Anticipatory Coarticulation and Aphasia: Implications for Connectionist Models of Speech Production

_____________________________________________________________

NOTE: In the email version, there are no special characters
for IPA symbols. Thus, phonetic transcriptions include symbols
which should be translated into the following:

[*] = schwa
[E] =  epsilon (lax, central vowel)
[o:] = mid front rounded vowel, half-close
--------------------------------------------------------------

Introduction

     Much of human behavior may be  described  as  involving
serially  ordered  processes.  This  is  true for both motor
behavior as well as perception and comprehension. What makes
serially  ordered  aspects of behavior difficult to model is
that although discrete stages of activity may be psychologi-
cally  perceptible,  actual behavior is rarely context-free.
That is, we have the ability to perform the "same" types  of
motor  movements  in vastly differing real world situations,
as well as to decode "single" units of meaning out of highly
parallel information streams.

     Nowhere has this phenomenon been more apparent than  in
speech  production and perception. Even within the component
of  language  traditionally  thought  to  involve  the  most
"bottom-up"  information, i.e., phonology, one is confronted
with information present in a highly context-dependent form.
Speech  has  remained one of the more problematic aspects of
human communication to study because it has  been  extremely
difficult  to identify the units corresponding to components
of phonological analysis in either articulatory gestures  or
in  the  acoustic waveform. At the core of this problem is a
phenomenon known as  "coarticulation"  (or  "co-production",
Fowler,  1980). This refers to the fact that speakers do not
string together  discrete  sound  segments  as  beads  on  a
string,  but rather overlap speech sounds in a graded, time-
compressed manner.

     Coarticulation may  be  generally  classified  as  per-
severatory  ("backwards",  "left-to-right")  or anticipatory
("forwards, "right- to-left").  Anticipatory  coarticulation
is  of  special interest to speech researchers because it is
considered to be a  measure  of  the  planning  of  upcoming
speech  segments.  By examining speakers' ability to antici-
pate articulatory configurations it  has  been  possible  to
gain  insight  into the nature of the speech sequencing pro-
cess.

     Speech production has traditionally been  divided  into
planning  and execution processes. These processes have been
commonly given names borrowed from linguistic theory,  i.e.,
"phonemic"  (or selectional), and "phonetic" (motor output).
Because coarticulation involves the translation of  linguis-
tic  targets  into speech articulator movements, the process
entails both  phonemic  and  phonetic  information.  It  has
therefore  remained controversial whether these regularities
should be captured in a linguistic grammar  (cf.  Browman  &
Goldstein,  1986;  Keating,  in  press;  Fowler,  in press).
Importantly, the phonemic/phonetic nature of  coarticulation
has also made it necessary to study cognitive and linguistic
representations, as well as a broad range of  physical  data
(e.g.,  the kinematic properties of the speech articulators)
in order to best capture the facts.

     A recent breakthrough in modeling cognitive aspects  of
coarticulation has been made by Jordan (1986), who simulated
anticipatory rounding and nasalization in a small corpus  of
French and English words and phrases. Jordan proposed a con-
nectionist model which receives as input a  "plan"  (in  the
form of featural specifications for individual phonemes) and
yields  as  output  a  list  of  coarticulatory  constraints
presented  occurring serially across time. Jordan's model is
based  upon  previous  PDP   frameworks,   incorporating   a
recurrent  network  of  processing  units, "hidden units" to
capture nonlinear response  patterns  (Hinton  &  Sejnowski,
1983),  and a back propagation learning algorithm to provide
training (Rumelhart, Hinton, & Williams, 1986). This  model,
however,  has  provided  two  key innovations: First of all,
input was separated into state and plan units. This  permit-
ted  nominally  serial  properties  to  be  modelled without
recall to  explicit  temporal  order  or  action  sequences.
Secondly,  plan  units  (representing  phoneme strings) were
designed to include "don't care" conditions. This  permitted
anticipated features (e.g., rounding) to spread over certain
phonemic segments and not others.

     Jordan was careful to point out that the model was  not
intended to be a "realistic model of speech production", due
to the fact that complex lower-level (motoric) processes are
also  involved in actual speech. Moreover, it was noted that
the same regularities captured in the connectionist approach
could  also  be  explained  by means of "traditional" models
(e.g.,  feature  spreading,  Henke,   1966).   Nevertheless,
Jordan's  model  was  able to predict the degree of featural
spread (i.e., the "boundary conditions" for articulators) in
a  manner consistent with the empirical data, and was there-
fore claimed to provide a more parsimonious account of coar-
ticulation.  The  success  of  this initial attempt suggests
that more elaborated connectionist models might  be  greatly
useful in coarticulation research.

     Numerous empirical studies have been conducted  analyz-
ing the acoustic, perceptual, and kinematic details of anti-
cipatory coarticulation in the speech of normal adults  (see
Sharf  and  Ohde,  1981; Lubker and Gay, 1982, for reviews).
These  studies  have  addressed   important   claims   about
language-dependent  features  of  speech  planning, and have
yielded  important  information  about  the  properties   of
individual  speech  articulators.  In  addition, researchers
have recently begun to examine the development of  coarticu-
lation  in normal children (Repp, 1986; Sereno et al., 1987;
Nittrouer, 1989a; 1989b; Katz, Kripke, and  Tallal,  1989a),
as  well  as in children presenting with language impairment
(Hewlett, 1988; Katz, Kripke, and Tallal, 1989b).

     This article will focus primarily upon studies of adult
aphasic  subjects  (i.e.,  subjects presenting with specific
damage  to  brain  regions  known  to  subserve  speech  and
language function).  This research has sought to make use of
"experiments in nature" to analyze  a  number  of  important
questions  about  speech  behavior.   First of all, the data
address whether the sound sequencing capabilities  necessary
for   naturally  coarticulated  speech  are  linked  to  the
integrity of particular neural structures in the  brain.  It
is  possible to investigate, for example, hypotheses stating
that anterior brain regions (esp. "Broca's Area")  are  pre-
ferentially  involved  in  speech  motor  planning (Mlcoch &
Noll, 1980, Kimura and Watson, 1989). In addition, the coar-
ticulatory  patterns of aphasic speech have been viewed as a
possible means of determining the extent  of  normal  speech
motor  planning  processes.   That is, by observing the con-
straints governing aphasic speech breakdown, it may be  pos-
sible  to  infer  specific planning processes at work in the
normal brain. Finally, it  has  also  been  of  interest  to
determine  whether  coarticulatory  data  comport with known
patterns of  "phonemic"  and  "phonetic"  disintegration  in
aphasic speech.

     In this paper, I shall first provide a brief summary of
the facts known about coarticulation in normal adult speech.
This will be followed by  a  short  description  of  aphasic
speech  characteristics, including facts known about coarti-
culation. The body of this report will be concerned with new
research  addressing labial, lingual, and velar anticipatory
coarticulation in normal and  aphasic  German-speaking  sub-
jects.  The  results  of  these investigations indicate that
anterior aphasic subjects show essentially intact  anticipa-
tory  coarticulation.  These data will be discussed in light
of current models of speech production.

Coarticulation in normal, adult speech

     The most widely studied forms of anticipatory coarticu-
lation  involve  the  motion of the lips, tongue, and velum.
Anticipatory labial coarticulation typically  involves  ini-
tiation  of  a  rounding gesture during consonant production
preceding a rounded vowel. For example, a speaker  producing
the  English  syllable [su] will round his lips at the start
of the [s] in anticipation of the rounded vowel [u]. In con-
trast,  no rounding occurs when he is producing the syllable
[si].

     Studies of anticipatory  lingual  coarticulation  typi-
cally  focus  on the front/back positioning of the tongue in
velar stop closure as a function of the  feature  specifica-
tion  of  the  following  vowel.   For instance, the English
phoneme /k/ has a front allophone with a relatively anterior
vocal  tract constriction, and a back allophone with a rela-
tively posterior constriction.

     Anticipatory  velar  coarticulation  describes  effects
upon  velar height as a function of the nasality features of
upcoming phones.  For example, the presence of the nasalized
consonant  [n]  in the English word "pent" may begin shortly
after the initial [p], effectively nasalizing the vowel [E].

     Coarticulation has been studied by means  of  acoustic,
perceptual  and kinematic analyses. These studies have indi-
cated that the timing and extent of anticipatory coarticula-
tion is language-particular. Thus, one finds that lip round-
ing is anticipated earlier and  with  greater  precision  in
Swedish  (a  language  having  an  elaborate  set of rounded
vowels) than in English (which has relatively  little  vowel
rounding).  It  has  also  been  shown  that  within a given
language there may be substantial individual variation, with
some  subjects  showing  more  "feature- based" anticipatory
patterns, and others showing evidence of anticipation over a
"phase-locked",  single  window of time (see Lubker and Gay,
1982, for details).

Speech production in aphasia

     The major  subdivision  between  aphasic  syndromes  is
based  upon  the  character  of speech output (Goodglass and
Kaplan, 1982).  Aphasic subjects with lesions in  the  ante-
rior  portion  of  the brain generally present speech marked
with difficulties in the initiation and sequencing of  arti-
culatory movements.  These symptoms are generally considered
to be an integral part of a nonfluent aphasia,  referred  to
as  "Broca's  aphasia." There may also be agrammatism, which
is a reduction  and  simplification  of  grammatical  forms,
including  the  loss of small function words.  These factors
add up to a speech pattern which is halting  and  effortful,
generally termed "dysfluent."

     In contrast, aphasics with posterior lesions  generally
present  with  speech  which is fluent and well-articulated,
though semantically impoverished. These  deficits  are  gen-
erally  considered  to  be an integral part of two posterior
aphasia syndromes (Wernicke's aphasia, Conduction  aphasia).
Wernicke's  aphasia involves a severe impairment of auditory
and written comprehension, and speech which is  semantically
"empty"  and  difficult to understand. There is an abundance
of high-frequency, low-content words (e.g., "thing",  "it"),
and  a  reduction  of  substantive nouns and verbs to convey
meaning. The speech of these patients typically shows verbal
and phonemic paraphasias. There may be a phenomenon known as
"press for speech" (or "logorrhea") in  which  the  patient,
even  in a conversation setting, produces copious amounts of
speech without stopping. The syndrome of Conduction  aphasia
is  qualitatively  similar to Wernicke's aphasia in a number
of respects, however, the chief difference is  that  repeti-
tion is greatly impaired in relation to the level of fluency
of spontaneous speech.

     Traditional clinical descriptions of  aphasia  consider
the errors in speech produced by anterior (Broca's) aphasics
to reflect phonetic  or  articulatory  errors,  whereas  the
errors  of posterior (e.g., Wernicke's) aphasics are thought
to originate at the phoneme planning level. In recent years,
fine-grained  acoustic  analyses  have  uncovered additional
data which generally support this dichotomy. With respect to
anterior aphasic subjects, these data all suggest impairment
in the timing or integration of movements of  the  articula-
tory  system. These anterior aphasic impairments do not seem
to reflect a global weakness or discoordination of the arti-
culators.  Rather,  articulatory  disabilities  appear to be
best characterized as affecting two "independent"  articula-
tors,  e.g.,  coordinating  vocal  fold  vibration  with the
tongue tip release of stop consonants.

Coarticulation in aphasia

     Ziegler and von Cramon (1985, 1986, 1986b), based  upon
acoustic  and perceptual studies of German-speaking anterior
aphasic subjects, argue that anterior aphasia may involve  a
"delayed  onset  of  anticipatory vowel gestures relative to
the labial occlusion." A similar conclusion was  tentatively
reached  by Tuller and Story (1986), who conducted an acous-
tic analysis of coarticulatory information  present  in  the
speech of English-speaking anterior aphasic subjects.

     In contrast, Katz (1987, 1988)  conducted  an  acoustic
and  perceptual  study  of  English-speaking  aphasics which
showed a "mixed" pattern of  results  for  anterior  aphasic
subjects.  The  acoustic  data showed no differences between
the coarticulation present in normal  and  anterior  aphasic
speech,   while   the  perceptual  data  indicated  somewhat
degraded coarticulatory information in the speech  of  ante-
rior  aphasic  subjects.  Moreover, group differences in the
perceptual data varied as a function of  stimulus  type.  It
was  concluded that a "uniform delay" in coarticulation does
not adequately characterize anterior aphasic speech. Rather,
coarticulatory "planning" was considered to be intact, while
the degree of actual coarticulatory behavior was  considered
to  vary  as  a function of the complexity of the motor ges-
tures involved.

     Little direct kinematic exists concerning  this  issue.
Sussman,  Marquardt,  MacNeilage,  &  Hutchison  (1988) have
reported kinematic findings concerning labial coarticulation
in  aphasia. It was found found that for correct productions
of CV, CCV, and VC CCV stimuli,  anterior  aphasic  subjects
exhibited  lip  and  jaw  coarticulatory behavior similar to
normal subjects.

     With respect  to  velar  coarticulation,  a  series  of
kinematic  experiments  have been conducted by Itoh and col-
leagues in Japan.  Using both fibroscopic  (Itoh,  Sasanuma,
and  Ushijima,  1979)  and X- ray microbeam (Itoh, Sasanuma,
Hirose, Yoshioka,  and  Ushijima,  1980)  techniques,  these
authors  analyzed  the speech of a Japanese- speaking, ante-
rior aphasic subject. The results showed  a  great  deal  of
variability in the apraxic patient's speech kinematics, par-
ticularly with respect to the successional patterns of velar
movement.  However,  despite  occasional  deviations, it was
concluded that anticipatory coarticulation was intact.

     As the preceding, somewhat equivocal pattern of results
indicates,  it  is difficult to obtain a coherent picture of
coarticulation in aphasic speech based solely upon  acoustic
or  perceptual  data.  Rather,  it  is  essential to combine
acoustic and perceptual investigations with direct kinematic
measurement.   Moreover,  studies  which provide information
about the simultaneous motion of several articulators  offer
a greater advantage for understanding coarticulatory impair-
ment than do analyses of  individual  articulator  movement.
For  these  reasons,  Katz,  Machetanz,  Orth,  and Schoenle
(1989c, 1989d) conducted kinematic and acoustic analyses  of
the  speech  of  German-speaking  anterior aphasic subjects.
This work made use of the recent technology  of  electromag-
netic  articulography, which affords real-time, simultaneous
tracking of several articulators in the vocal tract.   Also,
by  using German-speaking subjects, it was possible to study
lip-rounding contrasts which  did  not  involve  changes  in
tongue position (as found in English).

Experiment

                           METHOD

Subjects

     Subjects included two anterior aphasic and two  normal,
control  speakers.  All  subjects  were  adult, right-handed
native speakers of German,  from  similar  dialect  regions.
Aphasic  subjects  presented  with  single,  clearly defined
anterior lesions (see Katz et al., 1989c for  details),  and
were  classified based upon clinical exam and speech pathol-
ogy assessment. Control subjects had no history of neurolog-
ical disease, and no speech or language difficulties.

Procedure

     Speech kinematics were measured using  the  electromag-
netic  articulography  system developed at the University of
Goettingen Department of Clinical Neurophysiology (Schoenle,
Grabe,  Wenig, Hohne, Schrader, & Conrad, 1987). This device
allows for the simultaneous recording of multiple points  in
and  outside  of  the vocal tract. Subjects were seated in a
quiet testing room and were fitted with a helmet  containing
three  magnetic  transmitter  coils.   Minute receiver coils
were attached to the upper lip (UL) and lower lip (LL)  (for
the  labial  study) and to the tongue tip and velum (for the
nasal study). A computer was used to sample positional  data
for two receiver coils, and to record acoustic data.

Speech material

     Speakers  were  asked  to  produce  real  word  stimuli
designed  to  probe the timing of anticipatory labialization
during consonant production  and  anticipatory  nasalization
during  vowel  production.  There was a total of 10 stimulus
items   (4   labial,   6   nasal).   The   labial    stimuli
([g*li:g*]/[g*ly:g*];  [g*lez*]/[g*lo:z*])  consist  of word
pairs contrasting minimally in the rounding feature  of  the
vowel   following  the  consonant  [l].  The  nasal  stimuli
([ti:g*]/[tingl]; [ti:d*]/[ti:n*]; [ti:b*]/[ti:m*])  consist
of  word  pairs  differing  in the nasality features of each
word's initial vowel and middle consonant. The nasal stimuli
were  also selected to represent velar, alveolar, and labial
place of articulation for middle (word-internal) consonants.
All  stimuli were embedded in the carrier phrase: "Ich sagte
_______ zweimal" ("I said ________ twice").

     Subjects' productions were perceptually screened by the
researchers,  and  speech  errors  were identified. Detailed
classification of speech errors are listed in  Katz  et  al.
(1989c).   For the normal control subjects, no speech errors
were noted. For the aphasic subjects, a total of  22  speech
errors were detected (= 7.3% of the 300 total repetitions by
aphasic subjects). Errors  were  separately  classified  for
further  analysis.  From  a total of 15 repetitions recorded
for each stimulus type (per speaker), the first ten  correct
repetitions were used for kinematic and acoustic analyses.

Analysis

     Kinematic data were analyzed from  graphic  representa-
tions  of  articulator  position and tangential velocity. In
addition,  quantitative  records  of  speech   timing   were
obtained  using  interactive  software  designed  for speech
movement analysis.

     Speech acoustics were analyzed using the a speech  pro-
cessing  program for microcomputer. Speech samples were low-
pass filtered and digitized.  Speech  segment  regions  were
identified  in  the waveform from an oscillographic display,
and segment durations were recorded. Due to the known diffi-
culty  of  analyzing  cues for vowel nasalization present in
prevocalic consonantal  spectra  only  labial  stimuli  were
analyzed.  For the labial [g*lVg*] stimuli, 5 segments ([g],
schwa, [l], vowel, final [g*]) were delineated (see Katz  et
al., 1989c for details).

     For each stimulus, an analysis window was  placed  over
specific  areas  of each segment. Spectra were then obtained
using Fourier analysis and linear predictive  coding  (LPC).
Analysis of anticipatory lip-rounding focussed upon spectral
peaks in the liquid portion of the waveform anticipating the
second formant of the vowel. Additional details are provided
in Katz et al., 1989c.

Kinematic Results

I. Correct productions

A. Labial stimuli

i. Articulator displacement

     Lip rounding may be  characterized  in  terms  of  both
extension   and  vertical  movement  (raising  or  lowering,
depending upon the region of  the  lip  examined,  and  upon
individual subject characteristics).  The two control speak-
ers produced robust UL protrusion (extension  and  lowering)
for rounded (as compared with unrounded) stimuli.  These lip
protrusion gestures were rapid, concise,  and  were  clearly
related to production of the rounded vowel in the utterance.
In addition, speaker PS  showed  substantial  extension  and
lowering  of  LL  (and  jaw)  for both stimuli series, while
speaker EO demonstrated only slight LL (and jaw)  protrusion
for [g*ly:g*] as compared with [geli:ge].

     Considering next the anterior aphasic data, UL  protru-
sion was also found for rounded (as compared with unrounded)
stimuli. The overall time course of these gestures  differed
somewhat between the two aphasic subjects. Speaker AW showed
rapid UL movements corresponding with production of  rounded
vowel  segments  (i.e.,  resembling  the data of the control
subjects). In contrast, speaker EG showed  more  gradual  UL
movements, with a much lesser amount of net displacement. In
terms of LL (and jaw) movement, both aphasic speakers showed
articulator  protrusion  (extension and raising) for rounded
(as compared with unrounded) stimuli.

     In order to determine the extent to which speakers dif-
fered  in  the regularity of articulator movement, a measure
of item-to- item displacement variation  was  computed  (see
Katz  et  al.,  1989c).  The aphasic speakers showed greater
overall variation than the control speakers,  with  particu-
larly  high  variation  for  LL  (and jaw) displacement. The
results of two-way (Group x Articulator) analyses  of  vari-
ance  (ANOVA)  confirmed  that these patterns were statisti-
cally significant.

ii. Articulator timing

     Because it was  considered  important  to  examine  the
extent  to  which  anticipatory  coarticulation  varied with
speaking rate, the overall duration of speakers'  utterances
were  analyzed.  It was found that the aphasic speakers pro-
duced slower speech, with greater variation in segment  tim-
ing,  than  is  found  in the speech of normal, control sub-
jects. Further analysis of aphasic speakers'  temporal  pat-
terns  indicated  that these subjects showed a prevalence of
intersyllabic pauses, as well as occasional segment  prolon-
gations  (particularly  for  vowels)  in their speech. These
findings are in accord with previous descriptions of  verbal
apraxic speech.

     In order to analyze the point of anticipatory coarticu-
lation  onset,  individual  repetitions  were inspected with
attention paid to the exact point in the  acoustic  waveform
at  which  kinematic, coarticulatory effects could be noted.
The results demonstrated that lip protrusion  was,  for  all
subjects,  confined  to  a  region  proximate to the rounded
vowel. The beginning of labial  protrusion  generally  began
either  shortly before, or during production of the syllable
[g*], i.e.,  the  syllable  preceding  that  containing  the
rounded vowel.

     Although speakers were quite consistent across  repeti-
tions  (with  control  subjects  showing greater consistency
than aphasics),  there  were  notable  Speaker-  and  Group-
dependent differences in coarticulation onset. These indivi-
dual patterns  are  discussed  at  length  in  Katz  et  al.
(1989c).  In  general, the data may be summarized as showing
that anterior aphasic subjects show more variation in  their
onset positions, although this does not fit the pattern of a
uniform delay. Rather, aphasic speakers' variation generally
involved  unusually  early  lip  protrusion in comparison to
that found for control subjects.

B. Nasal stimuli

i. Articulator displacement

     The two control speakers showed little movement of  the
velum   during  production  of  the  non-nasalized  stimuli,
whereas robust velar port opening (i.e., velar extension and
lowering)  was  observed  for  productions  of the nasalized
stimuli. In comparison with the normal speakers, the aphasic
speakers  showed  highly impaired patterns of velar movement
(detailed in Katz et al.,  1989c).  These  impairments  were
more  marked  for  aphasic  subject  EG than for subject AW.
However, despite these imprecise movement patterns,  aphasic
subjects  showed  clear  evidence  of  correct, anticipatory
velar port opening before nasalized consonants.

     As with the labialization stimuli, variation  in  velum
displacement  was quantified by calculating the RMS distance
between  individual  utterances  and  averaged  displacement
waveforms.    Statistical  analysis  revealed  that  aphasic
speakers produced greater overall  variation  in  velum  and
tongue displacement than the control speakers. This was par-
ticularly true for the velar movement of aphasic subject EG,
who  showed  the  opposite pattern (i.e., greater velar than
lingual variation). Taken together, these data indicate that
the two aphasic speakers, considered as a "group", were more
variable than controls, and that aphasic subject EG showed a
particularly high degree of velar movement variation.

ii. Articulator timing

     The time course of correct  velar  stimuli  productions
was  investigated  by conducting comparisons of the acoustic
and kinetic data. As with the  labialization  results,  sub-
jects  were  found  to be quite consistent from utterance to
utterance, though aphasics  showed  greater  variability  in
onset   position.  Both  control  speakers  showed  context-
dependent differences at a region  of  the  speech  waveform
located  between the end of the aspiration following the [t]
segment, and the first 40 ms of  vowel  pulsing.  Similarly,
the  first  noticeable  context-dependent  difference in the
movement traces  of  the  aphasic  subjects  (i.e.,  visible
changes  in  the  rates  of  velar lowering) was noted to be
within this same  temporal  region.  In  other  words,  even
though  the  displacement  of velar position was noted to be
qualitatively different and more variable for aphasic speak-
ers,  the aphasic subjects appeared to initiate nasalization
gestures at approximately the  same  point  in  time  during
speech as normal controls. These data suggest that, although
the overall ability to control spatial  positioning  of  the
velum was clearly compromised for aphasic speakers, temporal
aspects of velar,  coarticulatory  movement  appeared  rela-
tively preserved.

II. Error-prone productions

     An investigation was made into the  claim  that  speech
production  errors  perceptually  resembling phone substitu-
tions might in fact be the result of discrete,  interarticu-
latory  phasing  difficulties  (e.g.,  Mlcoch  & Noll, 1980;
Ziegler and von Cramon, 1985). There were, however, very few
speech  errors  in the database containing the target struc-
tures of interest. These consisted  of  instances  in  which
aphasic  subject  EG  produced substitutions of "[tid*]" for
[tin*] targets.  For these cases, there were found to be [d]
stop  consonant  bursts in the acoustic waveforms. Kinematic
analyses showed that in two of  the  cases,  velar  movement
more  closely resembled correctly-produced [d] than [n] ges-
tures, while in the third  case  the  displacement  patterns
fell  midway  between  those  typical of nasal and non-nasal
consonants. In  sum,  these  data  do  not  suggest  that  a
slightly  mistimed [n] production resulted in a [d] percept,
but rather indicate that these errors likely  resulted  from
phone selectional ("phonemic") difficulties.

Acoustic Results

I. Correct productions (labial stimuli)

Vowel-anticipatory peaks in [l] spectra

     Data for the vowel pairs [i:]/[y:] and  [e:]/[o:]  were
grouped  for  comparison of coarticulatory effects. For both
sets of stimuli, the distribution of peaks  for  productions
by aphasic subjects was more variable than for those of nor-
mal subjects. This was particularly true for the higher fre-
quency  regions  (above 2 kHz).  Aphasia-related generaliza-
tions about coarticulatory shift were difficult to make  for
[i:]/[y:],  because  the  effect  did not seem clearly esta-
blished for the two normal subjects in  the  F2  range,  and
because  data  from  the aphasic speakers showed substantial
variation in the F3 frequency range. For [e:]/[o:], spectral
peaks in the F2 frequency range showed evidence of coarticu-
latory effects for both aphasic and control  subjects.  Con-
sidering both stimulus sets together, aphasic speakers' pro-
ductions appear  to  provide  evidence  for  vowel  context-
dependent  spectral shift in a manner similar to (or exceed-
ing that of) productions by normal subjects.

Discussion

     Although the present data must  be  considered  prelim-
inary  because of the small number of subjects investigated,
the findings address a number of important issues concerning
the neurological bases of speech production. To briefly sum-
marize the results, the kinematic data showed that for  both
labial  and  nasal  (correct) productions, aphasic speakers'
coarticulatory patterns were more highly variable than those
of  control subjects. These differences, however, were noted
chiefly for spatial displacement characteristics, while  the
temporal aspects of articulator movement involved in antici-
patory coarticulation appeared largely intact. It  was  also
found that velum mistiming did not appear to explain a small
corpus of stop/nasal substitution errors produced by one  of
the aphasic speakers.

     The acoustic data  largely  agree  with  the  kinematic
findings.  To the extent that vowel formant frequency energy
could be traced  back  into  the  portion  of  the  waveform
corresponding  to the prevocalic consonant, [l], the correct
productions of the anterior aphasic subjects showed patterns
of  labial  anticipation  similar  to (or exceeding that of)
normal speakers.

     A major empirical issue which these data  address  con-
cerns  whether  anterior  aphasic  subjects  show systematic
delays in  the  production  of  anticipatory  coarticulation
information  during speech. With respect to labial anticipa-
tion, uniform coarticulatory delays on the order of 20-30 ms
have  been  proposed  as a possible speech characteristic of
German (Ziegler and von Cramon,  1985;  Ziegler,  1989)  and
English-speaking  (Tuller  and Story, 1987) anterior aphasic
subjects. In contrast, Katz (1987, 1988) and Sussman et  al.
(1988)  have provided evidence suggesting that temporal con-
trol of anticipatory labial coarticulation is largely intact
in  anterior aphasic speakers' correct productions. The data
from the present experiment replicate the findings of  Suss-
man  et  al.  (1988),  and support the acoustic data of Katz
(1987, 1988), in that anterior aphasic subjects  were  found
to  produce  coarticulatory gestures as early as (or earlier
than) matched control speakers.  These data suggest that  if
listeners  show  uniform delays in picking up coarticulatory
information present in anterior aphasic speech, this may  be
due to complicating factors other than actual coarticulatory
cues (see Katz et al., 1989c for discussion).

     With  respect  to  anticipatory  velar  movement,   the
present study addresses the question of whether discoordina-
tions in velar movement might correspond  to  coarticulatory
impairments  (e.g.,  Mlcoch and Noll, 1980) and whether such
discoordinations  might  also   account   for   perceptually
apparent  "substitution"  errors  in  the speech of anterior
aphasic subjects (Itoh et al., 1979, 1980; Ziegler  and  von
Cramon,  1986b).  The present data replicate the findings of
Itoh et al. (1979, 1980), in that  they  demonstrate  essen-
tially  intact  anticipatory  coarticulation  in the correct
speech of anterior aphasic subjects.

     As for theories concerning aphasics'  error-prone  pro-
ductions, the present findings do not rule out the possibil-
ity that interarticulatory phasing difficulties may  account
for occasional substitution errors. In the present database,
however, there was very little evidence  for  this.  Of  the
three stop/nasal substitution errors examined, all contained
clearly  identifiable  stop  consonant  bursts,   suggesting
selectional   ("phonemic")   rather  than  interarticulatory
discoordination in motor output. Only one of the three  sub-
stitution  errors showed kinematic patterns allowing for the
possibility  of  interarticulatory  discoordination   (i.e.,
velar  displacement patterns midway between those typical of
nasal and non-nasal consonants).

     A key theoretical aim of this research  is  to  explore
how  patterns of coarticulation in aphasic speech can inform
models of normal speech production. With respect to  locali-
zation  of  function  issues,  the present data suggest that
anticipatory coarticulation capabilities of the adult  brain
do  not  critically  rely  upon  anterior  structures (e.g.,
Broca's area). Rather, it appears that anterior regions  are
involved  in coordinating the timing of the articulators for
producing individual phones (e.g., VOT values for stop  con-
sonants)  and possibly single syllables (see Kimura and Wat-
son, 1989), but not for effecting  anticipatory  transitions
between phones. If future experimentation confirms the find-
ing that highly automatized behavior  such  as  anticipatory
coarticulation  does  not  require the integrity of anterior
structures  in  adult  subjects,  then  models   implicating
specific  anterior  regions  (e.g., Broca's area) as general
"speech programming" centers will  certainly  require  revi-
sion. It may instead be the case that anticipatory coarticu-
lation is a property more  globally  represented  throughout
the  language  "zone" of the brain (i.e, dominant peri- Syl-
vian cortex and subcortical structures). This type  of  dif-
fuse  neural representation has been proposed for other pro-
perties of language, such as the representation  of  indivi-
dual lexical items (Ojemann, 1983).

     The present findings agree with  the  traditional  view
that  anterior  aphasics demonstrate problems chiefly at the
"phonetic" level, while posterior  aphasic  evidence  mainly
"phonemic"   (selectional)   deficits.   That  is,  anterior
aphasics appear  to  have  difficulty  in  interarticulatory
coordination,  which  impinges  on their ability to initiate
and produce a variety of speech sounds.  The problem is  not
in  phoneme  selection, it is in outputting selected sounds.
In a similar  fashion,  it  may  be  assumed  that  anterior
aphasics retain representations containing information about
the coarticulatory  spread  of  featural  information.  Such
representations would be qualitatively similar to the "boun-
dary conditions" yielded as the output of the Jordan  (1986)
model.  In  keeping  with  this  view, one could reason that
where the system fails for anterior aphasics is in the  map-
ping of coarticulated representations into motor output.

     Future investigations might  examine  these  issues  by
"lesioning"  models  of coarticulation in speech production,
and observing the  manner  in  which  feature  spreading  is
affected.  Results from "lesioned" connectionist models have
recently been used by investigators  to  simulate  cognitive
breakdown  in  adult  aphasia  (Gigely,  1988)  and acquired
dyslexia (Hinton & Shallice, in press).  The  present  data,
however,  suggest  that  "lesion"  experiments  using models
similar to Jordan (1986) would be most relevant to  deficits
in   posterior  aphasic  speech,  i.e.,  deficits  involving
"phonemic" (selectional) errors.  In  order  to  best  model
anterior  aphasic  speech,  it is essential to first develop
connectionist models which incorporate information about the
kinematic properties of the articulators. Models designed to
capture the "lower-level" inertial properties of the articu-
lators  are  currently  under  development  by  a  number of
researchers  (e.g.,  Browman  and  Goldstein,  1985;  Kelso,
Saltzman, and Tuller, 1989). [1]
____________________
  1  These models  assume  that  the  articulators  may  be
viewed as a series of mass-spring oscillators, whose "dynam-
ic" patterns may be described  mathematically  in  terms  of
system  constraints  upon  oscillatory  properties.  Because
"dynamic" models offers a high degree of mathematical rigor,
they have been viewed favorably by a number of connectionist
researchers.  However,  these  models  remain  controversial
within the speech research community. For example, see Kell-
er (in press) for an alternative view.
_____________________


     Additional information about the role of  brain  struc-
tures  in speech motor programming might be obtained by com-
paring the present results with data concerning the develop-
ment of coarticulatory patterns in children. Recent findings
have suggested that the ability  to  sequence  intersyllabic
anticipatory  coarticulation  information is present from an
early age, and may be more extensive in young children  than
in  older  children  (Nittrouer,  1989a; 1989b; Katz et al.,
1989a). These data suggest that intrasyllabic coarticulation
patterns might develop during early stages of brain develop-
ment, at which point  they  are  relatively  susceptible  to
disruption. However, once mature coarticulatory capabilities
are established, they may be far less susceptible to disrup-
tion,  even  in the face of massive damage to anterior brain
structures.


                      Acknowledgments

This research was sponsored by grant NS 08176-01A1  to  Wil-
liam  Katz,  University  of  Goettingen/DFG  grant to Jochen
Machetanz, and BMFT grant 0706839A to Paul-Walter  Schoenle.
The author thanks  Karen  Yummier,  Virginia  Marchman,  and
Teenie Matlock for their helpful comments.

                       References

Browman, C. & Goldstein, L. (1985). Dynamic modeling of phonetic
     structure.In V. Fromkin (Ed.) Phonetic Linguistics. New York: 
     Academic Press.

Browman, C. & Goldstein, L. (1986). Towards an  articulatory
     phonology.  Phonology Yearbook 3, 219-252.

Fowler, C.A. (1980).  Coarticulation  and  theories  of
     extrinsic timing. J. Phonetics 8, 113-133.

Fowler, C. (in press). Some regularities in speech  are not 
     consequences of formal rules. A commentary on Keating. In J.   
     Kingston  and M.  Beckman  (Eds.)  Papers in Laboratory Phonetics, 
     I. London: Cambridge University Press.

Gigely, H. Process synchronization, lexical  ambiguity, and ambiguity
     and aphasia.  In S. Small, G. Cottrell, M. Tannenhaus (Eds.) Lex- 
     ical ambiguity  resolution: Perspectives  from  psycholinguistics,  
     neuropsychology,  and  artificial  intelligence.  San  Mateo,
     Calif.: Morgan Kaufmann Publishers, Inc.

Goodglass, H. and Kaplan, E. (1982). The assessment  of aphasia and
     related disorders. Philadelphia: Lea and Febinger.

Hinton, G. and Shallice, T. (1989). Lesioning a connectionist network:
     Investigations of acquired dyslexia.  CRG  Technical  Report
     89-3,  University of Toronto, Canada.

Hinton, G. & Sejnowski, T. (1983).  Optimal  perceptual inference.  
     Proceedings of the IEEE Computer Society Conference on  Computer  
     Vision and Pattern Recognition, 448-453.

Hewlett, N. (1988). Acoustic properties of /k/ and  /t/ in normal and
     phonologically disordered speech. Clinical Linguistics and  Phonetics,  
     Vol. 2., No. 1, 29-45.

Itoh, M., Sasanuma,  S.,  Ushijima,  T.  (1979).  Velar movements during
     speech in a patient with apraxia of speech. Brain and Language,  7,
     227-239.

Itoh, M., Sasanuma, S., Hirose, H., Yoshioka, H., Ushijima, T. (1980).   
     Abnormal articulatory dynamics in a patient with apraxia  of  speech:
     X-ray  microbeam  observation.  Brain and Language, 11,
     66-75.

Jordan, M. (1986). Serial order: A parallel distributed processing
     approach.(Tech. Rep. 8604). La Jolla: University  of  California  
     San Diego, Institute for Cognitive Science.

Katz, W.F.  (1987).  Anticipatory  labial  and  lingual coarticulation in  
     aphasia. In J. Ryalls (Ed.), Phonetic Approaches to Speech Production
     in  Aphasia  and Related Disorders. San Diego: College-Hill Press.

Katz,  W.F.  (1988).  Anticipatory  coarticulation   in aphasia: 
     Acoustic and perceptual data. Brain and Language, 35, 340-368.

Katz, W.F., Kripke, C., and Tallal, P. (1989a). Anticipatory labial
     coarticulation in the speech of young children (age 3 to 8).
     Submitted.

Katz, W.F., Kripke, C., and Tallal, P. (1989b). Anticipatory labial
     coarticulation in the speech of normal and language-impaired
     children. Submitted.

Katz, W.F., Machetanz, J., Orth, U.  and  Schoenle,  P. (1989c) A
     kinematic analysis of  anticipatory  coarticulation  in  the
     speech  of  anterior aphasic subjects using electromag-
     netic articulography. Submitted.

Katz, W.F., Machetanz, J., Orth, U.  and  Schoenle,  P. (1989d)
     Anticipatory labial coarticulation  in  two  German-speaking
     anterior    aphasic    subjects:   Acoustic   analyses.
     Manuscript.

Keating, P. (in press). The window model of coarticulation:
     Articulatory evidence. In J. Kingston and M. Beckman  (Eds.)
     Papers  in  Laboratory  Phonetics, I. London: Cambridge
     University Press.

Keller, E. (in press). Speech  motor  timing.  In  W.J. Hardcastle &
     Marchal (Eds.) Speech production and speech modelling.

Kelso,  J.A.,   Saltzman,  E.L.,  and  Tuller,  B.  (in press).  The
     dynamical perspective on speech production: Data and theory.
     Journal of Phonetics 14, 29-59.

Kimura, D., & Watson, N. (1989). The  relation  between oral movement
     control and speech. Brain and Language, 37 (4), 565-590.

Lubker, J.F. and Gay, T. (1982). Anticipatory labial coarticulation:
     Experimental, biological, and  linguistic variables.
     Journal of the Acoustical Socety of America, 17(2), 437-447.

Mlcoch and Noll, J.D. (1980). Speech production  models as related 
     to the concept of apraxia of speech.  In  N.J.  Lass  (Ed.),
     Speech  and  language:  Advances  in basic research and
     practice, Vol. 4, 201-239.

Nittrouer, S., Studdert-Kennedy, M. and  McGowan,  R.S. (1989a).  
     The emergence of phonetic segments: Evidence from  the  
     spectral structure  of fricative vowel syllables spoken by 
     children and adults. Journal of Speech and Hearing Reseach, 
     32,  No.  1,  pg. 120-132.

Nittrouer, S. and  Whalen  D.  (1989).  The  perceptual effects of 
     child-adult differences in  fricative-vowel  coarticulation.
     J.  Acoust. Soc. Am. 86 (4), 1266-1276.

Ojemann, G. (1983).  Brain  organization  for  language from the
     perspective of electrical  stimulation  mapping.  Behavioral
     and Brain Sciences, 189-230.

Repp, B. (1986). Some observations on  the  development of
     anticipatory coarticulation. J. Acoust.  Soc.  Am.  79  (5),
     1616-1619.

Rumelhart, D.E., Hinton, G.E., & Williams, R.J. (1985). Learning
     internal representations by error  propagation  (Tech,  Rep.
     8506).   La  Jolla: University of California San Diego,
     Institute for Cognitive Science.

Sharf, D. and Ohde, R. (1981). Physiological,  acoustic and 
     perceptual aspects of coarticulation: Implications  for  
     the remediation  of  articulatory  disorders. In N. J. Lass
     (Ed.), Speech and language: Advances in basic research,
     Vol. 5, New York: Academic Press.

Schoenle,  P.W.,  Grabe,  K.,  Wenig,  P.,  Hohne,  J., Schrader, J., 
     Conrad, B. (1987). Electromagnetic  Articulography:  Use  of
     alternating  magnetic  fields for tracking movements of
     multiple points inside and  outside  the  vocal  tract.
     Brain and Language, 31, 26-35.

Sereno, J.A., Baum, S.R., Marean, G.C., and P.  Lieberman (1987).
     Acoustic analyses and perceptual data on anticipatory labial
     coarticulation  in  adults and children. J. Acous. Soc.
     Am. 81, 512-519.

Sussman, H., Marquardt, T., MacNeilage, P.,  Hutchison, J. (1988).
     Anticipatory coarticulation in aphasia: Methodological  con-
     siderations. Brain and Language, 35, 369-379.

Tuller, B. and Story, R. (1987). Anticipatory coarticulation in
     aphasia. In J. Ryalls (Ed.), Phonetic Approaches  to  Speech
     Production in Aphasia and Related Disorders, San Diego:
     College-Hill Press.

Ziegler, W., von Cramon, D. (1985) Anticipatory coarticulation in
     a patient with apraxia of speech.  Brain  and  Language  26,
     117-130.

Ziegler and Von Cramon (1986a).  Disturbed  coarticulation in apraxia
     of speech: Acoustic evidence. Brain and Language, 29, 34-47.

Ziegler and Von  Cramon  (1986b).  Timing  deficits  in apraxia of
     speech. European  Archives of  Psychiatry  and  Neurological
     Sciences, 236, 44-49.