase mode, are initiated near-synchronously; the vowel and the coda, being synchronized in an anti-phase mode, are produced sequentially (Browman & Goldstein, 1988; Goldstein et al., 2006; Nam et al., 2009; Saltzman & Byrd, 2000; but see Tilsen, 2016). Syllable position also modulates the aspiration of a stop: English voiceless stops are typically aspirated, and have long positive VOT, in the onset of a stressed syllable, but not in the coda (Jones, 1922; Lisker & Abramson, 1967).

Within the syllable, two sub-units have been distinguished, the syllable rhyme and the syllable body (Derwing & Eddington, 2014; Selkirk, 1982). As shown in Fig. 1, the rhyme consists of the nuclear vowel followed by any coda; the body comprises the vowel and any onset consonant. Note that arboreal phonological descriptions cannot allow both body and rhyme to be structures within the same syllable (Goldsmith, 2011). We propose that the syllable body provides a parsimonious unit for describing the articulatory origin of coda-to-onset VOT effects, as argued below.

Fig. 1.

Structure of a /CVC/ syllable, including the syllable body (onset and vowel together) and the rhyme (vowel and coda together).

The syllable rhyme and body are important units for a variety of phonological processes. The rhyme is involved in well-studied linguistic patterns such as stress assignment (Goedemans & Hulst, 2013; Gordon, 2007; McCarthy, 1979), rhythm variation (Cooper et al., 1988), and phonotactics (Fikkert, 1994; Paramore, 2025). Rhyme duration has also been related to VOT variation (Allen et al., 2003), with longer VOT before longer rhymes.

The syllable body is much less studied than the rhyme. Within the syllable body, the articulatory gestures realizing the onset consonant and the vowel are tightly synchronized to each other (Goldstein et al., 2006; Liu et al., 2022), suggesting that the syllable body functions as an articulatory unit. Moreover, electrical brain responses specific to perception of the syllable body have been observed, suggesting a distinct role in lexical access (C. Feng et al., 2019). The syllable body also relates to phonological awareness and literacy: Kindergarteners from linguistic backgrounds as different as Hebrew, Russian, and Korean show increased phonological awareness in word-splitting tasks targeting the syllable body (/CV-C/) compared to the rhyme (/C-VC/) (Kim, 2007; Kogan & Saiegh-Haddad, 2023). Tilsen (2016) reviews different lines of empirical evidence from the literature on speech development in children, emphasizing the importance of grouping onset and vowel together (/CV-C/) as a critical stage in the emergence of articulatory coordination.

These crosslinguistic results highlight the phonological salience of the syllable body at the psycholinguistic level. However, it is unclear what mechanism maintains such timing relations within the syllable body. In our view, this mechanism relates to articulatory stiffness: The syllable body may emerge as a unit because onset and vowel gestures vary in stiffness cohesively to anticipate different stiffness degrees in the coda (see below).

The phonological literature has emphasized the syllable structure in which there is only one major division, between the onset and the rhyme (Goldsmith, 2011; Selkirk, 1982). The psycholinguistic evidence just reviewed indicates that some languages may instead need a syllable body element. Arguments have also been made for a flat structure (Davis, 1982; Derwing & Eddington, 2014). It may be that no formalism would allow a language to have both rhyme and body organizations for the same word (as argued by Davis, 1982). Similarly, a dynamic computational account might allow for either structure to emerge (Goldsmith, 1995), but it is not clear that this model would allow two different structures for the same input. Even a proposal for bidirectionality in a higher-order process, stress assignment in Lenakel, would not apply to a single word form but rather separately for nouns and verbs (Goldsmith, 1995). By exploring the coda-to-onset aspiration effect, the present study is expected to provide further data for the representations of syllables, as will be addressed in the Discussion.

1.2. Coda effects on onset aspiration: A possible stiffness effect

The effect of codas on onset aspiration has been reported in four published studies on English (Choi et al., 2016; Mielke & Nielsen, 2018; Port & Rotunno, 1979; Weismer, 1979) and also touched on in an unpublished dissertation on Thai (Onsuwan, 2005). Weismer (1979) and Port and Rotunno (1979) found that onset aspiration in American English monosyllables, measured as positive VOT, is shorter when the coda is a voiceless stop than when it is a voiced stop or nasal. Additionally, in both studies, vowel duration varied in the same direction as onset VOT, with shorter vowels before coda voiceless stops. Port and Rotunno's (1979) results further showed that coda-driven onset VOT variation is not merely a byproduct of vowel duration changes: Although both onset aspiration and vowel duration shortened before voiceless codas, the two measures were only weakly correlated, and the VOT effect (13%) was considerably smaller than the vowel effect (43%; 1979: 659, Table IV). They concluded that the coda “directly” controls onset VOT (1979: 658), without the mediation of the vowel. However, the mechanism underlying the direct coda-to-onset control remained unclear, as they proposed that a phonological rule deliberately alters onset VOT to cue coda voicing. A similar treatment is found in Choi et al. (2016), who also observed shorter onset VOT in forms like pet than ped and proposed that changes in coda voicing reshape “the temporal structure of the entire syllable” (2016: 13). Yet the origin of this temporal restructuring remains ambiguous: If the duration of onset aspiration is a parameter that the phonology deliberately manipulates to cue coda information, why is it specifically a shorter aspiration that anticipates a voiceless coda rather than a longer aspiration? It is unclear whether duration can be meaningfully treated as an independent parameter of action, since timing at relatively small temporal scales such as VOT is typically understood to emerge from motor coordination rather than being explicitly specified (Todorov & Jordan, 2002). These phonological accounts therefore do not fully explain the origin of coda-driven aspiration differences in the onset.

Port and Rotunno (1979) do, however, acknowledge that phonological explanations may ultimately give way to “a carefully developed mechanical explanation” (1979: 654) of the coda-to-onset effect. Here, we explore the possibility that onset aspiration varies with the coda because the speech production system varies the stiffness of syllable-body gestures to anticipate different degrees of coda constriction. We test this hypothesis by comparing onset aspiration before a highly or mildly constricted coda (coda /ɹ/ vs. voiceless obstruents, /T/), partially replicating Mielke and Nielsen’s (2018) design and results.

Mielke and Nielsen (2018) reported that onset aspiration in English is longer before coda liquids /ɹ, l/ than before voiceless coda obstruents. Their finding adds a new class of sounds to the list of codas associated with a relatively long onset aspiration, alongside voiced stops and nasals. They measured word-initial positive VOT from 120 mono- and disyllabic words containing /ɹ, l/ in various positions (2018: 2168, Fig. 1), with a sample of 122 speakers. Analyses with Linear Mixed-Effects (LME) models were “based on a single repetition per speaker” for each word (2018: 2168). In their best-fitting model (2018: 2171, Table II), word-initial VOT was longer when liquids occurred in the coda (e.g., pull, pair, peal) or even later in the word (e.g., period, coolant, power). Conversely, VOT was shorter before voiceless obstruents, especially voiceless stops (e.g., pastry, puppy). Short VOT was also observed when another stop intervened between the onset stop and the liquid (e.g., paper or pantry). They further assessed whether VOT variation was merely a byproduct of changes in rhyme duration: Models with categorical liquid presence outperformed those with liquid duration included in the rhyme (2018: 2170). Moreover, the effect of coda type remained significant after controlling for vowel duration, suggesting that liquids and voiceless obstruents directly influenced onset VOT, thus excluding a byproduct of vowel duration changes. This finding supports Port and Rotunno's (1979) account of direct coda-to-onset control but raises the question of what properties of coda voiceless obstruents lead to shorter onset aspiration, and what shared property of coda liquids, voiced stops, and nasals lead to longer onset aspiration. As Mielke and Nielsen (2018) is the only report of a coda rhotic effect on onset VOT, replication is necessary to confirm the finding and clarify its underlying causes. They propose an aerodynamic account of onset VOT variation, which we review next in terms of its strengths and limitations.

For Mielke and Nielsen (2018), onset VOT may lengthen or shorten to anticipate the “airflow requirements” (2018: 2175) of later consonants. In their initial LME models, a constriction-based continuum emerged: A shorter onset VOT was predicted before upcoming fully occluded voiceless stops than before glottalized and flapped stops, or fricatives. Thus, the more constricted the following voiceless obstruent, the shorter the onset aspiration. This continuum was interpreted as an “anticipation of airflow” (2018: 2175). That is, tightly constricted voiceless stops reduce airflow more than glottalized and flapped stops or fricatives. In the authors’ view, the aspiration of word-initial stops may shorten before voiceless obstruents to mirror the airflow reduction (2018: 2169, 2175).

However, to explain the opposite effect (longer onset VOT before liquids /ɹ, l/), they propose an argument that is quite similar to that used for VOT shortening: /ɹ/ and /l/ would lengthen the preceding onset aspiration because the liquids, by “being produced with a relatively closed vocal tract”, may delay voicing onset by reducing airflow (Mielke and Nielsen, 2018: 2175). Thus, in both cases (shorter and longer onset aspiration) the same premises are invoked: vocal tract constriction and airflow reduction. This creates a conundrum: If liquids lengthen onset aspiration due to constriction and airflow reduction, then voiceless obstruents, which involve major constriction and airflow reduction, should also lengthen aspiration, yet voiceless obstruents shorten the aspiration in the preceding onset. Additionally, the idea that a postvocalic liquid may significantly delay voicing in the preceding vowel and onset by reducing airflow contradicts previous aerodynamic studies showing that liquids, with nasals and approximants, have the lowest intraoral pressure amplitudes of all consonants (Subtelny et al., 1966: 505; Table 1). Moreover, the notion that speakers deliberately manipulate aspiration airflows to mirror future airflows conflicts with the well-established view that aspiration is not an air puff added to the stop, but results from laryngeal relative to supralaryngeal timing (Lisker & Abramson, 1964): Any changes in aspiration more likely reflect changes in the articulatory maneuver from which aspiration originates.

Table 1.

Summary of published studies indicating that the onset consonant and nuclear vowel (i.e., the syllable body) undergo qualitatively similar changes triggered by the same coda types.

Reference	Acoustic segment type	Value before coda /D/, /L/, or /N/	Value before Coda/T/
Weismer (1979)	Onset aspiration	Longer	Shorter
Port and Rotunno (1979)	Onset aspiration	Longer	Shorter
Choi, Kim, and Cho (2016)	Onset aspiration	Longer	Shorter
Mielke and Nielsen (2018)	Onset aspiration	Longer	Shorter
Slater and Coleman (1996)	Onset sonorant	Longer	Shorter
Hawkins and Nguyen (2004)	Onset /l/	Longer	Shorter
Hawkins and Nguyen (2004)	Onset fricative	Longer	Shorter
Heffner (1937)	Vowel	Longer	Shorter
House and Fairbanks (1953)	Vowel	Longer	Shorter
Peterson and Lehiste (1960)	Vowel	Longer	Shorter
Chen (1970)	Vowel	Longer	Shorter
Wolf (1978)	Vowel	Longer	Shorter & peripheral
Summers (1987)	Vowel	Longer	Shorter & peripheral

These issues challenge the airflow anticipation account and highlight the need for both replication and an alternative framework. We retain the insight that onset aspiration relates to properties of upcoming postvocalic consonants but reinterpret the effect in terms of syllable-level stiffness and articulatory dynamics.

1.3. Stiffness effects seen in articulatory and acoustic measures

In the speech literature, stiffness is a property of gestures with biomechanical consequences. A stiffer gesture more rapidly achieves and releases its target (Kelso & Tuller, 1984; Saltzman & Munhall, 1989), observable as steeper articulatory transitions and higher peak velocities in point-tracking studies, or increased muscular activation in electromyographic (EMG) recordings (Beckman, 1992; Chen, 1970; Du & Gafos, 2023; Edwards et al., 1991; Fujimura & Miller, 1979; Öhman, 1967; Raphael, 1975).

In models based on dynamical systems, such as the Task Dynamics and Articulatory Phonology (Browman & Goldstein, 1992; Saltzman, 1986), stiffness is the abstract parameter k in a mass-spring system modulating the restoring force that moves the articulator toward its target, such that higher stiffness produces stronger restoring force (Iskarous & Pouplier, 2022; Kirkham, 2025). Stiffness shapes gestural timing: Gestures planned with higher stiffness produce more rapid movements. Because movement velocity covaries with amplitude, amplitude-normalized peak velocity (kˈ) has been proposed as a more valid empirical index of the abstract stiffness k (Roon et al., 2021). However, stiffness as a parameter can only be indirectly inferred from articulatory or acoustic measures, and empirical indices relate to but do not equal the algorithmic k parameter (Du et al., 2025: 6; Du & Gafos, 2023: 12–13). In the present study, our acoustic data will not allow for a discrimination among these possibilities. In the acoustic domain, evidence that muscular activity and acoustic duration are strongly related for the same segment, with increased muscular activity producing shorter acoustic durations (Raphael, 1975), suggests that increases in stiffness and articulatory rapidity of laryngeal and supralaryngeal onset gestures should correspond to shorter onset VOT durations in acoustic signals measured in the present study. Note that the prediction of coda stiffness being anticipated during the onset may not be derived from the canonical Task Dynamics model, wherein a gesture’s stiffness has a fixed value (Saltzman & Munhall, 1989). However, early studies conducted within the Task Dynamics literature already proposed that “stiffness remains effectively constant for a given gesture but is specifiable by the nervous system and can therefore be altered between gestures to produce the desired kinematic effects” (Ostry & Munhall, 1985: 641). Thus, the biomechanical correlates of algorithmic gestural parameters, such as stiffness, may be updated to respond to contextual factors. From this perspective, it appears possible that a stiff coda closure be anticipated with a stiff syllable body. This also aligns with more recent proposals suggesting that a gesture’s stiffness may be updated dynamically depending on gestural activation during planning (Tilsen, 2018).

Importantly, under the prediction that stiffness increases in the onset to anticipate coda stiffness, shortening should not be expected across all sections of the onset consonant. Stiffness primarily affects the rapidity of transitions into and out of constriction targets rather than the duration of sustained closures. Specifically, a gesture can be analyzed into four movement “landmarks”: onset, target achievement, release, and offset, defining a trapezoidal profile (Gafos, 2002: 270). Increased stiffness steepens the transitions toward and away from the target, and changes in stiffness are less readily observable in the closure plateau. Indeed, studies of consonant stiffness in languages such as English, German, Moroccan Arabic, and Spanish have measured the slopes of closing and opening transitions as indices of gestural stiffness, with steeper slopes indexing increased stiffness (Du et al., 2025; Du & Gafos, 2023; Roon et al., 2021). Additionally, it has been proposed that closure and release may entail two distinct gestures (Browman, 1994), the release having its own stiffness specification (Nam, 2007).

From this perspective, onset aspiration appears as a transition phenomenon embedding two motor transitions: the release of the supralaryngeal gesture and the closure of the glottal gesture. By our hypothesis, stiff codas should increase syllable-body stiffness and thus produce stiffer, more rapid supralaryngeal and laryngeal transitions from the onset into the vowel, with the faster supralaryngeal and laryngeal transitions resulting in shorter aspiration. More specifically, our claim is that variation in onset aspiration before different codas may arise primarily from changes in syllable-body stiffness rather than from specified differences in gestural phasing or targets. Note that this prediction (stiffer→shorter) applies specifically to transitions: stop closure intervals may behave differently. Stop closures often have slightly longer duration than voiced ones (Byrd, 1993; Crystal & House, 1988), likely due to greater compression (stiffer/faster transitions result in more rapid movements that in turn result in tighter, longer constrictions). Moreover, voiced stop closure may undergo shortening because of aerodynamic constraints on voicing (Ohala, 2011). Thus, the effect of coda voicing on the durations of preceding onset stop closure and glottal pulsing (Crystal & House, 1988; Jacewicz et al., 2021; Lisker, 1957) fall outside the present study focused on the stiffness of voiceless stop transitions, inferred from the duration of aspiration (positive VOT).

Differences in stiffness, inferred from differences in the rapidity of target achievement and release, have been invoked to characterize different linguistic contrasts. For example, the approximant /w/ reaches its labial constriction more rapidly than the vowel /u/, the timing difference reflecting the greater stiffness of /w/ (Iskarous & Pouplier, 2022: 23, Fig. A2-C; Nam, 2007: 626). Similarly, contrasts such as /p/-/v/-/w/ have been modeled in terms of differing stiffness values, with /p/ more rapidly reaching its target lip closure due to its greater stiffness (Iskarous & Pouplier, 2022: 23; Fig. A2-B). Stiffness variation has also been linked to temporal variations due to prosodic boundaries (Beckman, 1992; Byrd & Saltzman, 1998; Edwards et al., 1991) and speech rate effects (Kelso et al., 1985; Kühnert & Hoole, 2004).

In the present study, we extend this line of reasoning to coda-driven variation in onset aspiration: Highly constricted voiceless codas are hypothesized to increase syllable-body stiffness, thereby accelerating onset transitions and shortening aspiration. To test this hypothesis, we compared onset aspiration in /TVC/ words with a highly constricted vs. a mildly constricted coda (voiceless stop /T/ vs. /ɹ/). Our prediction is that shorter onset VOT will be observed before coda /T/ than /ɹ/ due to the stiffness of coda /T/ being anticipated with a rapid onset-to-vowel transition, resulting in shorter onset aspiration.

1.4. Aspiration variation indicates subsyllabic changes in glottal and supraglottal dynamics

Considering aspiration dynamics in the onset of a syllable like [paɹ] is critical to understanding onset VOT variation due to the coda. Articulatorily, during VOT, the stop’s oral constriction is released and the vocal folds transition from an abducted to an adducted position to initiate vibration for the vowel. While the folds are abducted after the stop release, the airstream generates aspiration noise, either at the glottal or higher locations, this noise being typically measured as positive VOT in the acoustic signal (Abramson & Whalen, 2017; Lisker & Abramson, 1964). From this description, aspiration noise emerges as an aerodynamic and acoustic byproduct of the time elapsing between stop release and vocal fold adduction, how rapidly these two gestures unfold likely relates to aspiration duration.

In fact, since both supraglottal and glottal gestures shape VOT, Mielke and Nielsen's (2018) finding of shorter aspiration before coda voiceless stops than before coda /ɹ/ should raise the question of why the timing of stop release and fold adduction in the syllable onset is modulated by the coda. For Choi et al. (2016), coda-driven onset VOT variation stems from general coda effects on vowel duration: VOT varies because it embeds the vowel’s “mouth opening” gesture (2016: 4). However, this view is incomplete, since VOT embeds not only the oral opening gesture, but also the vocal-fold adduction gesture. Thus, an account of coda-driven onset VOT variation should explain changes in both the supraglottal and glottal gestures that shape the VOT acoustic interval. Either a consistent laryngeal gesture starts earlier, or the laryngeal gesture itself is stiffer, and therefore more rapid, before voiceless codas. A physiological experimental paradigm would be necessary to distinguish these two possibilities. By our hypothesis, the increase in stiffness reported for voiceless codas (cf., Chen, 1970) may not only affect the vowel’s articulatory and acoustic durations, but also extend to the onset and modulate the timing of both supraglottal and glottal onset gestures, resulting in shorter aspiration. This modulation may be achieved by operating on the stiffness parameters of the laryngeal and supralaryngeal gestures individually, or via a shared stiffness parameter that may govern both onset and vowel gestures in the syllable body (see Discussion).

To further support the view that the coda modulates stiffness early in the syllable, we next review two distinct lines of research that report similar coda effects on the duration of various types of onset consonants and on nuclear vowels, temporal patterns that mirror the coda-to-onset aspiration effect and point to a broader relationship between coda constriction and the dynamics of syllable-body gestures.

1.5. Parallel coda effects on onsets and vowels within the syllable body

The same type of temporal effects that codas exert on onset aspiration have been observed in other onset consonants and in nuclear vowels. In Slater and Coleman (1996), sonorant onsets were shorter before voiceless than voiced codas. Hawkins and Nguyen (2004) reported similar shortening in onset /l/ and voiceless fricatives before voiceless codas. It appears that it is not only onset stops that are sensitive to the coda, but various onsets are, with the acoustic duration of onsets being shorter before voiceless codas and longer before coda sonorants and voiced obstruents.

Like onset consonants, vowels are shorter before voiceless codas than before coda voiced obstruents and sonorants, a well-established pattern studied for more than a century across different language families (Chen, 1970; Heffner, 1937; House & Fairbanks, 1953; Metz, 1914; Meyer, 1904; Navarro Tomás, 1916; Peterson & Lehiste, 1960; Raphael, 2021; Sanker, 2019; Zimmerman & Sapon, 1958). However, the magnitude of coda effects on the vowel depends on the language, being generally larger in English than in the majority of other languages, including French (Chen, 1970).

Importantly, it is not only the duration of the vowel that is modulated by the coda; vowel spectra also vary with the coda. Before voiceless codas, monophthongs and diphthong offglides tend to become more peripheral (Pycha & Dahan, 2016; Summers, 1987; Thomas, 2000; Wolf, 1978). Thus, vowels are not only shorter but also more fully achieve their phonological targets before voiceless codas (Moreton, 2004).

These findings highlight a broader pattern of coda influences on the onset consonant and the vowel, i.e., on the entire syllable body, as summarized in Table 1: Onsets (including onset aspiration, fricatives, and sonorants) and vowels vary systematically depending on coda type: Both are shorter before voiceless codas and longer before coda voiced obstruents, nasals, and liquids. Additionally, vowel articulations more fully reach their targets before voiceless codas. This combination (shorter vowel duration yet more fully reached target) has been described as a major phonetic puzzle, as it contrasts with the well-known pattern of undershoot, wherein vowels become shorter and more centralized under prosodic shortening, such as with increased speech rate or lack of stress (Pycha & Dahan, 2016). The presence of both shorter duration and enhanced target achievement suggests that the shortening of onsets and vowels before voiceless codas may emerge from a mechanism distinct from prosodically induced temporal reduction. One possible explanation is an increase in gestural stiffness in anticipation of a voiceless coda. This would predict the cascade of temporal and spatial adjustments just reviewed: To anticipate the forceful closure movement required by a voiceless coda, gestures in the vowel and onset become stiffer, their articulations becoming more rapid and more fully reaching their targets due to the increased restoring force. Targets may thus be more fully reached despite the shorter time slot due to higher stiffness. From this view, since aspiration is a transition phenomenon, encompassing the laryngeal and supralaryngeal transitions from the onset stop to the vowel, a shorter onset aspiration before a voiceless coda may reflect an increase in the stiffness of the laryngeal and supralaryngeal onset gestures as a strategy to prepare the forceful closure gesture of the voiceless coda. In what follows, we assess evidence that voiceless obstruents are indeed produced with stiff gestures. We also examine the potential aerodynamic basis of this stiffness increase.

1.6. Voiceless coda stiffness inferred from articulatory force, vowel duration and intraoral pressure

That voiceless obstruents involve stiff closure gestures can be traced to earlier debates on articulatory force and muscular effort, where articulatory rapidity, an index of stiffness in later models, was already under discussion (Delattre, 1953; Grammont, 1933; Malécot, 1955; Straka, 1964). Rousselot (1897) first observed that the articulatory trajectories of voiceless obstruents exhibit greater rapidity and more complete target achievement than other consonants (1897: 586–600). Thus, he mapped manner and voicing onto a single continuum of “force of articulation” (1897: 586–597), with voiceless obstruents showing more force than voiced ones. This idea was later related to vowel shortening (Belasco, 1953; Chen, 1970; Delattre, 1941; Malécot, 1977; Moreton, 2004). In one of the earliest reports of coda effects on vowel duration, Locke and Heffner (1940: 75) noted an impression of “added energy expended on the articulation of the final [t],” speculating that the force required for the [t] closure might shorten the preceding vowel, a concept later connected to transition slopes into voiceless codas.

Steep vowel-to-coda transitions have previously been interpreted as evidence of forceful closure gestures in the coda. Chen (1970: 152–157) recorded audio signals of words ending with a voiced or voiceless bilabial coda (/læp/ vs. /læb/) and tracked lip movements with a photoelectric transducer. He observed that shortening in acoustic durations before the voiceless coda was accompanied by steeper lip-closure slopes in the transition from the vowel /æ/ into the coda closure (1970: 155–156; Fig. 6 and Tables XIII-XIV). For the author, the increase in “rate of closure transition” suggests that voiceless codas are produced with greater “muscular effort,” producing an earlier, faster closure movement that shortens the vowel (1970: 152, 157). Similarly, Fujimura and Miller (1979) studied jaw movements during the vowel /ɛ/, reporting that the jaw opens more and “moves more vigorously” before /t/ than before /d/ (1979: 270). Furthermore, Summers (1987) observed shorter vowel duration, larger lip-opening amplitude, and steeper transition slopes before /p/ than /b/ (1987: 850, Fig. 2). A similar scenario is later found in Moreton (2004), who argues that the “forceful” gesture in a voiceless coda closure shortens the preceding vowel through “neuromuscular coupling” (2004: 29). These results point to greater target achievement and faster movements before voiceless codas.

Fig. 6.

Boxplot distributions of word-initial VOT in French word pairs where the onset stop is followed by /ʁ/ (red boxes) or a voiceless stop (blue boxes). Y-axis: VOT of word-initial stops. X-axis: words.

Further support of voiceless obstruents being articulatorily forceful can be found in physiological measures of muscular activity and intraoral pressure. In Chen's (1970) study just reviewed, the author compared his kinematic and acoustic evidence of a forceful /p/ closure to EMG data from (Öhman, 1967: 38, Fig. I-B-16). In that study, EMG peak amplitude during lip closure was 20% greater for /p/ than /b/, paralleling the roughly 20% increase in closure velocity observed for /p/ by Chen (1970: 156, Table XIII). This correspondence suggests that voiceless consonants involve greater articulatory force for their closure than other consonants.

This increased force is, according to Chen (1970), a response of the articulatory system to an intraoral pressure buildup. Specifically, because voiceless obstruents are articulated with an open glottis, intraoral pressure increases due to the air volumes of lungs and mouth both exerting pressure behind the supralaryngeal constriction (1970: 152), the pressure buildup requiring greater force to establish and maintain the closure.

Differences in intraoral pressure have been related to phonological classes, with high pressure characterizing voiceless obstruents (Hayes, 2009; Ladefoged & Johnson, 2011). Moreover, Subtelny et al. (1966) measured substantially lower pressure for nasals and liquids (0.2–1 cm H₂0) than for voiceless obstruents, whose pressure was nearly six times higher (1966: 505). Numerous subsequent studies reported higher pressure in voiceless than voiced obstruents, and higher pressure in voiceless stops compared to fricatives (Cho et al., 2002; Flege, 1983; Fuchs & Koenig, 2009; Koenig et al., 2011; Malécot, 1966b, 1966a; Ünal-Logacev et al., 2018). These findings provide some support to Chen’s (1970) view that the articulatory system responds to the pressure increase in an upcoming voiceless obstruents with an increase in articulatory force.

Perceptual evidence also suggests that voiceless codas involve a forceful closure gesture. Listeners have been shown to utilize vowel-to-coda transition dynamics to perceive coda voicing, with voiceless codas being more reliably perceived after diphthongs with a steeper rising slope (Pycha & Dahan, 2016: 26, Fig. 5; Thomas, 2000). From this perspective, the phenomenon investigated in the present paper, i.e., the coda-to-onset aspiration effect, can be reframed in terms of onset aspiration being longer before coda consonants with low intraoral pressure (liquids, nasals, and voiced obstruents) and shorter before high-pressure codas that involve a forceful gesture to establish and maintain closure during the pressure buildup (voiceless stops and fricatives). The shortening of onset aspiration before voiceless codas thus suggests that the increase in articulatory force involved in voiceless codas is anticipated not only during the vowel, but also in the onset.

Fig. 5.

Boxplot distributions of onset VOT in English CVC words with coda /ɹ/ (red boxes) and with coda voiceless stop (blue boxes). Y-axis: VOT of the onset stop, in ms. X-axis: words.

Indeed, increases in articulatory rapidity before voiceless codas have been observed earlier in the syllable. In Wolf (1978), coda voicing modulated formant structures as early as in the first 50 ms of the vowel in CVC words. Moreover, in Chen's (1970) study we reviewed, /læp/ showed not only a steeper vowel-to-coda transition but also a steeper onset /l/-to-vowel transition than /læb/ (1970: 155; Fig. 6), a pattern also found in Summers (1987: 850, Fig. 2).

These findings indicate a relation between the rapidity of a vowel-to-coda transition and the rapidity of gestures in the preceding onset. Notably, Chen’s approximate 20% increase in the velocity of coda voiceless closures (1970: 155–156) intriguingly relates to the 20% EMG amplitude increase reported by Öhman (1967) for the closure movement of voiceless obstruents. This approximate 20% increase in articulatory rapidity and articulatory force also relates to the 13% shortening of onset aspiration reported by Port and Rotunno (1979: 659, Table IV) before voiceless codas. Although drawn from different domains, these physiological, kinematic and acoustic effects show comparable directional modulation, consistent with a syllable-level increase in articulatory rapidity on the order of 10–20% before a voiceless coda. In a dynamical framework, such steeper and faster movements are compatible with increased gestural stiffness. We therefore predict higher stiffness for voiceless obstruent codas compared to the rhotic /ɹ/, a stiffness increase that may be anticipated in the onset, resulting in shorter onset aspiration before voiceless codas.

The possibility that onset aspiration anticipates properties of the coda is also suggested by more recent models of speech motor planning and control (cf., Floegel et al., 2023; Kasper et al., 2026; Kent, 2024; Parrell et al., 2019; Parrell & Houde, 2019). Building on both experimental data and computer simulations, these models view speech planning and production as a control problem whose solution relies on predicting and compensating for perturbations arising from future articulatory and acoustic events. To this goal, the cognitive system relies on external and internal feedback loops to scan and monitor the unfolding utterance and anticipate perturbations associated with upcoming consonants and vowels, adjusting online production. From this perspective, our proposal that voiceless codas are anticipated with an increase in articulatory force across the entire syllable points to a mechanism that prepares the motor system to manage the perturbation associated with a stiff movement establishing a voiceless coda closure.

One of these models, Tilsen’s Selection and Coordination theory (Tilsen, 2013b, 2013a, 2016, 2018, 2019), has pointed to coda information being present in the articulatory and acoustic signals of the onset (Tilsen, 2020), a result that highlights the need to further investigate what properties of the coda may be actually anticipated in the onset. The Selection and Coordination model makes nuanced predictions about this anticipation process, which we will return to in the Discussion.

1.7. Present study: The Syllable-Body Stiffness Hypothesis

Building on the acoustic, articulatory, and aerodynamic findings just reviewed, we hypothesize that the articulatory properties of the coda modulate the stiffness of gestures composing the syllable body, including the onset and vowel. According to this Syllable-Body-Stiffness Hypothesis, we predict that onset aspiration varies as a function of coda stiffness, expecting shorter onset aspirations before voiceless codas due to stiffer, more rapid onset-to-vowel transitions.

Figure 2 illustrates this hypothesis with schematized representations of the gestures and audio signal for the syllable /pa/. The figure draws on supraglottal kinematic data (e.g., Goldstein et al., 2006). For laryngeal dynamics, the figure is based on evidence that the glottis is open at the release of a voiceless aspirated stop and progressively closes afterward (Sawashima, 1970). The onset /p/ in /pa/ requires an open glottis for voicelessness and bilabial closure and release gestures; the vowel /a/ requires a closed glottis for voicing and oral aperture. The transitions from closed to open lips and from open to closed glottis take time, and together they shape the duration of aspiration measured as VOT in the audio signal. In our model, both glottal and supraglottal gestures are governed by stiffness parameters k that determine their movement durations. Importantly, as long as the vocal folds remain abducted, aspiration persists in the acoustic signal. Thus, the duration of the devoicing gesture is an important contributor to VOT duration.

Based on the reviewed evidence that voiceless codas are associated with increased rapidity of supraglottal movements in vowel-to-coda and onset-to-vowel transitions, we hypothesize that the glottal stiffness in syllable-body gestures is also modulated by the coda. Thus, stiff codas, i.e., codas involving tight closure and high intraoral pressure, should trigger stiffer and faster glottal and supraglottal gestures in the preceding onset. For onset voiceless stops, this predicts faster onset-to-vowel transitions, and thus shorter aspiration, before voiceless codas. Conversely, less constricted codas, involving only a minor or no increase in stiffness, may not increase syllable-body stiffness, thereby predicting no shortening of onset aspiration. In this way, the dynamics of onset stops reflects the stiffness of the upcoming coda: Stiffer syllable onsets, indexed as faster onset-to-vowel transitions, are expected before stiffer codas, such as voiceless obstruent codas.

Coda rhotics allow for testing this hypothesis crosslinguistically. The English rhotic approximant /ɹ/ involves mild constriction and relatively unimpeded airflow, resulting in low intraoral pressure (Boyce et al., 2016; Subtelny et al., 1966; Zhang et al., 2003, 2005; Zhou et al., 2008). Therefore, no substantial stiffness increase is expected in the syllable body before /ɹ/, predicting no onset VOT shortening. In contrast, voiceless obstruents involve substantial intraoral pressure buildup, predicting an increase in onset stiffness and a resulting VOT shortening.

Comparing this effect with different coda rhotics would be of interest to phonetic and phonological theories. To this respect, French rhotics are phonologically voiced (typically /ʁ/ or /ʀ/, see Russell Webb, 2009), but are often realized phonetically as voiceless fricatives (Gendrot et al., 2015; Ladefoged & Maddieson, 1996). Thus, after replicating Mielke and Nielsen's (2018) finding that the English coda /ɹ/ is associated with relatively long onset aspiration, studying the effect in French would allow for an interesting crosslinguistic comparison. The comparison may also allow for investigating the phonetic or phonological origin of the effect. If the phonological identity of the coda rhotic is what influences onset VOT, then French VOT may also be longer before a rhotic than before a voiceless stop. If, however, onset VOT is sensitive to the phonetic realization of the coda, then no significant difference in French VOT may emerge before coda rhotics vs. voiceless stops, both of which are voiceless obstruents requiring a stiff and rapid onset-to-vowel transition. An alternative possibility is that French has simply no coda effects on onset VOT, in line with the VOT of unaspirated stops remaining relatively stable across contextual variations (Kessinger & Blumstein, 1997).

To test the Syllable-Body-Stiffness Hypothesis, we conducted two speech production experiments, comparing word-initial positive VOT before coda rhotics vs. voiceless stops in English and French. We partially replicated Mielke and Nielsen's (2018) design, using a smaller set of word types, a smaller number of participants, and a greater number of repetitions to obtain a more precise estimate of the mean difference in VOT as a function of coda type.

2. Methods

2.1. Participants in the English and French studies

Participants in the English experiment were twenty-five adult native speakers of American English (19 women, 6 men; mean age of 34.9 years, standard deviation = 6.5). They were from different US regions; 8 from the West, 5 from the Midwest, 8 from the Northeast/Mid-Atlantic, and 4 from the South. Since they had all lived in more than one US state, origins were not included in the analyses. Participants in the French experiment were forty-four adult native speakers of French from the Paris area (31 women; mean age of 42.7 years, standard deviation = 11.8).

All participants reported no history of bilingualism that may have caused interferences between aspirated and unaspirated stops (e.g., English-French, English-Spanish, French-German). All participants had normal or corrected to normal vision and reported normal hearing and no history of speech, language, or cognitive disabilities. Participants gave informed consent prior to starting the experiment. This study was reviewed and approved by the Institutional Review Board of The Graduate Center of The City University of New York.

2.2. English stimuli

The stimuli for the English experiment were three pairs of CVC words with /p/, /t/, or /k/ as onset and the rhotic /ɹ/ or a voiceless stop as coda: par /pɑɹ/ and pat /pæt/, tar /tɑɹ/ and tap /tæp/, car /kɑɹ/ and cap /kæp/. These CVC words were selected to replicate the CVC condition from Mielke and Nielsen (2018: 2168, Fig. 1) in a more controlled setting. Specifically, their dataset included words with heterogeneous syllable structures, and several words (e.g., portal, coolant, corny, porcelain) included either both liquids /ɹ/ and /l/, or even liquids and nasals. Thus, several consonants that are assumed to lengthen aspiration could occur in the same word, complicating interpretation. To increase internal validity, we focused on a limited set of CVC words with coda /ɹ/ vs. a voiceless stop to establish whether /ɹ/ relates to longer onset aspiration.

In the three stimulus pairs we selected, the vowels /ɑ/ or /æ/ occupied the syllable nucleus. Only low vowels were selected to control for vowel effects on VOT, as high vowels have been related to longer VOT than low vowels (e.g., Nearey & Rochet, 1994). Both vowels were coded as low tense vowels in Mielke and Nielsen (2018: 2173). Both have shown similar F1 and durations, and have been characterized as phonetically tense vowels in previous research (Bradlow, 1995; Crystal & House, 1988; Feng & Wang, 2024; House & Fairbanks, 1953; Peterson & Lehiste, 1960). These similarities suggest that any VOT pattern may reasonably be attributed to the coda rather than the vowel.

A carrier phrase was used to minimize inter-trial influences on VOT. Specifically, in Mielke and Nielsen (2018), VOT was influenced by the VOTs of up to three preceding trials, likely due to the close sequencing of isolated words in their list-based design. A carrier phrase may help reset motor specifications from previous target words, making interpretations of coda effects on VOT more reliable. We selected the carrier phrase “Do say ‘X’ constantly”, which contains sounds of different phonetic classes, including vowels, obstruents, and sonorants, limiting the saliency of any particular class.

The six stimulus sentences were audio recorded to serve as prompts in a speech repetition task. We used auditory rather than written prompts to minimize orthographic influences on phonological and phonetic processing (e.g., Rastle et al., 2011), particularly with regard to potential orthographic effects on the planning of consonant voicing, central to the present study (cf., Mousikou et al., 2015).

The audio prompts were recorded in a sound booth using a PCB Piezotronics 378B02 condenser microphone connected to a PowerLab 8/35 acquisition device and a computer, with 44.1 kHz sampling rate and 16-bit resolution.

The speaker was an adult male native speaker of American English with training in phonetics. He was instructed to maintain a consistent intonation pattern and avoid the sentence-final rising intonation typical of lists (Ladd, 1980, 2008). The speaker was unaware of the purposes of the experiment. Multiple tokens of each sentence were recorded and the productions with neutral intonations were selected. In the audio prompts, the onset VOTs in the target words were as follows: par (58 ms), pat (55 ms), tar (50 ms), tap (62 ms), car (64 ms), cap (69 ms).

Each audio prompt was intensity scaled (65 dB) and silence intervals were added to the beginning and end to avoid clipping during prompt distribution, using custom scripts in Praat (Boersma & Weenink, 2023). Ten repetitions of each prompt were included in a randomized list and distributed as auditory stimuli in a speech repetition task using the software PsychoPy (Peirce et al., 2019).

2.3. French stimuli

The French stimuli were five pairs of words with initial /p/, /t/, or /k/, differing by the presence or absence of /ʁ/ in the coda or in the following onset (Table 2).

Table 2.

Stimulus word pairs used in the French experiment, grouped by syllable number and place of articulation (POA) of the initial stop. In each pair, a word containing /ʁ/ is matched with a word that does not contain /ʁ/. The table includes orthographic forms, phonemic transcriptions, and English translations.

Syllable number	Word pair	POA	Word with /ʁ/	Word without /ʁ/
Monosyllabic	1	Bilabial	PART /paʁ/ (part)	PATTE /pat/ (paw)
	2	Alveolar	TARD /taʁ/ (late)	TAPE /tap/ (hit)
	3	Velar	CAR /kaʁ/ (because)	CAPE /kap/ (cape)
Disyllabic	4	Bilabial	PATRIE /pa.ˈtʁi/ (homeland)	PÂTÉ /pa.ˈte/ (pâté)
	5	Bilabial	PATRAQUE /pa.ˈtʁak/ (unwell)	PATATE /pa.ˈtat/ (potato)

Specifically, the French CVC words (ex. /paʁ/ vs. /pat/), were selected to test whether coda type influences onset VOT within the same syllable, replicating the CVC condition from our English experiment and from Mielke and Nielsen (2018). Additionally, since this was the first study of this effect in French, we took the opportunity to explore an additional disyllabic condition. The disyllabic word pairs began with /p/ and differed in the presence or absence of /ʁ/ in the subsequent onset cluster. A key feature of this condition was the voiceless stop /t/ intervening between the onset /p/ and /ʁ/, as in patrie /pa.ˈtʁi/. This structure parallels Mielke and Nielsen's (2018) English items such as paper, where a stop intervenes between the rhotic and the onset stop. As in that study, we were interested in whether the intervening stop would neutralize any effect of /ʁ/ on onset aspiration.

The ten French words were embedded in the carrier phrase “Dis le mot X comme ça” /di.lə.mo.X.kɔm.sa/ (“Say the word X like this”) to serve as auditory prompts in a speech repetition task. A carrier phrase with different sounds was selected, as in our English study, to limit the saliency of any particular phonetic class. Audio recording of the French prompts was made by an adult male native speaker of French with training in phonetics. The speaker was unaware of the purposes of the experiment. He was instructed to produce each sentence with neutral intonation and relatively constant rhythm, ensuring that the utterances would not display the rising intonation typical of continuation in French (Delattre, 1966). The recording took place in a sound booth, using an H2 Zoom Handy Recorder with a 44.1 kHz sampling rate and 16-bit resolution. Each sentence was produced three times, and the production with the most neutral intonation was selected as auditory prompt. The word-initial VOTs in the audio prompts were as follows: part (10 ms), patte (7 ms), tard (32 ms), tape (26 ms), car (34 ms), cape (33 ms), patrie (6 ms), pâté (7 ms), patraque (7 ms), patate (7 ms). Intensity scaling (65 dB) and silence intervals (250 ms) were applied. Experiment procedure and audio processing were identical in both the English and French experiments.

2.4. Procedure

In the testing session, the participant sat in front of a laptop computer, listened to the sentences played through headphones and repeated them aloud. Trials progressed automatically: the software PsychoPy presented the acoustic prompt and recorded the speech response, with no action required from the participant. Speech was recorded at a 44.1 kHz sampling rate and 16-bit resolution using a Shure SM-137 condenser microphone connected to a Motu M2 soundcard and a computer. The microphone was positioned approximately 10 cm (3.9 in) from the participant’s mouth, 30° off axis, with a pop filter. There were three initial practice trials (excluded from analyses). Each participant produced ten repetitions per word, in a randomized list. One full experiment lasted approximately 20 minutes, with breaks at fixed intervals.

In the English experiment, a total of 1,500 utterances were recorded. Of these, 2.3% (35 tokens) were discarded due to speech production errors. Analyses were conducted on 1,465 English utterances. In the French experiment, a total of 4,350 utterances were recorded. One participant produced five instead of ten repetitions per word due to a software error. Less than 1% of the data (35 tokens) was discarded due to speech production errors. Analyses were conducted on 4,315 French utterances.

2.5. Acoustic measures

Acoustic analyses entailed segmentation, annotation, and extraction of segmental durations. The audio recordings were automatically annotated with the Montreal Forced Aligner (MFA; McAuliffe et al., 2017), followed by manual adjustment of boundaries. Each audio file was segmented by MFA into a two-tier TextGrid containing word and phone boundaries. Each boundary was then visually inspected in Praat, adjusting duration outliers. Adjustments were guided by visual inspection of the oscillogram and broadband spectrogram (5-ms window length; Figs. 3–4).

Fig. 3.

English words tap and tar, example segmentations of (a) vowel-stop sequences into vowel and closure intervals, and (b) vowel-/ɹ/ sequences into vowel and /ɹ/ intervals.

Fig. 4.

French words tape /tap/ and tard /taʁ/, example segmentations of (a) vowel-stop sequences into vowel and closure intervals, and (b) vowel-/ʁ/ sequences into vowel and /ʁ/ intervals.

Accurate segmentation is critical but involves the experimenter’s decisions, especially for transitions from the vowel into a coda rhotic in English (Peterson, 1955; Peterson & Lehiste, 1960). To address this, the segmentation criteria used with the English and French data are defined in the following section.

2.6. Segmentation criteria: VOT, vowel, coda closure, and rhotics

The VOT interval for the onset voiceless stop began at the stop release, signaled by an energy burst in the oscillogram and corresponding spike in the spectrogram (Figs. 3–4). VOT ended at the onset of laryngeal vibration for the vowel, indicated by the appearance of periodicity in the waveform and low-frequency energy (<200 Hz) and F1 energy (~700 Hz for /ɑ/, /æ/, and /a/) in the spectrogram. These visual criteria were supplemented by auditory inspection of each token and by the software’s f0 detector.

Vowels in vowel-stop sequences (e.g., English pat /pæt/ and French patte /pat/) were segmented into a vocalic interval (henceforth, “vowel”) and a coda closure interval. The vowel began at the end of VOT and was defined by glottal pulses in the oscillogram and formants in the spectrogram. The coda closure interval began at the vowel’s amplitude drop and formant offset (signaling vocal tract closure for the upcoming coda stop) and included the silent portion visible in the oscillogram up to any final release (Figs. 3a, 4a).

Segmenting the English coda /ɹ/ is challenging due to its smooth formant transitions. We located its onset where F3 and F2 converge (Fig. 3b), following convention (Peterson & Lehiste, 1960; Zhou et al., 2008). The MFA output generally aligned with this scenario. When it did not, we adjusted the /ɹ/ interval to match this criterion.

Segmentation criteria for the French /ʁ/ were defined as follows. Unlike the segmentation of English coda /ɹ/, which was challenging due to smooth formant transitions into the approximant, the segmentation of French coda /ʁ/ was relatively straightforward. In our data set, /ʁ/ tokens were predominantly voiceless fricatives, characterized by discernible acoustic markers in the waveform, such as turbulence and reduced amplitude (Fig. 4b), typical of the friction noise caused by narrow constrictions (Gendrot et al., 2015; Ladefoged, 2005). The /ʁ/ interval began at the onset of friction and included the entire fricative portion up to the word-final boundary.

2.7. Speech rate variables

Speech rate was not varied deliberately, but overall duration was taken as a proxy for rate. The variables chosen replicated Mielke and Nielsen’s word- and vowel-duration variables (2018: 2167–2168), ensuring consistency with that study. Word duration averaged by participant and vowel duration averaged by word captured VOT variation linked to speech rate differences at the individual and word level. Residual word duration measured speech rate variability unexplained by participant or word characteristics¹.

2.8. Statistical analyses

All descriptive and inferential statistics were computed with the R software (R Core Team, 2021). LME models were built using the lme4 package (Bates et al., 2015). The lmerTest package (Kuznetsova et al., 2017) was used to display p values in the LME summary tables. Power analyses were conducted with the simr package (Green & MacLeod, 2016).

3. RESULTS

3.1. English results

3.1.1. English VOT distributions

The mean VOT and standard deviation (SD) of bilabial, alveolar, and velar stops measured in our English experiment are displayed in Table 3. The table highlights longer VOT for lingual stops (alveolar and velar) compared to bilabials, as reported in many studies (Byrd, 1993; Klatt, 1975; Lisker & Abramson, 1964). Note that the lingual stops display similar VOTs, consistent with Chodroff and Wilson (2017: 33, Table 1). This finding also aligns with the pattern observed by Chodroff et al. (2019) across multiple languages, wherein “dorsal stops have the highest values while the relative order of coronals and labials is more variable” (2019: 110). Next, we address the research question of whether coda type (/ɹ/ vs. /T/) influences VOT distributions.

Table 3.

Average VOT and standard deviation (SD) of word-initial bilabial, alveolar, and velar voiceless stops in American English CVC words measured in the present study.

Stop consonant	Mean VOT (ms)	SD VOT (ms)
/p/	66.5	20.8
/t/	78.6	22.5
/k/	80.7	19.9

Boxplot distributions of word-initial VOT in words with coda /ɹ/ vs. coda voiceless stops are displayed in Fig. 5. For each pair of words sharing the same onset stop, the red box (word with coda /ɹ/) consistently appears above the blue box (word with coda /T/). This pattern indicates that onset aspiration was longer in words with coda /ɹ/ compared to words with a coda voiceless stop, a finding that mirrors the effect reported by Mielke and Nielsen (2018).

Importantly, all outliers in Fig. 5 were inspected and confirmed to be accurate productions of individuals with inherently long VOTs. This aligns with previous research reporting considerable individual VOT variability (Allen et al., 2003; Chodroff & Wilson, 2017; Theodore et al., 2009). Individual variation in baseline VOT was addressed with random intercepts and slopes in the LME models (Barr et al., 2013).

Table 4 further quantifies onset VOT differences by coda type, showing the mean onset VOTs and SDs for each word pair, alongside the mean VOT shortening and percentage difference in the voiceless coda condition. The table shows that onset VOT consistently shortened before coda voiceless stops compared to coda /ɹ/, for all Places of Articulation (POA). The magnitude of shortening ranged from 9 to 15%, depending on the onset stop’s POA. In the next section, VOT variation is further analyzed with LME models.

Table 4.

Average VOT and standard deviation (SD) of word-initial /p, t, k/ in American English, by word type and coda condition (coda /ɹ/ vs. voiceless stop). The last two columns show the difference in onset VOT between words with a voiceless coda and words with a coda /ɹ/, calculated as [Onset VOT with voiceless coda - Onset VOT with coda /ɹ/], and the corresponding percentage change. Negative values indicate shortening of onset VOT in the voiceless coda condition.

Word	Onset	Coda	M VOT (ms)	SD	VOT diff. (ms)	% diff.
par	/p/	rhotic	69.6	20.7	—	—
pat	/p/	v.less stop	63.3	20.5	−6.3	−9
tar	/t/	rhotic	83.9	22.1	—	—
tap	/t/	v.less stop	73.6	21.8	−10.3	−12
car	/k/	rhotic	87.0	18.8	—	—
cap	/k/	v.less stop	74.4	19.0	−12.7	−15

3.1.2. Model 1 – English VOT – Structure

An LME model (Model 1) was built to examine the relationship between onset VOT and coda type, alongside well-known linguistic (POA, speech rate) and individual factors (age, sex). The dependent variable was the VOT of onset /p, t, k/, measured in milliseconds (ms). To facilitate model convergence, in line with previous studies, VOT was centered at zero by subtracting the grand mean from each VOT token; all continuous factors were scaled by subtracting the grand mean from each token and dividing by the standard deviation computed across all productions (Chodroff & Wilson, 2017; Mielke and Nielsen 2018).

Model 1 was based on the best-fitting model in Mielke and Nielsen (2018: 2171, “Model 5”), with adjustments to accommodate the present dataset. Fixed effects modeled phonological and phonetic influences on VOT (POA, coda type, and speech rate) and experimental context (list position and previous VOT residuals; cf., Mielke and Nielsen, 2018). Participant-level factors were age, sex, and the interaction between sex and word duration. Random effects for speaker included an intercept, modeling baseline VOT differences (Chodroff & Wilson, 2017; Theodore et al., 2009), and slopes for word duration residuals and coda type, modeling possible individual differences in gestural timing (Byrd, 2000) and in how VOT responds to speech rate (Allen et al., 2003). Random intercepts for lexical item (i.e., the six stimulus words) and word pair (item with and without the rhotic) were tested but excluded, as they led to non-convergence, likely because our small and controlled stimulus set left minimal residual variability for word type after accounting for POA and coda type.

To summarize, the fixed effects included the following predictors: POA; Coda type; Word duration (by-subject); Word duration residuals; Vowel duration (by-word); Sex; Age; List position; Lagged VOT Residuals n-1; Coda type x Word duration (by-subject); Word duration residuals x Sex. Random effects for speaker included a random intercept and random slopes for Word duration residuals and Coda type².

3.1.3. Model 1 – English VOT – Output

The output of Model 1 is summarized in Table 5. The model revealed significant effects on English VOT from known phonological and prosodic factors. POA was the primary source of VOT variation: VOT was longer for alveolar /t/ and velar /k/ compared to bilabial /p/. Crucially for our study, coda type had a significant effect on onset VOT: The VOT of onset stops before coda /ɹ/ was, on average, 9.32 ms longer than before voiceless stop codas, after controlling for POA and other factors. This finding replicates Mielke and Nielsen's (2018) result of longer VOT before coda liquids, with a larger estimated effect (5.07 ms in their study).

Table 5.

Model 1. Linear mixed-effects (LME) model output showing the relationship between word-initial VOT (dependent variable) and phonological, phonetic, and demographic factors in American English CVC words.

Predictor	Estimate	SE	t-value	p-value	sign.
(Intercept)	−12.71	5.21	−2.44	0.022	*
POA Alveolar	13.62	0.88	15.50	< 0.001	***
POA Velar	17.89	1.17	15.25	< 0.001	***
Coda type – Coda /ɹ/	9.32	1.41	6.63	< 0.001	***
Vowel duration (By-word)	2.23	0.49	4.52	< 0.001	***
Word Duration Residuals	6.31	1.71	3.69	0.001	**
Word Duration (By-subject)	9.62	2.71	3.56	0.002	**
List Position	0.80	0.34	2.37	0.018	*
Lagged VOT Residuals n-1	0.47	0.34	1.40	0.161
Age	3.07	2.29	1.34	0.194
Coda type – Coda /ɹ/ x Word Duration (By-Subj)	1.73	1.40	1.24	0.230
Sex – Woman	−3.11	5.59	−0.56	0.583
Word Duration Residuals x Sex – Woman	0.37	1.91	0.19	0.850

Baseline levels for categorical variables are: Place of Articulation (POA; bilabial); Coda type (voiceless stop); and Sex (man). Significance codes (sign.): p < .001, ***; p < .01, **; p < .05, *.

Speech rate, indexed by word and vowel durations, had a significant effect on VOT, suggesting that onset VOT was longer in longer words and before longer vowels. This effect was smaller than those of POA and coda type. The positive estimate for vowel duration suggests that vowel and VOT durations covaried, consistent with previous studies showing that onset VOT and vowel duration in English vary in the same direction, although by different proportions (Port & Rotunno, 1979). The interaction between coda type and by-subject word duration was not significant, indicating that the coda effect was not moderated by speakers’ overall speaking rate.

We also replicated Mielke and Nielsen’s exploration of several ways to encode the presence of a coda liquid vs. voiceless obstruent, either as a categorical factor or as duration change (2018: 2170; Model 3f). In their study, the binary liquid factor explained VOT variation better than durations. To replicate that design choice, we tested other segmental duration variables: one incorporating vowel duration plus the duration of coda /ɹ/ in the rhotic condition, and another variable encompassing the entire rhyme duration in both coda conditions. Results with these vowel and rhyme duration variables were inconclusive; they either worsened the model (BIC increase) or led to non-convergence.

The experimental context had only a minor or null effect on VOT. Specifically, list position had a small but significant positive estimate, suggesting that VOT slightly increased as the experiment unfolded. However, this effect is likely too small to be meaningful. VOT residuals from the preceding trial had no significant effect on VOT. We also tested the residuals and the z-scored VOTs from up to three preceding trials, which yielded nonsignificant results, in contrast to Mielke and Nielsen's (2018) report of effects from up to three preceding trials. The use of a carrier phrase in the present study likely countered the influence of previous VOTs on online VOT production. Finally, the effects of speaker factors such as age, sex, and the interaction between sex and speech rate were all nonsignificant.

In sum, Experiment 1 showed that the VOT of the voiceless stops /p, t, k/ in the onset of American English CVC words is longer when the coda is the rhotic approximant /ɹ/ than when it is a voiceless stop (e.g., par /pɑɹ/ vs. pat /pæt/), even after controlling for speech rate, phonological, and individual factors. This result replicates Mielke and Nielsen (2018), who first reported effects of coda liquids (/ɹ, l/) on English onset VOT.

3.2. French results

3.2.1. French VOT distributions

The mean VOT and standard deviations (SD) of French onset bilabial, alveolar, and velar stops are displayed in Table 6. French VOT was longer in alveolar and velar than bilabial stops, in line with the well-documented /k>t>p/ VOT pattern reported for many languages, where VOT increases as the place of stop occlusion moves further back in the oral cavity (Cho et al., 2019). Next, we addressed the research question of whether French onset VOT differs before /ʁ/ compared to a voiceless stop.

Table 6.

Average VOT and standard deviations (SD) for word-initial /p, t, k/ in the French words produced in the present study.

Stop consonant	Mean VOT (ms)	SD VOT
/p/	13.5	6.6
/t/	25.4	9.6
/k/	40.4	12.3

Fig. 6 shows the boxplot distributions of word-initial French VOTs in words with /ʁ/ vs. voiceless stops. A vertical gray line separates the three monosyllabic pairs (on the left) from the two disyllabic pairs (on the right). Importantly, the figure shows no consistent difference in VOT between the /ʁ/ and voiceless stop conditions, in either the monosyllabic or disyllabic words. The outliers in Fig. 6 were inspected and confirmed to be accurate productions of individuals with longer VOTs than their companions, consistent with reports of individual differences in baseline VOT (Allen et al., 2003; Chodroff & Wilson, 2017; Theodore et al., 2009).

To further quantify the differences in word-initial VOT in words with and without /ʁ/, Table 7 displays mean VOTs by word, alongside VOT differences between the rhotic and no-rhotic conditions. The scenario emerging from this table is drastically different from the English data in Table 4. Word-initial French aspiration showed little variation between the rhotic and the stop conditions: differences were all under 2 ms, with no consistent direction of change. Such small differences are difficult to interpret in phonetic terms, suggesting that French word-initial VOT remained relatively stable across conditions. Next, we analyzed French VOT variation with LME models.

Table 7.

Average VOT and standard deviation (SD) of word-initial /p, t, k/ in our French dataset, organized by word, onset stop, and rhotic condition (with /ʁ/ vs. without /ʁ/). The last two columns show onset VOT differences between words with and without /ʁ/, calculated as [VOT before voiceless stop - VOT before /ʁ/], and the corresponding percentage change. Negative values indicating VOT shortening in the no-/ʁ/ condition.

Word	Onset	Condition	M VOT (ms)	SD	VOT diff. (ms)	% diff.
Monosyllabic
part /paʁ/	/p/	rhotic	14.6	6.9	—	—
patte /pat/	/p/	v.less stop	14.5	7.3	−0.1	−0.7
tard /taʁ/	/t/	rhotic	26.2	10.3	—	—
tape /tap/	/t/	v.less stop	24.7	8.7	−1.5	−5.7
car /kaʁ/	/k/	rhotic	40.3	12.5	—	—
cape /kap/	/k/	v.less stop	40.5	12.0	0.2	0.5
Disyllabic
patrie /pa.ˈtʁi/	/p/	rhotic	12.7	5.8	—	—
pâté /pa.ˈte/	/p/	v.less stop	13.8	7.0	1.1	8.7
patraque /pa.ˈtʁak/	/p/	rhotic	12.2	5.5	—	—
patate /pa.ˈtat/	/p/	v.less stop	13.5	6.3	1.3	10.7

3.2.2. Model 2 – French VOT – Structure

An LME model (Model 2) was built to examine the relationship between onset VOT, presence of /ʁ/ and the linguistic and experimental factors already described in Model 1. A syllable factor was included here because the French words contained mono- and disyllables. Centering and scaling of predictors were applied as in Experiment 1.

The fixed effects of Model 2 thus included the following predictors: POA, /ʁ/ Presence, Syllable type, Word duration (by-subject), Word duration residuals, Vowel duration (by-word), Sex, Age, List position, Lagged VOT Residuals n-1, /ʁ/ Presence x Word duration (by-subject), /ʁ/ Presence × Syllable type, Word duration residuals x Sex.

Random effects for speaker included a random intercept and random slopes for Word duration residuals, /ʁ/ Presence, and Syllable type³.

3.2.3. Model 2 – French VOT – Output

The output of Model 2 is summarized in Table 8. The model reveals a scenario markedly different from that observed in Experiment 1 with American English stops. French VOT was overall less sensitive to phonetic and prosodic factors than English VOT. Apart from POA and residual word duration, no other factor significantly influenced the VOT of French stops in our dataset.

Table 8.

Model 2: Linear mixed-effects (LME) model output showing the relationship between word-initial VOT (dependent variable) and phonological, phonetic, and demographic factors in French. Predictors are ordered by t-value magnitude, from largest to smallest.

Predictor	Estimate	SE	t-value	p-value	sign.
(Intercept)	−9.49	1.61	−5.90	< 0.001	***
POA – Velar	25.95	0.29	88.65	< 0.001	***
POA – Alveolar	11.02	0.29	38.66	< 0.001	***
Word Duration Residuals	0.70	0.22	3.12	0.004	**
Sex – Woman	−2.97	1.67	−1.78	0.083
Word Duration (By-subject)	1.28	0.90	1.43	0.161
Vowel Duration (By-word)	1.02	0.77	1.32	0.186
List Position	−0.09	0.09	−1.02	0.308
/ʁ/ Presence– With /ʁ/	−1.25	1.31	−0.95	0.341
Lagged VOT Residuals n-1	0.07	0.09	0.81	0.417
Age	−0.59	0.89	−0.66	0.516
/ʁ/ Presence – With /ʁ/ x Word Duration (By-Subj)	0.07	0.21	0.33	0.742
/ʁ/ Presence – With /ʁ/ x Syllable Type – Disyl.	0.20	1.46	0.14	0.890
Word Duration Residuals x Sex – Woman	−0.01	0.27	−0.03	0.975
Syllable Type – Disyl.	−0.01	0.99	−0.01	0.990

Baseline levels for categorical variables are: Place of Articulation (POA; bilabial), /ʁ/ Presence (without /ʁ/), Syllable type (monosyllable), and Sex (man). Significance codes (sign.): p < .001, ***; p < .01, **; p < .05, *.

The primary source of French VOT variation was POA, with velar stops having longer VOTs than alveolar and bilabial stops. Syllable type had no effect on VOT, suggesting that /p/ did not vary between the mono- and disyllable conditions. French VOT showed overall low sensitivity to speech rate variation. Among the three speech rate variables, only residual word duration showed a significant but small effect, indicating that VOT was slightly longer in words that were longer than expected. This effect was much weaker than in English, which aligns with previous reports of VOT being more sensitive to speech rate variations in aspirated than unaspirated stops (Kessinger & Blumstein, 1997). The effects of vowel duration and by-subject word duration on French VOT were nonsignificant.

Crucially for our research question, /ʁ/ presence had no significant effect on onset VOT in our French dataset, suggesting that VOT was similar before the rhotic /ʁ/ compared to a voiceless stop. Interactions between /ʁ/ Presence and by-subject word duration, and between /ʁ/ Presence and syllable type, were also nonsignificant. Thus, unlike in English, the VOT of French onset stops in the present dataset was unaffected by the presence of a rhotic in the coda or in the following onset.

French VOT remained relatively stable across the course of the experiment. The effect of VOT residuals from the previous trial was nonsignificant. To replicate Mielke and Nielsen (2018), we also tested the residual and the z-scored VOTs from up to three preceding trials; all were nonsignificant. List position also had no significant effect. Age, sex, and the sex by residual word duration interaction also had no significant effect on French VOT. Importantly, we also ran the analysis on the mono- and disyllables separately. In both cases, VOT did not vary significantly before /ʁ/ vs. a voiceless stop.

Given the absence of a significant /ʁ/ effect, we conducted a simulation-based power analysis using the powerSim() function of the simr package (Green & MacLeod, 2016), which allows for running Monte Carlo simulations based on the output of an LME model. Following previous phonetic applications of this technique (Kirby & Sonderegger, 2018; Proctor et al., 2019), we ran two tests assessing our design’s power to detect a significant 5 ms or 9.3 ms effect of French /ʁ/ on VOT based on the LME model reported in Table 8 (1,000 simulations; alpha = 0.05; default Wald z-test; 4,315 observations from 44 speakers). When the /ʁ/ coefficient was set to 5 ms, corresponding to the effect of English rhotics on VOT reported by Mielke and Nielsen (2018), statistical power was 98.00% (95% CI: 96.93, 98.77%). When the coefficient was set to 9.3 ms, corresponding to the effect observed in our English experiment, power was 100% (95% CI: 99.63, 100.0). These simulations suggest that, given our dataset and our model’s residual variability (SD = 5.74 ms) and between-subject random intercept variability (SD = 7.06 ms), our design would have been likely to detect an effect of the magnitude reported previously, had such an effect been present in our French dataset. However, as with all null effects, caution is warranted in interpreting the absence of a significant effect as evidence that no effect exists in the population (See Limitations in the Discussion). To summarize, we found no significant effect in word-initial French VOT in words with vs. without /ʁ/ (e.g., /paʁ/-/pat/; /pa.ˈtʁi/-/pa.ˈte/).

4. GENERAL DISCUSSION

4.1. Summary of findings: Links between syllable coda and onset aspiration

Coda effects on the duration of onset consonants, including effects on the aspiration of onset voiceless stops, have been observed in several studies, but the phenomenon remains largely understudied and unexplained. In contrast, coda effects on vowel duration have been extensively studied for over a century. Notably, onsets and vowels vary in similar ways before the same coda types: shorter durations are observed before voiceless compared to voiced codas (Table 1).

The present speech production study investigated coda-to-onset effects in English and French, focusing on the influence of coda rhotics vs. coda voiceless stops on the aspiration of onset stops, indexed as positive VOT. In Experiment 1, word-initial VOT in English CVC words was significantly longer before coda /ɹ/ than before coda voiceless stops (e.g., par /pɑɹ/ vs. pat /pæt/). VOT and vowel duration covaried, but coda effects on VOT remained significant after controlling for vowel, rhyme, and word durations, ruling out vowel duration and speech rate as explanations for the coda effect on onset VOT. In contrast, Experiment 2 found no significant difference in French VOT before the uvular rhotic /ʁ/ compared to a voiceless stop, in mono- and disyllabic words (e.g., part /paʁ/ vs. patte /pat/, and patrie /pa.ˈtʁi/ vs. pâté /pa.ˈte/).

The fact that onset VOT in English is shorter before a voiceless coda than before coda /ɹ/ supports our Syllable-Body Stiffness Hypothesis, which posits that the syllable body (onset and vowel) varies in stiffness to anticipate the constriction degree and stiffness of the coda. Compared to the longer VOT observed in onsets before a mildly constricted coda (e.g., the approximant /ɹ/), the short onset VOT observed before the stiff and highly constricted voiceless obstruents may reflect a general increase in stiffness in the syllable body. This adjustment may be necessary to anticipate a tight voiceless coda closure gesture, thus compensating for the intraoral pressure increase associated with voiceless obstruents (cf., Chen, 1970). Increased syllable-body stiffness would result in more rapid articulatory gestures and shorter acoustic durations, including a shorter onset aspiration. Syllable-wide planning of stiffness may thus play a critical role in the emergence of the syllable as a unit of production, perception, and literacy. These interpretations are further discussed in Sections 4.2-4.4.

The present study also provides new insights into which coda classes are associated with a long vs. short onset positive VOT. We investigated the influence of coda rhotics on English VOT by partially replicating Mielke and Nielsen (2018)—the only prior report of English aspiration being longer before coda /ɹ/ compared to coda voiceless obstruents. In that study, the effect of liquids (rhotics and laterals) on VOT was studied by grouping both liquids in a single factor. Also, the liquids appeared in different syllable structures, and some words contained multiple consonants of interest (liquids and nasals), complicating interpretation. By replicating the effect in a study focused on coda /ɹ/, we established that the rhotic approximant /ɹ/ does indeed participate in the coda-to-onset VOT effect: Coda /ɹ/ is associated with long onset positive VOT, patterning with coda nasals, liquids, and voiced obstruents.

Our results indicate that the coda-driven variation in onset VOT is not a byproduct of changes in vowel duration. The effect remained robust even after controlling for vowel, word, or rhyme durations. This result relates to the direct coda-to-onset control proposed by Port and Rotunno (1979) and Mielke and Nielsen (2018), who interpreted the independence of onset VOT from vowel duration as evidence that the coda directly influences onset duration. Thus, while the durations of the onset consonant and vowel nucleus do vary in the same direction depending on the coda (Table 1), this pattern likely reflects both segments responding to some articulatory property of the coda, rather than the vowel influencing the onset. The onset VOT in English par vs. pat likely varies to anticipate the different intraoral pressure and stiffness demands of the /ɹ/ and the voiceless stop (cf., Chen, 1970).

4.2. Implications for the Syllable-Body-Stiffness Hypothesis

The pattern of onset VOT being longer before coda /ɹ/ and shorter before coda voiceless stops is compatible with stiffness increasing across the syllable body to anticipate the coda’s articulatory states, the latter being modulated by aerodynamic factors.

Voiceless obstruents require an intraoral pressure buildup that demands increased stiffness, and consequently greater articulatory velocity, in the consonant closure and the preceding vowel (Chen, 1970; Kozhevnikov & Chistovich, 1965; Öhman, 1967; Moreton, 2004; Summers, 1987). If the onset also participates in the anticipation of coda stiffness, we would expect to see increases in rapidity of both stop release and vocal fold adduction in the onset, generating shorter onset aspiration before voiceless coda stops, as found for English in our Experiment 1. Even if the adduction is a gesture separate from the supralaryngeal movement (Browman & Goldstein, 1992), it is still part of the syllable body. On our account, it should also increase in stiffness, resulting in faster adduction and shorter aspiration.

Voiceless codas appear to be the critical condition triggering onset aspiration shortening. We showed that onset positive VOT is shorter before coda voiceless obstruents and longer before several different coda types—voiced obstruents, nasals, and liquids (Table 1). The magnitude of VOT shortening before voiceless codas relative to coda /ɹ/ in our English data (Table 4) was comparable to previously reported effects across different coda types. Specifically, /t/ VOT shortened by 12% in our data (before coda /ɹ/ vs. /p/) and by 13% in Port and Rotunno (1979; before coda /n/ vs. /pt/). Moreover, /p/ VOT shortened by 9% in our data (coda /ɹ/ vs. /t/) and by approximately 10% in Choi et al. (2016; coda /d/ vs. /t/). Thus, onset VOT appears to be similar before coda nasals, liquids, and voiced stops. In contrast, onset VOT appears to shorten before coda voiceless obstruents.

Given that a wide range of codas (voiced obstruents, nasals, liquids) is associated with longer onset aspiration, while shortening is observed consistently before voiceless obstruents, a more parsimonious account is that onset aspiration shortens specifically before voiceless codas and otherwise remains near a default level. Under this view, voiced codas may not induce sufficient stiffness increase to trigger syllable-wide modulation. This interpretation differs from previous claims that aspiration lengthens before coda liquids (Mielke & Nielsen, 2018). Rather than treating liquids as actively lengthening onset aspiration, we propose that onset aspiration remains relatively stable before coda liquids, nasals, and voiced obstruents, i.e., when intraoral pressure does not rise critically and thus no stiffness increase is necessary to realize the closure gesture. The apparent lengthening reported in previous studies likely reflects comparison with the shortening effect observed in the voiceless coda condition. In our study, this pattern held for onset positive VOT before the English rhotic /ɹ/, while the results for the French rhotic require further analysis.

In French, we found no significant difference in onset VOT before /ʁ/ vs. a voiceless stop. A power analysis confirmed that our design could have detected an effect of 5 ms in our dataset. This suggests several possible explanations. The French rhotic in our dataset may have patterned as a phonetically voiceless obstruent (cf., Gendrot et al., 2015). As such, /ʁ/ would significantly raise intraoral pressure, thereby shortening onset VOT similarly to voiceless stops.

Nonetheless, it remains possible that differences in stiffness, and consequently in onset VOT, exist between the /ʁ/ and voiceless stops in the general population, although they were not detected in the present study. For example, Koenig et al. (2011) showed that voiceless stops increase intraoral pressure more than voiceless fricatives, which opens to the possibility of stiffness differences between coda voiceless stops and uvular fricatives in French being reflected in the onset. Moreover, earlier studies reported shorter French vowels before coda voiceless stops than coda /ʁ/ (Chen, 1970; Delattre, 1941, 1962, 1969), suggesting stiffness differences between the two codas. Thus, the French rhotic uvular fricative could, in principle, impose less of a stiffness increase to the syllable body compared to a voiceless stop, in which case a difference in onset VOT could appear, possibly under different prosodic conditions.

An alternative explanation for the absence of /ʁ/ effects on VOT is that French simply has no significant coda effect on onset aspiration. In such a case, we may envision two possibilities. First, unaspirates are much less sensitive than aspirates to phonetic manipulations, such as speaking rate variation (Kessinger & Blumstein, 1997). From this view, the French unaspirated stops in the present study may not have varied enough for a coda effect to be observed. That is, borrowing a concept from Klatt (1976), the short aspiration of French voiceless stops may be relatively “incompressible” (Klatt, 1976: 1215). Second, the absence of an effect in French may reflect crosslinguistic differences in syllable organization. It is conceivable that in languages displaying a coda-to-onset aspiration effect, such as English, stiffness planning operates over a domain corresponding to the syllable-body, whereas in languages without such an effect, such as French, stiffness planning aligns more closely with a syllable-rhyme stiffness constituent. Further experimental and theoretical work on the interaction between syllable structure and gestural stiffness is needed to evaluate these possibilities.

4.3. Implications for syllable theory: the syllable as a speech unit

The syllable is known to determine the duration and relative timing of consonant and vowel gestures (Goldstein et al., 2006; Saltzman & Byrd, 2000; Sproat & Fujimura, 1993). Our model with syllable-level stiffness triggering coda-to-onset spatio-temporal effects relates to the long-standing issue of the syllable as a speech production unit (Kozhevnikov & Chistovich, 1965; Rousselot, 1897; Stetson, 1928).

Our finding that English onset aspiration shortens before voiceless codas supports a model in which syllable-wide stiffness is planned in response to the biomechanical demands of the coda. In our dataset, the VOT of English onset /p/, /t/, and /k/ shortened before coda voiceless stops by 9%, 12%, and 15%, respectively. Comparable VOT reductions were reported in previous studies with different coda voicing contrasts (Choi et al., 2016; Port & Rotunno, 1979). The magnitudes of these coda-driven VOT differences align with the magnitude of increases in articulatory force and rapidity before voiceless codas observed in previous studies. Chen (1970) observed an approximate 20% increase in lip closure velocity in the vowel-to-coda transition before voiceless codas (1970: 155, Fig. 6). Similarly, Öhman found a 20% increase in EMG peak amplitude during voiceless lip closure (1967: 38, Fig. I-B-16). Taken together, these findings suggest that voiceless codas may trigger an increase in articulatory stiffness across the entire syllable, with increases falling in the range of approximately 10–20% across conditions. Indeed, Summers (1987) showed jaw movement trajectories for CVC syllables differing by coda voicing (1987: 850, Fig. 2): In the voiceless coda condition, the trajectory slopes were steeper from onset to coda, indicating that stiffness at the syllable level is reshaped depending on the coda.

Such a model of syllable-wide planning of stiffness may offer a unified articulatory explanation for previously reported coda effects on onset and vowel durations. A wide range of explanations has been proposed for voicing-related temporal asymmetries in vowels and onsets, including aerodynamic constraints, gestural timing and articulatory force differences, perceptual cue reweighting, and listener-oriented enhancement (Chen, 1970; Choi et al., 2016; Delattre, 1962; Hawkins & Nguyen, 2004; Lisker, 1957, 1977; Port & Rotunno, 1979; Raphael, 1972; Sanker, 2019; Westbury & Keating, 1986; Wright, 2004). However, considering our results alongside earlier findings (Table 1), it appears that a wide range of coda types (voiced obstruents, liquids, nasals) is associated with longer onset VOT and vowel. Given the number and variety of potential coda candidates, listeners may not reliably predict the coda based on the temporal change.

In fact, these earlier findings can be reinterpreted through the lens of stiffness, shifting the focus to production constraints: In presence of a stiff voiceless coda, stiffness is anticipated in the syllable body to prepare the vocal tract for upcoming biomechanical demands, producing faster articulations in the syllable body before voiceless codas. From this perspective, Hawkins and Nguyen’s (2004) finding that onset /l/ is shorter before voiceless codas is compatible with our view that voiceless codas increase stiffness through the entire syllable body, shortening different onset types. Similarly, stiffness provides a further and unified explanation for Raphael's (1975) finding that, as the vowel’s acoustic signal shortens before a voiceless coda, so does the EMG signal reflecting genioglossus activity: The faster genioglossus movement is compatible with speakers increasing syllable-body stiffness before a voiceless coda. Finally, Mielke and Nielsen (2018) observed a continuum in which the more constricted the upcoming voiceless obstruent, the shorter the word-initial aspiration. Rather than interpreting this pattern as onset aspiration airflows mirroring “future airflow requirements” (2018: 2175), the pattern may reflect a stiffness continuum in terms of graded changes in the stiffness of the supralaryngeal and laryngeal gestures associated with the onset stop. That is, the more constricted the upcoming consonant, the stiffer its closure gesture, and the greater the stiffness anticipated by supraglottal and glottal gestures in the syllable body, including the onset. This may explain Mielke and Nielsen’s pattern of onset aspiration shortening the most before fully occluded voiceless stops: Producing these sounds requires a tight oral constriction under rising intraoral pressure, a constriction established with a stiff closure gesture. This stiffness may be anticipated in the onset, leading to faster, more forceful supralaryngeal release and laryngeal adduction, and consequently shorter aspiration.

From this perspective, anticipating future stiffness requirements may lead the speakers to treat the syllable as a unit of production and perception. In production, the propensity of speakers to anticipate future vocal tract states is known from the coarticulation literature. For example, jaw and lip height during consonant closure reflects the aperture of an upcoming vowel (Sussman et al., 1973), and upcoming tongue anteriority is anticipated largely in advance (Whalen, 1990). Stiffness may thus be anticipated similarly to the location and degree of future constrictions. In dynamical models, this capacity to anticipate future requirements makes the production system efficient in achieving constriction goals under perturbations (Kelso et al., 1984). Similar to a perturbation, the intraoral pressure buildup associated with a voiceless coda likely requires an increase in closure force, force that may be anticipated by increasing stiffness in the syllable body. This aligns with the observation that vowels in CV syllables become progressively shorter as codas with increasing degrees of constriction are added (Belasco, 1953; Delattre, 1940, 1941, 1962; House & Fairbanks, 1953; Peterson & Lehiste, 1960), a shortening effect that suggests and increase in vowel stiffness in anticipation of progressively stiffer coda constrictions. Notably, Lehman (1993) compared the duration of the vowel /i/ in an open syllable (/CV/, bee) and in syllables with a voiced coda (/CVD/, bead) and a voiceless coda (/CVT/, beet). The difference in vowel duration between bee and bead was not significant, whereas both were significantly longer than beet. Lehman interpreted this result as a replication of the finding reported by Rositzke (1939). From this perspective, while /CV/ and /CVD/ syllables pattern similarly, the presence of a voiceless coda appears to be the critical factor associated with temporal shortening in the preceding syllable onset. This is consistent with our proposal that voiceless codas may be responsible for the stiffness-related adjustments observed in our data and reviewed in Table 1. Thus, anticipation of upcoming stiffness may shape the syllable at both the macroscopic level (/CV/ vs. /CVC/, etc.) and the microscopic level (/CVC/ with different coda classes), with voiceless codas significantly increasing stiffness compared to the default level of /CV/ and /CVD/ syllables.

In perception, the fact that onsets and vowels shorten systematically before voiceless codas suggests that the syllable body may function as a perceptual unit. Whalen (1989) found that percepts of onset and nucleus are mutually dependent: perceptual decisions made about the vowel influenced those about the onset fricative, and vice versa. This reciprocity may be elucidated by the concept of syllable-body stiffness. It may be that the relation between onset and vowel percepts is reciprocal, rather than simply anticipatory, because both share a common stiffness setting within the syllable body. This shared stiffness node could enable information exchange between onset and vowel during planning and perception. Our Syllable-Body Stiffness Hypothesis predicts that the syllable functions as a window over which a certain stiffness range is active in planning and production, shaping both articulatory and acoustic signals. The cohesiveness of stiffness in the signal may then be perceived as a unit. In Whalen’s (1989) /CV/ syllables, the stiffness domain likely encompassed only a syllable body. In /CVC/ syllables, this domain may extend from onset to coda. Such unitary onset-to-coda dynamics may support the listener in speech processing and acquisition (cf., Xu, 2025). This point is illustrated in Fig. 7a, which shows schematized audio signals for a word pair distinguished by coda sonorant (/TVL/; panel a) vs. coda voiceless obstruent (/TVT/; panel b). Dynamic changes in constriction (Manner Dynamics) and glottal state (Voicing Dynamics) are represented in the figure by an arrow and a schematized glottis, respectively. Differences in coda pressure are also highlighted. Compared to the voiced coda condition (upper panel), in the voiceless coda condition (lower panel), vowel periodicity begins earlier after the stop release, signaling more rapid glottal adduction. A similar dynamics is found in the coda: The coda-to-vowel transition is more abrupt before the voiceless coda, signaling more rapid and energetic glottal and supraglottal gestures. During speech acquisition, listeners may associate the rapid syllable dynamics with the firm voiceless closure at the syllable’s end, learning to anticipate coda pressure buildups through stiffer movements.

Fig. 7.

(a) Example audio signals of the English words car /kɑɹ/ (top) and cap /kæp/ (bottom), with annotations representing the articulatory and aerodynamic configurations underlying the audio signal. Regions of expected low vs. high intraoral pressure are highlighted in green and red, respectively. The horizontal line at zero-crossing serves as a reference for vocal tract closure. Arrows illustrate manner dynamics as changes in constriction degree. Voicing dynamics is represented as changes in glottal aperture. (b) A schematic illustration of the coupling between syllable-body and coda stiffness nodes in the cognitive representation of a CVC syllable, as proposed in our Syllable-Body Stiffness Hypothesis. The schematic notation k_C → k_B → {k_β1, k_β2} represents the directional influence of stiffness, with coda stiffness (k_C) being anticipated in the syllable-body stiffness node (k_B). Then, the onset (k_β1) and vowel (k_β2) receive stiffness information from the syllable-body node (k_B).

Thus, the speech system may spread coda information to the onset, but not in terms of the coda’s phonological identity. Through stiffness anticipation, the system may signal to the listener what level of articulatory effort the coda closure demands. This may explain why different coda classes, spanning from voiced obstruents to liquids, are similarly related to a long onset aspiration, while coda voiceless obstruents are specifically anticipated with a short aspiration. In fact, early in the syllable, acoustic dynamics may inform the listener of upcoming significant changes in articulatory dynamics, such as the presence or absence of a constriction in the coda, its type, and how quickly and forcefully it will be established. Supporting the idea that the syllable transmits perceptually valid information about articulatory dynamics, studies have shown that listeners across different English dialect backgrounds rely on vowel-to-coda transition slopes to perceive coda voicing (Pycha & Dahan, 2016; Thomas, 2000).

Stiffness may also help explain the role of the syllable body in phonological awareness and literacy. For example, Kim (2007) showed that, for early readers of Korean, “body-coda awareness is an important predictor of word decoding and spelling” (2007: 69). Importantly, kindergartners from other linguistic backgrounds, such as Hebrew and Russian, also show greater sensitivity to CV-C divisions than to rhyme-based divisions (Kogan & Saiegh-Haddad, 2023). The presence of this pattern in children acquiring languages with different writing systems is consistent with an important role for the syllable body at the cognitive, phonological level. Moreover, this literacy pattern relates to the finding that onset and nucleus percepts are interdependent (Whalen, 1989). In fact, phonetic objects within the syllable body may be closely linked as a cognitive routine in perception and in literacy because they are tightly synchronized at the motor level through the assignment of a cohesive stiffness value. This view suggests that a network of syllable-internal stiffness relationships must be represented at the speech planning level, as we propose here.

Our findings and our model of stiffness anticipation relate to recent models of speech production, in which the anticipation of future perturbations is an important property of articulatory control (Parrell et al., 2019). However, important differences exist between our proposal and existing models. In particular, although control models differ substantially in their approach to consonant and vowel properties, construed as acoustic features or articulatory gestures, and involving either discrete or dynamic processes (Parrell et al., 2019; Tilsen, 2016, 2019), the different models generally converge in viewing anticipation as a process that involves canonical properties of individual vowels and consonants, such as voicing, manner and place of articulation, or nasality. For example, Tilsen (2020) provided evidence that coda information is present in the onset signal, a finding replicated in our study. From Tilsen's (2013a) Selection and Coordination theory, the coda would influence the onset primarily because the voicing specification of the coda consonant is active simultaneously with that of the onset consonant, that is, glottal adduction and abduction are coactive. However, in a /TVD/ syllable, this leads to two predictions: Either the onset’s abduction gesture is dissimilated from the coactive coda adduction gesture, predicting longer onset VOT before voiced codas, or it is assimilated, predicting shorter onset VOT; both predictions are possible based on the current state of the theory. Thus, while the gestural coactivation of Selection and Coordination theory provides interesting insights into long-distance phenomena such as consonant and vowel harmonies (Tilsen, 2019), its application to the coda-to-onset aspiration effect reported here currently leads to underdetermination due to two symmetrical predictions being available within the model.

By contrast, we propose that the coda-to-onset aspiration effect is due to the speech motor control system anticipating not voicing, manner, or other canonically defined properties of consonants and vowels, but more simply how forceful a movement will be required by an upcoming consonant. Indeed, the large class of coda consonants associated with default onset long VOT is more consistent with shortening before an increase in stiffness required by an upcoming voiceless consonant. Therefore, rather than anticipation of canonical properties such as voicing, the coda-to-onset aspiration effect suggests a scenario in which the system anticipates the force and rapidity with which upcoming articulatory transitions are realized. From this view, more rapid and forceful onset-to-vowel transitions are expected before rapid and forceful vowel-to-coda transitions, in both supralaryngeal and laryngeal gestures. More specifically, coda voicing is known to have certain biomechanical consequences for movement force and rapidity (Chen, 1970), a more forceful and rapid closure being observed for voiceless than voiced codas. Thus, the shortening of onset aspiration before voiceless codas suggest that onset gestures may be planned in relation to the expected biomechanical consequences of upcoming consonant voicing (force, rapidity, stiffness), rather than to voicing or devoicing gestures themselves.

We should also consider the possibility that this stiffness anticipation operates across larger units than the syllable, including the foot or phrase, or within more gradient timing domains. This aligns with Fowler’s view that speech segments are broader and less sharply bounded than theoretical linguistic categories (1980, 2015). Indeed, Mielke and Nielsen (2018) found that onset VOT responds not only to the coda, but also to a following onset. They discuss Grassman’s Law in Ancient Greek and Sanskrit, where aspirated consonants lose aspiration if followed by another aspirate in the next syllable (2018: 2175). Under our model, Grassman’s Law could reflect a process through which the speech production system increases the stiffness of the first voiceless obstruent to anticipate the upcoming stiff voiceless obstruent, producing rapid supraglottal and glottal transitions into the vowel, thereby reducing aspiration.

A model in which syllable-body stiffness is cohesively planned at the production level is compatible with the tight synchronization between onset and vowel studied in dynamical approaches to speech. Noting that onset and vowel movements typically begin nearly simultaneously, researchers have proposed that they are synchronized (or “coupled”) in an in-phase fashion at the cognitive level (Browman & Goldstein, 1992; Goldstein et al., 2006; Saltzman & Byrd, 2000). However, the source of this synchronization remains unclear. Du et al. (2025) have suggested that stiffness relationships may underly the syllable-internal timing relationships long studied in the coarticulation literature, such as gestural overlap. In our framework, stiffness could serve as the currency mediating syllable-onset phasing: Onset and vowel movements may be tightly coordinated because they share a stiffness node in the syllable’s cognitive representation.

Fig. 7b illustrates this structure. In our theory, given a CVC syllable, the force and timing of onset and vowel gestures are controlled by the syllable-body stiffness node k_B. Since their respective stiffnesses k_β1 and k_β2 emanate from this common node, onset and vowel should display cohesive motor behaviors, consistent with their near-synchronous initiation. The syllable-body node k_B receives input from the coda’s stiffness node k_C, predicting stiffness anticipation and coherent changes in both onset and vowel durations before the same coda types. This aligns with the cohesive effects of codas on the durations of both onsets and vowels reported in the literature (Table 1). Importantly, consonant and vowel gestures have intrinsic differences in stiffness, with vowel being less stiff than consonants (Browman & Goldstein, 1989; Iskarous & Pouplier, 2022). Thus, a prediction is that stiffness modulation on the syllable body will have different effects on stiffer (consonantal) and less stiff (vocalic) gestures. This would explain the fact that onset and vowel durations vary in similar directions depending on the coda, but by different proportions, with the onset displaying about one third of the variation observed in the vowel (Port & Rotunno, 1979). If we treat the supraglottal and the glottal gestures as separate but tightly coupled, we could expect that their stiffness responses could differ as well. There is currently too little data in the literature to be certain of this. Also note that the proposed syllable-body node is intended only as a source of stiffness modulation shared by the onset and vowel gestures.

Our work has implications for defining the syllable within the Articulatory Phonology and Task Dynamics (AP/TD) framework. In AP/TD, the syllable has been characterized as a “molecular unit” (Iskarous & Pouplier, 2022: 6) emerging from time and phase relations between gestural atoms (Browman & Goldstein, 1988, 1995, 2000). That is, the syllable emerges as a consequence of coordination and timing constraints established by the linguistic system on individual gestures. In fact, our syllable-body unit entails another molecular unit, i.e., a node controlling onset and vowel stiffness, nested within the syllable, not a syllable-body gesture with its own tract variables or constriction target. In this respect, our proposal differs from applications of AP/TD in which prosodic units are modeled as gestures (e.g., Byrd & Saltzman, 2003; Saltzman & Byrd, 2000). From our view, the syllable may emerge as an articulatory unit because of the stiffness network previously described and represented in Fig. 7, without the need for dedicated syllable-body tract variables.

The stiffness anticipation arising from the concatenation of gestures represented in Fig. 7b may be independent of the prosodic adjustments of stiffness described in theories of prosodic gestures such as the π- and μ-gestures of Articulatory Phonology and Task Dynamics (Byrd et al., 2000; Saltzman et al., 2008). Specifically, previous work in speech biomechanics has suggested that a gesture’s stiffness is modulated by the nervous system depending on utterance-level properties (Ostry & Munhall, 1985: 641), consistent with our finding that onset stiffness depends on coda stiffness. This suggests a temporal ordering of control: During planning, the speech motor system first prepares subsyllabic stiffness relations across gestures, and only subsequently may prosodic gestures apply additional modulation.

In this view, prosodic gestures in Articulatory Phonology and Task Dynamics may not directly set the stiffness of individual gestures, but instead operate on a system that has already been biased by syllable-internal anticipatory stiffness relations (cf., Fig. 7). Thus, syllable-level organization in speech may emerge primarily from stiffness anticipation rather than from an independently specified syllable gesture or coupling architecture. That is, if the syllable functions as a speech unit, this may be a consequence of syllable-internal stiffness anticipation.

The glottal gesture also requires explicit formalization in our model. The VOT interval encompasses the acoustic outcomes of both supraglottal and glottal gestures. Glottal dynamics may vary because the glottal gesture receives information directly from the syllable-body node. Then, before the stiff voiceless codas, glottal stiffness in the onset also increases, producing more rapid adduction and shorter VOT. Another possibility is that when onset VOT shortens before a voiceless coda, it is only the supraglottal movement that receives the stiffness increase directly from the syllable-body node. Then, since the supraglottal and glottal movements are part of the same “gestural constellation” (Browman & Goldstein, 1992) for the stop, the glottal gesture follows the stiffness change of the coupled supraglottal gesture. A similar scenario may apply to the coda: Since the supraglottal closure is coupled with glottal abduction within the same segment, stiffness information may be exchanged between the two movements, producing a stiffer devoicing gesture before voiceless codas. A stiffer gesture being supposedly faster, this may align with (Klatt, 1976: 1214) claim that the glottis opens earlier before voiceless codas.

Note that the shorter onset aspiration before voiceless codas may suggest an anticipation of coda stiffness in the onset also from the perspective of gestural overlap. Increased stiffness has been linked to increases in overlap between two synchronized gestures (Byrd et al., 2000). Aspiration emerges from the synchronization of stop release and fold adduction (Lisker & Abramson, 1964), leading to different degrees of overlap between the glottal and supraglottal gestures. The extent of this overlap contributes to differences in aspiration (Abramson & Whalen, 2017), aspiration being shorter when release and adduction overlap more. Thus, if the stiffness of onset stops is increased before stiff codas, onset release and adduction should overlap more, shortening aspiration and VOT, as observed.

Anticipation of the dynamic requirements of the coda in the syllable body entails that the entire syllable functions as a “coordinative structure” of articulator groups that work together to anticipate upcoming constrictions and perturbations, such as those coming from different aperture and constriction degrees. When different motor structures are coordinated, the “challenge experienced by one (or more) members of the group [is] responded to by other members of the group at a site remote from the challenge” (Kelso et al., 1984: 813), i.e., from coda to onset. From this view, the syllable body node may receive input from the coda as part of a large motor “synergy”, i.e., a collaboration between multiple articulators to achieve the same goal (Iskarous et al., 2013). The synergy may indeed span the syllable level in both spatial and temporal terms, distributing motor tasks across articulatory networks to improve motor stability and flexibility, in a way similar to what proposed by Latash et al. (2007) for general motor control, and relied upon by Xu (2025) for syllable production and acquisition.

It is important to address alternative explanations for the findings reported in the present study and across the literature (Table 1). An alternative account of the syllable-body stiffness node is that the stiffness of the onset and nucleus are modified separately. The indication that there is a larger change in the nucleus stiffness than in the onset gestures (cf., Port and Rotunno, 1979) is consistent with such an approach. This would also be compatible with phonologies that allow a flat syllable structure (Clements & Keyser, 1983; Scheer & Cyran, 2017). Because our review suggests that stiffness is not treated identically in vowel and consonant gestures, it might be that addressing them separately is the most accurate treatment. If our assumption is correct that a single, proportional modification applies, then it would be simpler to model it via a syllable body node. Additionally, treating these findings without a shared syllable-body stiffness node may entail devising a process in which vowels and onset shorten before voiceless codas because the influence of the coda propagates regressively through the syllable, in a stepwise manner, so that the coda influences the nucleus, which in turn influences the onset. However, in such a scenario, the question would remain of why onset and vowel vary in the same direction but by different proportions before the same coda types (Port & Rotunno, 1979), a pattern that may be better addressed by a stiffness-node with adjustments due to intrinsic differences in stiffness between onset and vowel. Further experiments are needed to dissociate these possibilities.

Finally, the idea that a single command changes stiffness in the syllable body, even though the result is clearly different for the consonant and the vowel, is similar to the “elasticity” approach of Morley and Smith (2023). One difference between these approaches is that our model assumes that both the laryngeal and supralaryngeal gestures (C and V) are likely to differ, while the “elasticity” idea is attributed to the lingual gesture alone. Having to specify each of these gestures individually is possible but is potentially less parsimonious. When there are trade-offs between the location of complexities in the model, parsimony becomes challenging to evaluate. Future work will be needed to sort out the various approaches.

4.4. Limitations and future directions

Future research needs to combine acoustics, articulation, and modeling to better understand how consonants affect the stiffness of glottal and supraglottal gestures in preceding segments. To further test our syllable-stiffness model, our next step is to study glottal dynamics using high-speed laryngoscopy, to examine whether glottal adduction in the syllable onset has higher velocity before voiceless codas, relating to the observed VOT shortening. This study will be complemented with computer simulation: The approximate 20% stiffness increase previously discussed may serve as a starting point for testing whether supraglottal and glottal gestures display qualitatively similar changes in response to the coda.

Future work needs to extend our design to additional vowel and consonant contexts. The English low vowels /ɑ/ and /æ/ used in the present study are similar articulatorily and acoustically, but a residual influence of vowel quality on VOT cannot be entirely excluded. More research is thus needed, with both low and high vowels. Additionally, studies with other English coda consonants, including a constriction continuum from voiceless obstruents to voiced obstruents and sonorants, may further test the generality of the coda-to-onset VOT effect. We also note that our design relied on auditory prompts, which raises the question of whether the participants imitated the prompt. Our results provide evidence against a simple imitation account: The patterns observed across our English and French participants differed from the VOT values of the respective auditory prompts. Specifically, although the English /tVC/ auditory prompts displayed longer onset VOT in the coda-/T/ condition (counter to the known trend), participants did follow the predicted pattern, producing longer onset VOT in tar than tap (Fig. 2; cf., Section 2.2). Similarly, the pattern observed across the French participants differed from that in the French auditory prompts (Fig. 6; cf., Section 2.3). This mismatch suggests that the observed effect is unlikely to be driven solely by imitation of the prompts. However, future studies should compare auditory and written prompts to further assess the extent to which the effect could be modulated by perceptual factors and imitation. Additionally, we plan to compare French /CVC/ words with genuine voiced-voiceless coda contrasts (/tVT/-/tVD/), to further assess the extent to which onset aspiration responds to coda voicing in French.

Note that our results for French /ʁ/ may contribute to the open debate on the voicing status of /ʁ/. In our data set, the /ʁ/ was overwhelmingly voiceless in the coda of CVC words and in /Tʁ/ clusters, which aligns with Gendrot et al. (2015). However, other contexts have led theorists to classify /ʁ/ as voiced (Fagyal et al., 2006). Notably, the /ʁ/ does pattern as voiced in /Dʁ/ clusters (Russell Webb, 2009), where it does not devoice the voiced stop, as other voiceless fricatives do, as with /s/ devoicing /b/ in absurde [apsyʁd] (Hallé et al., 2000). Gendrot et al. (2015) highlight the constriction degree and resulting intraoral pressure buildup as critical factors in devoicing the /ʁ/. Thus, the /ʁ/ voicing puzzle remains unsolved, calling for further research on acoustics, articulatory dynamics, and intraoral pressure during the production of /ʁ/, possibly across different speaking rates.

Formant transitions before different codas is another aspect of coda stiffness we plan to investigate. Our stiffness-based account predicts that voiceless codas trigger steeper formant slopes in both onset-to-vowel and vowel-to-coda transitions. Additionally, by examining polysyllabic words, we aim to explore how far in advance syllable-body stiffness responds to upcoming constrictions. Relatedly, Mielke and Nielsen (2018) found that onset VOT responds not only to the coda, but also to a following onset.

5. CONCLUSION

This study showed that the duration of a stop’s aspiration in the syllable onset varies depending on the coda consonant. In English, onset voiceless stops had shorter aspiration before coda voiceless stops compared to coda /ɹ/. In contrast, in French, there was no significant difference in onset aspiration before coda /ʁ/ vs. a voiceless stop. These results indicate that onset release and vocal fold adduction are stiffer, and thus more rapid, before voiceless coda obstruents.

The finding that coda effects extend to the onset highlights the syllable body (onset and vowel together) as a unit of speech timing, affecting onset-vowel phasing. While studies have mainly focused on the syllable rhyme, the syllable body warrants more attention. By focusing on the syllable body, we bridged two previously separate lines of research: The well-established effects of codas on vowel duration, and the less-studied effects of codas on onset timing, highlighting similar coda effects across the entire syllable body. Onsets and vowels shorten before voiceless codas, though with different magnitudes, which may reflect the intrinsic compressibility of consonantal and vocalic gestures. This synthesis generates new predictions about the role of coda biomechanics in determining overall syllable timing: Speakers appear to anticipate future intraoral pressure and stiffness when planning syllable-wide temporal dynamics.

With coda stiffness largely anticipated through the syllable body, coarticulation may involve not only the anticipation of future spatial properties (e.g., place and manner) but also the anticipation of future temporal dynamics, such as how rapidly and forcefully an upcoming consonant is articulated. This large-scale anticipation of spatio-temporal properties suggests that the syllable functions as a speech production unit, with cohesive articulatory and acoustic dynamics from onset to coda. In turn, stiffness anticipation may support the syllable as a unit of perception and literacy, binding onset and vowel together as gestures, percepts, and written symbols.

Highlights.

VOT in the syllable onset is longer before coda [ɹ] than voiceless codas in English.

No significant effect of French coda /ʁ/ on onset VOT was observed.

As is typical, vowel duration was also shorter before voiceless consonants in English.

The Syllable-Body Stiffness Hypothesis accounts for coda effects on onset and vowel.