See discussions, stats, and author profiles for this publication at: [Link]

net/publication/239645822

A model of lingual coarticulation based on articulatory constraints

Article in The Journal of the Acoustical Society of America · July 1997

DOI: 10.1121/1.419727

CITATIONS READS
169 306

3 authors, including:

Daniel Recasens
Autonomous University of Barcelona
119 PUBLICATIONS 2,167 CITATIONS

SEE PROFILE

All content following this page was uploaded by Daniel Recasens on 04 June 2014.

The user has requested enhancement of the downloaded file.

A model of lingual coarticulation based on articulatory
constraints
Daniel Recasens
Laboratori de Fonètica, Institut d’Estudis Catalans, c/Carme 47, Barcelona 08001, Spain and Universitat
Autònoma de Barcelona, Bellaterra, Barcelona, Spain

Maria Dolors Pallarès and Jordi Fontdevila

Laboratori de Fonètica, Institut d’Estudis Catalans, c/Carme 47, Barcelona 08001, Spain

~Received 27 December 1995; accepted for publication 14 February 1997!

The magnitude and temporal extent of consonantal and vocalic coarticulation are reported for VCV
sequences with two vowels ~/{/, /a/! and seven consonants ~/!/, /'/, dark /(/, /2/, /b/, /F/,/%/!. Different
degrees of articulatory constraint, or DAC values, are assigned to the consonants and vowels based
on knowledge of their articulatory properties, in particular, the degree of involvement of the tongue
dorsum in closure or constriction formation. Mean results on dorsopalatal contact and F2 frequency
for five speakers of the Catalan language are presented. Predictions based on the DAC value for
consonants and vowels account satisfactorily for the C-to-V effects ~e.g., those for /Fa/ are more
prominent that those for /pi/!; moreover, vowel-dependent effects tend to be negatively correlated
with the DAC value for the consonant ~e.g., they are more prominent when the intervocalic
consonant is /!/ than when it is dark /(/!. V-to-C effects are also conditioned by the tongue-dorsum
position for the consonantal gesture. Coarticulatory directionality trends reveal that the extent to
which the vowel-dependent tongue-dorsum activity may be anticipated is closely linked to the
mechanico-inertial constraints associated with the tongue dorsum during consonantal production;
this observation explains the salience of the vowel-dependent anticipatory effects in VCV sequences
favoring C-to-V anticipation and of the vowel-dependent carryover effects in VCV sequences
giving special weight to C-to-V carryover. © 1997 Acoustical Society of America.
@S0001-4966~97!03106-8#
PACS numbers: [Link], [Link] @AL#

INTRODUCTION attention than those for more mobile articulators such as the
lips or the velum ~e.g., Krakow, 1989; Boyce, 1990! presum-
Two general principles of segmental overlap in a pho- ably because it was believed that labial and velar coarticula-
netic string have been proposed in the literature: The extent tion would be more extensive than lingual coarticulation. In
to which a vowel falls short of its target ~i.e., the amount of many respects the coarticulatory patterns presented in this
undershoot! increases with the distance between the articula- paper ought to have a universal status ~although some coar-
tory targets for the vowel and for the adjacent consonant ticulation aspects may differ across languages; Beddor and
~Stevens et al., 1966!, and coarticulation on consonants af- Yavuz, 1995! and may prove useful in predicting the fre-
fects mostly those articulators which are not involved in clo- quency of occurrence of some assimilatory processes ~i.e., it
sure or constriction formation ~Öhman, 1966!. There is, how- may turn out that phonetic segments more prone to undergo
ever, no model that predicts how much coarticulation is assimilations exhibit prominent coarticulatory effects along
allowed by a given phonetic segment, or the extent to which the relevant articulatory dimensions!.
a particular phonetic segment is likely to affect the surround-
A. Articulatory constraint
ing segments in the speech chain. The aim of this paper is to
present such a model by examining the relative salience of In the present study phonetic segments are characterized
the C-to-V, V-to-C, and V-to-V coarticulatory effects in in terms of articulatory gestures. An articulatory gesture is an
Catalan VCV sequences with several consonants ~bilabial actively controlled movement toward a presumed target con-
stop /!/, alveolar nasal /'/, dark alveolar lateral /(/, alveolar figuration ~e.g., lip rounding for rounded vowels, tongue-
fricative /2/, alveolopalatal fricative /b/, alveolopalatal nasal dorsum raising for palatal articulations! ~Browman and
/F/, velar stop /%/! and vowels ~high front /{/, low back /a/!. Goldstein, 1986!. Phonetic segments may be produced with
Our work is mostly concerned with tongue-dorsum activity1 one or two lingual gestures and thus with one or more lingual
since two of the variables under investigation, i.e., dorso- articulators ~e.g., the tongue tip for /t/, the tongue tip and the
palatal contact and F2, are known to be positively tongue dorsum for dark /(/!. The biomechanics of the lingual
correlated.2 articulator~s! may cause some displacement on other tongue
Spatiotemporal patterns of tongue-dorsum coarticulation regions. Thus the strength of the coupling effects between
should be of general interest for theories of articulatory con- the tongue tip or the tongue blade and the tongue dorsum
trol in speech production in so far as they have received less should increase with an increase in the retraction and the

544 J. Acoust. Soc. Am. 102 (1), July 1997 0001-4966/97/102(1)/544/18/$10.00 © 1997 Acoustical Society of America 544
extent of the apicolaminal closure or constriction, for alveo- To summarize, the DAC scale ranks phonetic sound cat-
lopalatals versus alveolars and for laminoalveolars versus egories from maximally to minimally constrained as follows:
apicoalveolars. /b/, /F/, /{/, /%/, dark /(/, ~/2/! (DAC53)./'/, /~/, ~/2/!
These notions are clearly related to a core concept (DAC52)./!/, /./ (DAC51). It needs to be emphasized
within our descriptive framework, i.e., degree of articulatory that this is a preliminary DAC classification which could be
constraint ~DAC!.3 The formulation of the concept DAC de- improved with a more accurate formulation of the articula-
rives from assumptions about the degree of involvement of tory constraints for consonants and vowels.
the speech articulators in the formation of a closure or con-
striction based on information obtained from experimental
production data, e.g., data on articulatory displacement and
linguopalatal contact. As shown below, the consonants and B. Specific coarticulatory effects
vowels of interest in this paper will be assigned different
DAC values depending on the degree of tongue dorsum con-
straint during their production. The DAC scale proceeds 1. C-to-V coarticulation
from a DAC minimum of 1 to a DAC maximum of 3, where DAC values should have a bearing on the extent to
the number of levels could be any other than 3; the assign- which a given segment will influence adjacent segments, and
ment process of DAC values to vowels does not differ from the final outcome will also depend on the DAC value for the
that used for consonants and implies that vowels may exhibit contextual segment and whether the constraints involved are
constriction locations along the vocal tract ~Wood, 1979!. compatible or opposing. Within this framework, a first goal
Dorsals, i.e., alveolopalatals /b/ and /F/, palatal /{/, velar of this study is to investigate the prominence of the
/%/, and dark /(/, are maximally constrained and thus speci- consonant-to-vowel effects ~C-to-V! in the three scenarios
fied for a maximal DAC value (DAC53). Considerable described below and to determine interactions between the
tongue-dorsum involvement in the production of all these magnitude and the temporal extent of C-to-V coarticulation
consonants renders them highly constrained: Alveolopalatals ~Sec. II A!.
and palatals are articulated with active tongue dorsum raising The first scenario is at work when C and V are produced
toward the palatal zone where they cause large amounts of with comparable articulatory trajectories, as for /F{/, /'{/, and
contact ~Recasens, 1990; Recasens and Romero, 1997! and /(~/ since /F/, /'/, and /{/ involve tongue-dorsum raising ~as a
dorsovelars share similar kinematic properties; the produc- consequence of coupling between the tongue tip and the
tion of dark /(/ involves two lingual gestures, i.e., tongue tip tongue dorsum for /'/! and /(/ and /~/ are produced with
raising for a primary apicoalveolar closure and active tongue tongue-dorsum lowering. Little C-to-V coarticulation is ex-
postdorsum retraction for a secondary dorsopharyngeal con- pected to occur for /F{/ and /'{/ since both phonetic segments
striction ~Sproat and Fujimura, 1993!. are either specified for the same DAC value ~i.e., DAC53
Labial /!/ is minimally constrained and thus specified for /F/ and /{/ in the sequence /F{/! or the vowel is more
for a minimal DAC value (DAC51). This is so since the constrained than the consonant ~i.e., DAC53 for /{/ and
tongue body may not be required to achieve an articulatory DAC52 for /'/ in the sequence /'{/!. Some C-to-V effects
target. Its obvious correlate in the vowel class is /./ which should occur when the consonant is more constrained than
appears to be more sensitive to contextual effects than other the vowel especially if highly constrained ~i.e., DAC53 for
vowels ~Dutch: Bergem, 1994; English: Kondo, 1994! al- /(/ and DAC52 for /~/ in the sequence /(~/!. An interesting
though it has been reported to exhibit an articulatory target at research topic is to find out whether such effects will be
a neutral tongue position ~Browman and Goldstein, 1992!. more or less salient than those occurring in CV combinations
An intermediate degree of tongue-dorsum constraint and of antagonistic phonetic segments exhibiting the same DAC
thus an intermediate DAC value (DAC52) can be tenta- relationship ~as for /F~/ below!.
tively assigned to phonetic segments for the production of C and V are produced with opposing articulatory trajec-
which the tongue dorsum is not directly involved in closure tories in a second scenario, as for /F~/, /'~/, and /({/ where
or constriction formation but is subject to coupling effects /F/, /'/, and /{/ involve tongue-dorsum raising and dark /(/
with the primary articulator. This would be the case for al- and /~/ involve tongue-dorsum lowering. The sequence /F~/
veolars /'/ and /2/ ~in accordance with tongue blade raising should show some C-to-V coarticulation since the DAC
for these consonants causing some tongue dorsum raising to value for the consonant (DAC53) exceeds that for the
occur; Kent and Moll, 1972; Lindblom, 1983! and for low vowel (DAC52). In principle, little C-to-V coarticulation is
back /~/ ~in view of the tongue root retraction gesture for this expected to occur if both segments are specified for the same
vowel bringing about some concomitant tongue dorsum low- DAC value, i.e., 2 in the sequence /'~/ or 3 in the sequence
ering!. /({/. It deserves to be seen, however, whether constriction
In addition to the place of articulation and interarticula- requirements for consonants yield C-to-V effects in these
tory coupling, the manner of articulation may also affect the circumstances and whether these effects increase with the
DAC specification for consonants. Thus, the precise forma- degree of antagonism between C and V ~i.e., for /({/ vs /'~/!.
tion of a medial groove for fricatives ~e.g., Stone et al., One of the two adjacent phonetic segments is specified
1992! should render /2/ more constrained than nonfricative for DAC51 in a third scenario, as for /!{/ where the conso-
alveolars which would cause this consonant to exhibit a nant involves no tongue dorsum activation. This segmental
DAC value of 3 instead of 2. combination should yield minimal C-to-V effects.

545 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 545
2. Vowel-dependent coarticulation active tongue-dorsum lowering such as dark /(/ are more sen-
The coarticulatory analysis presented in this paper also sitive to lowering effects associated with following /~/ in the
examines the salience of the vowel-dependent effects as a sequence /{(~/ than to raising effects exerted by following /{/
function of the seven consonants of interest ~Sec. II B!. For in the sequence /ali/.
that purpose, the magnitude of the effects from /{/ vs /~/ will It will also be investigated whether there is an inverse
be measured during the consonant ~V-to-C!, and their size correlation between the vocalic effects and the consonantal
and temporal extent as well as the interaction between the effects, and thus whether context-dependent sensitivity for a
two will be computed along the fixed transconsonantal vowel given phonetic segment varies inversely with coarticulatory
/{/ or /~/ ~V-to-V!.4 aggression on the adjacent phonetic segments ~Fowler and
Tongue-dorsum related data in the literature are in ac- Saltzman, 1993!.
cordance with the prediction that coarticulatory sensitivity
for consonants and vowels is inversely related to their DAC
value. Using the same dataset as in the current study, V-to-C C. Coarticulatory directionality
effects in symmetrical VCV sequences reported in Recasens A novelty of the present investigation is to analyze the
et al. ~1995! show that variability for consonants pro- relative prominence of the anticipatory direction ~right-to-
ceeds in the progression [Link] left! and the carryover direction ~left-to-right! for the conso-
.alveolopalatals, dark /(/ and /2/; many other production nantal and the vocalic effects in VCV production. This
studies reveal the same trend @see Recasens ~in press! for a analysis will be carried out with reference to the DAC value
review#. On the other hand, the degree of coarticulatory sen- for the phonetic segments involved and will evaluate
sitivity for vowels decreases for /././~/./i/ ~Kiritani et al., whether coarticulatory direction is conditioned by the ges-
1977; Recasens, 1985, 1991a!. More tongue-dorsum coar- tural demands for the intervocalic consonant ~Sec. II C!.
ticulation for velars than other dorsal consonants is in agree- There is some evidence in the literature in support of the
ment with their showing a palatovelar constriction in front notion that the prominence of the anticipatory and carryover
vowel contexts and a velar one in back vowel contexts component of consonantal coarticulation varies with the lin-
~Swedish: Öhman, 1966; American English: Kent and Moll, gual requirements for the consonant and determines the di-
1972!. This finding does not run against consonantal seg- rectionality of the vocalic effects @see Recasens ~in press! for
ments specified for a high DAC value allowing little vowel a review#. Evidence on C-to-V coarticulation reveals that,
coarticulation but appears to be due to the consonant and the among consonants specified for a high DAC value, some
vowel not being able to achieve adjacent tongue-dorsum tar- exert more prominent carryover than anticipatory effects
gets unless the consonantal constriction moves toward that ~dorso-alveolopalatals, in line with tongue-dorsum activity
for the vowel ~Wada et al., 1970!. proceeding more slowly at consonantal release than at con-
As indicated by data in the literature ~Gay, 1974, 1977; sonantal onset! while others require a strong anticipatory
Carney and Moll, 1971; Recasens, 1984, 1987, 1989!, the component ~dark /(/, in line with strong demands on the for-
relationship between coarticulatory sensitivity and DAC be- mation of a double lingual constriction!. Within this frame-
comes more complex when V-to-V effects are accounted for. work the direction of the vocalic effects shows the following
It appears that the occurrence of these effects is conditioned trends whose validity will be tested in the present paper: On
not only by the DAC specification for the intervocalic con- the one hand, vowel-dependent carryover effects exceed
sonant but also by that of the fixed vowel. The prediction vowel-dependent anticipatory effects across consonants ex-
will be tested that consonant-dependent differences in erting more prominent carryover than anticipatory C-to-V
V-to-V coarticulation ~e.g., for effects across bilabials ex- coarticulation; on the other hand, consonants requiring a
ceeding those across alveolopalatals! should occur when the strong anticipatory component allow more prominent vowel-
fixed vowel is not highly constrained ~e.g., /~/! but could be dependent anticipatory than carryover effects or exhibit a
cancelled out if the fixed vowel is specified for a maximal significant reduction of the latter in comparison to the
DAC value ~e.g., /{/!. former. Special attention will be given to those consonants
While DAC is a potential predictor of coarticulatory ~bilabials, dentoalveolar stops, fricatives, clear /(/! which
sensitivity, the present study will also test for a large number have been reported to favor either coarticulatory direction in
of consonants whether articulatory position determines the previous studies.
articulatory dimension along which coarticulation occurs
~Sec. II B 3!. Preliminary data from the literature indicate
that C-to-V effects on high vowels involve tongue lowering I. METHOD
while those on low vowels cause the tongue dorsum to raise
A. Articulatory and acoustic analysis procedure
and that V-to-C effects are determined by the target lingual
position for the consonant ~Recasens, 1987, 1991a, b!. Thus F2 and linguopalatal contact data were collected for
the fact that the production of /F/ ~also /%/! requires the /{C{/, /~C~/, /{C~/, and /~C{/ with stress on the first syllable.5
achievement of a dorsal closure justifies why this consonant As indicated in the Introduction, the consonants were /!/, /'/,
allows larger tongue-dorsum raising and fronting effects as- ~dark! /(/, /2/, /b/, /F/, and /%/. Sequences were read five times
sociated with V25/i/ in the sequence /~F{/ than tongue- each by five Catalan speakers ~DR, JP, JS, DP, JC!. In Cata-
dorsum lowering and backing effects associated with V2 lan, unstressed /~/ ~i.e., V2 in the sequences of the present
5/~/ in the sequence /{F~/; also, consonants produced with study! is realized as @.# systematically. Differences in quality

546 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 546
 the case of /'/, /2/, /b/, and /F/, since a central closure or
constriction at the frontmost row of the palatal zone is gen-
erally absent for the alveolar lateral but frequently present
for the two nasals and for the two fricatives.7 The Qp index
for /%/ and /!/ has also been calculated without taking into
account the aforementioned row given that the contact pat-
tern at the three back rows of the artificial palate provides an
adequate and sufficient articulatory characterization of velars
and bilabials in the context of /{/.
The acoustic data were recorded at a 20-kHz sampling
rate. F2 values were obtained at the same temporal intervals
as the EPG data using LPC ~25-ms Hamming window, 12
coefficients! and visual inspection of spectrographic displays
on a Kay CSL analysis system. The latter method was ap-
plied instead of the former when an unsatisfactory resolution
was provided by the all-pole peak picking LPC method, i.e.,
at the VC boundary of VCV sequences with nasal conso-
nants ~because of the presence of spectral discontinuities as-
sociated with nasal poles and zeros! and during the conso-
nantal period in VCV sequences with fricatives /s/ and /b/
~because of the presence of frication noise at low spectral
regions!. F2 frequencies were measured using a cursor mov-
ing in 20-Hz steps after careful inspection of the spectro-
graphic patterns by two experimenters.
An EPG criterion was used in order to determine the
consonantal boundaries for /l/ and for /n/ and /F/, i.e., from
onset to offset of central alveolar contact for the lateral and
of complete closure for the nasal stops. However, the seg-
mentation of /s/, /b/, /p/, and /k/ was performed on waveform
and spectrographic displays, from onset to offset of the fri-
cation noise for the two fricatives ~it was often hard to iden-
tify a well-defined period based on the degree of constriction
narrowing on the linguopalatal contact patterns! and from
offset of V1 formant structure to closure offset at the stop
burst for the two stops ~bilabial /p/ shows no lingual activity
FIG. 1. Linguopalatal configurations at consonantal midpoint for different and complete closure /k/ was not visible when occurring be-
consonants in symmetrical VCV sequences ~speaker DR!. Data have been
averaged across repetitions. Percentages of electrode activation: ~black! hind the palatal zone!.
80%–100%; ~dotted! 40%–80%; ~white! less than 40%.
B. Criteria for measuring coarticulatory effects
between V1 and V2 do not render the /~C~/ sequences fully
Coarticulatory effects were calculated in size ~C-to-V,
symmetrical, although we will refer to them as such all
V-to-C, V-to-V! and temporal extent ~C-to-V, V-to-V! ac-
throughout the paper.6
cording to the measurement criteria summarized next.
Linguopalatal contact was gathered every 10 ms using
artificial palates equipped with 62 electrodes @Reading elec-
tropalatographic ~EPG! system; Hardcastle et al., 1989#. Fig-
ure 1 displays average linguopalatal configurations for each 1. C-to-V coarticulation
consonant. The four front rows at the top of each EPG dis- C-to-V temporal effects were taken to occur during the
play belong to the alveolar zone ~extending from the teeth to period when a significant consonant-related acoustic or ar-
the alveolar ridge! and the four back rows belong to the ticulatory difference extends into the vowel. In order to ob-
palatal zone ~from the alveolar ridge back to the soft palate!. tain such effects, Qp and F2 frequency values were first
The graphic representations in the figure do not capture the computed for all repetitions of /iCi/ and /aCa/ for each
fact that the distance between adjacent rows is smaller at the speaker in 10-ms steps starting at the consonantal onset back
former zone than at the latter. to V1 onset and at the consonantal offset until V2 offset. No
The EPG data reported in this paper will be expressed in data were collected for Qp effects from /l/ on /a/ since little
Qp values, where Q p stands for the percentage of contact or no linguopalatal contact occurs along this vowel in this
activation over the palatal zone, i.e., number of activated consonantal environment. The resulting values were then
palatal electrodes/total number of palatal electrodes. The compared statistically ~one-way ANOVAs, Scheffé; dfbg
Qp index has been computed using the four back rows of the 51, dfwg54! with a measure of steady-state V1 of the same
artificial palate in the case of /(/ but the three back rows in symmetrical VCV sequence type for studying C-to-V1 an-

547 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 547
FIG. 2. F2 trajectories showing anticipatory effects ~left! and carryover effects ~right! in a single repetition of /VIV/ sequences with fixed /{/ ~top! and fixed
/a/ ~bottom!. Data correspond to speaker JP. The zero point on the x axis occurs at /l/ central closure onset for the anticipatory effects and at central closure
offset for the carryover effects. See text for details.

ticipation and of steady-state V2 for studying C-to-V2 carry- ~bottom!. A dotted line has been traced at the estimated F2
over. For each repetition of a given symmetrical VCV se- value for the steady-state portion of V1 and for that of V2. A
quence, steady-state vowel values were established at the statistical comparison between the steady-state value and all
highest Qp and F2 for /{/ and at the lowest Q p and F2 for other F2 frequency values along V15/i/ of the sequence /ili/
/a/, starting at the vowel midpoint backward in the case of ~top left! yields a significant difference starting at the 0 tem-
V1 and at two-thirds of the vowel duration onward in the poral point at central closure onset back to 280 ms. The
case of V2. Those maxima were found either at a single application of the same procedure to V25/i/ yields a 80-ms
temporal frame or along a plateau lasting several frames; the long C-to-V effect after the 0 temporal point at central clo-
selection of a different temporal criterion for V1 and V2 is sure offset ~top, right!. Anticipatory and carryover effects on
justified by the fact that ~sentence-final! V2 was longer than /a/ ~bottom! were found to last until 270 ms and 40 ms,
V1 and the V2 steady-state period was located further away respectively.
from the consonantal period than the V1 steady-state period. The largest Qp or F2 difference between V1 or V2 of a
The steady-state V1 and V2 values for each VCV sequence symmetrical VCV sequence and the steady-state vowel value
type represent the average across values for the individual is taken to be the ~positive or negative! size of a C-to-V
repetitions of that sequence. The last significant difference effect. This maximal size difference usually occurs near clo-
obtained in the statistical analysis procedure ( p,0.05) was sure onset for the anticipatory C-to-V effects and near clo-
taken to be the onset of C-to-V1 anticipatory coarticulation sure offset for the carryover C-to-V effects. C-to-V size ef-
when the vowel being measured was V1 and the offset of fects reported in this paper correspond to average values
C-to-V2 carryover coarticulation when the vowel subject to across repetitions and speakers for each symmetrical VCV
measurement was V2. Onsets and offsets of the C-to-V ef- sequence. In Fig. 2 the size of the C-to-V anticipatory effect
fects for each VCV sequence type were separately averaged for V15/i/ ~i.e., the largest F2 difference between steady-
across speakers; the resulting averages will be used as a mea- state V15/i/ and V15/i/ of /ili/! amounts to 811 Hz, while
sure of C-to-V temporal coarticulation in this paper.8 that of the C-to-V carryover effect for V25/i/ equals 736 Hz.
Figure 2 exemplifies the method of analysis of the As for the /a/ condition, C-to-V size effects measure 243 Hz
C-to-V effects exerted by dark /l/ on /{/ ~top! and on /a/ ~anticipatory! and 254 Hz ~carryover!.

548 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 548
2. V-to-C coarticulation Analogously to the C-to-V effects, the size of a V-to-V
V-to-C effects were measured in Q p and F2 size at the effect is taken to be the largest Qp or F2 frequency differ-
midpoint of the consonantal period for all consonants except ence and usually occurs near closure onset for the anticipa-
for /p/ and /k/ ~since oral stops exhibit no formant structure- tory direction and near closure offset for the carryover direc-
during closure!. Differences were obtained for the sequence tion. When this difference was not significant, the size of the
pairs /iCi/–/iCa/ and /aCi/–/aCa/ ~anticipatory coarticula- V-to-V effect was taken at the first temporal frame showing
tion! and /iCi/–/aCi/ and /iCa/–/aCa/ ~carryover coarticula- the expected coarticulatory difference. V-to-V size effects
tion!, and averaged across speakers and repetitions. This ex- reported in this paper correspond to averages across speakers
perimental setting allows studying the extent to which /a/ and repetitions. In the data of Fig. 2, the size of V-to-V
affects the consonant when /{/ stays constant ~in the sequence anticipation is 35 Hz ~/ili/–/ila/! and 79 Hz ~/ali/–/ala/!, and
pairs /iCi/-/iCa/ and /iCi/–/aCi/! and how much V-to-C coar- that of V-to-V carryover amounts to 0 Hz ~/ili/–/ali/! and to
ticulation is exerted by /{/ when the fixed vowel is /a/ ~in the 54 Hz ~/ila/–/ala/!.
sequence pairs /aCa/–/iCa/ and /aCa/–/aCi/!. Differences
~positive! were computed when /{/ caused a higher Qp or C. Other statistical analyses
F2 than /a/ at the consonantal midpoint. V-to-C coarticula- The size and the temporal extent of the consonantal and
tion values reported in this paper correspond to averages vocalic effects for the different consonants across speakers
across differences for each speaker; negative values ~i.e., were also submitted to one-way ANOVAs ~Scheffé! in order
when /a/ yielded a higher Q p or F2 than /{/! were equated to to determine the existence of consonant-dependent signifi-
zero in this averaging procedure. Data for each type of VCV cant differences (dfbw56, dfwg528; p,0.05!. For that pur-
pair were submitted to one-way ANOVAs Scheffé ~dfbg51, pose, negative C-to-V size effects were transformed into
dfwg54! with vowel context as the independent variable and positive values and separate tests were conducted for the
were considered to be significant at the p,0.05 significance anticipatory and for the carryover component in each vowel
level. condition.
Four correlation analyses involving the consonantal ef-
fects or the vocalic effects were separately performed for
3. V-to-V coarticulation
each vowel condition and coarticulatory direction across
V-to-V temporal effects occur during the period along consonants and speakers: between the size and the temporal
which a significant vowel-dependent articulatory or acoustic extent of the C-to-V effects; between the size and the tem-
difference extends into the transconsonantal vowel. In order poral extent of the V-to-V effects; between the size of the
to single out such effects, Q p and F2 frequency differences V-to-C effects and the size of the V-to-V effects; between
as a function of /{/ vs /a/ ~for /{/./a/! were calculated for the the size of the V-to-C effects and the temporal extent of the
same VCV pairs given in Sec. I B 2 every 10 ms, starting at V-to-V effects. The purpose of this analysis was to investi-
consonantal onset back to V1 onset and from consonantal gate whether an increase in magnitude for a given coarticu-
offset until V2 offset. Essentially the sequence pairs with latory effect was matched by an increase in duration and, less
symmetrical /iCi/ allow studying the effect of /a/ along fixed importantly, if V-to-C and V-to-V effects were strongly cor-
/{/ and those with symmetrical /aCa/ allow measuring the related. Four other correlations were carried out between the
effect of /{/ along fixed /a/. Those differences were submitted consonantal effects and the vocalic effects which should bear
to one-way ANOVAs Scheffé (dfbg51, dfwg54) at each on the issue of whether vocalic coarticulation varies in-
temporal point during fixed V1 for studying V-to-V antici- versely with the DAC value for the consonant: between
pation and during fixed V2 for studying V-to-V carryover. C-to-V size and V-to-C size; between C-to-V size and
The last significant difference ( p,0.05) counting backward V-to-V size; between C-to-V size and V-to-V time; between
during V1 was taken to be the onset of a V2-to-V1 anticipa- C-to-V time and V-to-V time. In this paper positive or nega-
tory effect and the last one counting onward during V2 was tive correlations will be taken to occur when the r value is
taken to be the offset of a V1-to-V2 carryover effect. Data on relatively low ~equal or larger than 0.70, whether significant
V-to-V temporal coarticulation presented in this paper refer or not! so as to make sure that no meaningful interactions
to the mean onset and offset times of the V-to-V effects between coarticulatory effects are left out.
across speakers for each VCV sequence pair.
Figure 2 illustrates the procedure for measuring V-to-V II. RESULTS
coarticulation. V2-to-V1 anticipatory effects occur when a
A. C-to-V coarticulatory effects
significant difference between the pairs of F2 trajectories
/ili/–/ila/ ~top left! and /ali/–/ala/ ~bottom left! extends be- Figure 3~a! and ~b! displays Qp and F2 data on C-to-V
fore the 0 temporal point at central closure onset; significant coarticulation for all seven consonants and the two vowels /{/
tests for these trajectories yielded 0 ms for the former se- @Fig. 3~a!# and /a/ @Fig. 3~b!#. Effects are plotted separately
quence pair and 240 ms for the latter. On the other hand, for C-to-V size and temporal extent and for the anticipatory
V1-to-V2 carryover effects take place when a significant dif- direction ~A; stippled bars! and carryover direction ~C; black
ference between the F2 trajectory pairs /ili/–/ali/ ~top right! bars!. The height of the bars indicates whether C-to-V effects
and /ila/–/ala/ ~bottom right! extends after the 0 temporal are larger/smaller ~size effects! and longer/shorter ~temporal
point at central closure offset; significant effects for both effects!.9 For example, according to Fig. 3~a!, the anticipa-
pairs were nonexistent ~0 ms!. tory and carryover effects in dorsopalatal contact (Qp) from

549 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 549
 that this consonant causes Qp and F2 to lower during V1
and V2.

1. /i/ context
According to the predictions in the Introduction section
~see Sec. B 1!, coarticulation should not be too prominent in
sequences made of compatible consonants involving tongue-
dorsum raising whether specified for the same DAC value as
/{/ ~i.e., 3 in the case of alveolopalatals and velars! or for a
lower DAC value ~i.e., 2 for alveolars excluding /(/!. C-to-V
effects in Qp and F2 in the figure are indeed small and short
for alveolopalatals /F/ and /b/ and for velar /%/ although this
trend is less clear for the carryover effects in Qp for reasons
pointed out in Sec. II C 1 a.10 According to Table I, these
effects are positive ~in Qp! and slightly positive or negative
~in F2! which is in accordance with these consonants being
articulated with a similar degree of tongue-dorsum height to
that for /{/. Size and temporal effects in Qp and F2 for the
alveolars /'/ and /2/ are often somewhat more prominent than
those for alveolopalatals and velars which does not accord
with the DAC value for alveolars being lower than that for
/{/. The intermediate status of the C-to-V size effects for /'/
and /2/ is reflected by their being significantly different from
those for other consonants in some cases. The negative sign
of the C-to-V size effects in Qp and F2 for /'/ and /2/ ~see
Table I! may account for this finding: It appears that the
production of these two consonants involves some lowering
of the tongue predorsum ~presumably because of manner re-
quirements in the case of /2/ and of language-specific con-
straints in the case of /n/! which renders them not fully com-
patible with the tongue-dorsum raising gesture for /{/.11
Little C-to-V coarticulation was expected to occur as a
function of /p/ since bilabials do not involve tongue-dorsum
activity (DAC51) and /{/ is specified for a maximum DAC
value ~3! ~see Sec. B 1!. Qp effects for bilabial /p/ in the
figure are indeed small and short. C-to-V effects in F2 ~also
in F3! for /p/ were unexpectedly large and long; more spe-
cifically, size effects were found to be significantly larger
than those for most other consonants. This outcome is prob-
ably unrelated to tongue-dorsum activity but to F2 and F3
of /{/ being highly sensitive to variations in lip rounding, i.e.,
to a protrusion of the lips, a decrease of their area or both
FIG. 3. ~a! Mean C-to-V effects across repetitions and speakers for each
~Fant, 1960!. The negative sign of the F2 size effects in
consonant when the contextual vowel is /{/. Size effects in % of electrode Table I is indeed related to lip closing for /p/; Qp size effects
activation and in Hz ~left! and temporal effects in ms ~right! are plotted ~which could be exclusively attributed to tongue predorsum
separately for Q p ~above! and F2 ~below!, and for each coarticulatory di- lowering! are also negative but very small.
rection, i.e., anticipatory ~stippled bars! and carryover ~black bars!. ~b! Same
representation as in ~a! when the fixed vowel is /a/.
One scenario deserves special attention, i.e., the se-
quence /li/ with the participation of lingual gestures which
are at the same time highly constrained ~dark /l/ and /{/ are
/(/ on /{/ are larger and longer than those exerted by other both specified for DAC53! and antagonistic ~the vowel is
consonants and the situation for the F2 effects is essentially produced with tongue-dorsum raising and fronting and the
the same as that for the Q p effects referred to. consonant with tongue-dorsum lowering and backing! ~see
The numerical values for the coarticulatory effects rep- Sec. B 1!. The figure reveals the existence of large and long
resented in the figures are presented in Table I ~means and C-to-V effects in this case with size effects being signifi-
standard errors!. Size effects in the table are assigned a posi- cantly larger than those for most other consonants. This coar-
tive or a negative sign depending on whether the consonant ticulatory outcome indicates that the requirements for a
causes an increase or a decrease in the Q p or F2 value for highly constrained consonant override those for a highly
the vowel. Thus the table assigns a negative sign to the an- constrained vowel when the two phonetic segments are pro-
ticipatory and carryover size effects from /(/ on /{/ meaning duced with antagonistic lingual gestures. The negative sign

550 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 550
TABLE I. Mean Q p and F2 coarticulation values across speakers and repetitions for different consonants when the fixed vowel is /{/. Data are separately
displayed for the size and the temporal extent of the C-to-V, V-to-C, and V-to-V effects at the anticipatory and carryover level. Size effects are given in %
of electrode activation (Q p) and in Hz (F2); temporal effects are given in ms. C-to-V size effects can be positive or negative ~i.e., the consonant contributes
to a raising or a lowering of the Q p or F2 frequency value for the vowel!. An asterisk indicates the presence of significant V-to-C size effects (p,0.05). A
variability measure (se5standard error) is also given.

C to V V to C V to V

Anticipatory Carryover Anticipatory Carryover Anticipatory Carryover

Vowel /{/ Size Time Size Time Size Sice Time Size Time

/p/ X̄ 21.83 0.00 24.83 8.00 26.67* 31.00* 3.00 0.00 11.67 70.00
se 1.52 0.00 1.78 5.83 1.99 1.43 1.07 0.00 1.73 16.43
/n/ X̄ 29.33 16.00 28.33 22.00 13.33* 15.83* 5.50 10.00 5.00 30.00
se 2.73 10.30 3.11 10.68 1.86 1.92 1.20 10.00 1.21 20.00
/l/ X̄ 227.63 52.00 222.13 54.00 13.38* 13.38* 6.50 14.00 4.13 2.00
se 0.64 3.74 3.71 12.49 3.40 3.39 2.86 14.00 2.18 2.00
/s/ X̄ 210.50 26.00 28.83 20.00 8.83* 4.66 2.50 0.00 1.50 0.00
Qp
se 2.07 14.70 2.31 12.25 1.80 2.12 1.18 0.00 0.55 0.00
/b/ X̄ 2.17 16.00 2.00 30.00 4.00 6.83* 1.50 8.00 5.33 28.00
se 0.68 16.00 2.72 23.24 1.57 1.61 0.93 8.00 2.21 20.83
/F/ X̄ 6.83 14.00 7.67 30.00 0.17 12.83* 1.00 6.00 6.94 23.33
se 1.67 8.72 1.27 12.65 0.17 2.89 0.81 6.00 3.20 12.02
/k/ X̄ 4.46 2.00 15.33 36.00 11.15* 4.50 3.75 6.00 1.00 0.00
se 2.94 2.00 4.11 9.27 2.80 2.00 2.07 4.00 0.49 0.00

/p/ X̄ 2353.00 24.00 2587.00 40.00 ••• ••• 89.60 8.00 72.00 18.00
se 109.43 6.78 84.35 6.32 21.67 8.00 28.02 11.14
/n/ X̄ 15.20 28.00 2130.40 38.00 632.13* 646.10* 309.20 26.00 415.60 46.00
se 85.40 11.58 38.15 15.30 68.32 74.20 56.94 6.78 58.52 18.60
/l/ X̄ 2732.00 64.00 2552.40 68.00 193.32* 86.97* 118.64 12.00 77.96 10.00
se 31.29 6.00 98.84 8.60 22.34 13.81 36.49 5.83 50.44 10.00
/s/ X̄ 2116.00 32.00 2149.60 38.00 99.20* 137.40* 41.60 2.00 52.80 12.00
F2
se 51.74 22.67 43.90 13.56 33.11 37.26 24.68 2.00 8.62 9.70
/b/ X̄ 20.80 0.00 2104.00 38.00 104.00 116.40 48.80 0.00 74.00 0.00
se 12.99 0.00 42.01 20.35 43.67 54.59 15.56 0.00 39.28 0.00
/F/ X̄ 20.80 12.00 248.80 0.00 22.40 105.48* 26.40 4.00 33.67 0.00
se 18.82 12.00 7.09 0.00 12.24 11.63 10.32 4.00 10.59 0.00
/k/ X̄ 86.00 10.00 10.40 6.00 ••• ••• 41.00 8.00 94.20 50.00
se 42.17 10.00 58.25 4.00 23.11 5.83 52.44 27.93

of the C-to-V size effects in Q p and F2 from /l/ on /{/ ~Table mum, i.e., alveolopalatals /F/ and /b/ and velar /k/. These
I! reveals that the consonant contributes to much tongue dor- effects are often significantly different from those for other
sum lowering during the vowel. consonants. They are positive ~Table II! meaning that the
According to Table III, correlations between C-to-V size consonant causes tongue dorsum raising to occur during the
and C-to-V temporal extent, i.e., CVs 3CVt , yield high posi- vowel.
tive r values for Qp and F2 and for the anticipatory and for In comparison to alveolopalatals and velars, the size and
the carryover direction meaning that the duration of the con- temporal effects in the figure are usually smaller and shorter
sonantal effect varies with its magnitude in the /{/ context for the alveolars /n/, /s/, and dark /l/ and for the bilabial /p/.
condition. A close correspondence between C-to-V size and This is the expected outcome for consonants such as /n/ and
temporal extent can be seen for fixed /{/ in Fig. 3~a!. Larger /s/ involving different articulatory trajectories from that for
effects are also longer ~as those for /l/! and smaller effects the vowel and specified for the same moderate DAC value
are also shorter ~as those for /p/ in Q p and for alveolopala- ~see Sec. B 1!. In this case, C-to-V effects can be somewhat
tals and velars in F2!. more prominent than predicted since the tongue-dorsum po-
sition for the consonant is higher than that for the vowel ~as
revealed by Qp and F2 effects for /na/ and /sa/ showing a
2. /a/ context positive sign; see Table II! and consonants tend to override
vowels when the two phonetic segments are specified for the
According to predictions in Sec. B 1, prominent consonant- same DAC value and are antagonistic ~as for /li/!. C-to-V
dependent effects on /a/ should occur in scenarios where the effects of little prominence for /p/ accord with this consonant
consonant and the vowel are produced with opposing articu- exhibiting a minimal DAC value ~see Sec. B 1!. These ef-
latory trajectories and the DAC value for the consonant ex- fects often exhibit a negative sign due to lip closing and/or
ceeds the DAC value of 2 for /a/. Figure 3~b! shows large some tongue lowering ~Table II!. The existence of smaller
and long C-to-V effects in Q p and F2 for consonants requir- C-to-V effects in F2 ~also in F3! from /p/ on /a/ than on /{/
ing tongue-dorsum raising and specified for a DAC maxi- is in accordance with predictions in Fant’s nomograms relat-

551 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 551
TABLE II. Same contents as in Table I when the fixed vowel is /a/.

C to V V to C V to V

Anticipatory Carryover Anticipatory Carryover Anticipatory Carryover

Vowel /a/ Size Time Size Time Size Size Time Size Time

/p/ X̄ 2.00 0.00 25.00 42.00 17.17* 21.50* 3.67 30.00 2.83 6.00
se 0.50 0.00 1.97 28.00 2.91 2.89 1.87 30.00 1.76 4.00
/n/ X̄ 13.00 22.00 2.00 16.00 9.67* 12.17* 1.83 6.00 1.50 2.00
se 0.86 2.00 1.41 16.00 3.28 1.59 0.81 4.00 1.30 2.00
/l/ X̄ ••• ••• ••• ••• 5.50 4.75 1.28 22.00 2.97 10.00
se 1.62 1.85 0.86 22.00 1.34 7.75
/s/ X̄ 15.67 54.00 20.33 14.00 6.33* 2.67 2.33 8.00 1.50 0.00
Qp
se 1.43 6.00 3.15 8.72 1.57 1.63 1.27 8.00 0.93 0.00
/b/ X̄ 38.50 68.00 17.67 68.00 0.17 2.17 1.50 0.00 0.33 0.00
se 2.90 5.83 3.76 11.58 0.17 1.33 0.55 0.00 0.33 0.00
/F/ X̄ 36.50 48.00 26.83 86.00 3.17 14.83* 0.67 0.00 11.00 24.00
se 2.20 4.90 2.90 13.27 1.52 4.37 0.41 0.00 3.26 10.30
/k/ X̄ 23.00 26.00 32.84 68.00 20.00* 12.83 10.67 6.00 6.50 6.00
se 5.76 7.48 3.91 12.00 2.06 4.76 3.24 4.00 3.37 6.00

@p# X̄ 2179.20 40.00 2137.00 46.00 ••• ••• 151.20 78.00 81.40 18.00
se 27.67 4.47 34.00 15.68 29.81 15.94 18.62 8.00
@n# X̄ 124.00 56.00 22.40 36.00 158.94* 172.91* 124.44 92.00 112.56 50.00
se 31.62 14.35 28.86 20.64 45.74 38.49 14.68 13.19 30.75 22.14
@l# X̄ 2187.00 52.00 2196.00 40.00 197.27* 90.92* 99.39 52.00 29.09 6.00
se 40.72 13.93 31.21 3.16 21.66 21.08 13.02 17.44 13.21 6.00
@s# X̄ 198.40 58.00 14.40 2.00 83.50 123.30 58.40 2.00 67.20 66.00
F2
se 59.82 12.81 23.45 2.00 28.84 49.23 21.45 2.00 23.10 29.26
@b# X̄ 364.80 94.00 151.20 44.00 114.80* 117.20 132.80 44.00 60.00 25.00
se 25.66 5.10 30.92 19.13 27.62 44.36 27.05 15.03 16.57 16.58
@F# X̄ 472.00 84.00 475.20 126.00 89.32 186.80* 76.00 24.00 131.20 22.00
se 29.31 5.10 115.89 18.06 43.83 34.53 43.89 17.49 40.86 7.35
@k# X̄ 228.00 64.00 160.80 38.00 ••• ••• 148.00 42.00 87.00 30.00
se 9.38 6.00 86.69 17.15 39.27 18.81 53.86 18.97

ing variations in lip rounding to changes in constriction lo- /la/! than when they are antagonistic ~as for /Fa/!. Negative
cation. As expected ~Sec. B 1!, dark /l/ exerts moderate F2 C-to-V effects in F2 for dark /l/ ~Table II! are consistent
effects on /a/ in spite of the consonant being highly con- with this consonant requiring active tongue-dorsum lower-
strained since the two phonetic segments are produced with ing.
comparable articulatory trajectories and small differences in Analogously to the /{/ condition, C-to-V size effects and
tongue-dorsum position. Thus for consonants specified for C-to-V temporal effects, i.e., CVs 3CVt , are highly corre-
DAC53, C-to-V effects appear to be less prominent when lated in the /a/ condition ~see Table III!. High positive r
the consonant and the vowel gestures are compatible ~as for values were obtained both for Qp ~anticipatory effects! and

TABLE III. Correlations for C-to-V, V-to-C, and V-to-V effects in Qp and F2 across data for seven consonants and five speakers; lowercase s and t indicate
size and temporal effects, respectively. r values at and above 0.70 are given independently for the anticipatory and carryover direction and for each fixed
vowel condition. An asterisk indicates the presence of significant correlations (p,0.05).

Vowel /{/ Vowel /a/

Qp F2 Qp F2

Anticipatory Carryover Anticipatory Carryover Anticipatory Carryover Anticipatory Carryover

C effects
CVs 3CVt 0.94* 0.79 0.84 0.73 0.81 0.84 0.94*
V effects
VCs 3VVs 0.91* 0.99* 0.99* 0.82 1.00*
VCs 3VVt 0.87 0.95 0.96 0.72
VVs 3VVt 0.94* 0.95* 0.91* 0.71

C effects3V effects
CVs 3VCs
CVs 3VVs 0.71 0.71
CVs 3VVt 20.85
CVt 3VVt 20.70 20.79

552 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 552
for F2 ~anticipatory and carryover effects!. A close match value! allow shorter V-to-V anticipatory temporal effects in
between the C-to-V effects in size and temporal extent is F2 frequency than consonants /p/ and /n/ requiring lesser
apparent across consonants in the graphs of Fig. 3~b!. tongue-dorsum activation.
V-to-C size effects in Qp are also most prominent for
consonants /p/ and /n/ ~for /p/./n/! specified for low DAC
B. Vocalic coarticulatory effects values, as well as for /k/ which may be due to vowel-related
changes in velar closure fronting; these effects for /p/ and /k/
1. Fixed /i/ ~also V-to-V size effects in Qp for /k/! are significantly
Figure 4~a! displays the size and the temporal extent of larger than those for other consonants. V-to-C size effects in
vocalic coarticulation in the sequence pairs /iCi/–/iCa/ ~an- F2 are small and show a less coherent picture, with antici-
ticipatory! and /iCi/–/aCi/ ~carryover!, i.e., the size of the patory size effects decreasing in the progression /l/./n/,
vocalic effects at the consonantal midpoint ~V-to-C! and /b/./s/, /F/ for most speakers. V-to-V size effects in Qp and
their size and temporal extent during the fixed vowel /{/ ~V- F2 and V-to-V temporal effects in Qp are very small and
to-V!. Effects are always positive and occur when the Qp or short.
F2 value for /{/ exceeds that for /a/. Correlation data for the /a/ context condition in Table III
The data in Fig. 4~a! show that vowel-dependent effects yield similar results to those for the /{/ condition. High posi-
in Qp and/or F2 tend to be larger and longer for consonants tive r values occur at least once for correlations involving
specified for lower DAC values ~bilabial /p/, alveolar /n/! vocalic effects exclusively, i.e., VCs 3VVs , VCs 3VVt ,
than for consonants requiring more active tongue-dorsum VVs 3VVt , meaning that there is a good correspondence
control ~dark /l/, fricative /s/, alveolopalatals, velars!. This between vowel-dependent size and temporal coarticulation.
finding is in accordance with the degree of coarticulatory An inverse correlation between vocalic coarticulation and
resistance varying directly with the DAC value for the con- consonantal coarticulation was found to hold just for 2 out of
sonant and with the DAC value for /s/ being higher ~3! than 16 correlation pairs ~i.e., CVs 3VVt , CVt 3VVt ). The ab-
that for /n/ ~2!. The size effects for /p/ and /n/ are signifi- sence of high negative correlations between the consonant-
cantly different from those for other consonants; moreover dependent and vowel-dependent effects appears to be mostly
the Qp effects are significantly larger for /p/ vs /n/ which related to the high variability in the coarticulatory effects
accords with DAC differences between the two consonants. across consonants ~given the considerable number of conso-
Correlation data for the /{/ context condition in Table III nants included in the correlation analyses! and to /a/ exhib-
often reveal high positive r values between the size of the iting little dorsopalatal contact ~Qp correlations only!.
V-to-C effects, and the size and the temporal extent of the
V-to-V effects ~i.e., VCs 3VVs , VCs 3VVt !, and between 3. Articulatory dimension of the V-to-C effects
the two latter magnitudes ~i.e., VVs 3VVt !. This trend oc- Data on V-to-C coarticulation in Tables I and II and
curs for F2 both at the anticipatory and carryover levels, and those for the individual speakers reveal a trend for the vowel-
for Qp at the carryover level. Figure 4~a! reveals that con- dependent effects during the consonantal period to favor the
sonants allowing large vocalic size effects are the same ones target articulatory position for the consonant.
allowing long vocalic effects in temporal extent ~/p/, /n/!, A comparison between V-to-C effects in the two tables
while vocalic effects are usually small and short for another reveals that consonants /F/ and /k/ requiring a high tongue-
set of consonants ~dark /l/, /s/, alveolopalatals, velars!. dorsum position allow larger effects in tongue-dorsum rais-
Negative correlation values between the consonantal and ing ~exerted by /{/ in the fixed /a/ context! than in tongue-
the vocalic effects should be taken as evidence that the latter dorsum lowering ~exerted by /a/ in the fixed /{/ context!. This
vary inversely with respect to the former ~see Sec. I C!. As is so for the coarticulatory effects in Qp ~/F/, /k/! and in
shown in Table III, only one high negative correlation ~i.e., F2 ~/F/!. Thus for example, the size of the anticipatory and
CVt 3VVt ! was obtained out of 16 possibilities. Low corre- carryover effects in F2 for /F/ is 89.32 Hz and 186.8 Hz in
lation values occur since the degree of vocalic coarticulation the fixed /a/ context but only 22.4 Hz and 105.48 Hz in the
is inversely related to the degree of C-to-V coarticulation in fixed /{/ context. F2 and dorsopalatal contact (Qp) trajecto-
some cases ~/l/! but not in others ~alveolopalatals, velars!. ries for /F/ in Fig. 5 suggest that the raising movement for
Thus while dark /l/ exerts maximal C-to-V coarticulation and /aFi/ occurs earlier ~it begins during V1 and ends about clo-
allows vowel-dependent effects of little prominence, alveo- sure offset! than the lowering movement for /iFa/ ~it begins
lopalatals and velars exhibit little C-to-V and vocalic coar- during closure or somewhat later and ends during the V2
ticulation. period!. It thus appears that the tongue dorsum early
achieves a high position when /F/ follows /a/ and holds this
position as long as possible when /F/ follows /{/.
2. Fixed /a/
Dark /l/, on the other hand, allows effects in tongue-
Vocalic coarticulatory effects for fixed /a/ are displayed dorsum lowering exerted by /a/ in the fixed /{/ context over
in Fig. 4~b!. F2 effects in temporal extent reveal the ex- those exerted by /{/ in the fixed /a/ condition which accords
pected trend for consonants specified for higher DAC values with this consonant involving active tongue-dorsum lower-
to allow lesser vocalic coarticulation than consonants with ing. This trend is at work for the Qp data but not so for the
lower DAC values. Indeed, dorsal /l/, /F/, /b/, and /k/ and F2 data ~see Tables I and II!. Qp and F2 data in Fig. 5
fricative /s/ ~which appears to be specified for a high DAC reveal indeed that the lowering trajectories for /ila/ begin

553 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 553
FIG. 4. ~a! Mean V-to-C and V-to-V effects across repetitions and speakers for each consonant when the fixed vowel is /{/. Size effects in % of electrode
activation and in Hz ~left: V-to-C, middle: V-to-V! and temporal effects in ms ~right: V-to-V! are plotted separately for Qp ~above! and F2 ~below!, and for
each coarticulatory direction, i.e., anticipatory ~dotted bars! and carryover ~black bars!. ~b! Same representation as in ~a! when the fixed vowel is /a/.

554 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 554
FIG. 5. Q p trajectories ~left! and F2 trajectories ~right! for the sequences /aCi/ ~dashed lines! and /iCa/ ~solid lines! lined up at the onset of the consonantal
period ~0-ms value along the temporal axis!. Short vertical marks crossing the F2 and Qp trajectories have been inserted at the offset of the consonantal
period. Data correspond to a single speaker, i.e., JP ~/l/, /n/, /F/! and DR ~/b/, /p/!.Trajectories for /p/ have been displayed for Qp only ~no formant structure
is available during the closure period for this consonant!; vertical marks have been indicated at closure offset and at burst offset in this case.

555 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 555
earlier ~during V1! and achieve an earlier endpoint ~during
closure! than the raising trajectories for /ali/ ~in this case
raising onset occurs toward the end of V1 or during the cen-
tral closure period and a raising maximum is not achieved
until the V2 period!. It is thus apparent that the tongue
reaches and maintains a low target position for the produc-
tion of dark /l/ which is more compatible with that for a low
vowel than with that for a high vowel.
According to data in Tables I and II, bilabial /p/ (Qp)
and dentoalveolar /n/ (Q p, F2) also allow more prominent
effects from /a/ in the fixed /{/ condition than effects from /{/
in the fixed /a/ condition. Q p trajectories for bilabial /p/ in
Fig. 5 show indeed that, although the increase in lingual
contact for /api/ and the contact decrease for /ipa/ begin more
or less at the same time ~at V1 offset!, the latter ends earlier
~at closure offset! than the former ~during V2!. This finding
suggests that bilabial /p/ is more sensitive to tongue-dorsum
lowering effects than to tongue-dorsum raising effects,
which is consistent with this lingual region exhibiting a low
position at rest. Trajectories for /n/ in the figure show that,
analogously to dark /l/, F2 lowering onset for /ina/ occurs
earlier ~during V1! than F2 raising onset for/ani/ ~at the end
of the closure period!; the offset of these F2 trajectories also
takes place much earlier after /{/ ~at closure onset! than after
/a/ ~during V2!. As for the Q p data, the lowering trajectory
for /ina/ and the raising trajectory for /ani/ start more or less
simultaneously during V1 while their endpoint is achieved
earlier for the former ~at closure offset! than for the latter
~during V2!. Predominance of the vowel-related lowering ef-
fects for /n/ is a quite unexpected outcome since the tongue
dorsum should exhibit a high position during the production
of this consonant if subject to coupling effects with the pri-
mary tongue tip articulator. This finding suggests that /n/ and
perhaps other dentoalveolar consonants may differ with re-
spect to tongue-dorsum height across languages ~i.e., /n/
would be specified for a rather low tongue-dorsum position
in Catalan! and that articulatory regions which are not in-
volved in the closure or constriction making process may be
actively controlled.
Data in Tables I and II reveal that fricatives /s/ and /b/
also favor slightly the effects from /a/ over those from /{/ ~for
Qp and F2 in the case of /s/, and for Q p in the case of /b/!. FIG. 6. ~a! Differences in coarticulatory direction for the C-to-V effects for
The situation for fricatives is exemplified by the Qp and each consonant across repetitions and speakers when the contextual vowel is
F2 trajectories for /b/ in Fig. 5 ~the /s/ trajectories show an /{/. Predominance of the anticipatory component yields a negative value
~below the 0 line!; predominance of the carryover component yields a posi-
analogous behavior to those for /b/ in the figure!. Table II tive value ~above the 0 line!. Size effects in % of electrode activation and in
indicates a slight predominance of the vowel-dependent low- Hz ~left! and temporal effects in ms ~right! are plotted separately for Qp
ering vs raising effects for these consonants ~thus, for ex- ~above! and F2 ~below!. ~b! Same representation as in ~a! when the fixed
ample, V-to-C effects for /s/ amount to 99.2 Hz and 137.4 vowel is /a/.
Hz in the fixed /{/ context, and to 83.5 Hz and 123.3 Hz in
the fixed /a/ context!. These V-to-C data are not in agreement
with the onset times of trajectory movement in Fig. 5 ~since extent of the anticipatory component have been subtracted
raising onset for /abi/ occurs earlier than lowering onset for from those of the carryover component ~see Tables I and II!
/iba/! and may be related to Q p and F2 raising trajectories and the resulting differences have been displayed in Fig. 6~a!
achieving a plateau while the corresponding lowering trajec- and ~b! ~consonant-dependent effects! and in Fig. 7~a! and
tories exhibit continuous motion. ~b! ~vowel-dependent effects!. Differences in sign have not
been taken into account in the subtraction procedure. Nega-
C. Coarticulatory direction tive values in the figures indicate that anticipation prevails
In order to evaluate the relative salience of the two coar- over carryover; positive values indicate the opposite relation-
ticulatory directions, values for the size and the temporal ship.

556 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 556
FIG. 7. ~a! Differences in coarticulatory direction for the V-to-C and V-to-V effects for each consonant across repetitions and speakers when the fixed vowel
is /{/. Predominance of the anticipatory component yields a negative value ~below the 0 line!; predominance of the carryover component yields a positive value
~above the 0 line!. Size effects in % of electrode activation and in Hz ~left: V-to-C, middle: V-to-V! and temporal effects in ms ~right: V-to-V! are plotted
separately for Q p ~above! and F2 ~below!. ~b! Same representation as in ~a! when the fixed vowel is /a/.

557 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 557
1. C-dependent coarticulation 2. V-dependent coarticulation
a. /i/ context. According to Fig. 6~a!, C-to-V effects in a. Fixed /i/. According to Fig. 7~a! highly constrained
Qp and F2 show some noticeable trends in coarticulatory dark /l/, which favors consonant-dependent anticipation, of-
direction. On the one hand, alveolopalatals and velars tend to ten gives priority to the anticipatory direction for the vocalic
favor carryover over anticipatory coarticulation in size and effects in size and temporal extent. On the other hand, con-
temporal extent; the absence of much C-to-V anticipation in sonants assigning more weight to carryover over anticipation
a sequence made of two comparable highly constrained pho- for the C-to-V effects also allow more prominent vowel-
netic segments ~such as /b/, /F/, or /k/ and the vowel /{/! may dependent carryover than anticipation, i.e., /F/ and /b/ ~in
lead to the prevalence of the carryover component associated Qp and F2! and /k/ ~in F2!.
with the mechanico-inertial properties of the tongue-dorsum Consonants giving much less clear priority to a specific
raising gesture. On the other hand, dark /l/ exerts larger ~but direction in consonant-dependent coarticulation favor no ob-
not longer! anticipatory than carryover effects; the salience vious direction when the vowel-dependent effects are taken
of the C-to-V anticipatory effects in this case appears to be into consideration. This is so for most speakers in the case of
linked to the performance of the highly constrained lingual /p/ ~which shows prevalence of the carryover direction in
gesture for the consonant when preceded by the antagonistic Qp and of the anticipatory direction in F2 size for four
vowel /{/. speakers! and of /s/ ~which favors the anticipatory compo-
Speaker-dependent Q p and F2 data reveal no clear nent for the Qp size effects and the carryover component for
coarticulatory direction pattern for the alveolars /s/ and /n/ the F2 effects!. While exhibiting a similar behavior to /p/
although most speakers favor the carryover direction in some and /s/ in C-to-V directionality, alveolar /n/ gives some more
instances. Size and temporal effects for the bilabial /p/ also weight to the carryover over the anticipatory component
show predominance of the carryover component which may component for the vocalic effects.
be related to the influence of the bilabial release and, as for b. Fixed /a/. Qp and F2 data in Fig. 7~b! reveal the
the other nondorsal consonants /n/ and /s/, to the inertial existence of maximal vocalic anticipation for consonants
~/p/, /n/, /l/! giving preference to no clear C-to-V coarticula-
requirements associated with the raising of the tongue dor-
tory direction ~/p/! or to C-to-V anticipation ~/n/, /l/!. There
sum induced by the vowel /{/.
is a trend to favor more prominent vocalic effects in F2 size
b. /a/ context. Q p and F2 data in Fig. 6~b! show that
and temporal extent and in Qp temporal extent at the antici-
fricatives resemble other constrained consonants in exerting
patory vs carryover level for /p/, /l/, and /n/; somehow
strong gestural anticipation: Size and temporal data reveal
V-to-C trends in Qp for /p/ and /n/ are larger at the carryover
that /s/ and /b/ often favor the C-to-V anticipatory compo-
level. Velar /k/ also tends to favor the anticipatory compo-
nent. To a lesser extent, predominance of anticipation over
nent for the vocalic effects whether related to a more salient
carryover is also at work for the alveolars /n/ ~size and tem-
consonant-dependent carryover component ~for Qp! or an-
poral coarticulation in Q p and F2! and /l/ ~coarticulation in
ticipatory component ~for F2!; preference for vocalic antici-
F2 temporal extent for four speakers!. Overall, alveolars
pation for /k/ in the fixed /a/ context condition follows from
may give preference to the anticipatory component in line specific constraints in the production of velars ~see Sec. III!.
with the tongue dorsum being ruled by apical activity and Alveolopalatal /F/, which exerts longer C-to-V effects at the
not being subject to strong mechanico-inertial constraints carryover level than at the anticipatory level, is the only con-
during their production in the /a/ context; consonants subject sonant clearly allowing more prominent vowel-dependent
to important manner requirements, i.e., fricatives, give par- carryover vs anticipatory coarticulation. Fricatives, which
ticular emphasis to the anticipatory direction. clearly exert more prominent anticipatory versus carryover
More retracted dorsal consonants exhibit a different be- effects, may give more weight to either the vowel-dependent
havior. Alveolopalatal /F/ favors anticipatory size effects in anticipatory or the carryover component: There is a trend for
Qp and also in F2 @for four speakers in spite of Fig. 6~b! not /s/ to favor Qp anticipation and F2 carryover; as for /b/,
showing so since one speaker exhibits very large carryover vocalic size effects are quite small for Qp and may give
effects#, presumably because the articulatory manifestation priority to the anticipatory direction for F2.
of the dorsal gesture is more /j/-like at consonantal release
than at consonantal formation; temporal effects in Qp and
F2 for /F/ are however longer at the carryover versus antici-
patory level. Regarding velar /k/, C-to-V effects are larger III. GENERAL SUMMARY AND DISCUSSION
and longer for the carryover vs anticipatory direction ~Qp!
and for the anticipatory versus carryover direction (F2). The Data reported in Sec. II confirm the validity of the DAC
finding of more prominent carryover than anticipatory effects model of coarticulation in many respects. Consonants and
for dorsals vs alveolars is in accordance with the mechanico- vowels were assigned different DAC values depending on
inertial properties associated with the tongue-dorsum raising the degree of tongue-dorsum constraint during their produc-
gesture for the former consonantal class. tion, i.e., 3 to alveolopalatals, palatals, velars, and dark /l/, 1
No clear coarticulatory direction holds for bilabial /p/ to bilabials and the schwa, and 2 to those consonants and
which favors anticipation in F2 size and perhaps carryover vowels the production of which involves some uncontrolled
in Qp size ~longer carryover vs anticipatory effects in Qp tongue-dorsum activity ~/n/, /a/!. It was also suggested that
for this consonant are associated with two speakers only!. manner requirements could raise the DAC value ~e.g., from 2

558 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 558
to 3 for /s/!. As shown next, this DAC scale accounts to a effects in the former vowel context are larger and longer for
large extent for the C-to-V, V-to-C, and V-to-V effects re- /p/ and /n/ and smaller and shorter for /s/ and dorsals includ-
ported in the present study. ing dark /l/!. It thus appears that the extent of the consonant-
An important finding was that the prominence of the and vowel-dependent effects along the time domain is re-
C-to-V effects depends both on the relative DAC specifica- lated to their magnitude ~see also Farnetani and Recasens,
tion for the two adjacent phonetic segments as well as on 1993!.
their articulatory trajectories being compatible or antagonis- The notion that the articulatory dimension of the C-to-V
tic. When the consonant and the vowel are specified for the effects should depend on the target articulatory position for
same DAC value, C-to-V effects are negligible in a situation the consonant was also confirmed. Articulatory activity asso-
of gestural compatibility ~e.g., /Fi/! and become more promi- ciated with the target tongue position often begins earlier and
nent as the degree of gestural antagonism increases ~e.g., for is held for a longer period than that for other tongue posi-
/li/ vs /na/, /sa/!. When the consonant is specified for a higher tions. Thus consonants requiring tongue-dorsum raising ~al-
DAC value than the vowel, C-to-V effects increase with the veolopalatals, velars! are more sensitive to effects from /{/ vs
degree of gestural antagonism involved ~e.g., for /Fa/ vs /la/!. /a/ while those involving tongue-dorsum lowering favor ef-
Negligible C-to-V coarticulation occurs when the DAC value fects from /a/ vs /{/ ~bilabial /p/, alveolar /n/, dark /l/!. Some
for the vowel exceeds that for the consonant ~e.g., for /pi/, justification for /p/ and /n/ involving tongue-dorsum lower-
/pa/!. More prominent C-to-V effects in tongue-dorsum low- ing is needed: Tongue-dorsum relaxation for /p/ may occur
ering than expected for /si/ indicate that manner of articula- concomitantly with lip closing ~see Engstrand, 1988!, while
tion may modify the basic DAC value ~from 2 to 3 in this tongue-dorsum lowering for /n/ is possibly a language-
case!. It thus appears that gestural antagonism does not only specific characteristic ~see Footnote 8 for other possible ex-
prevent coarticulatory effects from occurring ~as postulated planations!. Fricatives /s/ and /b/ were found not to clearly
by several coarticulation theories: Henke, 1966; Bell-Berti favor either tongue-dorsum raising effects or tongue-dorsum
and Harris, 1981! but, on the contrary, may cause C-to-V lowering effects associated with the adjacent vowel; this par-
coarticulation to increase. Our finding is consistent with dif- ticular behavior may be related to fricatives requiring the
ferences in the production mechanisms of consonants and formation of a precise medial groove which causes the
vowels: The formation of a closure or constriction requires tongue-dorsum surface to occupy an intermediate position
consonants to be actualized during the adjacent vowel and between that for high and low consonantal articulations.
such production requirements become especially relevant An important goal of this study was to test the implica-
with the difficulty involved in the making of the vowel-to- tions of the DAC model with respect to coarticulatory direc-
consonant transition. It can be claimed that maximal antago- tionality. C-to-V data presented in the paper confirm that,
nism results in maximal C-to-V coarticulation so as to ensure among consonants specified for a high DAC value, some
that the consonantal gesture is successfully realized and that, favor the anticipatory component while others favor the
for a given DAC value, consonants are more constrained carryover component. Prevalence of the anticipatory compo-
than vowels. nent for dark /l/ follows from this consonant involving the
V-to-C coarticulation data reported in Sec. II are consis- formation of two lingual constrictions; such anticipatory ef-
tent with the notion that V-to-C sensitivity should vary in- fects are especially salient when tongue-dorsum lowering
versely with the DAC value for the consonant, i.e., in the and retraction for dark /l/ needs to be made during the pre-
progression [Link], ve- ceeding antagonistic vowel /{/. Consonants produced with
lars, dark /l/, /s/ ~which is in support of /s/ being specified for active tongue-dorsum raising, i.e., the alveolopalatal /F/ and
DAC53!. This progression was found to hold in both fixed the velar /k/, favor the carryover over the anticipatory com-
vowel conditions in spite of the initial expectation that it ponent due to the tongue dorsum being lowered more slowly
should be at work when the fixed vowel is /a/ but not neces- at consonantal release than raised at consonantal onset, and
sarily so when it is /{/ since a highly constrained fixed vowel thus in line with the tongue dorsum mechanico-inertial re-
could prevent much V-to-V coarticulation from occurring. quirements for the consonant; moreover, this asymmetrical
However, there were very few high negative correlations be- relationship becomes more obvious as the articulatory dis-
tween the vocalic effects and the consonantal effects in both tance between the vowel and the consonant increases, i.e.,
vowel conditions, and thus no inverse relationship between for /F/ and /k/ with adjacent /a/ vs /{/. Coarticulatory direc-
the strength of the two coarticulatory types. This finding has tionality for other consonants appears to depend on whether
been attributed to several factors often acting concomitantly the vowel causes the tongue dorsum to be raised or not dur-
such as the high number of consonants submitted to analysis ing the consonant. Prevalence of the anticipatory direction
or the coexistence of little consonantal and vocalic coarticu- occurs in the context of /a/ when tongue-dorsum raising is
lation in some scenarios ~e.g., in sequences with alveolopala- absent, presumably in line with the flexibility of the apical
tal consonants and the vowel /{/!. articulator ~for the alveolar /n/! and with lingual grooving
High correlations were obtained between C-to-V effects associated with manner of articulation demands ~for the fri-
in size and temporal extent meaning that larger effects are catives /s/ and /b/!; tongue-dorsum raising for /n/, /s/, /b/, and
usually longer ~e.g., for /li/, /Fa/! while smaller effects are /p/ with adjacent /{/ contributes to an increase in the
often shorter ~e.g., for /p/, /n/, /l/, and /s/ with adjacent /a/!. A mechanico-inertial requirements which results in a more
similar relationship appears to hold for the vocalic effects, prominent dorsal release and a more salient carryover vs an-
more so when the fixed vowel is /{/ than when it is /a/ ~e.g., ticipatory component.

559 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 559
 Our model predicts that the direction of the vocalic ef- location and narrowing ~Fant, 1960!; they will be reported occasionally
fects should conform to the directionality trends for conso- when deviating from those for F2 in some interesting way. F1 coarticula-
nantal coarticulation. Predominance of the V2-dependent an- tory effects will not be commented in this paper since they differ from
those for F2 in many respects presumably because of reflecting jaw open-
ticipatory effects for /l/ occurs presumably since vowel ing variations in addition to changes in dorsopalatal constriction width.
anticipation is not much affected by the carryover compo- 6
The following piece of evidence suggests that differences in phonetic real-
nent for this consonant; on the other hand, the salience of the ization between V1 and V2 in the /aCa/ sequences ~i.e., V1 was realized as
vowel-dependent carryover effects in the case of /F/ accords stressed @a# and V2 as unstressed @.#! did not affect the analysis results
significantly. Had this been the case, V-to-V carryover effects would have
with the mechanico-inertial constraints associated with the exceeded the corresponding anticipatory effects in /VpV/ sequences ~where
tongue-dorsum raising gesture preventing much vowel- /p/ is specified for a minimal DAC value! since there are reasons to expect
related anticipation from occurring. Vocalic anticipation is stressed vowels to exert more coarticulation than unstressed ones and non-
also blocked when fixed /{/ contributes to the raising of the reduced vowels to allow less coarticulation than reduced ones. V-to-V coar-
ticulation data across /p/ on fixed /a/ reported in Sec. II reveal that the
tongue dorsum during the consonant; this would explain why opposite in fact happened, i.e., anticipatory size and temporal effects were
consonantal and vocalic effects for /n/, /b/, and perhaps /p/ found to be more prominent than carryover effects. More carryover than
are more prominent at the carryover level when adjacent to anticipatory V-to-V coarticulation across /!/ in the fixed /{/ context is prob-
ably independent of the stress pattern since stressed and unstressed /{/ differ
/{/ and at the anticipatory level when adjacent to /a/. The
very slightly in Catalan.
absence of a clear directionality pattern for the vocalic ef- 7
The exclusion of that row from the calculation of the Qp index for the
fects for /s/ in the two fixed vowel contexts may be attributed latter four consonants is also advisable when their place of articulation is
to the strong manner requirements for this consonant and more front than usual since the dorsal contact at the front palatal zone in
perhaps to the mechanico-inertial requirements on the tongue this case is presumably related to interarticulatory coupling effects with the
tongue front rather than to tongue-dorsum activation.
dorsum in the /{/ context. Analogously to dorsal /F/, dorsal 8
In a few instances, reinforcement of the tongue-dorsum raising gesture by
/k/ favors the carryover component for the consonantal and the action of successive dorsopalatal V1 and C in the sequences /{FV/ and
vocalic effects in the /{/ condition; prevalence of vocalic an- /{bV/ may cause articulatory overshoot, i.e., more dorsopalatal contact and
frequency displacement than usual for a long period of time during V2.
ticipation in the /a/ context condition appears to be related to
Carryover effects extending until vowel offset in these circumstances have
how the tongue dorsum is being controlled during velar been excluded from analysis.
closure.12 9
In this paper, a given coarticulatory effect will be often characterized in
comparison to other effects as ‘‘large’’ or ‘‘small’’ ~size! and ‘‘long’’ or
‘‘short’’ ~temporal extent!.
ACKNOWLEDGMENTS 10
C-to-V effects in F3 for dorsal /F/ and /%/ are somewhat larger and longer
than F2 effects. This finding is consistent with information provided by
This work was supported by ESPRIT Basic Research nomograms ~Fant, 1960! showing that a considerable reduction of the
Action 6975 project from EC ~‘‘Speech Maps’’!, CS93- constriction area for alveolopalatals causes more F3 than F2 raising as the
9.908 research grant of the Government of Catalonia, and constriction location moves forward from the mediopalate to the prepalate.
project CE93-0020 of the Ministry of Education and Science It also accords with negative F3 effects if, as indicated by electropalato-
of Spain. The authors would like to thank Anders Löfqvist graphic data ~Recasens, 1991a!, some constriction retraction for /{/ in this
specific consonantal environment is accompanied by a decrease in lingual
and three anonymous reviewers for comments on a previous contact at the prepalate.
version of the manuscript. 11
Other explanations could account for the Qp and F2 lowering for /'/ and
/2/. Apical consonants have been reported to be articulated with a some-
1
The term ‘‘tongue dorsum’’ is preferred to the term ‘‘tongue body’’ and what concave tongue surface behind closure location and some postdor-
refers to the upper surface of the tongue in front of the hard and soft palate sum retraction resulting into a narrower pharyngeal passage than that ex-
where palatal and velar sounds are articulated ~Catford, 1988!. hibited by laminoalveolar articulations ~Dart, 1991!. Some tongue-dorsum
2
In a previous paper ~Recasens et al., 1995!, the degree of dorsopalatal relaxation could also be related to nasality and the resulting lack of pres-
contact was correlated with F2 at the consonantal midpoint in the se- sure containment associated with it ~Perkell, 1969!.
12
quences /iCi/ and /aCa/ for the same consonants analyzed in the present The salience of the anticipatory component for the vocalic effects is con-
study. High positive correlations were found across consonants and speak- sistent with ultrasound data showing an early onset of V2 related tongue-
ers as well as across consonants for a given speaker. This finding is con- dorsum movement during V1 ~Parush et al., 1983!, and with electromag-
sistent with data collected with vocal tract analogues indicating an increase netic midsagittal articulometry data showing V2 anticipation in /~%V/
in F2 frequency with a decrease in dorsopalatal constriction width ~Stevens sequences during closure mostly for V25/i/ but also for V25/~/
and House, 1955; Fant, 1960; Gay et al., 1991!. ~Mooshammer et al., 1995!. Our EPG data generally show more dorso-
3
The DAC scale is not to be identified with the CR ~coarticulatory resis- palatal contact during V1 for /~%{/ than for /{%~/ which is consistent with a
tance! scale proposed elsewhere ~Bladon and Al-Bamerni, 1976!. The CR strong V2-dependent forward movement in the former sequence already
scale assigns different values to phonetic segments depending on the degree during V1. It is agreed that these anticipatory effects reflect passive for-
of coarticulation allowed. The following CR values correspond to different ward tongue-dorsum movement associated with the large mass of the
allophones of English /l/: 1 to the least resistant allophone, i.e., clear /l/ in tongue body being controlled in a continuous fashion ~Perkell, 1969!. This
leaf; 2 to an allophone specified for an intermediate CR degree, i.e., dark /l/ forward movement does not occur to the same extent when V15/i/ since
in feel; 3 to the most resistant allophone, i.e., syllabic /l/ ~fiddle!. The DAC the tongue dorsum is already located at a front location at closure onset in
scale characterizes phonetic segments according to the types of articulatory this case.
constraints involved in their production and can be used as a predictor of
coarticulatory resistance ~see Sec. B 2!.
4
The term ‘‘fixed vowel’’ is applied to VCV pairs used to investigate vowel-
Beddor, P. S., and Yavuz, H. K. ~1995!. ‘‘The relationship between vowel-
dependent coarticulation. It refers to the vowel which is kept constant in a
to-vowel coarticulation and vowel harmony in Turkish,’’ in Proceedings
given VCV pair, e.g., V15/a/ in the sequence pair /api/-/apa/ which allows
of ICPhS95, Vol. 2, edited by K. Elenius and P. Branderud ~Arne Ström-
studying right-to-left effects associated with V25/i/ vs /a/ in this particular bergs Grafiska, Stockholm!, pp. 44–51.
V1 condition. Bell-Berti, F., and Harris, K. S. ~1981!. ‘‘A temporal model of speech pro-
5
F3 and F1 coarticulatory effects were also measured for the same speech duction,’’ Phonetica 38, 9–20.
material. Results for F3 were highly similar to those for F2 which is in Bergem, D. R. van ~1994!. ‘‘A model of coarticulatory effects on the
accordance with both formants being related to dorsopalatal constriction schwa,’’ Speech Commun. 14, 143–162.

560 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 560
Bladon, R. A. W., and Al-Bamerni, A. ~1976!. ‘‘Coarticulation resistance in Lindblom, B. ~1983!. ‘‘Economy of speech gestures,’’ in The Production of
English /(/,’’ J. Phonetics 4, 137–150. Speech, edited by P. F. MacNeilage ~Springer-Verlag, New York!, pp.
Boyce, S. E. ~1990!. ‘‘Coarticulatory organization for lip-rounding in Turk- 217–245.
ish and American English,’’ J. Acoust. Soc. Am. 88, 2584–2595. Mooshammer, C., Hoole, P., and Kuhnert, B. ~1995!. ‘‘On loops,’’ J. Pho-
Browman, C. P., and Goldstein, L. ~1986!. ‘‘Towards an articulatory pho- netics 23, 3–22.
nology,’’ Phonology 3, 219–252. Öhman, S. ~1966!. ‘‘Coarticulation in VCV sequences,’’ J. Acoust. Soc.
Browman, C. P., and Goldstein, L. ~1992!. ‘‘ ‘Targetless’ schwa: An articu- Am. 39, 151–168.
latory analysis,’’ in Papers in Laboratory Phonology II. Gesture, Segment, Parush, A., Ostry, D. J., and Munhall, K. G. ~1983!. ‘‘A kinematic study of
Prosody, edited by G. J. Docherty and R. Ladd ~Cambridge U. P., Cam- lingual articulation in VCV sequences,’’ J. Acoust. Soc. Am. 74, 1115–
bridge!, pp. 26–56. 1125.
Carney, P. J., and Moll, K. L. ~1971!. ‘‘A cinefluorographic investigation of Perkell, J. S. ~1969!. Physiology of Speech Production:Results and Implica-
fricative consonant-to-vowel coarticulation,’’ Phonetica 23, 193–202. tions of a Quantitative Cineradiographic Study ~MIT, Cambridge!.
Catford, J. C. ~1988!. A Practical Introduction to Phonetics ~Clarendon, Recasens, D. ~1984!. ‘‘V-to-V coarticulation in Catalan VCV sequences,’’ J.
Oxford!. Acoust. Soc. Am. 76, 1624–1635.
Dart, S. N. ~1991!. ‘‘Articulatory and acoustic properties of apical and lami- Recasens, D. ~1985!. ‘‘Coarticulatory patterns and degrees of coarticulatory
nal articulations,’’ UCLA Working Papers in Phonetics 79. resistance in Catalan VCV sequences,’’ Language Speech 28, 97–114.
Engstrand, O. ~1988!. ‘‘Articulatory correlates of stress and speaking rate in Recasens, D. ~1987!. ‘‘An acoustic analysis of V-to-C and V-to-V coarticu-
Swedish VCV utterances,’’ J. Acoust. Soc. Am. 83, 1863–1875. latory effects in Catalan and Spanish VCV sequences,’’ J. Phonetics 15,
Fant, G. ~1960!. Acoustic Theory of Speech Production ~Mouton, The 299–312.
Hague!. Recasens, D. ~1989!. ‘‘Long range coarticulatory effects for tongue dorsum
Farnetani, E., and Recasens, D. ~1993!. ‘‘Anticipatory consonant-to-vowel contact in VCVCV sequences,’’ Speech Commun. 8, 299–312.
coarticulation in the production of VCV sequences in Italian,’’ Language Recasens, D. ~1990!. ‘‘The articulatory characteristics of palatal conso-
Speech 36, 279–302. nants,’’ J. Phonetics 18, 267–280.
Fowler, C. A., and Saltzman, E. ~1993!. ‘‘Coordination and coarticulation in Recasens, D. ~1991a!. ‘‘An electropalatographic and acoustic study of
speech production,’’ Language Speech 36, 171–195. consonant-to-vowel coarticulation,’’ J. Phonetics 19, 179–192.
Gay, T. ~1974!. ‘‘A cinefluorographic study of vowel production,’’ J. Pho- Recasens, D. ~1991b!. ‘‘Organización articulatoria durante la producción de
netics 2, 255–266. algunas secuencias VCV del castellano,’’ Rev. Argentina Lingü. 7, 73–88.
Gay, T. ~1977!. ‘‘Articulatory movements in VCV sequences,’’ J. Acoust. Recasens, D. ~in press!. ‘‘Lingual coarticulation,’’ in Coarticulation, edited
Soc. Am. 62, 183–193. by W. J. Hardcastle ~Cambridge U. P., Cambridge!.
Gay, T., Boë, L. J., Perrier, P., Feng, G., and Swayne, E. ~1991!. ‘‘The Recasens, D., Pallarès, M. D., and Fontdevila, J. ~1995!. ‘‘Co-articulatory
acoustic sensitivity of vocal tract constrictions: a preliminary report,’’ J. variability and articulatory-acoustic correlations for consonants,’’ Eur. J.
Phonetics 19, 445–452. Disorders Commun. 30, 203–212.
Hardcastle, W. J., Jones, W., Knight, C., Trudgeon, A., and Calder, G. Recasens, D., and Romero, J. ~1997!. ‘‘An EMMA study of segmental com-
~1989!. ‘‘New developments in electropalatography: A state-of-the-art re- plexity in alveolopalatals and palatalized alveolars,’’ Phonetica 54, 43–58.
port,’’ Clin. Ling. Phon. 3, 1–38. Sproat, R., and Fujimura, O. ~1993!. ‘‘Allophone variation in English /l/ and
Henke, W. L. ~1966!. ‘‘Dynamic Articulatory Model of Speech Production its implications for phonetic implementation,’’ J. Phonetics 21, 291–312.
using Computer Simulation,’’ Ph.D. Dissertation, Massachussetts Institute Stevens, K. N., House, A. S., and Paul, A. P. ~1966!. ‘‘Acoustical descrip-
of Technology. tion of syllabic nuclei: An interpretation in terms of a dynamic model of
Kent, R. D., and Moll, K. L. ~1972!. ‘‘Cinefluorographic analyses of se- coarticulation,’’ J. Acoust. Soc. Am. 40, 123–132.
lected lingual consonants,’’ J. Speech Hear. Res. 15, 453–473. Stevens, K. N., and House, A. S. ~1955!. ‘‘Development of a quantitative
Kiritani, S., Itoh, K., Hirose, H., and Sawashima, M. ~1977!. ‘‘Coordination description of vowel articulation,’’ J. Acoust. Soc. Am. 27, 484–493.
of the consonant and vowel articulations—X-ray microbeam study on Stone, M., Faber, A., Raphael, L. J., and Shawker, T. H. ~1992!. ‘‘Cross-
Japanese and English,’’ Ann. Bull. Res. Inst. Logopedics Phonatrics ~Uni- sectional tongue shape and linguopalatal contact patterns in @s#, @b#, and
versity of Tokyo! 11, 11–21. @l#,’’ J. Phonetics 20, 253–270.
Kondo, Y. ~1994!. ‘‘Phonetic underspecification in schwa,’’ in ICSLP94, Wada, T., Yasumoto, M., Iteoka, N., Fujiki, Y., and Yoshinaga, R. ~1970!.
Vol. 1 ~The Acoustical Society of Japan!, pp. 311–314. ‘‘An approach to the cinefluorographic study of articulatory movements,’’
Krakow, R. A. ~1989!. ‘‘The Articulatory Organization of Syllables: A Ki- Cleft Palate J. 7, 506–522.
nematic Analysis of Labial and Velar Gestures,’’ Ph.D. dissertation, Yale Wood, S. ~1979!. ‘‘A radiographic analysis of constriction locations for
University. vowels,’’ J. Phonetics 7, 25–43.

561 J. Acoust. Soc. Am., Vol. 102, No. 1, July 1997 Recasens et al.: A model of lingual coarticulation 561

View publication stats