the effect of stress and speech rate on vowel ... · vowel coarticulation in catalan...

Complimentary Author PDF: Not for Broad Dissemination

JSLHR

Research Article

aUniversitat A

Corresponden

Editor: Jody KAssociate Edi

Received JulyRevision receAccepted JunDOI: 10.1044

Journal of Spee

The Effect of Stress and Speech Rate onVowel Coarticulation in Catalan

Vowel–Consonant–Vowel Sequences
Daniel Recasensa
Purpose: The goal of this study was to ascertain the effect ofchanges in stress and speech rate on vowel coarticulation invowel–consonant–vowel sequences.Method: Data on second formant coarticulatory effects asa function of changing /i/ versus /a/ were collected forfive Catalan speakers’ productions of vowel–consonant–vowel sequences with the fixed vowels /i/ and /a/ andconsonants: the approximant /δ/, the alveolopalatal nasal/ɲ /, and /l/, which in the Catalan language differs in darknessdegree according to speaker.Results: In agreement with predictions formulated by thedegree-of-articulation-constraint model of coarticulation,the size of the vowel coarticulatory effects was inversely

utònoma de Barcelona, Spain

ce to Daniel Recasens: [email protected]

reimantor: Ewa Jacewicz

15, 2014ived February 2, 2015e 18, 2015/2015_JSLHR-S-14-0196

ch, Language, and Hearing Research • Vol. 58 • 1407–1424 • October 20

related to the degree of articulatory constraint for theconsonant, and the direction of those effects was mostlycarryover or anticipatory in vowel–consonant–vowel sequenceswith highly constrained consonants (/ɲ/, dark /l/) and morevariable whenever the intervocalic consonant was lessconstrained (/δ/, clear /l/). Stress and speech-rate variationshad an effect on overall vowel duration, second formantfrequency, and coarticulation size but not on the consonant-specific patterns of degree and direction of vowelcoarticulation.Conclusion: These results indicate that prosodicallyinduced coarticulatory changes conform to the basicprinciples of segmental coarticulatory organization.

Consecutive phonetic segments influence each otherat the articulatory and acoustic level in the speechchain. These coarticulation effects occur, for ex-

ample, between vowels and consonants in vowel–consonant–vowel (VCV) sequences. The present study deals with therole that the segmental composition of a VCV sequenceplays in the extent to which the first or second vowel (V1or V2) affects the other two phonetic segments. Two majorissues will be subject to investigation. First is the size of thevowel coarticulatory effects during the intervocalic conso-nant (V-to-C coarticulation) and the size and temporalextent of those effects during the transconsonantal vowel(V-to-V coarticulation). Thus, for example, coarticulationdata show that V2 = /a/ may cause the tongue dorsumto lower during the preceding consonant and V1 = /i/ inthe sequences /ipa/ and /ita/ and that these effects arelarger and last longer in the former sequence than in thelatter because /p/ involves no lingual activity, whereas /t/does. The second major research topic is the direction of

the vowel effects, namely, whether the V2-dependent antic-ipatory effects extending to the preceding consonant andtransconsonantal V1 are more or less prominent than the V1-dependent carryover effects extending forward to the follow-ing consonant and transconsonantal V2. Thus, for example,although /a/ may cause the tongue dorsum to lower duringpreceding /t/ and V1 = /i/ in the sequence /ita/ and during fol-lowing /t/ and V2 = /i/ in the sequence /ati/, the prominenceof the former (anticipatory) and latter (carryover) tongue-dorsum lowering effects may not be the same. The study ofthe direction of vowel coarticulation is of particular impor-tance because it provides some insight into the cognitive andperipheral mechanisms involved in segmental production inrunning speech (Farnetani & Recasens, 2010; Grosvald,2009; Guenther, 1995). In particular, anticipatory effects havebeen taken to reflect segmental planning and carryover effectsto depend mostly on inertia and the biomechanical charac-teristics of the vocal-tract articulators (Whalen, 1990).

The present study analyzed these research issuesusing second formant second formant (F2) data on coarti-culatory effects exerted by /i/ or /a/ (also referred to through-out as the changing vowel) in Catalan VCV sequences withthree consonants—the alveolopalatal nasal /ɲ/, the alveolarlateral /l/ (which may vary from strongly dark to less dark ormoderately clear according to speaker in present-day Catalan;

Disclosure: The author has declared that no competing interests existed at the timeof publication.

15 • Copyright © 2015 American Speech-Language-Hearing Association 1407


Simonet, 2010), and the dental approximant /δ/—and thetransconsonantal vowels /i/ and /a/ (also referred to as thefixed vowel). In principle, F2 coarticulation should be posi-tively correlated with tongue-dorsum coarticulation—thatis, the higher and more anterior the tongue body, the higherthe F2 frequency (Fant, 1960). Anticipatory effects from V2 =/i/ versus /a/ will be measured in the sequence pairs /aCi/–/aCa/and /iCa/–/iCi/, and carryover effects from V1 = /i/ versus /a/will be analyzed in the sequence pairs /iCa/–/aCa/ and /aCi/–/iCi/. These sequence pairings allow an evaluation of essen-tially how much changing /i/ affects the consonant and fixed/a/ and how much changing /a/ affects the consonant andfixed /i/ (e.g., in pairs such as /aCi/–/aCa/ and /iCi/–/iCa/,respectively) along the two coarticulatory dimensions. Anal-ysis results will be evaluated within the framework of thedegree-of-articulatory-constraint (DAC) model of coarticu-lation (see Recasens, Pallarès, & Fontdevila, 1997).

A novelty of the present investigation with respectto previous lingual coarticulation studies (for a summary,see Recasens, 1999) is elicitation of the extent to whichchanges in stress and speech rate affect the degree anddirection of vowel coarticulation in VCV sequences thatdiffer regarding segmental composition. It has been knownfor some time that mostly vowels, but also consonants,undergo articulatory undershoot and acoustic reduction inunstressed versus stressed position and in fast versus nor-mal (also in normal vs. overarticulated) speech (Agwuele,Sussman, & Lindblom, 2008; Farnetani, 1990; Gay, 1978;Kent & Netsell, 1971; Moon & Lindblom, 1994; but seeMok, 2011, for studies failing to report more vowel under-shoot as speech rate increases). The general prediction isthat, given that stressed vowels are produced with moreprominent articulatory gestures than unstressed vowels, theformer ought to exert more coarticulation on the latter inVCV sequences than vice versa. There is indeed supportingevidence in the literature that unstressed vowels are moresensitive than stressed vowels to V-to-V effects (Fowler,1981; Magen, 1997; Yun, 2007; but see Cho, 2004). Alsoin line with previous experimental evidence (Krull, 1989;Matthies, Perrier, Perkell, & Zandipour, 2001), phoneticsegments should be more reduced and more prone to coar-ticulate with other segments at faster versus slower speechrates and in less versus more formal speech.

The remainder of the introduction first describes howthe articulatory constraints for the consonants and vowelssubject to investigation in the present study affect the de-gree of vowel coarticulation and render vowel coarticula-tion more prominent at the anticipatory or carryover levelin VCV sequences. It also deals with the effect of changesin stress and speech rate on the degree and direction ofvowel coarticulation in the segmental conditions subject toanalysis. A set of testing hypotheses about these researchgoals is summarized at the end of the introduction.

Degree of CoarticulationThe DAC model predicts that the extent to which

vowels affect consonants in VCV sequences should vary

1408 Journal of Speech, Language, and Hearing Research • Vol. 58 •

inversely with the degree of tongue-dorsum constraint forthe intervocalic consonant and, more specifically, with theextent to which the consonantal primary articulator isinvolved in the formation of a closure or a constriction.The model was formulated by Recasens et al. (1997) afterseveral studies showing that the extent to which conso-nants resist coarticulation from adjacent vowels is posi-tively conditioned by closure size or degree of constrictionnarrowing.

Table 1 (top, second and third columns) indicates thepredicted degrees of coarticulatory resistance to the voweleffects for the consonants under investigation in the presentstudy and whether vowel effects should be exerted by /i/rather than by /a/ or vice versa. The consonants /ɲ/ anddark /l/ are expected to exhibit the highest degree of tongue-dorsum coarticulatory resistance and therefore to blockvowel coarticulatory effects in F2 frequency to a large ex-tent. This should be so because the two consonants imposestrict demands on the tongue dorsum: active tongue-bodyraising and fronting toward the alveolopalatal zone in thecase of /ɲ/ and some active tongue-predorsum loweringand postdorsum retraction, in addition to the formation ofan apicoalveolar contact, in the case of dark /l/ (Proctor,2009). Moreover, vowel coarticulation is expected to bemost prominent when the vowel is antagonistic to the con-sonant and thus makes the lingual constriction for thelatter especially hard to achieve, namely in the case of theeffects from /a/ on /ɲ/ (the tongue-dorsum position is lowand back for /a/ and high and front for /ɲ/) and from /i/on dark /l/ (the tongue dorsum occupies a high and frontposition for /i/ and a low and back position for dark /l/).

The approximant /δ/ and clear /l/, however, areless constrained because the tongue body is not activelycontrolled during their production and therefore oughtto allow more vowel coarticulation than /ɲ/ and dark /l/.In addition, some more predorsum lowering for the passageof airflow through the sides of the oral cavity for clear /l/versus /δ/ should cause the alveolar lateral to be somewhatmore constrained and thus more resistant to vowel coarti-culation than the dental approximant (on dental conso-nants involving possibly some front dorsum lowering aswell, see Dart, 1991). In this case, tongue-dorsum and F2raising effects from /i/ are expected to be more prominentthan tongue-dorsum and F2 lowering effects from /a/ be-cause, as pointed out in the following, the former vowelexhibits a higher degree of tongue-body constraint than thelatter.

The prominence of the vowel-dependent effects inVCV sequences ought to vary as a function of the transcon-sonantal vowel as well (see Table 1, bottom, second andthird columns). As reported for other languages (Cho, 2004;Mok, 2011), these effects should be smaller when the fixedvowel is /i/ (most resistant) than when it is /a/ (least resis-tant). The rationale for this outcome is that, as predictedfor /ɲ/, tongue-dorsum raising and fronting ought to render/i/ more constrained than /a/. In addition, effects on vowelsought to be more prominent if they are exerted by highlyconstrained consonants than by consonants that are less

1407–1424 • October 2015

Table 1. Predicted vowel-consonant-vowel coarticulation scenarios for different conditions of intervocalic consonant and fixed vowel.

Segmental conditionCoarticulatoryresistance

Coarticulatory effects Coarticulatory direction

Vowel Consonant Consonant-to-vowel Vowel-to-consonant/vowel

Intervocalic consonant/ / Minimal /i/ > /a/ — Carr > antdark /l/ Maximal /i/ > /a/ Ant > carr Ant > carrclear /l/ Intermediate /i/ > /a/ — Ant ≥ carr/ɲ/ Maximal /a/ > /i/ Carr > ant Carr > ant

Fixed vowel/i/ Maximal dark /l/ > / /, clear /l/, and /ɲ//a/ Minimal /ɲ/ > / / and clear and dark /l/

Note. Carr = carryover; ant = anticipation; — = unavailable data.


constrained, in particular if the vowel and consonant donot share the same or a similar articulatory configuration.Consequently, effects on /i/ ought to be exerted mostly bydark /l/ and those on /a/ by /ɲ/, whereas clear /l/ and /δ/ oughtto cause less C-to-V coarticulation to occur.

In sum, I predict that vowel coarticulation ought tovary inversely with the degree of tongue-body constraintfor the contextual phonetic segments as follows: for theintervocalic consonant in the progression /δ/ > clear /l/ >dark /l/ and /ɲ/ and for the transconsonantal vowel in theprogression /a/ > /i/.

Coarticulatory DirectionThe DAC model of coarticulation also makes spe-

cific predictions about whether vowel effects in VCV se-quences should be anticipatory rather than carryoveror vice versa (Recasens, 2007; Recasens et al., 1997). Ac-cording to the model, the extent to which vowel effectsoccur along one direction rather than the other depends onthe directionality pattern of the C-to-V effects. Thus, forexample, if the consonant exerts more carryover coarticulationon V2 than anticipation on V1, one would expect the voweleffects on the consonant and the transconsonantal vowelto be more prominent at the carryover than at the anticipa-tory level as well. This should be so because, in this partic-ular case, there will be more blocking of the V2-dependentanticipatory effects (by C-to-V2 carryover) than of theV1-dependent carryover effects (by C-to-V1 anticipation).As a consequence, one must know whether effects fromconsonants on vowels favor the anticipatory or the carryoverdirection in order to predict which one of the two directionsof vowel coarticulation will be favored in a given VCVsequence. As summarized in Table 1 (top, fourth and fifthcolumns), the following predictions may be made regardingthe directionality of the consonant and vowel coarticula-tory effects in VCV sequences with (a) highly constrained/ɲ/ and dark /l/ and (b) less constrained /δ/ and clear /l/:

1. Data on C-to-V coarticulation reveal that /ɲ/ shouldexert carryover rather than anticipatory effects onvowels (mostly on antagonistic /a/) in line with thefact that the tongue-dorsum activity for the consonant

is higher at closure offset and during V2 than at clo-sure onset and during V1 (Recasens, 2007). As forthe vowel effects, tongue-dorsum and F2 loweringexerted primarily by /a/ ought to be more prominentat the carryover than at the anticipatory level, giventhat the (more prominent) C-to-V carryover effectsshould interfere with vowel anticipation to a largerextent than the (less prominent) C-to-V anticipatoryeffects interfere with vowel carryover.However, dark /l/ is expected to exert more C-to-Vanticipation than C-to-V carryover (mostly onantagonistic /i/) because the lowering and retractionof the tongue body occurs in anticipation of thetongue front raising gesture for this consonant (Sproat& Fujimura, 1993). Regarding vowel coarticulation,the tongue-dorsum and F2 raising effects exertedmostly by /i/ ought to operate at the anticipatory ratherthan the carryover level because now the C-to-Vcoarticulatory effects are mostly anticipatory andtherefore should interfere with V1 rather than with V2.

2. It is hard to make predictions about the direction ofthe vowel coarticulatory effects on the basis of thedirection of the consonant-dependent effects in thecase of VCV sequences with /δ/ and clear /l/, giventhat the production of the two dentoalveolars involvesno active tongue-dorsum motion associated with theformation of a closure or constriction. In principle,and in parallel to effects exerted by other alveolopa-latal segments such as /ɲ/, effects from /i/ rather thanfrom /a/ should be greater at the carryover than atthe anticipatory level. Moreover, these vowel effectsought to take place in VCV sequences with /δ/ ratherthan in those with clear /l/ because the need to lowerthe tongue predorsum somewhat in order to formlateral channels for the passage of airflow during thepreceding vowel should already render the consonantand vowel anticipatory components especially salient(similarly to dark /l/).

In sum, VCV sequences ought to show less vowelcoarticulation for /ɲ/ and dark /l/ than for /δ/ and clear /l/in view of differences in degree of tongue-body involvementduring consonant articulation. There ought to be clear-cut

Recasens: Effect of Stress and Rate on Coarticulation 1409


differences in vowel coarticulatory direction for the highlyconstrained consonants /ɲ/ and dark /l/—that is, vowel carry-over should exceed vowel anticipation in VCV sequenceswith the nasal consonant, and the reverse should occur inVCV sequences with the alveolar lateral. Regarding VCVsequences with the less constrained consonants /δ/ and clear/l/, vowel carryover effects as a function of /i/ are expectedto prevail over vowel anticipatory effects for the formerconsonant but not for the latter.

Stress and Speech RateIn the present investigation, the predictions about

degree and direction of vowel coarticulation in VCV se-quences formulated by the DAC model were evaluatedin Catalan VCV sequences bearing stress on V1 (′VCV)and on V2 (V′CV) and produced at normal and fast speak-ing rates.

Regarding coarticulation degree, vowel coarticula-tion effects induced by the presence versus absence of stressand by different speech rates were expected to be smallerfor highly constrained consonants than for less constrainedones, the rationale being that the former ought to bemore resistant to variations in closure or constrictiondegree than the latter. Thus, in comparison to /δ/ and to alarge extent clear /l/, consonants subject to stricter articu-latory requirements such as /ɲ/ and dark /l/ should bemore resistant to different degrees of coarticulation thatmay arise when stress position and speech rate are subjectto change.

As for the effect of stress on coarticulatory direction,American English data for VCV sequences with /b/, /d/,/g/, and several vowels indicate that V-to-V effects fromstressed vowels onto unstressed vowels in VCV sequencesoccur at the carryover rather than the anticipatory level(Agwuele, 2004; Hertrich & Ackermann, 1995). Within theDAC model framework, patterns of vowel coarticulatorydirection in VCV sequences with highly constrained conso-nants were expected to be enhanced when stress falls onthe vowel (V1 or V2) that triggers more coarticulation. Ifso, there ought to be an increase in vowel carryover coarti-culation relative to vowel anticipation in VCV sequenceswith /ɲ/ when stress falls on V1 (/′VɲV/) compared withwhen it falls on the CV2 string (/V′ɲV/), and in vowel an-ticipation relative to vowel carryover in VCV sequenceswith dark /l/ when stress falls on CV2 (/V′lV/) comparedwith when it falls on V1 (/′VlV/). As for the more uncon-strained dentoalveolars /δ/ and clear /l/, changes in stressposition could yield more vowel carryover in /′VδV/ versus/V′δV/ sequences and more vowel anticipation or no pre-ferred direction in /V′lV/ versus /′VlV/ sequences. Regardingthe role of speech rate, the articulatory gesture for vowelsand consonants could be anticipated to a larger extent asspeakers speak faster, which should result in an increase ofanticipation relative to carryover for both the consonant-and the vowel-dependent effects in VCV sequences in gen-eral. This should be so because articulatory gestures forspeech segments are typically anticipated in time, and this


anticipatory activity is expected to become more prominentas undershoot causes the mechanico-inertial characteristicsof the articulators to diminish at faster speech rates.

In sum, if coarticulatory direction is affected bychanges in stress and speaking rate, the presence of stressin the crucial changing vowel ought to cause an increase inthe degree of vowel anticipation vis-à-vis vowel carryoverin VCV sequences with dark /l/ and the reverse for /ɲ/ andan increase in the preferred vowel coarticulatory directionin VCV sequences with /δ/ and clear /l/ (presumably invowel carryover for the former consonant but not for thelatter). However, there should be more consonant andvowel anticipation at faster versus slower speaking ratesas a general rule.

Summary of Testing HypothesesThe present study predicts that the degree and direc-

tion of coarticulation effects as a function of /i/ versus /a/in VCV sequences ought to vary with the degree of tongue-body constraint for the intervocalic consonant and thefixed vowel. Vowel coarticulation should decrease inthe progression /δ/ > /l/ > /ɲ/ (and for dark vs. clearervarieties of /l/) and for fixed /a/ more than for fixed /i/.Coarticulatory direction is expected to favor vowel carry-over in /VɲV/ sequences and vowel anticipation in /VlV/sequences with dark /l/ and perhaps carryover effects from/i/ in /VδV/ sequences, whereas effects for /VlV/ sequenceswith clear /l/ could favor vowel anticipation or neitherdirection. There should also be more vowel coarticulationfrom stressed versus unstressed vowels and in VCV sequencesproduced at fast versus normal speaking rates, and thisincrease ought to occur mostly when the intervocalic con-sonant is relatively unconstrained. Moreover, the preferredvowel coarticulatory direction could be enhanced whenstress falls on the crucial changing vowel (e.g., on V1 in/VɲV/ sequences), and vowel anticipation ought to gainweight at fast versus normal speech rates irrespective of thesegmental conditions involved.

MethodSpeech Materials

Acoustic recordings were made of 10 tokens of thesymmetrical and asymmetrical sequences /aCa/, /iCi/, /iCa/,and /aCi/, with the consonants /δ/, /ɲ/, and /l/ and thetwo stress conditions /′VCV/ and /V′CV/, where stress fallseither on the first syllable and thus on V1 or else on thesecond syllable and thus on the CV2 string. These VCV se-quences allow the study of vowel coarticulation over timein all symmetrical and asymmetrical combinations withchanging and fixed /i/ and /a/.

VCV sequences were embedded in an [əβ__βə] stringand thus were flanked by a voiced labial consonant (whichhappens to be an approximant if occurring intervocalicallyin Catalan), in order to maximize possible coarticulatoryeffects in lingual activity and thus in F2 frequency thatmay take place during the transconsonantal vowel. The

1407–1424 • October 2015


nonsense [əβVCVβə] sequences were uttered in isolation inorder to make sure that differences in vowel duration andintensity between the stressed and unstressed syllables inthe /′VCV/ and /V′CV/ conditions would not be canceled,at least in part, in the case that one or more words withspecific stress levels were placed before and after the targetnonsense word.

Speakers and Recording ProcedureFive 40- to 60-year-old Catalan speakers (DR, AL,

DP, CE, and LU) read the speech material 10 times attheir normal speech rate and subsequently another 10 timesat a fast speech rate. All five speakers are university orhigh school teachers and speak Catalan on a daily basis.Overall, 2,400 sequence tokens were recorded and sub-mitted to analysis (3 Cs × 4 VCV combinations × 2 stressconditions × 2 speech rates × 10 tokens × 5 speakers). Thesmall number of speakers renders this study a preliminaryinvestigation about the effect of stress and speech rate onvowel coarticulation in VCV sequences, though this limita-tion is offset by the large number of data points that weresubject to measurement.

Audio recordings were carried out at a 22050-Hzsampling rate in a soundproof room with a Rode NT1-Amicrophone and the CSL4500 (Computerized Speech Lab)system by Kay Pentax (Montvale, NJ). The acoustic signalwas downsampled to 11025 Hz and analyzed with the CSLacoustic analysis software.

Acoustic AnalysisBecause all the consonants (/δ/, /ɲ/, and /l/) under

study have formant structure, VC and CV segmentalboundaries were determined visually on spectrographic dis-plays on the basis of a change in formant intensity level—that is, formants are more intense for vowels than forapproximants, laterals, and nasals—and on spectral dis-continuity whenever available, as in the case of the /VɲV/sequences. For each speech file, F2 frequency values werecollected at C midpoint and at several temporal points dur-ing the vowel segments separately for V1 and V2: at vowelonset and offset; at vowel midpoint; at the midpoint be-tween vowel onset and vowel midpoint; and at the midpointbetween vowel midpoint and vowel offset. Frequency mea-surements were taken visually at these 11 time points onspectrographic displays by placing a cursor in the middleof F2. Figure 1 shows a spectrogram in which these pointsare coded with numbers 1 through 11. Overall, 26,400 mea-surements were taken (2,400 sequence tokens × 11 timepoints). This analysis method was preferred to the linearpredictive coding peak-picking method, which, accordingto results from some pilot work, often yielded unsatisfac-tory frequency values whenever F2 was not clearly visibleor exhibited a rapid spectral change. A reliability checkwas not carried out on the present data sample but wasperformed on F2 vowel measurements taken by the sameexperimenter in previous studies (Recasens & Espinosa,

2009). Some notion of the measurement precision can begathered from the F2 data at the consonant midpointreported in Table 2, where coefficients of variation do notexceed 5% for about 90% of the F2 values and range be-tween 5% and 10% for the remaining 10%. Special carewas taken to place the cursor at the center of the formantin relatively fast vowel transitions and whenever F2 washardly visible.

The sizes of the V2-dependent anticipatory effectsin the sequence pairs /aCa/–/aCi/ and /iCi/–/iCa/ and ofthe V1-dependent carryover effects in the sequence pairs/aCa/–/iCa/ and /iCi/–/aCi/ were obtained by subtractingF2 for changing /a/ from F2 for changing /i/ at seven pointsin time: at the edge of the changing vowel (i.e., at V2 onsetfor vowel anticipation and at V1 offset for vowel carry-over), at C midpoint, and at five equidistant points duringthe transconsonantal vowel (see Figure 1). The reason whyV2- and V1-dependent effects were measured at V2 onsetand V1 offset, respectively, is to be found in the fact thatthe F2 frequency at those temporal points is strongly de-pendent not only on the vowel but on the intervocalic con-sonant as well. The sign of the difference had to be positivebecause F2 is at about 2000 Hz for /i/ and at about 1200–1500 Hz for /a/; as in previous studies (Recasens, 2002;Recasens et al., 1997), negative values that occurred occa-sionally when F2 for /a/ exceeded F2 for /i/ during the con-sonant and/or the transconsonantal vowel were convertedinto 0 for statistical analysis.

Statistical AnalysisF2 differences between /i/ and /a/ were tested statisti-

cally with a repeated measures design using linear mixedmodels and five independent factors: Fixed Vowel (/a/, /i/),Consonant (/δ/, /ɲ/, /l/), Stress Condition for the changingvowel (stressed, unstressed), Coarticulatory Direction(anticipatory, carryover), and Speech Rate (normal, fast).Separate tests were run on the F2 differences betweenchanging /i/ and /a/ at seven equivalent temporal points forevaluating anticipatory vowel coarticulation (e.g., the dif-ference between V2 = /i/ and V2 = /a/ at V1 offset at Point 5)and carryover vowel coarticulation (e.g., the differencebetween V1 = /i/ and V1 = /a/ at V2 onset at Point 7). Thetemporal comparisons subject to statistical testing were asfollows (see Figure 1):

1. Between V2 anticipatory effects at V2 onset (Point 7)and V1 carryover effects at V1 offset (Point 5)

2. Between V2 anticipatory effects and V1 carryovereffects at C midpoint (Point 6)

3. Between V2 anticipatory effects at V1 offset (Point 5)and V1 carryover effects at V2 onset (Point 7)

4. Between V2 anticipatory effects at Point 4 and V1carryover effects at Point 8

5. Between V2 anticipatory effects at V1 midpoint(Point 3) and V1 carryover effects at V2 midpoint(Point 9)


Figure 1. Spectrographic representation of the sequence /ala/, with temporal points at which the F2 measurements weretaken. V1 = first vowel position; C = consonant; V2 = second vowel position.


6. Between V2 anticipatory effects at Point 2 and V1carryover effects at Point 10

7. Between V2 anticipatory effects at V1 onset (Point 1)and V1 carryover effects at V2 offset (Point 11)

The statistical model included all main effects andtwo- and three-way interactions. Bonferroni post hoc testswere performed on those significant variables that hadmore than two levels, and simple effects tests were carriedout on the significant factor interactions. A low thresholdfor significance was chosen in order to account for theapplication of multiple statistical tests on the temporalcoarticulation data. Because seven comparisons were car-ried out, the Bonferroni correction requires that the p valuebe .007. Although significant effects are reported forp < .05, only those differences approaching the p < .001level of significance are considered to be robust. Partialeta-squared (ηp

2) values and thus a measure of the strengthof the main effects and significant interactions will alsobe provided.

Additional repeated measures analyses were run onthe data for the symmetrical sequences /iCi/ and /aCa/ inorder to uncover differences in segmental duration andin F2 frequency triggered by changes in stress and speechrate. Two tests were performed on vowel duration and onthe F2 frequency values at the vowel midpoint, with Stress(stressed, unstressed), Position of the stressed vowel (V1,V2), Vowel Quality (/i/, /a/), and Speaking Rate (normal,fast) as factors. Another analysis of variance was carriedout on sentence duration, with Speaking Rate as the indepen-dent variable.

ResultsCoarticulatory Resistance

A first goal of this study was to analyze the extentto which /δ/, /ɲ/, and /l/ exhibit the predicted degrees ofcoarticulatory resistance to vowel effects. In order to inves-tigate this issue, Figure 2 plots the F2 frequency values at


consonant midpoint for the three consonants in the sym-metrical sequences /iCi/ and /aCa/ across stress and rateconditions, speakers, and tokens. According to the figure(see also Table 2) and in agreement with the initial pre-dictions, coarticulatory resistance is minimal for /δ/, withF2 occurring at 1900–2000 Hz in the /i/ context and at1200–1350 in the /a/ context. Except for speaker CE, thealveolopalatal /ɲ/, whose F2 frequency is about 2000 Hzor higher, is slightly more resistant than /l/—that is, theF2 difference between /i/ and /a/ is somewhat smaller forthe nasal than for the lateral.

Differences in vowel coarticulation between /l/ and/ɲ/ may be related to variations in /l/ darkness degreeamong speakers. Darkness degree may be evaluated in thesequence /ili/, where the lingual gestures for the vowel andthe consonant are more antagonistic if /l/ is dark than ifit is clear. Thus, although /i/ is articulated with a high fronttongue-body position, dark /l/—and much less so clear /l/—is produced with a low predorsal and back postdorsalconfiguration. These articulatory characteristics ought tocause F2 to be about 2000 Hz for /i/, 1000–1300 Hz fordark /l/, and 1400–2000 Hz for clear /l/ and cause dark /l/to be more resistant than clear /l/ to coarticulatory effectsin tongue-body position and F2 frequency associatedwith the high front vowel. Data for /l/ in the sequence /ili/plotted in Figure 2 show that darkness degree is higherfor speakers DR, AL, and DP than for CE and LU (F2for /l/ is at about 1250–1350 Hz for the three former subjectsand about 1550 Hz for the two latter ones). Moreover, theexpected positive relationship between coarticulatory resis-tance and darkness degree—that is, darker variants of /l/should be more resistant to F2 differences from /i/ to /a/ thanclearer variants of the consonant—holds only to someextent. As expected, coarticulatory resistance is higher forspeakers DR, AL, and DP than for speaker LU (the F2difference between /ili/ and /ala/ amounts to 200–350 Hz forthe three former subjects and to 433 Hz for the latter). Indisagreement with the expected positive relationship betweendarkness degree and coarticulatory resistance, however, /l/

1407–1424 • October 2015

Table 2. Mean F2 values (standard deviations) at consonant midpoint in symmetrical /VðV/, /VɲV/, and /VlV/ sequences.

Speechrate

Stressedvowel

position Vowel

Speaker DR Speaker AL Speaker DP Speaker CE Speaker LU

/ð/ /ɲ/ /l/ /ð/ /ɲ/ /l/ /ð/ /ɲ/ /l/ /ð/ /ɲ/ /l/ /ð/ /ɲ/ /l/

Normal V1 /i/ 2011.7(60.37)

2192.4(15.23)

1271.2(51.10)

1812.1(202.74)a

2310.8(30.62)

1250.3(35.98)

1973.1(31.70)a

2039.3(17.37)

1299.6(40.57)

1917.3(65.05)

2173.4(31.69)

1568.9(36.04)

1948.3(72.52)b

2074.6(83.28)

1534.5(32.05)

/a/ 1284.9(32.46)

2034.6(83.32)

1066.6(12.80)

1325.7(62.86)

2162.8(60.16)b

1070.1(26.56)

1254.2(32.11)

1947.1(28.15)

1007.9(25.11)

1335.4(30.59)

1834.5(93.56)

1355.3(53.16)

1269.0(107.86)

1969.0(66.45)

1101.2(59.43)

V2 /i/ 1956.5(66.57)b

2222.2(34.78)

1324.2(101.57)

1797.8(96.94)a

2327.2(28.99)

1252.0(45.97)

1913.4(92.17)a

1980.4(32.51)

1278.7(46.39)

1941.7(66.24)

2120.0(44.79)

1542.2(25.31)

1997.6(117.35)b

2171.4(50.61)

1524.9(43.25)

/a/ 1286.3(25.97)

1963.5(27.75)

1074.9(28.44)

1294.5(40.65)

2030.3(142.48)d

1038.0(37.40)

1198.2(17.04)

1897.5(57.86)

976.0(35.55)

1391.3(51.71)

1828.0(90.63)

1368.8(58.85)

1186.8(64.42)

1993.5(50.09)

1102.4(45.96)

Fast V1 /i/ 2012.2(61.80)a

2207.9(27.24)

1316.1(119.61)

1915.6(163.81)a

2321.8(44.42)

1276.7(44.85)

1954.7(84.89)d

2076.5(45.10)

1362.2(35.20)

1921.0(49.39)

2170.6(21.93)

1554.8(43.17)

1875.9(122.26)

2109.2(41.74)

1560.2(49.07)

/a/ 1320.5(34.03)

2049.1(69.15)a

1069.8(20.49)

1333.4(49.12)

2129.2(54.92)a

1074.2(38.21)

1307.4(35.93)

1988.0(42.76)

1000.1(17.90)

1312.8(74.16)

1821.9(77.72)

1374.6(51.35)

1157.5(37.12)

1707.2(74.27)a

1075.0(52.88)

V2 /i/ 2066.9(114.16)c

2227.3(47.72)

1342.9(104.64)

2044.8(102.20)a

2300.1(30.13)

1254.8(46.73)

1943.6(71.77)

2005.3(29.95)

1350.4(46.99)

1962.3(47.39)

2088.8(55.20)

1549.0(24.20)

1900.2(96.17)a

2140.7(27.79)

1492.1(90.17)

/a/ 1286.3(30.90)

1960.6(103.31)b

1072.5(16.01)

1343.2(36.33)

2171.2(167.24)e

1061.9(49.12)

1232.3(39.51)

1881.2(68.84)

1007.9(32.15)

1379.7(57.25)

1828.6(64.39)

1387.8(66.87)

1211.9(38.16)

1812.7(57.09)

1100.0(39.79)

Note. Number of repetitions for each sequence is 10 unless otherwise noted. V1 = first vowel position; V2 = second vowel position.aNumber of repetitions = 9. bNumber of repetitions = 8. cNumber of repetitions = 7. dNumber of repetitions = 6. eNumber of repetitions = 5.

Recasens:E

ffectof

Stress

andRate

onCoarticulation

1413


Figure 2. Mean F2 frequency values at the midpoint of /δ/, /ɲ/, and/l/ in symmetrical vowel–consonant–vowel sequences with /i/ and/a / across stress and rate conditions, plotted as a function ofconsonant, vowel context, and speaker.


exhibits comparable degrees of vowel coarticulation, oreven less, in the case of speaker CE (182 Hz) than of speakersDR, AL, and DP (200–350 Hz). As the figure shows, thisunexpected result appears to be related to the presence ofa higher F2 for /l/ next to /a/ for this subject than for theother four speakers, presumably meaning that the tonguebody is not pulled down much by the low vowel during theproduction of the alveolar lateral.

As a complement to these data on consonant coarti-culatory resistance, Figure 3 presents the F2 trajectoriesfor the symmetrical sequences /iCi/ and /aCa/ with all threeconsonants and stress falling on V1 and on V2 (normalspeaking rate only). Less V-to-C coarticulation from /i/ on/l/ than on /δ/ at the midpoint of the consonant is consistentwith the alveolar lateral exhibiting a much lower F2 targetthan the dental approximant in the sequence /iCi/; more-over, as pointed out already, a shorter-than-expected F2 dis-tance between /ili/ and /ala/ for speaker CE is related tothe entire F2 trajectory for /ala/ occurring at a higher fre-quency for this speaker than for the other four. As for theVCV sequences with /ɲ/, the alveolopalatal nasal exhibitsan F2 peak when flanked by /a/ (/aɲa/), which comes closeto the F2 frequency for the consonant in the context of/i/ (/iɲi/); the peak in question may occur at about closuremidpoint (for speakers AL and DP) or toward closureoffset (for speakers DR, CE, and LU), depending presum-ably on the point in time when maximal linguopalatal con-tact takes place. An aspect worth noting is the extremelyhigh F2 frequency for the consonant and the two vowelsin speaker AL’s productions of the sequence /iɲi/, whichsuggests that articulatory overshoot—that is, a consider-able F2 increase whenever the VCV sequence is composedof several alveolopalatal segments in succession—is atwork.

Regarding the effect of consonants on vowels, Fig-ure 3 also shows that F2 frequency changes associated withthe consonant have a relatively restricted temporal spanin the sequence /ili/ (i.e., F2 lowering starts at about V1


midpoint and F2 raising ends at V2 midpoint) and extendover a longer period of time in the sequence /aɲa/ (i.e., F2rising starts at V1 onset and F2 lowering ends at V2 off-set). Moreover, regarding the latter VCV sequence, the F2lowering trajectory during V2 = /a/ often proceeds moreslowly than the F2 rising trajectory during V1 = /a/, whichresults in higher F2 frequency values at comparable tem-poral points during V2 versus V1 and thus more prominentC-to-V carryover than anticipatory effects (compare, e.g.,the F2 frequency value at V2 midpoint with that at V1midpoint in the sequence /aɲa/ for all speakers).

Differences in the temporal extent of vowel coar-ticulation among VCV sequences with /δ/, /ɲ/, and /l/ areconsistent with the scenario at the midpoint of the conso-nant just described. As revealed by the statistical resultsfor the main consonant effect presented in Table 3 (leftcolumn), differences in vowel coarticulation among thethree intervocalic consonants were found to occur at con-sonant midpoint (Temporal Comparison 2) and extendeduntil the midpoint of the fixed transconsonantal vowel(Temporal Comparison 5) or even further. Moreover, theF2 difference between changing /i/ and /a/ over the timedomain turned out to be greater when the intervocalicconsonant was /δ/ than when it was /l/ or /ɲ/ and gener-ally for the lateral than for the nasal. This consonant-specific behavior may be inferred from the height of thebars in Figures 4–6 displaying the size of the vowel ef-fects in VCV sequences with each of the three consonantsat all temporal points.

As for differences in coarticulatory resistance be-tween the two fixed vowels, Table 3 (middle and rightcolumns) also reports a main fixed vowel effect and a sig-nificant Consonant × Fixed Vowel interaction lasting untilabout the vowel midpoint (Temporal Comparison 5). Asrevealed by Figures 4–6, this finding is related to the exis-tence of more vowel coarticulation during fixed /a/ in VCVsequences with /δ/ and /ɲ/ and the reverse (i.e., more co-articulation during fixed /i/ than during fixed /a/) in VCVsequences with /l/. Indeed, for the temporal comparisonsyielding a significant effect, bars are higher for the fixed /a/condition than for the fixed /i/ condition in VCV sequenceswith /δ/ (see Figure 4) and to a lesser extent in VCV se-quences with /ɲ/ as well (see Figure 5) and for fixed /i/than for fixed /a/ in VCV sequences with /l/ (see Figure 6).These results are only in partial agreement with the initialhypothesis that /a/ should be more sensitive to coarticulationeffects than /i/ and are interpreted in terms of gestural af-finity and antagonism for consonants and vowels in theDiscussion section.

Coarticulatory DirectionAs shown in Table 4, differences in coarticulatory

direction—that is, whether a higher F2 for changing /i/ ver-sus /a/ extends more toward the right (carryover) or towardthe left (anticipatory)—may also reach significance duringthe consonant and the transconsonantal vowel. In particular,Consonant and Fixed Vowel interact significantly with

1407–1424 • October 2015

Figure 3. F2 trajectories for symmetrical /′VCV/ sequences (solid lines) and /V′CV/ sequences (dashed lines) with /δ/, /ɲ/, and /l/ and /i/ and /a/, forall speakers. Data correspond to the normal speaking-rate condition. V1 = first vowel position; C = consonant; V2 = second vowel position.



Table 3. Significant Consonant and Fixed Vowel effects for F2 differences between changing /i/ and /a/ over time.

Temporal Comparison

Consonant(df = 2, 199)

Fixed Vowel(df = 1, 199)

Consonant × Fixed Vowel(df = 2, 199)

F ηp2 F ηp

2 F ηp2

1. V2 on/V1 off 137.4*** .57 14.2*** .07 15.9*** .142. C midpoint 193.8*** .66 84.5*** .29 7.2** .073. V1 off/V2 on 86.3*** .46 25.5*** .11 40.4*** .284. 77.3*** .43 52.9*** .20 61.8*** .375. V1 midpoint/V2 midpoint 15.9*** .13 16.2*** .07 10.7*** .096.7. V1 on/V2 off 4.6** .04 6.4* .03

Note. df = degrees of freedom; V1 = first vowel position; V2 = second vowel position; C = consonant.

*p < .05. **p < .01. ***p < .001.


Direction in ways that are interpreted next with referenceto the coarticulation data plotted in Figures 4–6.

As shown in Figure 4, vowel effects in VCV sequenceswith /δ/ are generally larger at the carryover level in thefixed /a/ context (effects from /i/ on /δ/ and on fixed /a/)and favor no specific direction in the fixed /i/ condition (ef-fects from /a/ on /δ/ and on fixed /i/). Indeed, the four rightbars in each graph are higher than the four left bars formost temporal comparisons when the fixed vowel is /a/,whereas there is no clear prevalence of the four right barsover the four left bars or vice versa in the fixed /i/ context.As shown by the F2 trajectories for /iδa/ and /aδi/ for arepresentative group of speakers in Figure 7, the presenceof more carryover than anticipatory effects from /i/ is con-sistent with F2 being located at a higher frequency at Coffset in the sequence /iδa/ than at C onset in the sequence/aδi/, and also during the first portion of V2 in the formersequence than during the last portion of V1 in the latter,for all three speakers. There may be two (perhaps concomi-tant) reasons for this finding: The mechanico-inertial con-straints involved in the tongue raising and fronting gesturefor the high front vowel may explain why effects from /i/

Figure 4. Cross-speaker F2 differences for changing /i/ versus /a/ at seven/a/ (bottom). In each graph, the four left bars correspond to the V2 anticip(Carr). Within each bundle of four bars, four stress/rate conditions may bUSTn (unstressed, normal), STf (stressed, fast), and USTf (unstressed, fast


favor the carryover direction, and V1 = /a/ may cause thetongue dorsum to stay relatively low during /δ/, thus block-ing to a greater or lesser extent the tongue-dorsum raisingand fronting anticipatory effects exerted by V2 = /i/.

As revealed by the graphs in Figure 5, F2 differencesbetween changing /i/ and /a/ are greater at the carryoverthan the anticipatory level in VCV sequences with /ɲ/ andthe two fixed vowels /i/ and /a/ mostly in Temporal Compari-sons 1, 2, and 3. The vowel carryover effects also exceededthe vowel anticipatory effects in Comparisons 4–7 duringfixed /i/, though this directionality difference appears to beassociated with speaker AL only, presumably because heovershot the vowel (/i/) when it was preceded by anothersegment involving much dorsopalatal contact (/ɲ/). Inspec-tion of the F2 trajectories for the two asymmetrical VCVsequences /iɲa/ and /aɲi/ plotted in Figure 8 may shedsome light on why effects from /a/ on the consonant andfixed /i/ are greater at the carryover versus the anticipatorylevel. Analogous to the scenario for /δ/, data for all speakersindicate the presence of a much higher F2 frequency at Coffset and during the first half of V2 = /a/ in the sequence/iɲa/ than at C onset and during the second half of V1 = /a/

temporal points in /VδV/ sequences with fixed /i/ (top) and fixedatory effects (Ant) and the four right ones to the V1 carryover effectse identified: STn (stressed changing vowel, normal speech rate),). V1 = first vowel position; V2 = second vowel position.

1407–1424 • October 2015

Figure 5. Cross-speaker F2 differences for changing /i/ versus /a/ at seven temporal points in /VɲV/ sequences with fixed /i/ (top) and fixed/a/ (bottom). See Figure 4 caption for details.


in the sequence /aɲi/, which follows from the presence ofmore tongue-dorsum raising toward the palate at closureoffset than at closure onset. This may explain why tongue-body lowering effects exerted by /a/ are blocked at theoffset of the consonant rather than at its onset and, conse-quently, why carryover effects from V1 = /a/ are largerthan anticipatory effects from V2 = /a/.

Patterns of coarticulatory direction for sequenceswith /l/ are to a large extent determined by whether theconsonant is darker or clearer. Speakers AL and DP, withstrongly dark variants of the consonant, favored vowelanticipation in the fixed /i/ condition up to temporal com-parison 4 (see Figure 9, top graphs) and in the fixed /a/context mostly for Comparisons 1 and 2 (see Figure 9,bottom graphs). Coarticulation data across speakers pre-sented in Figure 6 also reveal the prevalence of vowel an-ticipation for Comparisons 2–4 in the context of /i/ andfor Comparisons 1 and 2—but not at temporal points situ-ated further away from the changing vowel—in the contextof /a/. F2 trajectories for the sequences /ila/ and /ali/ pro-duced by speakers AL and DP allow a determination insome detail of why effects from /i/ on dark /l/ and fixed /a/

Figure 6. Cross-speaker F2 differences for changing /i/ versus /a/ at seven(bottom). See Figure 4 caption for details.

are greater at the anticipatory versus the carryover level (seeFigure 10). According to the figure, the two speakers showa lower F2 for /ila/ than for /ali/ at C midpoint and atequivalent temporal points during the vowels—for exam-ple, at C onset and during V1 for /ila/ compared with at Coffset and during V2 for /ali/. This finding should be attrib-uted to a considerable blocking of the carryover effectsfrom V1 = /i/ by the tongue lowering/backing anticipatoryactivity for dark /l/; this becomes especially salient when-ever the consonant is followed by a low vowel because bothdark /l/ and /a/ share an analogous low/back tongue-dorsumconfiguration. Speakers CE and LU, with clearer variantsof /l/, as well as speaker DR, favor no major pattern ofvowel coarticulatory direction, perhaps because, at least forthe former subjects, anticipatory effects for the consonantcounteract the carryover action of V1 = /i/ to some extent,though less so than when the alveolar lateral is dark.

Stress and Speech RateSentence duration was affected significantly by Speech

Rate, F(1, 114) = 185.29, p < .001, ηp2 = .67. As shown in

temporal points in /VlV/ sequences with fixed /i/ (top) and fixed /a/


Table 4. Significant Coarticulatory Direction effects for F2 differences between changing /i/ and /a/ over time.

Temporal Comparison

Direction(df = 1, 199)

Direction ×Consonant(df = 2, 199)

Direction ×Fixed Vowel(df = 1, 199)

Direction × Consonant ×Fixed Vowel(df = 2, 199)

F ηp2 F ηp

2 F ηp2 F ηp

2

1. V2 on/V1 off 5.3* .03 13.5*** .12 9.8** .052. C midpoint 11.9** .06 8.7*** .083. V1 off/V2 on 6.1** .06 20.3*** .09 4.2* .044. 4.8** .04 6.8** .03 5.7** .055. V1 midpoint/V2 midpoint 6.3* .03 4.3* .04 11.1*** .106. 4.4* .047. V1 on/V2 off 5.9* .03

Note. df = degrees of freedom; V1 = first vowel position; V2 = second vowel position; C = consonant.

*p < .05. **p < .01. ***p < .001.


Figure 11 (top), sentences were shorter when produced atfast versus normal speech rates, and shortening degree wasspeaker dependent—that is, maximal for speakers LU andDR and minimal for speaker DP.

Figure 7. F2 trajectories for the sequences /iδa/ and /aδi/ for speakers DRV1 = first vowel position; C = consonant; V2 = second vowel position.


Vowel duration yielded a main effect of Stress,F(1, 221) = 94.75, p < .001, ηp

2 = .30; Speech Rate, F(1, 221) =21.25, p < .001, ηp

2 = .09; and Vowel Quality, F(1, 221) =24.49, p < .001, ηp

2 = .10, but not of Vowel Position, and

, DP, and CE (stressed V1, normal speaking-rate condition only).

1407–1424 • October 2015

Figure 8. F2 trajectories for the sequences /iɲa/ and /aɲi/ for speakers DR, DP, and CE (stressed V1, normal speaking-rate condition only).V1 = first vowel position; C = consonant; V2 = second vowel position.


no significant interactions. It was greater for /a/ than for /i/and, as shown by Figure 11 (bottom graph), it varied withstress and speech rate in the progression stressed/normal >stressed/fast > unstressed/normal > unstressed/fast formost speakers and therefore as a function of changes instress position rather than of changes in speech rate.

The vowel F2 frequency also varied significantly asa function of Stress, F(1, 221) = 4.06, p < .05, ηp

2 = .02;Vowel Position, F(1, 221) = 21.91, p < .001, ηp

2 = .09; andVowel Quality, F(1, 221) = 2,696.38, p < .001, ηp

2 = .92,but not Speech Rate. As shown by the F2 trajectories for/iCi/ and /aCa/ displayed in Figure 3, F2 was slightly higherfor stressed vowels than for unstressed ones for essentiallyall speakers (in the figure, continuous lines correspond tothe stressed V1 condition and discontinuous lines to thestressed V2 condition). F2 frequencies were also higher forvowels placed in the V2 versus the V1 position and for /i/versus /a/. Moreover, vowel duration and F2 frequencyturned out to be moderately correlated across Stress,Speech Rate, Consonant, and Vowel Position and speakerconditions—that is, F2 increased with vowel duration

mostly for /i/, r = .420, p < .001, but also for /a/, r = .252,p < .01.

As shown in Table 5, Stress also had an effect onthe degree of vowel coarticulation according to TemporalComparisons 4–7, whereas the effect of Speech Rate wasbarely significant and held for Temporal Comparison 4,for the most part. Figures 4–6 illustrate graphically thestress- and rate-dependent differences in vowel coarticula-tion occurring during VCV sequences composed of /δ/, /ɲ/,and /l/ and fixed /i/ and /a/. Regarding stress, the size ofthe vowel coarticulation effects appears in dark bars fortriggering stressed vowels and in gray bars for triggering un-stressed vowels. For VCV sequences with all three conso-nants, results for Temporal Comparisons 4–7 reveal atrend for the degree of coarticulation to increase whenthe triggering vowel is stressed (and thus the target vowelis unstressed) than when the triggering vowel is unstressed(and the target vowel is stressed). A relevant finding is thatthe effect of Stress on vowel coarticulation operates dur-ing the transconsonantal vowel but not for Comparison 2and thus during the consonant (see also the Discussion).


Figure 9. F2 differences for changing /i/ versus /a/ at temporal points 1–4 in /VlV/ sequences with fixed /i/ (top) and fixed /a/ (bottom) forspeakers AL and DP. See Figure 4 caption for details.


The effect of Speech Rate on vowel coarticulation maybe traced by comparing the two left bars and the tworight bars within each bundle of four bars in each graphof Figures 4–6 (i.e., the two left bars correspond to thenormal speech rate condition and the two right bars tothe fast speech rate condition). The figures reveal thepresence of slightly more vowel coarticulation for fastversus normal speech rates, mostly in the case of TemporalComparisons 2–5.

A relevant finding was the lack of significant Stress ×Consonant, Speech Rate × Consonant, Stress × Coarticula-tory Direction, and Speech Rate × Coarticulatory Direc-tion interactions (Speech Rate and Stress did not interactwith each other either). Therefore, an increase in coar-ticulation associated with the presence of a stressed versus


unstressed changing vowel and an increase in speech rateappears to take place independent of the intervocalic con-sonant and coarticulatory direction—namely, of whetherthe consonant is /δ/, /ɲ/, or /l/ and whether vowel effectsare mostly carryover or anticipatory, depending on the seg-mental composition of the VCV sequence. Indeed, datapresented in Figures 4–6 show a similar increase in coarti-culation as a function of stressed versus unstressed vowelsin the anticipatory and carryover directions for VCVsequences with all three consonants. Consistent with thisfinding, F2 trajectories for the asymmetrical sequences/aCi/ and /iCa/ such as those plotted in Figures 7, 8, and10 exhibit essentially the same temporal patterns independentof whether they are produced in the stressed/normal,unstressed/normal, stressed/fast, or unstressed/fast conditions.

1407–1424 • October 2015

Figure 10. F2 trajectories for the sequences /ila/ and /ali/ for speakers AL and DP (stressed V1, normal speaking-rate condition only). V1 =first vowel position; C = consonant; V2 = second vowel position.


DiscussionThe first hypothesis tested in the present study was

whether coarticulatory resistance varied with the degreeof lingual constraint involved in the production of theintervocalic consonant and the transconsonantal vowel(Recasens et al., 1997). VCV coarticulation data reportedin the Results turned out to be in general agreement withthis prediction regarding consonants: Vowel coarticulatory

Figure 11. (Top) Effect of changes in speech rate on sentenceduration across consonant, vowel quality, and stress position foreach speaker. (Bottom) Effect of changes in stress position andspeech rate on vowel duration in symmetrical vowel–consonant–vowel sequences across consonant and vowel quality for eachspeaker. Error bars correspond to one standard deviation.

effects from /i/ versus /a/ were greater for /δ/ than for /l/and /ɲ/, and these consonant-dependent differences invowel coarticulation extended essentially until about themidpoint of the fixed vowel. Moreover, the alveolopalatalnasal turned out to be somewhat more resistant than thealveolar lateral during the consonant and the fixed vowel.Coarticulatory resistance for /l/ was related to speaker-dependent differences in darkness degree to a large extent:As expected, /l/ was highly coarticulation resistant forthe three speakers showing a dark consonantal variety;among subjects showing a clearer variety of the consonant,coarticulatory resistance was relatively high for one speakerand as low as for dark /l/ in the case of the other subject.

Also as predicted, vowel effects in VCV sequenceswith /δ/ and /ɲ/ were greater when the fixed vowel was lessconstrained (/a/) than when it was more constrained (/i/).It thus appears that there is more room for /δ/ and /ɲ/ toadapt to coarticulatory effects when flanked by a vocalicsegment that does not cause the tongue-body requirements

Table 5. Significant Stress and Speech Rate effects for F2 differencesbetween changing /i/ and /a/ over time.

Temporal comparison

Stress(df = 2, 199)

Rate(df = 1, 199)

F ηp2 F ηp

2

1. V2 on/V1 off2. C midpoint3. V1 off/V2 on 4.5* .024. 11.6** .05 6.8** .035. V1 midpoint/V2 midpoint 13.5*** .06 5.5* .036. 13.8*** .067. V1 on/V2 off 10.5** .05

Note. df = degrees of freedom; V1 = first vowel position; V2 = secondvowel position; C = consonant.

*p < .05. **p < .01. ***p < .001.



for the consonant to increase considerably. Vowel effectsin sequences with /l/ were, however, greater in the fixed /i/versus the fixed /a/ condition—that is, there was moreanticipatory F2 lowering induced by V2 = /a/ on the con-sonant and preceding fixed /i/ than anticipatory F2 raisingexerted by V2 = /i/ on the consonant and preceding fixed/a/. This finding may be related to the joint tongue-dorsumlowering and retraction action of dark /l/ and following/a/ during preceding /i/.

As shown earlier in the literature, V-to-V effects inVCV sequences decrease as one gets away from the vowelthat exerts the coarticulatory effect. Moreover, as alsoreported in some studies (Fowler & Brancazio, 2000;Recasens, 2002), vowel effects may occur during the trans-consonantal vowel but not during the intervocalic con-sonant, mostly when, as for alveolopalatal consonants, alarge portion of the tongue is involved in making a closureand thus is left not too free to adapt to the contextualphonetic segments. Several predictions on coarticulatorydirection were also confirmed by the data:

1. Sequences with /ɲ/ showed more carryover than an-ticipatory vowel coarticulation (triggered essentiallyby low back /a/), which should be related to the ar-ticulatory configuration for the alveolopalatal nasalbeing more /j/-like at closure release than at closureonset. The coarticulation data for VCV sequenceswith the alveolopalatal nasal also provided evidencefor articulatory overshoot in the case of the sequence/iɲi/ produced by speaker AL, who exhibited thehighest F2 frequency trajectory of all subjects; theovershoot effect in question appears to result fromtongue-dorsum raising and fronting for both V1 = /i/and following /ɲ/, causing dorsopalatal contact andF2 during V2 = /i/ to increase above the expectedthreshold.

2. Dark /l/ favored vowel anticipation over vowel carry-over. This finding was attributed to the fact that thetongue-dorsum lowering and retraction movementfor this consonantal variety is anticipatory ratherthan carryover and therefore allows coarticulatoryeffects from V2 = /i/ rather than from V1 = /i/ to takeplace. Following /a/ appears to assist the anticipatorytongue-dorsum lowering and backing motion forthe consonant to become especially prominent dur-ing preceding /i/. However, VCV sequences with aclearer variety of /l/ did not show consistent patternsof vowel coarticulatory direction: Unlike sequenceswith dark /l/, they did not favor vowel anticipation,which appears to be in line with the fact that clear /l/does not involve much anticipatory tongue-dorsumlowering and backing; unlike sequences with /δ/(see the next point), they did not favor vowel carry-over effects from V1 = /i/ either, presumably becausesome anticipatory predorsum lowering associated withlaterality conflicts with the vowel effects in question.

3. Vowel coarticulatory effects in sequences with /δ/ fa-vored the carryover direction in the fixed /a/ context


and no specific direction in the fixed /i/ context. Thisfinding appears to be related primarily to the tongue-dorsum raising/fronting carryover effects exertedby V1 = /i/ on consonants such as /δ/ not involvingactive tongue-dorsum activity. This vowel carryoveraction may be assisted by V1 = /a/ causing thetongue dorsum to occupy a relatively low positionduring /δ/ and some delay of the anticipatory tonguefronting/raising motion for V2 = /i/ during theconsonant.

A main goal of this investigation was to study theeffect of changes in stress and speech rate on vowel coarti-culation in VCV sequences. Changes in stress turned out toaffect vowel duration and F2 frequency to a larger extentthan speech-rate variations: Vowels were longer and lesscentralized when stressed versus unstressed for essentiallyall speakers, and an increase in speech rate did not inducethe same degree of vowel shortening in all subjects andyielded small or no F2 changes. Vowel coarticulation wasalso greater and extended over a longer temporal periodwhen induced by stressed versus unstressed vowels ratherthan by an increase in speech rate. These Catalan data arein agreement with data for other languages showing thatvowels produced with different stress levels and at differentspeech rates may be hypoarticulated or hyperarticulated(de Jong, Beckman, & Edwards, 1993).

The initial prediction that stress and rate effects couldbe more prominent in VCV sequences with less constrainedconsonants than in those with more constrained conso-nants did not hold. Thus, although constrained consonantsturned out to be more resistant than unconstrained onesto vowel effects in tongue-dorsum raising/lowering, this dif-ference in vowel coarticulatory resistance applied largelyindependent of whether the changing vowel was stressedor unstressed and sentences were produced faster or moreslowly.

Another relevant finding is that stress and speechrate did not interact with consonant-specific patterns ofvowel coarticulatory direction and therefore affected thetwo coarticulatory directions alike—that is, changes invowel coarticulation over time for specific VCV sequencesoccurred more or less equally in the anticipatory and carry-over directions rather than in the direction that wasmostly favored by the consonantal articulatory gesture.Thus, for example, the prevalence of vowel carryover overvowel anticipation in /VɲV/ sequences did not becomemore evident when stress fell on V1 than when it fell on V2and therefore in /′VɲV/ versus /V′ɲV/ sequences; instead,stress caused a similar increase in vowel coarticulationin the target unstressed vowel in both /′VɲV/ and /V′ɲV/structures and therefore at both anticipatory and carryoverlevels.

To summarize, changes in vowel coarticulation asso-ciated with stress and speaking rate were found not to alterthe patterns of degree and direction of vowel coarticula-tion in VCV sequences composed of phonetic segments spec-ified for different degrees of articulatory constraint. This

1407–1424 • October 2015


finding allows the conclusion that the contribution of seg-mental and prosodic factors to VCV coarticulation is orga-nized hierarchically: Coarticulatory changes induced byprosodic variations conform to segment-dependent patternsof coarticulatory degree and direction. It also implies thatcoarticulatory fields for specific phonetic segments may beresized depending on prosodic condition.

The VCV coarticulation data reported in the Resultsalso support the view that prosodic variations affect vowelsrather than consonants. Indeed, stress-dependent voweleffects were found to occur during the transconsonantalvowel rather than during the consonant—that is, for Tem-poral Comparisons 4–7 rather than for Comparison 2 inFigures 4–6. This finding is consistent with data from theliterature showing that vowel articulation is affected to alarger extent than consonant articulation by an increase inspeech rate and by a shift from more formal to spontaneousspeech styles (see van Son & Pols, 1999) and also with thenotion that consonants but not vowels require the formationof invariant articulatory constrictions.

Future work needs to be carried out in order to ex-plore the extent to which the patterns of coarticulation re-ported in this study hold for more speakers, for real wordswith the same segmental and stress patterns, and forother languages besides Catalan. The perceptual effective-ness of these coarticulatory patterns should also be testedwith real speech stimuli excised from VCV sequences.

AcknowledgmentsThis research was funded by project FFI2013-40579-P from

the Spanish Ministry of Economy and Competitiveness, by ICREA(Catalan Institution for Research and Advanced Studies), andby the research group 2014SGR61 from the Catalan government.I would like to thank Meritxell Mira for assistance with the analy-sis of the acoustic data.

ReferencesAgwuele, A. (2004). Effect of stress on coarticulation. In J. Slifka,

S. Manuel, & M. Matthies (Eds.), From sound to sense: 50+years of discoveries in speech communication (pp. C109–C114).Cambridge, MA: Massachusetts Institute of Technology. Re-trieved from http://www.rle.mit.edu/soundtosense/conference/pdfs/fulltext/Saturday%20Posters/SB-Agwuele-STS.pdf

Agwuele, A., Sussman, H. M., & Lindblom, B. (2008). The effectof speaking rate on consonant vowel coarticulation. Phonetica,65, 194–209.

Cho, T. (2004). Prosodically conditioned strengthening and vowel-to-vowel coarticulation in English. Journal of Phonetics, 32,141–176.

Dart, S. (1991). Articulatory and acoustic properties of apical andlaminal articulations. UCLA Working Papers in Phonetics, 79,1–155.

de Jong, K., Beckman, M. E., & Edwards, J. (1993). The interplaybetween prosodic structure and coarticulation. Language andSpeech, 36, 197–212.

Fant, G. (1960). Acoustic theory of speech production. The Hague,the Netherlands: Mouton.

Farnetani, E. (1990). V-C-V lingual coarticulation and its spatio-temporal domain. In W. J. Hardcastle & A. Marchal (Eds.),

Speech production and speech modelling (pp. 93–130). Dordrecht,the Netherlands: Kluwer.

Farnetani, E., & Recasens, D. (2010). Coarticulation and connectedspeech processes. In W. J. Hardcastle, J. Laver, & F. E. Gibbon(Eds.), The handbook of phonetic sciences (2nd ed., pp. 316–352).Malden, MA: Wiley-Blackwell.

Fowler, C. A. (1981). A relationship between coarticulation andcompensatory shortening. Phonetica, 38, 35–50.

Fowler, C. A., & Brancazio, L. (2000). Coarticulation resistanceof American English consonants and its effects on transconso-nantal vowel-to-vowel coarticulation. Language and Speech,43, 1–41.

Gay, T. (1978). Effect of speaking rate on vowel formant move-ments. The Journal of the Acoustical Society of America, 63,223–230.

Grosvald, M. A. (2009). Long-distance coarticulation: A productionand perception study of English and American Sign Language(Unpublished doctoral dissertation). University of California,Davis.

Guenther, F. H. (1995). Speech sound acquisition, coarticulation,and rate effects in neural network model of speech production.Psychological Review, 102, 594–621.

Hertrich, I., & Ackermann, H. (1995). Coarticulation in slowspeech: Durational and spectral analysis. Language and Speech,38, 159–187.

Kent, R. D., & Netsell, R. (1971). Effects of stress contrast on cer-tain articulatory parameters. Phonetica, 24, 23–44.

Krull, D. (1989). Second formant locus patterns and consonant-vowel coarticulation in spontaneous speech. Phonetic Experi-mental Research at the Institute of Linguistics, University ofStockholm, 10, 87–108.

Magen, H. S. (1997). The extent of vowel-to-vowel coarticulationin English. Journal of Phonetics, 25, 187–205.

Matthies, M., Perrier, P., Perkell, J. S., & Zandipour, M. (2001).Variation in anticipatory coarticulation with changes in clarityand rate. Journal of Speech, Language, and Hearing Research,44, 340–353.

Mok, P. P. K. (2011). Effects of vowel duration and vowel qualityon vowel-to-vowel coarticulation. Language and Speech, 54,527–545.

Moon, S.-J., & Lindblom, B. (1994). Interaction between duration,context, and speaking style in English stressed vowels. TheJournal of the Acoustical Society of America, 96, 40–55.

Proctor, M. I. (2009). Gestural characterization of a phonologicalclass: The liquids (Unpublished doctoral dissertation). YaleUniversity, New Haven, CT.

Recasens, D. (1999). Lingual coarticulation. In W. J. Hardcastle& N. Hewlett (Eds.), Coarticulation: Theory, data and tech-niques (pp. 80–104). Cambridge, United Kingdom: CambridgeUniversity Press.

Recasens, D. (2002). An EMA study of VCV coarticulatory direc-tion. The Journal of the Acoustical Society of America, 111,2828–2841.

Recasens, D. (2007). Patterns of VCV coarticulatory directionaccording to the DAC model. In P. Prieto, J. Mascaró, &M.-J. Solé (Eds.), Segmental and prosodic issues in Romancephonology (pp. 25–39). Amsterdam, the Netherlands: JohnBenjamins.

Recasens, D., & Espinosa, A. (2009). Dispersion and variability inCatalan five and six peripheral vowel systems. Speech Com-munication, 51, 240–258.

Recasens, D., Pallarès, M. D., & Fontdevila, J. (1997). A model oflingual coarticulation based on articulatory constraints. TheJournal of the Acoustical Society of America, 102, 544–561.



Simonet, M. (2010). Dark and clear laterals in Catalan andSpanish: Interaction of phonetic categories in early bilinguals.Journal of Phonetics, 38, 663–678.

Sproat, R., & Fujimura, O. (1993). Allophonic variation inEnglish /l/ and its implications for phonetic implementation.Journal of Phonetics, 21, 291–311.


van Son, R. J. J. H., & Pols, L. C. W. (1999). An acoustic descrip-tion of consonant reduction. Speech Communication, 28, 125–140.

Whalen, D. H. (1990). Coarticulation is largely planned. Journalof Phonetics, 18, 3–35.

Yun, G. (2007). Word frequency, stress and coarticulation in English.Studies in Phonetics, Phonology and Morphology, 13, 315–331.

1407–1424 • October 2015

the effect of stress and speech rate on vowel ... · vowel coarticulation in catalan...

Documents