At the juncture of prosody, phonology, and phonetics—

The interaction of phrasal and syllable structure

in shaping the timing of consonant gestures.*

Dani Byrd

Susie Choi Department of Linguistics, University of Southern California

Los Angeles, California USA

The Laboratory Phonology research paradigm has established that abstract phonological structure has direct reflections in the spatiotemporal details of speech production. Past exploration of the complex relationship between linguistic structural properties and articulation has resulted in a softening of a severance between phonetics and phonology, which in recent years has begun to likewise encompass the phonetics-prosody interface. The abstract organization not only of words, but also of phrases, has a complex and rich effect on the articulatory details of speech production. This paper considers how phonological and prosodic structure interact in shaping the temporal realization of consonant sequences. Specifically, we examine the influence on articulatory timing of syllable structure effects in onset, coda, and heterosyllabic positions, and prosodic effects in the vicinity of a variety of phrase boundary conditions that grade in strength. Simultaneously, this work seeks to test and refine the π-gesture theoretical model of prosodic representation in speech production (Byrd & Saltzman 2003).

1. Introduction Three of the founders of Laboratory Phonology—Beckman, Ladd, and

Pierrehumbert—describe its research paradigm as “involving the cooperation of people…who share a concern for strengthening foundations of phonology through improved methodology, explicit modeling, and cumulation of results” (Pierrehumbert, Beckman, and Ladd 2001). Over the preceding nine Laboratory Phonology conferences, leading research in linguistic speech experimentation has clearly established that many aspects of abstract phonological structure is directly reflected in the spatiotemporal details of speech production. In fact, a softening of the clear severance between phonetics and phonology and an appreciation of the complexity of their relationship has been a prominent contribution of the Laboratory Phonology tradition. In recent LabPhon volumes, the study of the phonetics-phonology interface has been extended to the phonetics-prosody interface, and we have, again, seen that abstract structural properties, this time of sentences rather than words, have a complex and rich effect on the * The authors are grateful for the support of the NIH NIDCD (DC03172). The authors thank an anonymous reviewer, Cécile Fougeron, Vince Gracco, Dr. Masanobu Kumada, Sungbok Lee, Christine Mooshammer, Janet Pierrehumbert, Elliot Saltzman, the organizers of LabPhon 10, and Haskins Laboratories, where this data was collected.

Byrd and Choi

2

articulatory details of speech production (see e.g., Papers in Laboratory Phonology II, V). This has led to new theoretical conceptions regarding how phrase boundaries should be represented and to new computational models of their realization in speech (e.g., Papers in Laboratory Phonology V and VIII; Byrd and Saltzman 2003). In the study described below, we move from two areas of laboratory phonology that are reasonably well-explored—the effects of syllable structure and of phrasal structure on articulatory overlap—to a consideration of how these phonological and prosodic structural properties interact in shaping articulatory timing.

1.1. Background regarding structural effects on articulatory timing Articulatory timing refers to both the duration of and overlap among articulatory gestures. Articulatory timing is affected by both syllable structure, which we consider to be phonological, and phrase structure, which we consider to be prosodic. In previous articulatory experiments in English, word boundaries have generally been used to define syllable edges (since the syllable status of medial consonants in English can be ambiguous); so we will refer to syllable structure in considering the position of articulatory gestures as final in a word—coda—or initial in a word—onset—or spanning a word boundary—heterosyllabic. Gestural position within the prosodic structure of an utterance can also be characterized as phrase-final, phrase-initial, or phrase-medial (i.e., at a word boundary that does not coincide with a phrase boundary). Of course, two gestures could span a phrase boundary such that both are boundary-adjacent, one phrase finally and the other phrase initially. Articulatory gestures at syllable (word) edges and at phrase edges exhibit longer duration than those located word- or phrase-internally (e.g. Pierrehumbert and Talkin 1992; Byrd 1996; Fougeron and Keating 1997; Byrd and Saltzman 1998; Cho and Keating 2001; Fougeron 2001; Keating, Cho, Fougeron and Hsu 2003; Tabain 2003; Byrd et al. 2005; Bombien et al. 2006; Cho 2006). Both single consonants and consonant clusters in different word positions show specific duration and relative timing patterns (Krakow 1993, 1999; Delattre 1971; Sproat and Fujimura 1993; Hardcastle 1985; Harrington et al. 2004; Byrd 1996; Hardcastle and Roach 1979). Position in a word has been shown to influence gestural overlap in consonant clusters such that onsets are less overlapped (and less variable in their timing) than like clusters in coda or heterosyllabic positions (Byrd 1996 see also Nam and Saltzman 2003, Nam, Goldstein, & Saltzman to appear, Loevenbruck et al. 1999). Further, there is some evidence that articulatory gestures spanning a phrase juncture are less overlapped than those phrase medially and that this overlap is further decreased for prosodically stronger phrase boundaries (McLean 1973; Hardcastle 1985; Holst and Nolan 1995; Browman 1995; Byrd et al. 2000; Byrd 2000; Cho 2004; Bombien et al. 2006). However, it remains unknown how syllable and phrasal structure interact in determining intra- and intergestural coordination for consonant clusters preceding (in coda position), spanning (heterosyllabic), and following (in onset position) a phrase boundary.

Byrd and Saltzman (2003) (see also Byrd et al. 2000; Byrd 2000, 2006) have proposed a model of prosodic boundaries that makes specific predictions as to the temporal behavior of articulation near phrase boundaries. The prosodic (π-) gesture model views phrase boundaries as extending over an interval and acting to slow the time course of articulatory gestures that are active during that interval. Thus, the π-gesture

Byrd and Choi

3

model extends to the suprasegmental level a fundamental insight of Articulatory Phonology (e.g., Browman and Goldstein 1992), namely, that phonological units are inherently temporal. Articulatory Phonology posits atomic phonological units that are tasks or gestures defined in the constriction domain. One of the inherent properties of a gesture is that it is active over a temporal interval and that the activation intervals of gestures are patterned or choreographed in an overlapping fashion. The π-gesture model extends this notion to the level of prosodic structure. Figure 1 shows a representative partial gestural score schematizing the overlapping arrangement of a prosodic gesture, whose activation waxes and wanes, with two co-active constriction gestures.

Figure 1: A schema of the π-gesture model indicating two constriction gestures

overlapping both one another and a prosodic boundary (π) gesture.

While a prosodic gesture does not have a constriction task, it has an effect on the pacing of constriction tasks during its interval of activation. The activation strength of the π-gesture, and therefore the degree of local slowing it gives rise to, corresponds to the strength of a phrasal juncture. Many of the empirical findings on intra and intergestural timing at phrase boundaries have been successfully simulated using the π-gesture model (see Byrd and Saltzman 2003), and a number of the predictions of the π-gesture model regarding the general symmetricality of prosodic effects on articulation and the waxing and waning nature of such effects have recently found empirical support (Byrd et al. 2005; Bombien et al. 2006; Lee, Byrd and Krivokapić 2006; Byrd, Krivokapić, and Lee 2006.).

1.2. Hypotheses regarding consonant sequences The prosodic-gesture model of prosodic timing predicts longer durations and less

overlap under the influencing, coproduced interval of a phrase boundary (i.e. prosodic gesture), due to slowing of the central clock pacing the unfolding of gestural activation intervals (Byrd and Saltzman 2003). Specifically for consonant sequences at phrase boundaries, we will test the following hypotheses motivated by the π-gesture model as seen in Figure 2.

Byrd and Choi

4

Figure 2: A schema representing hypotheses motivated by the π-gesture model: (1) lengthening of both C1 and C2 and less C1-C2 overlap in heterosyllabic clusters (top-center), (2) more

lengthening of C2 than C1 in coda clusters (bottom-left), and (3) more lengthening of C1 than C2 in onset clusters (bottom-right). Internal timing changes in onset and coda clusters are also

possible.

The first hypothesis is that phrasal lengthening of constriction formation will affect both C1 and C2 in heterosyllabic clusters C##C, more C1 in onset clusters ##CC, more C2 in coda clusters CC##. This is because, at least as a simplest starting assumption, the strong activation portion of the π-gesture will be generally centered between the phrases. This supposition has received support in Byrd et al. (2005). This means that the constriction events closest to the end of the first phrase and to the beginning of the second phrase will be most affected because they are within the strongest activation of the π-gesture.

The second hypothesis is that there will be an interaction of boundary type and syllable position on CC overlap. Specifically, heterosyllabic clusters (C##C) spanning a phrase boundary are of course expected to exhibit large decreases in CC overlap. This has been observed in earlier experimental studies (see, e.g. Byrd, et al. 2000) and captured in simulation work (Byrd and Saltzman 2003). Coda (CC##) and onset clusters (##CC) are expected to show smaller but significant effects. That is, the effects of prosody on the relative timing of articulation are expected, within the π-gesture model, to ‘reach into’ the word-internal timing of gestures adjacent to the boundary even though they do not themselves span a phrasal juncture. Other accounts of the implementation of prosodic boundaries that only consider the effects of boundaries on phrase-initial or -final segments cannot explain relative timing changes farther downstream/upstream.

Further, we will evaluate a third prediction that timing effects will grade with respect to boundary strength. We are not concerned in this particular study with whether the phrasal junctures themselves are gradient or categorical in nature. Rather, “graded,” here simply refers to the prediction that the manipulation of the independent variable of phrase juncture—from no-phrase-boundary to a large utterance boundary—will engender

Byrd and Choi

5

a range of statistically different strengths of timing effects, ordered such that larger amounts of gestural lengthening and/or smaller amounts of gestural overlap will be associated with larger boundaries. This is in contrast to a situation in which only a two-way distinction between lengthening and no-lengthening obtains.

Finally, in addition to prosodic effects and their general interaction with syllable structure, we are interested in a qualitative evaluation of whether the segmental composition of a syllable onset will influence its prosodic patterning. Based on other studies, s+stop clusters are expected to be less overlapped as word onsets than as coda or heterosyllabic sequences; the potential patterning of other types of cluster such as [kl] is less clear. Gestural conglomerations such as [s+stop] with a characteristic and stable pattern of coordination observed repeatedly in a language’s lexicon are referred to as gestural “molecules” (Saltzman et al. 1998, Saltzman 1995; Saltzman, Löfqvist, and Mitra 2000; Goldstein, Byrd, and Saltzman 2006; Browman and Goldstein 2000). Word-initial clusters of s+stop in English (and some other languages) have long been recognized as having special articulatory and phonotactic properties such as a single laryngeal gesture, no voicing contrast permitted, and a lack of stop aspiration (Browman and Goldstein 1986, 1992; Hoole, Fuchs, and Dahlmeier 2003; Petursson 1977; Löfqvist and Yoshioka 1980, 1981; Goldstein 1990 citing Yoshioka, Löfqvist, and Hirose 1981, and Fukui and Hirose 1983; see also Harrington et al. 2004). It remains to be seen whether syllable position and boundary type might interact in a qualitatively different pattern for [s+stop]-type onsets and as compared to non-[sC] onsets. If English [s+stop] onsets have a special status as gestural molecules, they might be less susceptible to or prone to prosodic perturbation. We will examine whether these particular onset clusters are insulated from prosodic timing effects by virtue of their status as gestural molecules and what pattern in comparison is shown by non-[sC] onset clusters. It is interesting to note that the predicted decrease in overlap near a boundary, described above, would be somewhat in tension with a resistance to overlap changes by gestural molecules. One force might be resolved while sacrificing the other, or the two may accommodate to some degree. It is hoped that the qualitative data patterns below will be informative on this matter. 2. Method

2.1. Stimuli We investigated these hypotheses by examining tongue and lip movements during the production of two-member consonant sequences adjacent to phrasal boundaries of varying strengths, ranging from no phrase boundary—i.e., a simple Word Boundary, to an Intermediate Phrase Boundary, to an Intonational Boundary, to an Utterance Boundary. The Word condition was created by having a modifier-noun boundary. The Intermediate Phase condition was created by using a long initial subject noun phrase; the Intonational Phrase boundary condition was created by an introductory ‘when’ clause followed by an orthographic comma, and the Utterance Boundary condition was created between two complete clauses/sentences separated by an orthographic period. The stimuli are given in Table 1. In the results below, we designate the four boundary types with the abbreviations w for Word, im for Intermediate Phrase, IP for Intonational Phrase, and Utt for Utterance. (The intermediate phrase is often abbreviated ip, but below is

Byrd and Choi

6

abbreviated im to avoid confusion with IP in figures.) These terms for the boundary conditions are used because they are common in the literature but could, for our purposes, be understood completely atheoretically as: a control condition, a small phrase boundary, a large phrase boundary (without pausing), and an utterance boundary with an acoustic pause (note that in the study of speech production, articulation is still continuously varying during acoustic pauses when there is upcoming material). For each boundary condition, three two-member consonant sequences occurred: [sp], [sk], and [kl]. These occurred as word-final sequences, which we have labeled codas, word-initial sequences labeled onsets, and cross-word sequences, which we have labeled heterosyllabic. Two qualifications about this description of the stimuli must be made. First, the [kl] sequence in the coda condition was not actually a coda cluster but rather a [k] followed by a word-final [l] usually described as syllabic. For this reason, it is important to remember that the coda condition is actually more accurately described as a word-final condition. The [kl] was included here in order to allow for a fully-crossed design. The second note about the stimuli is that in the heterosyllabic condition the [sC] clusters were voiced, that is [zb] and [zg]. This was done to preserve the stop’s unaspirated status that existed in the other two conditions (onset and coda [sC]). Since we are interested in relative timing here, the decision to preserve the stop’s unaspirated status but voice the fricative was preferable over allowing a mismatch in the heterosyllabic stop quality as compared to the other two syllable conditions. So, with these two notes regarding the stimuli, Table 1 presents a 3-factor fully-crossed design in which the independent variables are consonant sequence (3-level), syllable position (3-level), and prosodic boundary (4-level). The boundary either fell before the sequence (for onsets), in the middle of the heterosyllabic sequence, or after the sequence (for codas).

This design will allow us to investigate the main effect of syllable type on the intra and intergestural timing of consonants in sequence and the main effect a prosodic boundary has on members of a neighboring cluster. Further, the design allows us to examine interactions of boundary and syllable in order to ask if the relative timing of clusters is differentially affected depending on the placement of a prosodic boundary either before, after, or in the middle of the cluster.

Byrd and Choi

7

Table 1: Stimuli sentences presented according to cluster, syllable position and boundary type conditions (W=control, word only boundary; IM=intermediate or small phrase boundary; IP=intonational or large phrase boundary; U=utterance or large phrase boundary with anticipated pausing; see text for details)

cluster syllable position boundary type sentence

[sp] coda W He saw the wasp eyes above the ground. [sk] coda W He saw the kiosk eyes above the ground. [kl] coda W He saw the focal eyes above the rest. [zg] heterosyllabic W He saw those guys above the ground. [zb] heterosyllabic W He saw those buys above the ground. [kl] heterosyllabic W He saw the oak lines upon the trail. [sp] onset W He saw no spies above the ground. [sk] onset W He saw no skies above the ground. [kl] onset W He saw no climb above the ground. [sp] coda IM The east wall wasp eyes over the prey. [sk] coda IM The east wall kiosk eyes over others. [kl] coda IM The dull soft yokel eyes over the barn. [zg] heterosyllabic IM Lean tall Rose guides a muffler around. [zb] heterosyllabic IM Lean tall Rose buys a muffler mound. [kl] heterosyllabic IM A lean tough oak lies above the pond. [sp] onset IM Lean tall Theo spies above the crowd. [sk] onset IM Lean tall Theo sky-dived into town. [kl] onset IM Lean tall Theo climbs above the ground. [sp] coda IP When he saw the wasp, eyes appeared all around. [sk] coda IP When he saw the kiosk, eyes appeared all around. [kl] coda IP When he saw the yokel, eyes appeared all around. [zg] heterosyllabic IP When he saw Rose, guys appeared all around. [zb] heterosyllabic IP When he saw Rose, buys appeared all around. [kl] heterosyllabic IP When he saw the folk, lines appeared all around. [sp] onset IP When he saw Theo, spies appeared all around. [sk] onset IP When he saw Theo, skies appeared all around. [kl] onset IP When he saw Theo, climbs appeared all around. [sp] coda U First, he saw the wasp. Eyes appeared then all around. [sk] coda U First, he saw the kiosk. Eyes appeared then all around. [kl] coda U First, he saw the yokel. Eyes appeared then all around. [zg] heterosyllabic U First, he saw Rose. Guys appeared then all around. [zb] heterosyllabic U First, he saw Rose. Buys appeared then all around. [kl] heterosyllabic U First, he saw the folk. Lines appeared then all around. [sp] onset U First, he saw Theo. Spies appeared then all around. [sk] onset U First, he saw Theo. Skies appeared then all around. [kl] onset U First, he saw Theo. Climbs appeared then all around.

Three native Speakers of American English read the 36 sentences (in this 4x3x3

design) with seven repetitions of each sentence blocked by sentence in a pseudorandomized order, yielding 756 tokens, with the constraint that no adjacent blocks be the same phrasal condition. (For two speakers, this dataset was the first collected in a

Byrd and Choi

8

slightly longer recording session.) The speakers reviewed the sentences before the experiment. The speakers are identified below as Speakers K, J, and N.

2.2. Data collection The electromagnetic midsagittal articulometer (EMMA) system (Perkell et al. 1992) was used to track the movement of the tongue and upper and lower lips during the speech. The EMMA magnetometer system transduced the horizontal (x) and vertical (y) movements of small receiver coils attached to the articulators in the midsagittal plane. Receiver coils were placed on the upper and lower lip, the tongue tip, and the tongue body. For Speakers N and J, three receivers were placed on the tongue body, and for Speaker K two were. For all speakers, the second receiver back from the tongue tip receiver was used as the relevant receiver for transducing [k] and was placed in a comparable position across receivers (i.e. as far back as comfortable for the subject and for placement by the experimenter). Receiver coils were also placed on the maxilla and nose for use in head-movement correction, and the data was rotated to bring the x-axis into alignment with the sampled occlusal plane.1 749 tokens were available for analysis; 7 tokens were lost to error.

2.3. Data analysis

The three types of consonant trajectories2 to be analyzed were the tongue tip (TT) vertical (y) trajectory for [s] and [l], a tongue body (TB) y trajectory for [k], and the Euclidean lip aperture (LA) trajectory for [p]—defined as:

!

lip aperture = lip aperture x( )2

+ lip aperture y( )2( )

where : lip aperture x = upper lip x - lower lip x;

lip aperture y = upper lip y - lower lip y

The timepoints of consonant gesture onsets and peaks were recorded, as determined by velocity zero-crossings in Lip Aperture ([p]), TTy ([s, l]), and TBy ([k]). This is shown schematically in Figure 3.

Figure 3: A schema of y-position (or dimensionless Lip Aperture) trajectories for C1 and C2 indicating the recorded timepoints and measurements of the consonant sequences: C1 onset and

peak, C2 onset and peak, and the resulting C1 and C2 constriction formation durations

Byrd and Choi

9

The algorithmic identification of these velocity zero-crossing timepoints for each consonant in the sequence allowed the calculation of four dependent variables of interest, as they identify the initiation of a constriction movement and the peak or end of the constriction formation.

• C1 Duration: C1 constriction formation duration=(time of C1 peak – time of C1 onset) (ms)

• C2 Duration: C2 constriction formation duration=(time of C2 peak – time of C2 onset) (ms)

• Absolute Latency: ∆Peaks=(time of C2 peak – time of C1 peak) (ms) • Relative Latency: ((C2peak-C1onset)/C1duration)

These four variables paint a picture of the absolute and relative timing of the consonants in the CC sequence. The first three are straightforward; the fourth is a measure that captures the proportion of the way through the first consonant that the second consonant peaks (Oliveira 2004; see also Byrd 1996; Byrd et al. 2000). Both latency measures are useful metrics of timing changes in the sequence but must be interpreted in light of individual changes in C1 Duration and C2 Duration. Importantly, Relative Latency as calculated above is particularly sensitive, in a non-linear way, to C1 duration (C. Mooshammer, p.c.). Lastly, at one point below, we will use the time between C1 and C2 onsets (∆Onsets) to evaluate a particular data issue that arises. Data analysis focuses on general patterns observed for all subjects; however, significant individual differences are also reported. The dependent variables outlined above are used in a repeated-measures ANOVA. The model uses pooled data with [Speaker] added as a random independent variable and the [Speaker x Variable] interaction as the error term in the test for [Variable], as described by Winer (1971) to provide control over individuals between experimental units (see e.g. Choi 1992; Byrd 1996). (Thus, the degrees of freedom reported for the error term are those of the [Speaker x Variable] interaction.) The SuperAnova package (Abacus Concepts, 1989) is used to perform the statistical tests. A full-interaction ANOVA with the independent variables of Speaker (3 level), Cluster (3-level), Syllable Position (3-level), and Boundary Type (4-level) is performed for each of the dependent variables. All and only results at the p<.01 level are reported as significant. The discussion will focus on the main effects and interaction of Syllable Position and phrase Boundary Type since these address the hypotheses we have laid out. Because of the many interactions in this model, we will not explore other interactions, except in the case where they bear on the particular hypotheses outlined above, such as those related to the status of onset clusters. Speaker effects and speaker interactions were uniformly present and will not be detailed below. Post-hoc and contrast of means tests are reported as significant at a criterial level of p<.05.

Byrd and Choi

10

3. Results

3.1. C1 and C2 duration Let’s first consider the duration of each of the consonants in the sequence as the syllable and phrasal position varies. For C1 Duration there is a significant main effect of Boundary Type (F[3,6]=41.48, p=.0002) and Speaker, and no other main effects. All interactions were also significant, with the exception of [Cluster x Syllable Position x Boundary Type]. For C2 Duration there are significant main effects of Boundary Type (F[3,6]=35.938, p=.0003), of Syllable Position (F[2,4]= 146.403, p=.0002), and of Speaker. All three- and four-way interactions with Speaker were significant, as was the two-way interaction of Speaker and Cluster. Turning first to the main effect of Boundary Type, which existed for both the first and second members of the CC clusters, Figure 4 presents these data pooled across subjects. The primary observation is that a consonant is in the vicinity of a prosodic boundary lengthens. Further, this lengthening increases in a graded way as the strength of the boundary (at least as intended for the stimulus sentences) increases. Fisher’s PLSD post-hoc tests find all pairwise comparisons of Boundary Type significant for both consonants (p<.05).

Figure 4: Pooled C1 and C2 Duration as a function of Boundary Type

Of course, it is important to consider the behavior of each of the consonants in all three syllable positions, as this determines the placement of the consonant either immediately at the phrase edge or slightly removed from it, and either before or after the boundary. Recall that in Figure 2, it was hypothesized that C1 would be least affected in codas and most subject to lengthening in heterosyllabic and onset sequences, while C2 would be least affected in onsets and most subject to lengthening in coda and heterosyllabic sequences. A two-way interaction of Syllable Position x Boundary Type existed for C1 supporting the prediction, and a 3-way interaction with subject also existed

Byrd and Choi

11

for C2 in support of the prediction. Figures 5 and 6 presents each speaker’s lengthening pattern as a function of syllable type for C1 and C2, respectively.

Figure 5 shows that all speakers exhibit a strong and graded prosodic lengthening effect on the first consonant in the onset cluster. They also show this strong lengthening for the phrase-final C1 in the heterosyllabic sequence. When the consonant is farther removed from the phrase edge as it is in the word-final (i.e. coda) sequence, Speakers N and K do show prosodic lengthening but of a smaller and less differentiated amount.

For C2 (Figure 6), all subjects show the most lengthening, as predicted, for the coda and heterosyllabic sequences. When C2 is further removed from the phrase-edge in the onset cluster case, one subject (K) shows graded lengthening of C2 Duration; Speaker J shows a small amount of C2 lengthening for the largest boundary.

Figure 5: C1 Duration as a function of Boundary Type for individual Speakers

Byrd and Choi

12

Figure 6: C2 Duration as a function of Boundary Type for individual Speakers In sum, we see that lengthening of the durations of the individual consonants in a CC sequence is sensitive to whether the consonant is at a phrase edge. However, even consonants removed slightly from the phrase edge can exhibit lengthening, though to a lesser degree. Further, when prosodic lengthening is strong, graded effects of boundary strength are observed.

3.2. Overlap We turn next to the patterns of temporal latency between the two consonants. The measure of Absolute Latency (∆Peaks) reflects the time from when the first consonant reaches its target to the time when the second reaches its target, and, as such, it is a useful measure of overlap. For Absolute Latency, there are significant main effects of Boundary Type (F[3,6]=22.297, p=.0012) and of Cluster (F[2,4]=18.28, p=.0097) and Speaker, though no main effect of Syllable Position. All interactions with Speaker are significant except [Speaker x Cluster x Boundary Type], and the two-way interactions of Syllable Position with both Boundary Type (F[6,12]=9.105, p=.0007) and Cluster (F[4,8]=77.306, p=.0001) are also significant.

Byrd and Choi

13

Figure 7: Absolute Latency (∆Peaks) (excluding Speaker J’s [k#l] which was in same direction of effect but even longer)

Figure 7 presents the change in overlap as measured in absolute peak latency for

all subjects pooled [one cell has been excluded—Speaker J’s [k#l] which behaved just as the pooled data shown but with larger values—in order to preserve the ability to compare differences on a single scale for all syllable conditions]. The unsurprising result is of course that when a consonant sequence spans a boundary, it becomes less overlapped. This effect is very large in the case of the Utterance boundary condition, where a pause is occurring, but is present in a graded way across boundary types. What is more interesting is that in a sequence of consonants preceding a boundary (CC#), overlap also decreases in a graded way with boundary strength. This could be due to the lengthening observed for C2 Duration. We note also that for all subjects the ∆Peaks changes due to Boundary Type for [kl] in both the coda and heterosyllabic conditions were quite large and one source of the 3-way interaction. The model was re-run with the [kl] cluster removed (leaving only the [sC]) clusters) and the main effect of Boundary Type remains significant (F[3,6]=12.758, p=.0052). The individual patterns for codas and onsets across cluster type and speaker are further explored below in Figure 9. In order to explore these latency changes further, another type of overlap measure will be evaluated: Relative Latency—the proportion of the way through C1 that C2 peaked. For Relative Latency, all two-way interactions and all interactions with Speaker are significant, as well as a significant Speaker main effect; there is also a 3-way Cluster x Syllable x Phrase interaction. The Syllable Position x Boundary interaction (F[6,12]=6.752, p=.0026) is shown in Figure 8, pooled across speakers.

Byrd and Choi

14

Figure 8: Relative Latency ((C2peak-C1onset)/C1duration)) pooled While the gradations are less consistent here—presumably because of the more derived nature of the variable which includes several timepoints—the effect of prosodic lengthening is still observed for coda and heterosyllabic sequences. The apparent small reversal of effect for the onsets is due to the fact that the C1 lengthening that occurred in this condition was large relative to the relative timing adjustments between the consonants.

In sum, intergestural timing in clusters both wholly before or after a boundary, as well as those spanning a boundary, show a sensitivity to the presence of an adjacent phrase boundary on the internal overlap of consonants in a cluster.

3.3. Specific cluster patterns At this point it will be useful to consider the patterning of individual clusters for each of the speakers. One purpose this serves is to ensure that the grading of lengthening and overlap effects observed in the pooled data actually reflect patterns found for individuals. Figure 9 presents this data for onsets and codas (the heterosyllabic data is very robust and individual patterns do not qualitatively differ from the pooled data); a checkmark has been placed above all subject-by-cluster patterns in which at least three of the four prosodic category differences go in the expected direction of lesser overlap at larger boundaries.

Byrd and Choi

15

Figure 9: Individual Speaker’s overlap patterns (∆Peaks and ∆Onsets) for Onsets (#CC) and Codas (CC#) as a function of Boundary Type

The frequency of the graded patterns of overlap changes in the expected direction

for individual speakers and individual clusters indicates that the pattern of gradation is fairly robust. Further, at least for onset clusters, the increase in ∆Peaks is not explained solely by C2 lengthening, as in many instances, ∆Onsets (the time between the initiation of the constriction movement for each consonant) is also affected (main effect of boundary type on ∆Onsets (F[3,6]=20.719, p=.0014); nor does C2 Duration always pattern like ∆Peaks for individual subjects. Finally, it is important to note that the effects on onset clusters are quite small relative to those same clusters in coda position (and obviously both coda and onset timing changes are much smaller than the large heterosyllabic effects).

3.4. Special behavior of onsets? Two issues regarding onsets were raised in Section 2. First, are onset clusters less

overlapped than like coda clusters, as has been observed at the word level by Byrd 1996 and Hardcastle and Roach 1979 using EPG? Second, are onset clusters, particularly [#sC] clusters, less sensitive to prosodic variation than other clusters?

Data across the range of prosodic conditions demonstrates that onset clusters are indeed less overlapped than like coda clusters. For each speaker, for each comparison of [sp] and [sk] in coda and onset in comparable prosodic positions, Absolute Latency (∆Peaks) is longer in every case for the onset [sC] cluster than the coda clusters.3 (Of

Byrd and Choi

16

course, this pattern does not obtain for [kl] since the [l] was syllabic in word-final position and had very long ∆Peaks.) These results can be seen in Figure 10.

Figure 10 : Pooled Absolute Latency (∆Peaks) for [sC] Onsets and Codas

Thus, the pattern of lesser overlap in onset clusters observed by Byrd (1996) and Hardcastle and Roach (1979) is evidenced not only at a word boundary but also when the cluster follows/precedes other prosodic boundaries. Let’s next turn to the question of whether or to what degree onset clusters are sensitive to perturbation in their internal timing caused by a preceding phrasal boundary. Figure 11 shows Absolute Latency for the [sC] clusters for each speaker at each boundary type.

Figure 11: Absolute Latency (∆Peaks) for [sC] Onsets and Codas as a function of Boundary Type

Byrd and Choi

17

In this figure, for Speakers K and N, the onsets tend to show two qualitative groupings (Utt vs. other), while their codas tend to show three groupings (w vs. IP vs Utt). That is, the onset cluster seems to show less overlap after an Utterance boundary but not to show much prosodic influence on its internal timing after the smaller boundaries. The coda [sC] clusters, on the other hand, tend to show more gradation across the boundary conditions. For Speaker J, there is no particular pattern of overlap as a function of boundary strength for his onsets, though his codas show two groupings (w/im vs. IP/Utt). This portrait suggests that indeed the [sC] ‘molecules’ may be less sensitive to prosodic perturbation than like clusters in coda. Lastly, we should consider any distinctions exhibited between [sC] and [kl] sequences. Certainly, for all speakers the onset [#kl] was relatively insensitive to prosodic effects on intergestural timing for all speakers compared to the heterosyllabic [k#l] sequence and the word-final [kl#]. This is seen in Figure 12 for each speaker.

Figure 12: Absolute Latency (∆Peaks) for [kl] as a function of Syllable Position for individual Speakers

However, this does not address whether the onset [#kl] behaved differently than the onset [#sC] clusters. There was a marginal Cluster x Boundary Type effect on ∆Peaks (F[6,12]=3.74, p=.0247) shown in Figure 13 (top) with subjects pooled, but there was a significant three-way interaction with Speaker (F[12, 641]=5.94, p=.0001). This data does not offer much support for the notion that the [#sC] cluster in particular has a status distinct from the [#kl] with regard to prosodic effects; though onsets do generally appear to behave somewhat specially. However, since there was a 3-way interaction with Speaker, it is worth also considering Figure 13 (bottom), which presents Speaker N’s onsets for each cluster in each prosodic condition. In the case of this speaker, there might

Byrd and Choi

18

be the possibility that intergestural timing in [#sC] is less sensitive to boundary type than that of this subject’s [#kl] clusters.

Figure 13: (top) Pooled Absolute Latency (∆Peaks) for [kl], [sk], and [sp] Onsets as a function of Boundary Type; (bottom) The same shown just for Speaker N

3.5. Summary The predictions of the prosodic-gesture model regarding individual consonant

lengthening in clusters are supported. Both consonants lengthen in CC clusters, but the consonant closest to the boundary lengthens most. There is an interaction of boundary type and syllable position on articulatory overlap. Clusters neighboring a prosodic boundary are sensitive to the strength of that boundary, and, of course, heterosyllabic sequences spanning a boundary show overlap to be strongly affected by the intervening boundary. Timing effects grade with respect to boundary type. Finally, [sC] onset clusters are less overlapped than [sC] coda clusters at all boundaries. There is mixed evidence that [sC] onset clusters (gestural molecules) are less sensitive to prosodic perturbation than [kl] onset clusters.

Byrd and Choi

19

4. Discussion Articulatory overlap in consonant sequences is sensitive to both the internal

phonological structuring of a word—namely, the composition and organization of gestural sequences—and to the placement of words in the larger informational structure of the utterance. On this Tenth Anniversary of the Laboratory Phonology conferences, it is fitting to consider the state of our understanding as to the relations between formal, hierarchical linguistic structure and the articulatory realization of that structure. The communication process requires vastly more than the production and recognition of individual words (despite the complexity of that apparently simple process); successful communication also requires that the information content created by syntactic structure and discourse factors be articulated and understood. In order to avoid missing the forest (of communication as social interaction) for the trees (of word production and perception), language scientists must work to address the dynamic interplay of these processes. In this paper, we have been interested in particular in experimentally investigating the relation between the phonological structure of words (specifically, the subsyllabic constituents of coda and onset composed of atomic phonological primitives) and the informational structure defining relations among words (specifically, phrasal structure).

We’d like to outline here one potentially interesting way in which to view the relation of phonological and informational structure. To do this, let us first be explicit about the information requisite for the specification of a word (aside from its semantic and syntactic properties). We suggest, as have many others in the Laboratory Phonology tradition, that a word’s phonological specification is inherently spatiotemporal—that is, that its compositional units are defined in space and time. In order to specify and speak a word, that word’s component units must be selected and associated with qualities defined by targets, activation, and coordination in an abstract coordinate system (or systems) appropriate for defining phonological tasks. Different approaches as to coordinate systems have been offered for defining those targets (e.g. constriction-based [e.g. Saltzman and Munhall 1989] and oro-sensory or perception-based [e.g. Guenther 1994, 1995]), and some progress has been made regarding the most cognitively satisfying way of understanding the internal temporal properties of these units. But the fundamental insight that we believe can be leveraged in understanding the dynamic interplay of phonology and prosody has been the conceptualization of linguistic representation as spatiotemporal in nature.

With regard to the organization of atomic phonological units of words into larger phonological structures, we suggest, in a way consonant with proposals of Browman and Goldstein, that other (larger) compositional units of words are instantiated by characteristic timing patterns among the atomic primitives (phonological gestures) (Browman and Goldstein 1995, 2000). Segments and syllables are molecules comprised of phonological gestures that recur through the lexicon with a characteristic and stable pattern of coordination (Saltzman and Munhall 1989; Saltzman and Byrd 2000; Nam and Saltzman 2003; Goldstein, Byrd, and Saltzman 2006; Nam, Goldstein, and Saltzman to appear). The laboratory phonology tradition of leveraging careful and quantitative analysis of speech data to inform an understanding of phonological structure has given us much information on the durational and coordination properties of gestures and gestural structures.

Byrd and Choi

20

More recently, however, this tradition has turned to a consideration of the nature of the informational or prosodic structure of utterances, particularly focus and phrasal structure, starting with the seminal works of Edwards, Beckman and Fletcher (1991), Beckman, Edwards and Fletcher (1992); Beckman and Edwards (1994); and Pierrehumbert and Beckman (1988). Again, the detailed examination of the phonetic properties of spoken language has provided insight into the nature of the informational structure that situates words in a potential discourse. We would suggest that while the phonological (i.e. lexical) structure of words is defined in terms of the spatiotemporal properties of phonological gestures (spatial [target], durational [internal timing] and coordination [intergestural timing]), the informational structure of an utterance is defined in terms of the local warping or modulation by prosodic gestures of the coordinates within which these spatial and temporal properties are defined (e.g., Byrd and Saltzman 2003; Byrd 2006). For example, phrasal prosodic structure has been proposed to be instantiated in articulation as a slowing of the central clock controlling the pace of unfolding of gestural activation trajectories in the neighborhood of phrasal boundaries (Byrd and Saltzman 2003; see also Edwards, Beckman and Fletcher 1991). Other types of informational structure such as prominence structure might warp spatial properties as well as temporal ones (Saltzman, Goldstein, Hot, Luzik and Nam to appear; for related proposals see also de Jong, Beckman and Edwards 1993, de Jong 1995).

An in-depth experimental and model-supported approach to such an account of prosodic structure, one that would rival our understanding of word-internal phonological structure, is an outstanding challenge for the next decade of Laboratory Phonology work. The field stands to benefit from revealing interactions between the spatiotemporal specification of phonological structure and the time-varying, flexible properties of the abstract space and time coordinate-systems within which speech is represented and carried out. It can be expected that the approach advanced by the Laboratory Phonology coalition of exploring these connections in an empirically rigorous and quantitative way will illuminate the interaction of phonological and prosodic structure in ways we are only now beginning to glimpse. References Beckman, Mary E. and Edwards, Jan. 1994 Articulatory evidence for differentiating stress categories. In Patricia A. Keating (ed.), Phonological Structure and Phonetic Form: Papers in Laboratory Phonology III, pp. 7-33. Cambridge: Cambridge University Press. Beckman, Mary E., Edwards, Jan, and Fletcher, Janet. 1992 Prosodic structure and tempo in a sonority model of articulatory dynamics. In: Gerard J. Docherty and D. Robert Ladd (eds.), Papers in Laboratory Phonology II: Segment, Gesture, Prosody, pp. 68-86. Cambridge: Cambridge University Press. Bombien, Lasse, Mooshammer, Christine, Hoole, Phil, Kühnert, Barbara and Schneeberg, Jennifer. 2006. An EPG study of initial /kl/ clusters in varying prosodic

Byrd and Choi

21

conditions in German. In Proceedings of the 7th International Seminar on Speech Production, Ubatuba, Brasil. Browman, Catherine P. 1995 Assimilation as gestural overlap: comments on Holst and Nolan. In: Bruce Connell and Amalia Arvaniti (eds.), Phonology and Phonetic Evidence: Papers in Laboratory Phonology IV. (pp. 334-342). Cambridge: Cambridge University Press. Browman, Catherine P. and Goldstein, Louis. 1986 Towards an articulatory phonology. Phonology Yearbook, 3, 219-252 Browman, Catherine P. and Goldstein, Louis. 1992 Articulatory phonology: An overview. Phonetica, 49 (3-4), 155-180. Browman, Catherine P. and Goldstein, Louis. 1995 Dynamics and articulatory phonology. In: Timothy van Gelder (ed.), Mind as Motion Explorations in the Dynamics of Cognition. (pp. 175-194). Cambridge, MA: MIT Press. Browman, Catherine P. and Goldstein, Louis. 2000 Competing constraints on intergestural coordination and self-organization of phonological structures. Les Cahiers de l'ICP, Bulletin de la Communication Parlée, 5, 25-34. Byrd, Dani. 1996 Influences on articulatory timing in consonant sequences. Journal of Phonetics, 24, 209-244. Byrd, Dani. 2000 Articulatory vowel lengthening and coordination at phrasal junctures. Phonetica, 57: 3-16. Byrd, Dani. 2006 Relating prosody and dynamic events: Commentary on the papers by Cho, Navas, and Smiljanić. Papers in Laboratory Phonology 8. Goldstein, Louis, Whalen, Douglas H., Best, Catherine (Eds.) Mouton de Gruyter, 549-561. Byrd, Dani, Kaun, Abigail, Narayanan, Shrikanth and Saltzman, Elliot. 2000 Phrasal signatures in articulation. In Michael B. Broe and Janet B. Pierrehumbert, (Eds.). Papers in Laboratory Phonology V. Cambridge:Cambridge University Press, 70 - 87. Byrd, Dani, Krivokapić, Jelena and Lee, Sungbok. 2006 How far, how long: On the temporal scope of phrase boundary effects. Journal of the Acoustical Society of America, 120, 1589-1599. Byrd, Dani, Lee, Sungbok, Riggs, Daylen and Adams, Jason. 2005 Interacting effects of syllable and phrase position on consonant articulation. Journal of the Acoustical Society of America 118, 3860-3873. Byrd, Dani and Saltzman, Elliot. 1998. Intragestural dynamics of multiple phrasal boundaries. Journal of Phonetics, 26:173-199.

Byrd and Choi

22

Byrd, Dani and Saltzman, Elliot. 2003 The elastic phrase: Dynamics of boundary-adjacent lengthening. Journal of Phonetics, 31,2, 149-180. Cho, Taehong. 2004 Prosodically-conditioned strengthening and vowel-to-vowel coarticulation in English. Journal of Phonetics, 32, 141-176. Cho, Taehong. 2006 Manifestation of prosodic structure in articulatory variation: Evidence from lip kinematics in English. In: Louis Goldstein, Doug H. Whalen, and Catherine T. Best (eds.), Laboratory Phonology 8: Varieties of Phonological Competence, Mouton de Gruyter, Berlin and New York, pp. 519-548. Cho, Taehong and Keating, Patricia. 2001 Articulatory and acoustic studies on domain-initial strengthening in Korean, Journal of Phonetics, 29, 155-190. Cho, Taehong and McQueen, James. 2005 Prosodic influences on consonant production in Dutch: Effects of prosodic boundaries, phrasal accent and lexical stress. Journal of Phonetics, 33, 121-157. Choi, John. D. 1992 Phonetic underspecication and target interpolation: An acoustic study of Marshallese vowel allophony. Ph. D. dissertation, UCLA, UCLA Working Papers in Phonetics, 82. De Jong, Kenneth. 1995 The supraglottal articulation of prominence in English: Linguistics stress as localized hyperarticulation. Journal of the Acoustical Society of America, 97(1), 491-504. De Jong, Kenneth, Beckman, Mary E., and Edwards, Jan. 1993 The interplay between prosodic structure and coarticulation. Language and Speech, 36, 197-212. Delattre, Pierre. 1971 Consonant gemination in four languages: An acoustic, perceptual, and radiographic study, Part I. International Review of Applied Linguistics in Language, IX/2. 31-52. Edwards, Jan and Beckman, Mary. E. 1988 Articulatory timing and the prosodic interpretation of syllable duration. Phonetica, 45, 156-174. Edwards, Jan, Beckman, Mary E. and Fletcher, Janet. 1991 Articulatory kinematics of final lengthening. Journal of the Acoustical Society of America, 89, 369-382. Fougeron, Cécile. 2001 Articulatory properties of initial segments in several prosodic constituents in French. Journal of Phonetics, 29, 109-135. Fougeron, Cécile and Keating, Patricia A. 1997 Articulatory strengthening at edges of prosodic domains", Journal of the Acoustical Society of America, 101, 3728-3740.

Byrd and Choi

23

Fukui, N. and Hirose, Hajime. 1983 Laryngeal adjustments in Danish voiceless obstruent production, Annual Bulletin Research Institute of Logopedics and Phoniatrics, 17, 61 – 71. Goldstein, Louis. 1990 On articulatory binding : comments on Kingston’s paper. In: John Kingston and Mary Beckman (eds.) Papers in Laboratory Phonology I: Between the Grammar and Physics of Ppeech, pp. 445 – 450. Goldstein, Louis. 2003 Emergence of discrete gestures. In Maria-Josep Solé, Daniel Recasens, and Joaquín Romero, (eds). Proceedings of the 15th International Congress of Phonetic Sciences. pp. 85-88. Goldstein, Louis, Byrd, Dani and Saltzman, Elliot. 2006 The role of vocal tract gestural action units in understanding the evolution of phonology. In: Michael Arbib (ed.) From Action to Language: The Mirror Neuron System. Cambridge University Press, pp. 215-249. Goldstein, Louis and Fowler, Carol. A. 2003 Articulatory Phonology: A phonology for public language use. In N. Schiller and A. Meyer (eds.) Phonetics and Phonology in Language Comprehension and Production. Berlin: Mouton de Gruyter. pp. 159-208. Goldstein, Louis, Pouplier, Marianne, Chen, Larissa, Saltzman, Elliot, and Byrd, Dani. in press Gestural action units slip in speech errors. Cognition. Guenther, Frank. H. 1994 A neural network model of speech acquisition and motor equivalent speech production. Biological Cybernetics, 72, 43-53. Guenther, Frank. H. 1995 Speech sound acquisition, coarticulation, and rate effects in a neural network model of speech production. Psychological Review, 102, 594-621. Hardcastle, William J. 1985 Some phonetic and syntactic constraints on lingual coarticulation during /kl/ sequences. Speech Communication, 4, 247-263. Hardcastle, William. J. and Roach, Peter. 1979 An instrumental investigation of coarticulation in stop consonant sequences. In: Harry Hollien and Patricia Hollien, (eds.) Current Issues in the Phonetic Sciences, pp. 531 – 540. Amsterdam: John Benjamins. Harrington, Jonathon, Butcher, Andy, Fletcher, Janet, Guzik, Karita, and Stephenson, Lisa. 2004 Production, perception, sound change and the lexicon: evidence from consonant clusters. Presented at the German-French Summerschool on "Cognitive and physical models of speech production, perception and perception-production interaction". September 2004, Lubmin, Germany (Baltic Sea).

Byrd and Choi

24

Holst, Tara and Nolan, Francis. 1995 The influence of syntactic structure on [s] and [S] assimilation. In: Bruce Connell and Amalia Arvanti (eds.) Phonology and Phonetic Evidence: Papers in Laboratory Phonology IV. (pp. 315-333) Cambridge: Cambridge University Press. Hoole, Phil, Fuchs, Susanne, Dahlmeier, Karl. 2003 Interarticulator timing in initial consonant clusters. Proc. Sixth Speech Production Seminar, Sydney, pp. 101-106. Jun, Sun-Ah. 1995 Asymmetrical prosodic effects on the laryngeal gesture in Korean. In: B.Connell and A. Arvanti (eds.), Phonology and Phonetic Evidence: Papers in LaboratoryPhonology IV. (pp. 235-253) Cambridge: Cambridge University Press. Keating, Patricia, Cho, Taehong, Fougeron, Cécile and Hsu, Chai S. 2003 Domain-initial strengthening in four languages. In: John Local, Richard Ogden, and Rosalind Temple (eds.) Papers in Laboratory Phonology 6, London: Cambridge University Press, pp. 143-161. Krakow, Rena A. 1993 Nonsegmental influences on velum movement patterns: Syllables, sentences, stress, and speaking rate. In: Huffman, Marie. A., and Krakow, Rena. A. (eds.) Nasals, Nasalization, and the Velum (Phonetics and Phonology V). New York: Academic Press, 87-116. Krakow, Rena A. 1999 Physiological organization of syllables: a review. Journal of Phonetics, 27, 23-54. Lee, Sungbok, Byrd, Dani and Krivokapić, Jelena. 2006 Functional data analysis of prosodic effects on articulatory timing. Journal of the Acoustical Society of America, 119, 1666-1671. Loevenbruck, Helene, Collins, M. J., Beckman, Mary E., Krishnamurthy, A. K., & Ahalt, S. C. 1999 An investigation of the temporal coordination of articulatory gestures in consonant clusters and sequences of consonants. In: Fujimura, Osamu, Joseph, B.D., & B. Palek, B. (eds.) Proceedings of LP’98, pp. 547-73. Prague: The Karolinum Press. Löfqvist, Anders and Yoshioka, Hirohide. 1980 Laryngeal activity in Swedish obstruent clusters, Journal of the Acoustical Society of America, 68, 792 – 801. Löfqvist, Anders and Yoshioka, Hirohide. 1981 Laryngeal activity in Icelandic obstruent production, Nordic Journal of Linguistics, 4, 1 – 18. McLean, Michael. 1973 Forward coarticulation of velar movement at marked junctural boundaries. J. Speech Hear. Res. 16: 286-296.

Byrd and Choi

25

Nam, Hosung. in press A competitive, coupled oscillator model of moraic structure: Split-gesture dynamics focusing on positional asymmetry. In: Cole, Jennifer and Hualde, Jose. (eds.) Papers in Laboratory Phonology 9. Nam, Hosung, Goldstein, Louis and Saltzman, Elliot. To appear Self-organization of syllable structure: A coupled oscillator model In: Pellegrino, T., Marisco, E., and Chitoran, I. (eds). Approaches to phonological complexity. Mouton. Nam, Hosung and Saltzman, Elliot. 2003 A competitive, coupled oscillator of syllable structure. In: Maria-Josep Solé, Daniel Recasens, and Joaquín Romero, (eds.) Proceedings of the XIIth International Congress of Phonetic Sciences, Barcelona, Oliveira, Leonardo, Yanagawa, Mariko, Chitoran, Ioana and Goldstein, Louis. 2004 Towards standard measures of articulatory timing. Journal of the Acoustical Society of America, 16, 2643-2644. Perkell, Joseph, Cohen, Marc, Svirsky, Mario, Matthies, Melanie, Garabieta, Iñaki and Jackson, Michel. 1992 Electromagnetic midsagittal articulometer (EMMA) systems for transducing speech articulatory movements. Journal of the Acoustical Society of America. 92:3078–3096. Petursson, Magnus. 1977 Timing of glottal events in the production of aspiration after [s] Journal of Phonetics, 5, 205 – 212. Pierrehumbert, Janet and Beckman, Mary. 1988 Japanese Tone Structure, Linguistic Inquiry Monograph 15, MIT Press, Cambridge. Pierrehumbert, Janet, Beckman, Mary E. and Ladd, D. Robert. 2001 Conceptual Foundations of Phonology as a Laboratory Science in Burton-Roberts, Noel, Carr, Phillip. and Docherty, Gerard. (eds) Phonological Knowledge, Oxford, UK: Oxford University Press, 273-304. Pierrehumbert, Janet and Talkin, David. 1992 Lenition of / h / and glottal stop. In: Gerard. J. Docherty and D. Robert Ladd (eds.) Papers in Laboratory Phonology II: Gesture, Segment, Prosody, pp. 90 – 117. Cambridge: Cambridge University Press Saltzman, Elliot. 1995 Dynamics and coordinate systems in skilled sensorimotor activity. In Port, R. and Van Gelder, T. (Eds.), Mind as Motion (pp. 149-173). Cambridge, MA: MIT Press. Saltzman, Elliot and Byrd, Dani. 2000 Task-dynamics of gestural timing: Phase windows and multifrequency rhythms. Human Movement Science, 19:499-526. Saltzman, Elliot, Goldstein, Louis, Holt, Kenneth, Kluzik, JoAnn, and Nam, Hosung. To appear Gait Wheels and Foot Cycles: A Parallel between the Dynamics of

Byrd and Choi

26

Locomotion and Speech. Poster submitted to the 11th Conference on Cognitive and Neural Systems, Boston, MA, May, 2007. Saltzman, Elliot, Löfqvist, Anders, Kay, Bruce, Kinsella-Shaw, Jeffrey, and Rubin, Phillip. 1998 Dynamics of intergestural timing: A perturbation study of lip-larynx coordination. Experimental Brain Research, 123: 412-424. Saltzman, Elliot, Löfqvist, Anders, and Mitra, Subhobrata. 2000 “Glue” and “clocks”—Intergestural cohesion and global timing. In: Michael Broe and Janet Pierrehumbert (eds.) Papers in Laboratory Phonology V, pp. 88-101. Cambridge: Cambridge University Press. Saltzman, Elliot L., and Munhall, Kevin. G. 1989 A dynamical approach to gestural patterning in speech production. Ecological Psychology, 1, 333-382. Sproat, Richard and Fujimura, Osamu. 1993 Allophonic variation in English /l/ and its implications for phonetic implementation, Journal of Phonetics, 21, pp. 291-311. Tabain, Marija. 2003 Effects of prosodic boundary on /aC/ sequences: articulatory results, Journal of the Acoustical Society of America. 113, 2834-2849. Winer, Benjamin J. 1971 Statistical principles in experimental design. New York: McGraw-Hill Book Company. Yoshioka, Hirohide, Löfqvist, Anders and Hirose, Hajime. 1981 Laryngeal adjustments in the production of consonant clusters and geminates in American English, Journal of the Acoustical Society of America, 70, 1615 – 1623. Notes 1 The movement data were sampled at 625 Hz after low-pass filtering at 200 Hz before voltage-to-distance conversion. After voltage-to-distance conversion (with a filter cutoff of 17 Hz), correction for head movement (using the nose and maxillary reference transducers), and rotation to the occlusal plane, the position signals were subject to 25 point smoothing. 2 The signals were differentiated and the resulting velocity signals smoothed with a 9th order Butterworth filter of cutoff frequency 15 Hz. 3 One way in which this qualitative evaluation of onset and coda means can be statistically supported is by a cell contrast of means comparison of the ∆Peaks measure. Pooling subjects, an analysis of the [sC] onsets and codas (following a two-factor ANOVA of syllable x boundary type with strong main effects and no interaction) shows the ∆Peaks cell comparisons of onset [sC] versus coda [sC] at every prosodic condition to be significant at the p<.0001 level.