THE ROLE OF THE OUTPUT PHONOLOGICAL BUFFER
IN THE CONTROL OF SPEECH TIMING:
A SINGLE CASE STUDY
Laurent Cohen 1•3 and Anne-Catherine Bachoud-LevF·3
eservice de Neurologie, Hopital de Ia Salpetriere, Paris; 2Service de Neurologie, Hopital
Henri Mondor, Creteil; 3Laboratoire de Sciences Cognitives et Psycholinguistique, CNRS
& EHESS, Paris)

ABSTRACT

A patient with a selective impairment of speech output, associating phonemic paraphasias
and pervasive preverbal pauses, is described. Experimental data are discussed in relation to
current models of speech production. On clinical evaluation, the pattern of paraphasias was
strongly suggestive of a functional lesion affecting the output phonological buffer. This
hypothesis received substantial support from a study of word and non-word repetition, as
well as from a study of picture naming latencies, with various imposed delays between
stimulus presentation and vocal response. In order to determine the origin of the preverbal
pauses, the patient was also submitted to other timed motor tasks implicity probing higher
levels of word production (gender decision and phonemic matching). We suggest that the
pauses probably resulted from the conjunction of several mechanisms, including a general
slowing apparent in all timed motor tasks, as well as a more specific time lag in speech
initiation originating from the output buffer or from the articulator itself.

CASE REPORT

The fundamental manifestation of the language faculty is the ability to easily
produce and understand continuous speech. The speech stream is normally
constituted of an uninterrupted sequence of phonological units, with no acoustic
discontinuity at word boundaries. Successive segments are produced at a fairly
regular rate (Lehiste, 1977), which speakers are able to control and adjust
depending on the immediate requirements of communication (Deese, 1984).
However there is an apparent discrepancy between the ability to maintain a
continuous, regular, and flexible speech output, and the complex psychological
machinery that leads to overt speech production. Activation of successive words
during sentence production does not occur serially but rather with some overlap
between words. As demonstrated for instance by anticipatory speech errors,
phonological encoding of a given word begins before processing of the foregoing
words is over (e.g. Shattuck-Hufnagel, 1979). Moreover, each of the semantic,
syntactic, lexical and phonological components which work together in language
production is likely to obey specific timing constraints, which should not be
expected to yield a steady and regular speech output. Finally, these various
components are largely automatic and their operating speed is probably
inaccessible to voluntary modulation (Fodor, 1983). How then should we account
for the apparent contradiction between continuous, regular and controlled speech
Cortex, (1995) 31, 469-486

470

Laurent Cohen and Anne-Catherine Bachoud-Levi

output on the one hand, and discontinuous, independent and automatic speech
production processes on the other hand?
Part of the answer is likely to come from a better understanding of the
operation of the output phonological buffer. It is thought that just before being
transmitted to the articulator for motor execution, phonological plan is stored
temporarily in this buffering facility (Levelt, 1989). This allows the speaker to
cope with asynchronies between formation of phonological plan and articulation
rate. Additionally, phonological words stored in the buffer may be combined
into a connected articulatory plan for entire phonological phrases, thus
contributing to speech continuity (Levelt, 1989, p. 420; Bub, Black, Howell et
al., 1987). Control over the output rate of the phonological buffer may also play
a direct role in the regulation of speech rate, inter alia through the insertion of
pauses.
A few patients with an impaired output phonological buffer have been
reported previously. An overview of some of the most recent and carefully
studied cases shows that all patients share a number of defining clinical features
(Caramazza, Miceli and Villa, 1986; Caplan, Vanier and Baker, 1986; Bub et
al., 1987; Belleville, Peretz and Arguin, 1992). Since words are assumed to be
represented in the buffer as sequences of phonemes, an impairment on this level
yields difficulties in correctly selecting and ordering the component phonemes.
Therefore speech errors almost exclusively consist of phonemic paraphasias,
phonologically close to the target phonemes (Caramazza et al., 1986). Similarly,
patients with an impaired graphemic output buffer make errors in selecting and
ordering the graphemes that compose the orthographic form of words (Caramazza
et al., 1987). Errors affect all oral production tasks (naming, reading, repetition,
etc.). Real words are better preserved than phonologically equivalent non-words,
presumably resulting from the interaction between the lexicon and post-lexical
phonological processing (Caramazza et al., 1986; Caplan et al., 1986; Bub et
al., 1987). Such patients are generally classified as "reproduction" conduction
aphasics, on the basis of the association of a preserved comprehension with
phonological errors affecting repetition, among other speech production tasks
(Shallice and Warrington, 1977). They usually display an associated reduction
of their verbal span. It is plausible that in such cases span reduction results, at
least partly, from the involvement of the output buffer in articulatory rehearsal
during short term memory tasks (Baddeley, Thompson and Buchanan, 1975;
Bub et al., 1987; Belleville et al., 1992). Finally, these patients are reported to
have a reduced articulation rate, and to pause frequently between words in
connected speech. This latter feature, which is generally mentioned as a marginal
observation, will be discussed more specifically here.
In the present study, we report the case of a patient whose most salient
anomalies in spontaneous speech consisted of phonemic paraphasias, associated
with a general slowing resulting from frequent pauses she made between words.
We will suggest that the patient's pattern of speech errors was suggestive of
an impairment of phonological encoding, most probably affecting the output
phonological buffer. The aim of the following experimental study will be
twofold. First, the buffer hypothesis will be confronted with a series of repetition
and timed naming experiments probing the time course of phonological encoding.

Phonological buffer and speech timing

471

Second, we will discuss to what extent the general slowing of the patient's
speech resulted from the impairment affecting her output phonological buffer,
and eventually propose that other additional sources of slowing should also be
considered in the present case.
CASE HISTORY

The patient was a 48-year-old right handed woman, who had had 15 years of schooling.
She had been working as a translator from German and English into French, which was her
mother tongue. She suffered a subarachnoid hemorrhage, revealing a large left parietal
arteriovenous malformation. Following a subsequent attempt to artificially embolize a feeding
artery, the patient suffered a recurrence of hemorrhage, accpmpanied by right sensory and
motor deficit, and by almost complete muteness. The present study was carried out 2 years
later. At that time, the patient still had severe right, mainly brachial, motor deficit. She could
accurately imitate 5/6 simple bucco-facial gestures. Visual field was normal on confrontation,
and language disorders had substantially receded.
LANGUAGE EXAMINATION

General Language Assessment

Spontaneous speech was rich and informative. The patient made use of
complex and diversified syntactic structures. However, speech rate was reduced
by frequent pauses occurring mainly between words. For instance, the patient
took as long as 32 seconds to utter the sentence "J'ai repense a Ia phrase que
vous avez dite a propos de mon bras" (I have thought again about the words
you said concerning my arm). The total duration could be broken down into
13 seconds of actual speech and 19 seconds of pauses between words.
Additionally, the patient made phonemic paraphasias, some of which she
attempted to correct.
The patient was submitted to a French version of the Boston Diagnostic
Aphasia Examination (Goodglass and Kaplan, 1972; Mazaux and Orgogozo,
1982). The overall severity rating was 4. Verbal agility was reduced (score: 6/
14). The patient scored at ceiling level in all subtests of auditory comprehension
(word discrimination: 72/72, body-part identification: 20/20, commands: 15/15,
complex ideational material: 12112). Although she did not make a single error
in the naming subtests, her scores were not quite perfect due to occasional delay
before response onset (responsive naming: 24/30, confrontation naming: 98/105,
naming of body-parts: 28/30). The patient produced 12 names of animals in the
fluency in controlled association subtest. Automatic speech was correct
(automatized sequences: 7/9, reciting: 1/2). Scores in repetition subtests were
high despite rather frequent phonemic paraphasias and pauses during sentences
(words: 9/10, high-probability sentences: 8/8, low-probability sentences: 8/8).
This observation applies equally well to oral reading (words: 27/30, sentences:
7/10). The patient was perfect in all subtests of reading comprehension (symbol
discrimination: 10110, word recognition: 8/8, word-picture matching: 9/10,
sentences and paragraphs: 10/10). Errors in comprehension of oral spelling (5/
8) may be related to her severely reduced verbal memory span (digits: 3,

472

Laurent Cohen and Anne-Catherine Bachoud-Levi

monosyllabic words: 4)_ 1 The patient could write quite effectively using her
non-dominant left hand, and her scores in the writing subtests were very high
(serial writing: 46/47, primer-level dictation: 14115, spelling to dictation: 10/10,
written confrontation naming: 10/10).
The patient scored 34/36 on the Token Test (De Renzi and Faglioni, 1978),
confirming that speech comprehension was virtually normal.
The patient was presented with 80 line drawings printed on cardboard,
selected from the pictures in Snodgrass and Vanderwart (1980). This set was
made up of ten pictures belonging to each of 8 semantic categories (animals,
tools, music instruments, body parts, vehicles, pieces of furniture, clothes,
vegetables). The patient could flawlessly sort the pictures into the 8 categories,
although she had not been informed of the number and identity of those
categories.
The patient was presented with 115 simple line drawings for oral naming. 2
She made 14 errors (12%), of which 13 were phonemic paraphasias. The
erroneous phoneme was generally close to the substituted phoneme (e.g.
champignon [mushroom] _.... /sampignon/). As noted previously, naming onset
was often preceded by a pause of abnormal length.
To summarize, the patient presented a selective impairment of verbal output,
consisting of an association of phonemic paraphasias with abnormal pauses
affecting connected speech as well as isolated words. Oral comprehension,
reading and writing, were virtually normal.
Repetition of Words and Non-words

The above clinical assessment revealed that the patient met several diagnostic
criteria for an impairment of the output phonological buffer. Similar patients
generally display a strong advantage of words over non-words in a variety of
verbal production tasks (Caramazza et al., 1986; Caplan et al., 1986; Bub et
al., 1987). The aim of the following repetition test was to search for such a
lexicality effect in the present case. Additionally, if the phonological buffer did
not function properly, the patient's performance should deteriorate in proportion
to the duration of the required storage. This prediction was tested by comparing
immediate and delayed repetition.
Method

The experimental material consisted of 54 words and 54 non-words, matched with respect
to syllabic length and initial phoneme. Stimuli were read aloud one by one by the experimenter
to the patient and 5 control subjects matched for age and education. In the immediate
condition, subjects were asked to repeat each item immediately. In the delayed condition,
they were instructed to wait until the experimenter knocked on the table after a 4 second
delay, and only then to repeat the stimulus. The block of 54 words as well as the block of

1 During span assessment, elementary phonemic paraphasias leaving no doubt on the identity of the produced word
were not scored as errors.
These pictures were selected from among the 260 pictures in Snodgrass and Vanderwart ( 1980). The original set
was presented to 15 normal subjects for naming. Those drawings for which there was perfect agreement among all
subjects were digitized and used in the present study.

2

Phonological buffer and speech timing

473

%correct

100

~ immed i ate

80

O delayed

Qs:

~

>s;

60
)(<

x;

'

'

40 r'

20 r­

0

'
·~{

words

<r

:

~>ON'

X666<

Qox,x

N'
)6

~

iS: Q<

0.

non-words

Fig. I - Percentage of correct responses in the repetition task. The patient produced significantly
more errors with non-words than with real words. With non-words, performance level was significantly
worse in the delayed than in the immediate condition.

54 non-words were successively presented in the immediate and in the delayed conditions.

Results

Controls made no errors over 540 trials with real words, and a total of 111
540 (2%) phonemic errors with non-words (2/270 [0.7%] in the immediate
condition and 9/270 [3%] in the delayed condition). All conditions pooled, the
patient made a total of 82 errors (40% ). All errors were phonemic paraphasias
similar to those produced in spontaneous speech and reading. The patient's
performance showed two important features, apparent in Figure 1. First, non­
words were significantly more error prone than words (67/108 [62%] and 15/
108 [14%] errors, respectively; x2 =53.15; d.f.=1; p<O.OOOl). Only three out
of 5 control subjects showed some tendency to make more errors with non­
words than with words (difference in error rates: 1.9%, 2.7%, and 5.6% ), albeit
of an amplitude much smaller than the patient's 48% difference. Second,
although the 4 second delay did not significantly modify the overall performance
level (x2 = 0.96; d.f. = 1), it affected differentially the repetition of words and
non-words. With non-words, errors were significantly more frequent in the
delayed than in the immediate condition (39/54 [72%] and 28/54 [52%] errors,
respectively; x2 = 4.76; d.f. = 1, p = 0.029), whereas these two conditions did not
differ with real words (6/54 [11%] and 9/54 [17%] errors, respectively; X2 = 0.70;

474

Laurent Cohen and Anne-Catherine Bachoud-Levi

d.f. = 1). There was a small non-significant tendency for 3 out of 5 control
subjects to show a similar effect of condition with non-words (difference in
error rates: 3.7%, 3.7%, and 5.6%).
Discussion

The discrepancies observed here between words and non-words in terms of
error rate, as well as with respect to the detrimental effect of delay, may seem
at odds with the post-lexical character of the hypothesized buffer impairment.
However, this observation is by no means an isolated finding, since comparable
patients have consistently shown a similar advantage of words over non-words
in repetition tasks (e.g. Caramazza et al., 1986; Caplan et al., 1986). During
the repetition of non-words, the buffer presumably receives its input from a
"surface" route directly translating input phonology into output phonology, with
no intermediate lexical stage. When repeating real words, while this direct route
may still be operative, it is supplemented by a lexical route allowing the buffer
to receive its input from the phonological output lexicon. Thus the buffer stands
at the confluence of the lexical and non-lexical routes, and it would seem natural
that an impairment at this level should affect similarly the repetition of words
and non-words. However, there are independent reasons to believe that the
interactions of the lexicon and the buffer may be somewhat more complex.
Spreading. activation models of phonological encoding propose that any sequence
of phonemes stored in the output buffer provides positive feedback to a number
of neighbouring word nodes in the lexicon. These nodes reciprocally activate
the corresponding phonemic sequences from top down (Levelt, 1989; Dell and
0'Seaghdha, 1992). In particular, the phonemic sequence corresponding to a
non-word activates a series of phonologically related words in the lexicon. These
word nodes reciprocally tend to modify the sequence of phonemes in the buffer,
yielding repetition errors through the activation of phonemes that are not present
in the stimulus. In contrast, this phenomenon is of little consequence on the
repetition of real words, due to the lateral inhibition of concurrent word nodes
by the actual target word. This may explain why the repetition of real words
should be easier than the repetition of non-words in normal subjects, and even
more so in patients whose content of the output buffer is exceedingly fragile
(see Caplan, 1992, p. 149).
More generally, and without commitment to any specific theory of
phonological encoding, the advantage of words over non-words may simply
result from the fact that non-words are only represented as transient phonological
sequences, while real words also enjoy permanent lexical and semantic
representations. Even if words and non-words are equally exposed to decay while
stored in the phonological buffer, these stable higher level representations may
act as a backup in the case of words, and help restore the adequate string of
phonemes. In good agreement with our findings, this advantage of words over
non-words may be expected to become larger with longer delays. Indeed long
delays increase the probability of phonological deterioration, without substantially
reducing the availability of these backup procedures only applying to real words.

Phonological buffer and speech timing

475

Summary of Language Examination
1t is now apparent that the main diagnostic criteria for an impairment of the
output phonological buffer were met. First, all production tasks were affected
irrespective of the input modality, indicating that the deficit was located in the
verbal output system. Second, errors consisted of the replacement of target
phonemes by phonologically close phonemes, suggesting that the impairment
affected a phonological level of representation. Third, error rate could be
increased by imposing a delay before overt production, confirming the lesion
of a short term memory system. Fourth, real words were demonstrably more
robust than non-words. Finally, our patient shared several related clinical
characteristics with similar, previously reported, cases: A reduced verbal span,
reduced articulation rate, and pervasive pauses before words. It must be stressed
that these additional clinical features need not result from the buffer impairment
itself. For instance, an additional impairment affecting an input phonological
store may in principle contribute to the span reduction, and a slowed articulatory
system may contribute to the speech flow abnormalities.
In the following section, we will further test the hypothesis of a buffer
disfunction, using a series of timed speech production tasks.

ExPERIMENTAL STUDY

Before describing the experimental study, it may be useful to present a brief
outline of the processes underlying word production, such as operating typically
in picture naming. For instance, naming the picture of an apple may be analyzed
into three successive stages (Levelt, 1989). First, the concept of an apple is
activated, as a consequence of visual processing and identification of the picture.
In a second stage, this conceptual representation triggers lexical access, making
all information concerning the word "apple" available. Actually, there is evidence
that different aspects of lexical information are not activated simultaneously but
serially and in a fixed order (e.g. Schriefers, Meyer and Levelt, 1990). First
come semantic features and syntactic information such as grammatical category
and gender, constituting the lemma of the word. Second, the phonological
characteristics of the word, or lexeme, are accessed (sequence of phonemes,
syllabic structure, stress pattern, etc.). As mentioned before, the resulting
phonological plan is stored in the output phonological buffer, to be eventually
transferred to the articulator for motor execution, which constitutes the third
and last stage of speech production. Although there is some disagreement as to
the precise temporal course of these processes, there is reasonable consensus
regarding their general architecture (Levelt, 1989; Dell and O'Seaghdha, 1992).
In the following experiments, we tried to tap selectively some of these
processing levels. The patient's performance was systematically compared to
that of the 5 matched controls mentioned above. The group of controls comprised
3 women and 2 men, all right handed, with an age ranging from 43 to 52 years,
and an education level comparable to the patient's (15 to 18 years of education).

476

Laurent Cohen and Anne-Catherine Bachoud-Levi

Experiment 1: Timed Picture Naming I
In this experiment, we studied the influence on naming latencies of imposing
an obligatory delay between stimulus presentation and overt naming. It has been
shown previously in normal subjects that picture naming latencies are
significantly longer if subjects are requested to respond immediately that if they
have to respond on a subsequent "go" signal (Klapp, Anderson and Berrian,
1973). It is thought that when producing their response in the delayed condition,
subjects do not have to go through all the early stages of the naming task. They
have ample time to formulate a phonological plan and store it in their output
buffer. Upon seeing the "go" signal, they simply transfer the available
phonological plan from the buffer to the articulator, and produce the
corresponding sounds (Levelt, 1989).
Thus the beneficial effect of delay on naming times directly results from the
operation of the output buffer. If indeed the patient's buffer did not function
normally, an imposed delay preceding the vocal response may fail to yield a
normal reduction of reaction times. The outcome of the repetition task reported
before may help us specify further our prediction. After a 4 second delay, real
words were repeated as accurately as when no delay was imposed, while the
repetition of non-words deteriorated severely over this delay. We suggested that
in the delayed condition, and with real words only, the patient could avoid using
the deteriorated phonological plan stored in her buffer by engaging again in a
complete phonological encoding on the basis of a higher level representation
of the target word. If this hypothesis is correct, then the patient should behave
identically whether a 4 second delay intervenes or not. In other terms, the patient
should not benefit from the normal latency advantage that normal subjects are
expected to show when naming with a 4 second delay, as compared to immediate
naming.
Method

Thirty pictures were chosen from the preselected corpus of 115 unambiguous line
drawings. Since it has been suggested that the number of syllables may affect differentially
immediate and delayed naming latencies (Klapp et al., 1973), the material was controlled
for length. The set of names corresponding to the selected pictures comprised 10 mono, 10
bi, and 10 trisyllabic words. The 3 classes of words were matched with respect to the
phonological category of their initial phoneme, as well as to frequency. The block of 30
drawings was presented twice for delayed naming, and then twice for immediate naming.
Presentation order was randomized within each block. Each target picture was displayed for
1500 msec. In the immediate condition, subjects were instructed to name the depicted item
as fast as possible. In the delayed condition, they had to refrain from overt naming. After
a subsequent 4000 msec blank screen, a black square appeared, indicating that subjects had
to respond as fast as possible. In both conditions, response was followed by a 1500 msec
blank screen, preceding the next trial. Pictures were presented in black on white, on a NEC
Multisync 2A screen controlled by a Toshiba T-5200 computer. The microphone was
connected to an OROS AU-22 digital board programmed to function as a voice key, which
measured naming time to the nearest millisecond. All testing sessions were recorded for
subsequent analysis of errors.

Phonological buffer and speech timing

477

RT(msec)

2000

1 500

1 000

500

------------------~t~----~-------~---~-~---:;;J·-------------------

0 ~----------------------------------------

immediate

4 sec delay

Fig. 2 -Mean picture naming latencies for the patient (triangles) and control subjects (squares)
in Experiment ]_ The patient was significantly slower than controls, and did not display the normal
latency difference between immediate naming and 4 second delayed naming.

Results

Reaction times corresponding to the production of an erroneous word, to
phonemic paraphasias affecting the initial phoneme or at least any two phonemes,
to a stuttering answer, to a self correction, to a technical malfunction, or differing
by more than 2 SD from the mean for the considered subject and experimental
condition, were excluded from the analyses. This procedure led to the rejection
of 28 values for the patient (23%) and 35 values for control subjects (6% ). In
the patient, an almost equal proportion of responses were rejected in the
immediate and in the delayed condition (12/60 and 16/60, respectively, x2 =0.42;
d.f. = 1; p = 0.52), preventing any problem of speed-accuracy trade-off in the
interpretation of reaction times.
Mean latencies for the patient and control subjects are displayed in Figure
2. The controls' mean reaction time was 561 msec in the immediate condition
(range: 511 msec to 629 msec), and 372 msec in the delayed condition (range:
286 msec to 465 msec). As could be expected, the patient was significantly
slower than controls in both the immediate (mean= 1494 msec; t = 17.27; d.f. = 4,
p=0.0001) and the delayed (mean= 1924 msec; t=20.16; d.f.=4; p<0.0001)
conditions. In addition to this overall slowing, the patient's performance was
also qualitatively abnormal. Control subjects were 189 msec faster in the delayed
than in the immediate condition (F= 273.26; d.f. = 1, 4; p = 0.0001). The patient's

478

Laurent Cohen and Anne-Catherine Bachoud-Levi

latencies did not obey this general tendency. Her results fell significantly off
the distribution of difference scores of control subjects between immediate and
delayed conditions (t = 24.26; d.f. = 4; p<O.OOOl). Actually, she even showed
an opposite tendency, since she was 430 msec slower in the delayed condition
(t= 1.63; d.f.=90; p=O.ll). Finally, it should be noted that no effect of syllabic
length on mean reaction times was observed for the control subjects nor for the
patient, in either immediate or delayed naming.
Discussion
In agreement with earlier studies, control subjects had shorter naming
latencies in the delayed than in the immediate condition (Klapp et al., 1973).
The fact that, as predicted, the patient did not show this pattern of performance
is compatible with the idea of an impaired output buffer. If the phonological
plan suffers a fast degradation while stored in the output buffer, it is natural
that the longer the interval before articulation, the more severe the degradation.
After a 4 second delay, the original phonological plan may be severely altered,
inducing the patient to engage in a new naming procedure on the basis of some
representation of the stimulus in memory. This procedure naturally produces
reaction times, as well as error rates (just as in the repetition task), not different
from immediate naming.
Experiment 2: Timed Picture Naming II
It may be objected that our interpretation of the patient's performance in
Experiment 1 is essentially based on negative results, i.e. the absence of a normal
latency difference between immediate and delayed naming. If, however, the
account we have submitted is correct, this abnormal pattern prevailed only
because the strip of phonological plan remained long enough in the buffer to
be deteriorated to a degree justifying a complete new encoding from the lexicon.
Therefore it should be possible to restore a normal beneficial effect of delay
on latencies simply by reducing the 4 second delay to some shorter duration.
If indeed the "go" signal is presented before the stored phonological plan
becomes severely deteriorated, this plan should be used by the patient for overt
utterance, just as it happens in normal subjects. In other words, resorting to a
delay shorter that 4 seconds should restore a normal advantage of delayed over
immediate naming in the patient. This prediction was tested in the present
experiment by comparing immediate naming and naming with a short delay.
Method

Fifty line drawings were chosen from the corpus of 115 preselected pictures. The
corresponding set of names comprised 20 monosyllabic, 20 bisyllabic and 10 trisyllabic words.
All words had a consonant in initial position. The procedure was identical to Experiment 1,
except that in the delayed condition, the "go" signal was preceded by a 2 second delay. The
block of 50 drawings was first presented for delayed naming, and then for immediate naming.

Phonological buffer and speech timing

479

RT(msec)

2000

1 500

·······························································································

1 000

500

0 ~------------------------------------

immediate

2 sec delay

Fig. 3 -Mean picture naming latencies for the patient (triangles) and control subjects (squares)
in Experiment 2. The patient was significantly slower than controls. However, she displayed the normal
latency difference between immediate naming and 2 second delayed naming.

Results

Data were filtered according to the same principles as in Experiment 1,
leading to the rejection of 36 values for the patient (36%) and 30 values for
control subjects (6%). In the patient, an equal proportion of responses were
rejected in the immediate and in the delayed condition (18/50 in both cases),
preventing any problem of speed-accuracy trade-off in the interpretation of
reaction times.
Mean latencies for the patient and control subjects are shown in Figure 3.
The controls' mean reaction time was 601 msec in the immediate condition
(range: 534 msec to 677 msec), and 337 msec in the delayed condition (range:
284 msec to 435 msec). The patient was significantly slower than controls in
both immediate (mean=1409 msec; t=15.76; d.f.=4; p=O.OOOl) and delayed
(mean= 699 msec; t = 5.98; d.f.= 4; p<0.005) conditions. As in Experiment 1,
controls were faster in the delayed than in the immediate condition (F = 663.59;
d.f. = 1, 4; p<0.0001). The patient, although slower than controls, nonetheless
showed a significant advantage of delayed over immediate naming (t = 7.715;
d.f. = 62; p<O.OOOl). Comparing Experiments 1 and 2, it appeared that controls'
mean latencies were not different between naming with a 2 second and with a
4 second delay (F=4.39; d.f. = 1, 4; p<0.15), whereas the patient was
significantly slower in the latter condition (t = 3.97; d.f. = 74; p = 0.0002).

480

Laurent Cohen and Anne-Catherine Bachoud-Levi

Discussion

In agreement with our prediction, the patient was significantly faster in 2
second delayed naming than in immediate naming, just as were normal subjects.
It is important to stress that the apparently opposite influences of a 2 second
delay and of a 4 second delay on latencies resulted in fact from a single
impairment affecting the output buffer. After a 2 second delay, the phonological
plan stored in the output -buffer is still usable for articulation and the patient
benefits from the same latency advantage as do normal subjects. However, after
a 4 second delay, the phonological plan is too severely deteriorated, and naming
latencies then return to the level of immediate naming. In sum, the pattern
emerging from the comparison of immediate, 2 second delayed and 4 second
delayed naming in Experiments 1 and 2 lends further support to the hypothesis
of an impaired output phonological buffer.
However, this conclusion should not be taken to imply that all the abnormal
time-dependent features of the patient's behavior, over and beyond the
differential effect of delays on naming times, resulted from the buffer deficit.
In particular, one should remember that the mean latency in immediate and in
4 second delayed picture naming was more than 1 second longer in the patient
than in control subjects, and that a difference of about 400 msec was observed
in the 2 second delayed condition. It is tempting to assume that this protracted
silence preceding the onset of words is functionally equivalent to the pauses
that pervaded the patient's connected speech. Four main possibilities should be
considered in principle regarding the origin of this abnormal time lag. First, the
delay may arise prior to the delivery of the phonological plan to the output
phonological buffer. Second, it may result from an impairment affecting the
buffer itself. Third, the starting of overt production may be abnormally delayed
due to an impaired articulator. Finally, the hypothesis of a non-specific cognitive
slowing affecting all timed motor tasks should be considered. In view of the
deficit documented so far, the buffer hypothesis stands out as the most
economical one. However this possibility should not be taken for granted so
far. The following experiments aim at identifying other candidate sources of
retardation upstream from the output buffer.
Experiment 3: Timed Gender Decision

The goal of this experiment was to evaluate the patient's speed of processing
up to and including the first stage of lexical access in speech production, namely
access to word lemmas. To this end, the patient was submitted to a timed gender
decision task. In French, all nouns are either feminine or masculine. The
grammatical gender of sexually neutral items is arbitrarily determined, and hence
must be stored in the lexicon. Gender actually constitutes the only obvious
grammatical feature that must be represented in the lemma of common nouns.
We assumed that the only way to determine the gender of the name associated
to a given picture is to go through all stages of the naming task, at least up to
accessing the lemma. Gender decision may therefore be used as a probe to the
grammatical stage of lexical access.

Phonological buffer and speech timing

481

RT(msec)

1 600
1 400
1 200
1 000

.............................................................................................

600
400

gender decision

immediate naming

phonemic matching

Fig. 4 -Mean latencies for the patient (triangles) and control subjects (squares) in Experiment 3
(left) and Experiment 4 (right). Immediate picture naming latencies from Experiment 1 are presented
again for comparison (center). The patient was moderately slower than controls in grammatical decision
and phonemic matching. Contrary to controls, she was faster in those tasks than in immediate naming.

Method

The patient was presented with a series of 30 line drawings chosen from the corpus of
115 preselected pictures. The corresponding names were masculine for half the pictures, and
feminine for the other half. Names containing any morphological mark of gender were
systematically avoided. Frequency and syllabic length were equated across the two gender
categories.
Each picture was displayed on the computer screen for 3000 msec. Subjects were
instructed to decide if the name of the depicted item was feminine or masculine, and to
respond as fast as possible by tilting a joystick with their left hand to the right or to the
left, respectively.

Results

Reaction times corresponding to an incorrect gender decision or differing
from the subject's mean by more than 2 SD, which made up 10% of the data
for the patient and 9% for control subjects, were excluded from the analyses.
Results are presented in Figure 4.
The controls' mean reaction time was 810 msec (range: 660 msec to 982
msec). For the purpose of our demonstration, it is important to note that normal

482

Laurent Cohen and Anne-Catherine Bachoud-Levi

subjects were significantly slower in this task than in immediate picture naming3
(F = 11.11; d.f. = 1, 4; p = 0.029). The patient's behavior was abnormal in two
respects. First, she was marginally slower than controls (mean= 1199 msec;
t = 2.739; d.f. = 4; p = 0.052). Second, she was significantly faster in the
grammatical decision task than in immediate picture naming (t = 2.52; d.f. = 73;
p=0.014).
Discussion
In the present experiment, the patient was marginally slower than controls.
This slowing may originate either from a processing stage upstream from access
to gender information, or from some non-specific component of the classification
task, such as decision making or motor response. At any rate, this non-significant
slowing cannot account for the abnormal delay observed in immediate picture
naming. First, the patient was more than 1000 msec slower than controls in
picture naming, a difference obviously larger than the 400 msec difference
obtained in the present experiment. Second, the fact that, in opposition to the
controls, the patient was significantly slower in immediate naming than in
grammatical decision directly points to some major slowing factor affecting the
naming process downstream from lemma retrieval.
In short, although the patient was somewhat slower than controls in a task
selectively tapping processing stages up to grammatical access, this difference
cannot account for the preverbal time lag observed in immediate picture naming.
Experiment 4: Timed Phonemic Matching
Obeying the same logic as in Experiment 3, we tried to evaluate the patient's
speed of processing up to and including the second stage of lexical access in
speech production, namely retrieval of word lexeme. Lexemes are thought to
specify the morphemic and syllabic structure of words, as well as their segmental
composition, i.e. the sequence of constituent phonemes. We assumed that in
order to determine the phonemic characteristics of the name associated to a
given picture, one has to go through all stages of the normal naming task, at
least up to accessing the lexeme. We therefore resorted to a timed phonemic
matching task in order to probe the patient's access to the phonemic content
of words, without eliciting overt speech production.
Method

The same 50 line drawings as in Experiment 2 were used. On each trial, the experimenter
orally produced a syllable consisting of a consonant followed by a schwa, and subsequently
triggered the display of a picture on the computer screen. On matching trials, the proposed
consonant was the initial phoneme of the picture name. On non-matching trials, the proposed
consonant was not correct, although it was always phonologically close to the actual initial
phoneme. Subjects were asked to judge whether the proposed consonant and the picture
3 This comparison was made on the basis of immediate naming latencies obtained in Experiment I. Data from
Experiment 2 would yield exactly equivalent results.

Phonological buffer and speech timing

483

were matching. They were instructed to respond as fast as possible upon seeing the picture,
by tilting the joystick with their left hand to the right for matching trials and to the left for
non-matching trials.

Results

Reaction times corresponding to an incorrect decision, or differing from the
subject's mean by more than 2 SD, were excluded from the analyses, yielding
the rejection of 11% of the data for the patient, and 9% for control subjects.
Results are presented in Figure 4.
The controls' mean reaction time was 839 msec (range: 766 msec to 905
msec). They were significantly slower in this task than in immediate picture
naming (F = 83.22; d.f. = 1, 4; p = 0.0008). Just as in Experiment 3, the patient's
behavior was abnormal in two respects. First, she was slower than controls
(mean= 1180 msec; t = 6.19; d.f. = 4; p<0.005). Second, she was significantly
faster in the phonemic matching task than in immediate picture naming (t = 3.94;
d.f. = 133; p = 0.0001).
Discussion

The present pattern of results is in all relevant respects closely similar to
that observed in Experiment 3. The patient was moderately but significantly
slower than controls, but also faster than in immediate naming. We may therefore
follow the same general argument as with grammatical decision, and safely
conclude that at least most of the patient's abnormal time lag in immediate
naming originated downstream from phonological access. However, it should
be acknowledged that the phonemic matching task only allows for an incomplete
evaluation of phonological access. It does not probe the retrieval of the complete
phonological specifications of words, but only of their very first phoneme. It is
in principle possible that an impaired access to further parts of the stored
phonological records could contribute to the patient's slowing. Beside this
possibility, the remaining candidates are logically the output buffer and the
articulator.
Discussion of Experiments 1 to 4

The experiments we have reported aimed (1) at confronting the hypothesis
of an impaired output phonological buffer with naming latency data, with a
special emphasis on the influence of delay, and (2) at shedding some light on
the origins of the general slowing of the patient's speech.
Since the first point was already discussed at length, it is sufficient to
remember that the hypothesis of an impaired phonological buffer, which was
first formulated following clinical language assessment and formal repetition
tasks, received additional support from the influence of different delays on picture
naming latencies (Experiments 1 and 2).
Regarding the second point, the origin of the patient's abnormal pauses
preceding speech onset, apparent in spontaneous speech as well as in

484

Laurent Cohen and Anne-Catherine Bachoud-Levi

Experiments 1 and 2, was explored using timed motor tasks implicitly probing
higher levels of word production. Experiments 3 and 4 showed that the cascade
of processing stages that lead from visual analysis up to and partly including
phonological access to the output lexicon could not be taken as responsible for
most of the pauses. However, the patient's mean latency exceeded the controls'
by about 400 msec in both Experiments 3 and 4. Phonemic matching, although
presumably tapping a later processing stage, did not reveal any additional slowing
as compared to grammatical decision. It is tempting to suggest that the equivalent
slowing affecting gender decision and phonemic matching resulted from a
process common to both tasks, such as decision making or motor response.
Some additional support to this hypothesis comes from Experiment 2. In the 2
second delayed naming task, the patient was about 400 msec slower than
controls, just as in Experiments 3 and 4. To summarize, the patient's slowing
may be considered as the conjunction of at least the two following factors: (1)
a general 400 msec slowing affecting all manual or vocal timed motor tasks, a
common observation with brain damaged patients (e.g. Benton, 1986), and (2)
an additional slowing specifically affecting immediate naming (as well as the
essentially equivalent 4 second delayed naming).
It is unfortunately difficult to elucidate precisely the mechanism that gave
rise to the excessive time lag in immediate naming. We have suggested that an
abnormally fast decay of the phonological plan in the buffer accounts for the
fact that the patient was faster in 2 second delayed naming but not in 4 second
delayed naming, as compared to immediate naming. However, this hypothesis
does not explain why immediate naming should be slowed down in the first
place. Actually, it is in immediate naming that the phonological plan is stored
in the output buffer for the shorter duration, and that it should therefore suffer
the least. Although our data do not provide an unambiguous answer to this
question, we may formulate two plausible hypotheses. First, the output buffer
may take an unduly long time to constitute a usable strip of phonological plan,
that can be transferred to the articulator. This could result either from slowed
feeding from the lexicon, or from some disfunction affecting the operation of
the buffer itself. Second, there is evidence from normal subjects that before
actual execution, motor programs must be retrieved and unpacked by the
articulator. The retrieving and unpacking processes take measurable time,
depending on the number and complexity of the required motor programs
(Sternberg, Monsen, Knoll et al., 1978; Levelt, 1989; Wright, 1990). In brief,
the patient's preverbal pauses in immediate naming may result from an abnormal
operation of the buffer or of the articulator. Both hypotheses seem compatible
with the results of Experiment 2, where it appeared that an additional 2 second
delay allowed the patient to normalize her naming latencies, except for the
general 400 msec slowing affecting all timed tasks.
CONCLUSION

We have reported the case of a patient presenting an isolated disorder of
speech production consisting of phonemic paraphasias and pervasive preverbal

Phonological buffer and speech timing

485

pauses. In the clinical section, we have shown that the pattern of errors was
consistent with an impairment of the output phonological buffer. This hypothesis
was further confirmed in the experimental section through the study of repetition
and of picture naming latencies, with various imposed delays.
We have also tried to determine the functional origin of the patient's frequent
preverbal pauses. These pauses probably resulted from the conjunction of several
mechanisms that could only be partially disentangled. First, a slowing of about
400 ms was apparent in all timed motor tasks, and may have contributed to the
overall delay in speech initiation. Second, a specific time lag prevailed in
immediate naming. We suggested that its origin should be found in the operation
of the output buffer or of the articulator itself. Third, and as a final point, we
may speculate on a further source of slowing, namely the specific role of the
phonological buffer in ensuring the continuity and fluency of connected speech.
It has been suggested that, in normal speech, the tendency to produce
phonological phrases of approximately equal duration may reflect constraints
imposed by the size of the output buffer (Levelt, 1989, p. 386). A reduced or
disfunctioning buffer would therefore naturally result in a severe reduction of
the size of phonological phrases, down to the dimension of single words. In
other terms, if the content of the buffer suffers an abnormally fast decay, it
becomes impossil:rle to prepare a ready-made phonological plan several words
in advance. Each word has therefore to be produced from scratch, just at the
moment when it must be uttered (Bub et al., 1987). This phenomenon may have
contributed to the patient's preverbal pauses in continuous speech. However,
our study was essentially limited to single word production tasks and such an
hypothesis is still purely speculative. Further experimental work is needed in
order to examine directly the timing of continuous speech in aphasic patients.
Acknowledgements. The authors gratefully thank Philippe Van Eeckhout for referring the
patient, Christophe Pallier for his invaluable technical help, Jacques Mehler and the staff of
the LSCP for their fruitful suggestions and comments.

REFERENCES
BADDELEY, A.D., THOMPSON, N., and BUCHANAN, M. Word length and the structure of short-term
memory. Journal of Verbal Learning and Verbal Behavior, 14: 575-589, 1975.
BELLEVILLE, S., PERETZ, I., and ARGUIN, M. Contribution of articulatory rehearsal to short-term memory:
Evidence from a case of selective disruption. Brain and Language, 43: 713-746, 1992.
BENTON, A. Reaction time in brain disease: some reflections. Cortex, 22: 129-140, 1986.
Bus, D., BLACK, S., HowELL, J., and KERTESZ, A. Speech output processes and reading. In M. Coltheart,
G. Sartori and R. Job (Eds.), The Cognitive Neuropsychology of Language. Hillsdale, NJ: Laurence
Erlbaum Associates, 1987, pp. 79-1 I 0.
CAPLAN, D. Language: Structure, Processing, and Disorders. Cambridge, MA: MIT Press, 1992.
CAPLAN, D., VANIER, M., and BAKER, C. A case study of reproduction conduction aphasia I: Word
production. Cognitive Neuropsychology, 3: 99-128, 1986.
CARAMAZZA, A., MICELI, G., and VILLA, G. The role of the (output) phonological buffer in reading,
writing, and repetition. Cognitive Neuropsychology, 3: 37-76, 1986.
CARAMAZZA, A., MICELI, G., VILLA, G., and ROMANI, C. The role of the graphemic buffer in spelling:
Evidence from a case of acquired dysgraphia. Cognition, 26: 59-85, 1987.
DEESE, J. Thought into Speech: The Psychology of a Language. Englewood Cliffs, NJ: Prentice-Hall,
1984.
DELL, G.S., and O'SEAGHDHA, P.O. Stages of lexical access in language production. Cognition, 42:
287-314, 1992.

486

Laurent Cohen and Anne-Catherine Bachoud-Levi

DE RENZI, E., and FAGLIONI, P. Normative data and screening power of a shortened version of the
Token Test. Cortex, 14: 41-49, 1978.
FoDOR, J.A. The Modularity of Mind. Cambridge, MA: MIT Press, 1983.
GOODGLASS, H., and KAPLAN, E. The Boston Diagnostic Aphasia Examination. Philadelphia: Lea and
Febiger, 1972.
KLAPP, S.T., ANDERSON, W.G., and BERRIAN, R.W. Implicit speech in reading, reconsidered. Journal
of Experimental Psychology, 100: 368-374, 1973.
LEHISTE, I. Isochrony reconsidered. Journal of Phonetics, 5: 253-263, 1977.
LEVELT, W.J.M. Speaking: From Intention to 1rticulation. Cambridge, MA: MIT Press, 1989.
MAZAUX, J.M., and 0RGOGOZO, J.M. Echelle d'Evaluation de l'Aphasie: Adaptation Fran~aise du Boston
Diagnostic Aphasia Examination. Issy-les-Moulineaux: Editions Scientifiques et Psychologiques,
1982.
SCHRIEFERS, H., MEYER, A.S., and LEVELT, W.J.M. Exploring the time course of lexical access in
language production: Picture-word interference studies. Journal of Memory and Language, 29: 86­
102, 1990.
SHALLICE, T., and WARRINGTON, E.K. Auditory-verbal short-term memory impairment and conduction
aphasia. Brain and Language, 4: 479-491, 1977.
SHATTUCK-HUFNAGEL, S. Speech errors as evidence for a serial-ordering mechanism in sentence
production. In W.E. Cooper and E.C.T. Walker (Eds.), Sentence Processing: Psycholinguistic Studies
Presented to Merril Garrett. Hillsdale, NJ: Laurence Erlbaum Associates, 1979, pp. 295-342.
SNODGRASS, J.G., and V ANDERWART, M. A standardized set of 260 pictures: Norms for name agreement,
image agreement, familiarity, and visual complexity. Journal of Experimental Psychology: Human
Learning and Memory, 6: 174-215, 1980.
STERNBERG, S., MONSELL, S., KNOLL, R.L., and WRIGHT, C.E. The latency and duration of rapide
movement sequences: Comparison of speech and typewriting. In G.E. Stelmach (Ed.), Information
Processing in Motor Control and Learning. New York: Academic Press, 1978.
WATERS, G.S., RocHON, E., and CAPLAN, D. The role of high-level speech planning in rehearsal:
Evidence from patients with apraxia of speech. Journal of Memory and Language, 31: 54-73, 1992.
WRIGHT, C.E. Controlling sequential motor activity. In D.N. Osherson, S.M. Kosslyn and J.M. Hollerbach
(Eds.), An Invitation to Cognitive Science: Visual Cognition and Action (Vol. 2). Cambridge, MA:
MIT Press, 1990, pp. 285-316.
Dr. Laurent Cohen, Service de Neurologie I, Clinique Paul Castaigne. Hopital de Ia Salpetriere. 47 Bd de l'Hopital. 75651 Paris
CEDEX 13. France.
Tel. (1)42161798. Fax (1)44245247. E-mail cohen@lscp.msh-paris.fr.